THE MORPHOSYNTAX OF THE TURKISH CAUSTIVE CONSTRUCTION by Gregory Key

THE MORPHOSYNTAX OF THE TURKISH CAUSTIVE CONSTRUCTION by Gregory Key

THE MORPHOSYNTAX OF THE TURKISH CAUSTIVE CONSTRUCTION by

Gregory Key

____________________________

A Dissertation Submitted to the Faculty of the

DEPARTMENT OF LINGUISTICS

In Partial Fulfillment of the Requirements

For the Degree of

DOCTOR OF PHILOSOPHY

In the Graduate College

THE UNIVERSITY OF ARIZONA

2013

1

2

THE UNIVERSITY OF ARIZONA

GRADUATE COLLEGE

As members of the Dissertation Committee, we certify that we have read the dissertation prepared by Greg Key, titled The Morphosyntax of the Turkish Causative

Construction, and recommend that it be accepted as fulfilling the dissertation requirement for the Degree of Doctor of Philosophy.

Simin Karimi _______________________________________________________

Date: (06/10/2013)

Heidi Harley _______________________________________________________

Date: (06/10/2013)

Adam Ussishkin ___________________________________________________

Date: (06/10/2013)

Rudolph Troike ____________________________________________________

Date: (06/10/2013)

Jorge Hankamer ___________________________________________________

Date: (06/10/2013)

Final approval and acceptance of this dissertation is contingent upon the candidate’s submission of the final copies of the dissertation to the Graduate College.

I hereby certify that I have read this dissertation prepared under my direction and recommend that it be accepted as fulfilling the dissertation requirement.

Dissertation Director: Simin Karimi ___________________ Date: (06/10/2013)

Dissertation Director: Heidi Harley ___________________ Date: (06/10/2013)

3

STATEMENT BY AUTHOR

This dissertation has been submitted in partial fulfillment of the requirements for an advanced degree at the University of Arizona and is deposited in the University

Library to be made available to borrowers under rules of the Library.

Brief quotations from this dissertation are allowable without special permission, provided that an accurate acknowledgement of the source is made. Requests for permission for extended quotation from or reproduction of this manuscript in whole or in part may be granted by the head of the major department or the Dean of the Graduate

College when in his or her judgment the proposed use of the material is in the interests of scholarship. In all other instances, however, permission must be obtained from the author.

SIGNED: Gregory Key

4

ACKNOWLEDGEMENTS

I have many people to thank for supporting me in various ways in this endeavor. I would like to begin with my classmates in the Department of Linguistics. I thank Hyun

Kyoung Jung (HK) and Jaehoon Choi for their helpful discussions of Korean causatives.

As it turned out, I did not deal with Korean causatives in this dissertation, but what I learned from HK and Jaehoon helped me develop many of the ideas presented here, and will form the basis of future work. I also thank Kara Hawthorne; our weekly study sessions in various coffeehouses around Tucson helped keep my academic progress on track, and the friendly conversation that accompanied these sessions was something that I looked forward to each week. I would like to give special thanks to Deniz Tat, my friend, classmate, and occasional co-author. Thanks go out as well to the rest of my classmates, including Alex Trueman, Chen-chun Er, Megan Stone, Sylvia Reed, Jeff Punske, and many others. I also thank members of the faculty, including but by no means not limited to Andy Wedel, Andy Barss, Mike Hammond, and Andrew Carnie.

I thank the members of my committee: Simin Karimi, Heidi Harley, Jorge

Hankamer, Adam Ussishkin, and Rudolph Troike. I owe a special debt to my advisers,

Simin Karimi and Heidi Harley. I have benefited from Simin’s guidance since 2004, when she agreed to join my MA thesis committee when I was in the Department of Near

Eastern Studies at the University of Arizona; it is largely because of her influence that I decided to pursue a degree in linguistics. Heidi is responsible for my adoption of the framework of Distributed Morphology, which has proved an especially valuable tool for the analysis of Turkish causatives. Both Simin and Heidi have provided invaluable academic and moral support over the years. I owe them an extra thanks for agreeing to my panicked request to review my implementation of post-defense revisions in the quite unreasonably short span of one week. My post-doctoral position depended on this time frame, and Simin and Heidi came through for me.

I am much indebted to Rudy Troike. Had it not been for his influence and encouragement, I never would have taken up the study of Turkish to begin with. In terms of my life’s trajectory, this dissertation represents not the endpoint but a significant milestone in a journey that began twenty-four years ago, in 1989, in a class he taught called ‘Modern Grammar and Usage.’ This was part of the undergraduate English

Literature program I was enrolled in at the University of Arizona. Rudy (or Dr. Troike, as

I called him in those days) introduced us to the fundamentals of Transformational

Grammar. Most of the analysis in the textbook was based on English, but Rudy would frequently use examples from other languages, and quite often these examples were from

Turkish. This provoked my first interest in the language. One day, in a conversation outside of class, he suggested that I study Turkish. I followed up on this suggestion, and since then I have devoted a significant part of my life to solving the puzzles that this language presents. I feel very fortunate to have had Rudy on my doctoral committee two decades later, and to benefit more profoundly from his expertise and insights.

I am grateful to Eszter Ótott-Kovács on different levels. I thank her for her extremely helpful discussions of Hungarian causatives, for helping me compose

Hungarian example sentences, for providing her own judgments, and for obtaining judgments from other Hungarian speakers. Without her, much of the analysis and discussion of Hungarian herein would have been impossible, and the dissertation as a whole would have been much the poorer. I am also extremely grateful for Eszter’s companionship, and for her support and tolerance during these last few weeks of completing my revisions here on the island of Cunda, Turkey.

Finally, I would like to express my gratitude to my Turkish friends who have provided grammaticality judgments and offered linguistic insights. They include (in no particular order) Zeynep Günal, Elçin Arabacı, Deniz Tat, Aysel Özduran, Aslı Postacı,

Deniz Gedizlioğlu, Seçil Uluışık, İlker Hepkaner, Didar Demir, Nurcan Abacı, Özlem

Akdoğan, Nilay Özlü, Efe Balıkçıoğlu, Şerife Yorulmaz, Selda Işık, Pelin Aşan, Hikmet

Kocamaner, Aslı Iğsız, Aslı Plail, Ayşe Baltacıoğlu-Brammer, Nihan Ketrez, Selin

Kalaycıoğlu, Sultan Türkan, and Rabia Harmanşah. I regret if I have neglected to mention anyone, but this seems sadly inevitable, given the large number of people who have helped me in numerous ways over the years.

5

DEDICATION

To my parents, Sylvia Willard, Linford Key, and Faye Key, and to the memory of

Arthur Willard.

6

7

TABLE OF CONTENTS

LIST OF TABLES.......................................................................................................10

ABSTRACT.................................................................................................................11

CHAPTER 1: INTRODUCTION................................................................................13

1.1 Goals of the Dissertation........................................................................................13

1.2 Theoretical Framework and Assumptions.............................................................15

1.3 The Chapters..........................................................................................................16

1.4 Contributions of This Work...................................................................................27

1.5 Orthography, Notations, and Glosses....................................................................29

CHAPTER 2: THE ROOT IN TURKISH...................................................................31

2.1 Introduction............................................................................................................31

2.2 The Acategorial Root.............................................................................................32

2.3 Motivating the Root in Turkish..............................................................................39

2.4 Identifying the Roots of Turkish Verbs.................................................................46

2.4.1 Allomorph Distribution.................................................................................46

2.4.2 Competition and Blocking............................................................................50

2.4.2.1 The Distribution of Causative Allomorphs....................................50

2.4.2.2 Underspecified Verbalizing Allomorphs.......................................54

2.4.2.3 Fully Specified and Underspecified Morphemes in Competition..59

2.4.3 Complex Monosyllables...............................................................................60

2.4.3.1 Causative Allomorphy...................................................................63

2.4.3.2 Aorist Allomorphy.........................................................................65

2.4.3.3 Historical Forms of the Aorist.......................................................68

2.4.3.4 Acquisition of the Aorist...............................................................70

2.5 Polysyllabic Stems.................................................................................................74

2.6 Conclusion.............................................................................................................84

CHAPTER 3: ROOT CAUSATIVES.........................................................................86

3.1 Introduction............................................................................................................86

3.2 Root Causative Allomorphy..................................................................................87

3.3 Verb Classes..........................................................................................................90

3.3.1 Change-of-State Verbs.................................................................................90

3.3.2 Motion Verbs................................................................................................94

3.3.3 Psychological Predicates...............................................................................95

3.3.4 Ingestive Verbs.............................................................................................97

3.3.5 Evaluative Causatives.................................................................................101

3.4 Another Type of Realization: –

t

/–

n

Alternating Verbs........................................103

3.4.1 Verb Classes................................................................................................104

3.4.2 The Problem with Blocking........................................................................106

3.4.3

CAUS

Fission...............................................................................................111

3.4.4

CAUS

fission beyond the –

t

/–

n

alternation..................................................113

8

TABLE OF CONTENTS —

Continued

3.5 Conclusion...........................................................................................................116

CHAPTER 4: THE CAUSATIVE/INCHOATIVE ALTERNATION......................122

4.1 Introduction..........................................................................................................122

4.2 The Causative/Inchoative Alternation.................................................................123

4.2.1 Causativization............................................................................................126

4.2.2 Anticausativization.....................................................................................128

4.2.2.1 Decausativization.........................................................................130

4.2.3 Equipollence...............................................................................................133

4.3 Anticausativization and Monotonicity.................................................................137

4.3.1 Anticausativization as Reflexivization.......................................................138

4.3.2.1 There is no Causative in Anticausative........................................142

4.3.2.2 Telling Apart the Look-Alikes.....................................................146

4.4 The Alternation in Turkish...................................................................................158

4.4.1 Middle Voice Inchoatives...........................................................................158

4.4.2 v

BECOME

Inchoatives..................................................................................164

4.5 Conclusion...........................................................................................................167

CHAPTER 5: VERB-SELECTING CAUSATIVES................................................168

5.1 Introduction..........................................................................................................168

5.2 Causatives and the Little-v Hypothesis................................................................169

5.2.1 Japanese Productive Causatives..................................................................169

5.2.2 Turkish Productive Causatives...................................................................175

5.2.3 Severing the External Argument from its Light Verb................................178

5.3 The Structure of Verb-Selecting Causatives........................................................182

5.3.1 Proposal.......................................................................................................182

5.3.2 Problems of Allomorphy.............................................................................191

5.3.3 Attachment Sites and the Elsewhere Principle...........................................192

5.3.4 Outer Causative Allomorphy......................................................................201

5.4 Causativization in the Lexicon.............................................................................203

5.5 Causee Animacy..................................................................................................206

5.6 Conclusion...........................................................................................................215

CHAPTER 6: CAUSATIVE ITERATION..............................................................216

6.1 Introduction.........................................................................................................216

6.2 Causative Iteration Cross-Linguistically..............................................................220

6.3 The Semantics of Indirect Causation...................................................................226

6.4 The Illusion in Turkish.........................................................................................228

6.5 Causative Iteration as Reduplication...................................................................232

6.5.1 The Coercive Reading................................................................................234

9

TABLE OF CONTENTS —

Continued

6.5.2 The Permissive Reading.............................................................................235

6.5.3 The number of repetitions...........................................................................240

6.5.4 The Vacuous Reading.................................................................................242

6.6 Formalization of Causative Focus Reduplication................................................242

6.7 Conclusion...........................................................................................................246

CHAPTER 7: CONCLUSION..................................................................................248

REFERENCES..........................................................................................................254

10

LIST OF TABLES

Table 1.

Verbs Based on Bound Stems.............................................................................41

Table 2.

Results from Nakipoğlu and Ketrez (2006) for error rates (%) in aorist formation with consonant-final verbs. ..............................................................................71

Table 3.

Onomatopoeic verbs, nouns, and reduplicated adverbs based on disyllabic bound stems..................................................................................................................................76

Table 4.

Monosyllabic Roots of Onomatopoeic Stems.....................................................81

Table 5.

Variable morphological marking of causative, anticausative, and passive.......157

11

ABSTRACT

This dissertation is an analysis of the morphosyntax of the Turkish causative construction within the framework of Distributed Morphology (DM). It is an attempt to capture a range of different phenomena in a principled way within this framework. Important aspects of DM for the analysis herein include the syntactic derivation of words; the existence of an acategorial Root from which all words are syntactically derived; and the late (post-syntactic) insertion of Vocabulary Items (VIs) into terminal syntactic nodes.

A distinction is made between two different levels of causative: Root (or inner) causatives, and productive (or outer) causatives. Root causatives are minimal structures in which a Root phrase (comprising a Root and its nominal complement) is merged with a verbalizing head, little-v (Harley 1995; Chomsky 1995, 2001; Marantz 1997). This domain is the locus of idiosyncratic allomorphy, and it is where the traditionally recognized ‘irregular’ causatives suffixes are found. In addition, another type of idiosyncratic Root-adjacent phenomenon is identified in this study: independent exponence of the verbalizing feature and of the causative feature (

CAUS

). This is analyzed as

CAUS

fission: the result of a post-syntactic operation that splits the terminal node [v,

CAUS

] into two positions of exponence.

Productive causatives are larger structures in which a vP is merged with a

CAUS

head. The identification of the Root causative head as v.

CAUS

but the productive causative head as simply

CAUS

is a departure from Harley’s (2008) analysis of Japanese causatives, and is a new proposal in this work. Following Pylkkänen (2002, 2008), the external argument is not introduced by either v.

CAUS

or

CAUS

, but by a higher projection,

12

Voice. This innovation makes it possible to model syntactic differences between Japanese and Turkish productive causatives. Japanese causatives embed Voice (i.e., they are

‘phase-selecting,’ in Pylkkänen’s terminology) while Turkish causatives embed little-v

(i.e., they are ‘verb-selecting’). Hence, the former behave as two clauses with regard to a range of diagnostics, while the latter behave as a single clause. Furthermore, it is proposed that productive causatives do not exhibit syntactic recursion, and that cases of causative iteration are actually morphological reduplication.

13

CHAPTER 1: INTRODUCTION

1.1 Goals of the Dissertation

This study is an analysis of the Turkish causative construction carried out within the framework of Distributed Morphology (Halle & Marantz 1993, Harley 1995, Marantz

1997). The immediate goal is a more sophisticated and accurate understanding of the

Turkish construction, while the broader goal is to produce results that can inform the study of causatives in other languages, and even suggest refinements to the theoretical model itself. There is thus an interaction between abstract theoretical assumptions and hypotheses on the one hand, and the concrete empirical details of the Turkish data on the other. The theory provides tools for analysis of the data, but in cases where the data do not submit to the assumptions that the theory brings with it, it is the assumptions that must be questioned.

The typological character of Turkish makes it an especially rich source of morphological information. Turkish belongs to the Southwest branch of Common Turkic, the main division of the Turkic language family. Like all languages of this family, it is an agglutinating language, which means that, to a great extent, morphemes with different functions have exponence as discrete suffixes. The derivational history of a verbal form, for example, is highly transparent. With this in mind, the empirical methodology used in this study consists primarily of the close examination of the morphological patterns found in causatives, as well as in non-causative verbs. I believe it is not an exaggeration to say that this study is the most detailed synchronic investigation yet conducted into the

14 morphological minutiae of Turkish causatives, or indeed of any construction in Turkish.

1

Nakipoğlu and Üntak (2008) represents an unprecedented effort to catalogue Turkish verbs exhaustively according to their morphological patterns. In the present study, I make extensive use of this valuable resource, but scrutinize the morphology of one particular domain—causatives—in far greater detail.

In this endeavor, one assumption is not questioned: the premise that morphemes are direct reflections of the hierarchic (syntactic) structure of words (as in Baker 1985).

This is contrary to Göksel (1993, 2001), who claims that morphology takes place in a different module than syntax, and that it does not reliably map to syntactic structure, even in such a morphologically rich language as Turkish. In this work, I do not question the premise that morphology maps to syntax because, in an important sense, this is the central hypothesis that is being tested. If it were false, then it should be impossible to build a coherent analysis of Turkish morphosyntax under the assumption that it is correct.

This dissertation shows that such an analysis is indeed possible.

Section 1.2 below presents the aspects of DM that are most relevant to this work.

Section 1.3 contains summaries of the main chapters of this dissertation. Section 1.4 is a summary of the theoretical contributions that this study makes to the field of morphosyntax. Section 1.5 provides an explanation of notations and glosses used throughout the dissertation.

1

I do not make this claim in a broader sense that would include diachronic studies, as such a claim would be falsified by such meticulous works as Clauson (1972), Tietze (2002), and Erdal (1991).

15

1.2 Theoretical Framework and Assumptions

The framework of Distributed Morphology (DM) is constructed within the

Minimalist Program (MP) (Chomsky 1995). Several features of DM distinguish it from other approaches within MP. The first is the hypothesis that all word formation is syntactic. It follows from this that the atoms of syntax are not themselves words, but rather Roots and affixes. The Root, the kernel of a word, has no grammatical category such as ‘noun’ or ‘verb,’ but acquires a category only upon the merger of a Root phrase

(√P) with a functional head such as

n

or

v

. Another distinguishing feature of DM is

late insertion

: When non-Root terminal nodes are built in the syntax, they are abstract feature bundles with no phonological content (and, on the approach of Harley (to appear (a)), the same holds true for Root nodes). On the Phonological Form (PF) branch of the derivation, Vocabulary Items (VIs) are matched with terminal nodes in the process of

Vocabulary Insertion. Each VI is specified for a set of features. According to the Subset

Principle (Halle 1997), the feature bundle of a terminal node may be realized by a VI specified for a subset of the node. Thus, a terminal node with the feature bundle [v,

CAUS

] may be realized by a VI fully specified for [v,

CAUS

], or by one underspecified for [v] or for [

CAUS

]. If a VI is specified for any feature not present in the feature bundle, it is disqualified from insertion. For example, a VI specified for [v,

CAUS

] could not realize a terminal node consisting of [v,

BECOME

]. Multiple VIs compete for insertion in each terminal node, and the VI that is most specific to the environment wins the competition, blocking less specific VIs. In general, the VI that has the most features without having any incompatible features will win the competition. However, a VI is also considered

16 highly specific if it is restricted to insertion in a position linearly adjacent to a particular

Root or other morpheme.

Preceding Vocabulary Insertion but following syntactic operations, at the level of

Morphological Form, adjustments may be made to the terminal nodes. Features may be deleted (impoverishment); multiple nodes and their corresponding bundles may be fused into a single node (fusion); or a single node may be split into multiple nodes, with the features of the original bundle distributed among the new nodes (fission).

1.3 The Chapters

Five central chapters make up the analytical body of this dissertation. Chapters 2 and 3 are concerned with the Root and the output of syntactic merger with the √P, both in general (Chapter 2) and with regard to Root causatives (Chapter 3). Chapters 4 and 5 expand the scope of analysis to non-causative (inchoative) verbs that alternate with Root causatives (Chapter 4), and to causatives formed via the merger of a

CAUS

head with vP rather than directly with √P (Chapter 5). These two chapters also explore the interaction of causatives with Voice (extending the work of Pylkkänen (2002, 2008)). Chapter 6 explores the iteration of causative suffixes, and reaches the perhaps surprising conclusion that there is no syntactic recursion of Turkish causatives.

The goal in Chapter 2, ‘The Root in Turkish,’ is to identify the acategorial Roots that underlie verbs and other words in Turkish. A key idea in DM is that word formation is syntactic, and that all functional and categorical properties are derived via the merger of functional heads with their complements. It follows from this that there can be no simplex words. Any entity of the category

noun

,

adjective

, or

verb

must minimally

17 consist of an acategorial base merged with a functional head

n

,

a

, or

v

(Marantz 1997,

Arad 2003). To realize the goal, it is necessary to analyze the phonological material of a word so as to demarcate Root material and affixal material. To this end, three tools are employed in this chapter: the occurrence of bound stems, the syllabic distribution of derivational morphemes, and allomorphy. The use of bound stems is based on the idea that a given Root may appear in various derivations, heading phrases that merge with different derivational morphemes. For example, the Root √Ö

D is clearly discernible in the words

öd-e-

‘pay (v)’ and

öd-ünç

‘loan (n).’ There is no common word from which each could plausibly be derived,

2

which makes sense only if the base is a Root. Furthermore, the semantics of ‘pay’ and ‘loan’ are clearly related but have also undergone substantial drift. This is because, when a functional head combines directly with a Root, it picks out a specific meaning from the semantic domain of that Root (Arad 2003). The addition of further derivational projections will retain the semantics of the initial Root merger, but other Root mergers pick out their own semantics. The pair

öd-e-

/

öd-ünç

also presents evidence that the Root from which each is derived is category neutral, as the first is derived with –

A

, traditionally and historically identified as a denominal verb formative, and the second with –

Inç

, a deverbal noun formative. Such cases are mysterious if derivation always proceeds from categories, but quite expected if the core of every word is an entity that lacks category.

2

There is a noun

öd

that means ‘bile.’ I do not consider this word to be related to

öd-e-

and

öd-ünç

, although it is conceivable that they are all based on a single Root that exhibits radical semantic drift in certain environments.

18

On this view of word formation, even a word that has no phonologically realized affix must still contain an affix, which must be null in such cases. Semantic drift will often indicate that this has happened. The words

top

‘ball’ and

baş

‘head’ have no overt suffixes, but the same Roots in the context of the verbal formative –

lA

exhibit semantic drift:

top-la-

‘gather’ and

baş-la-

‘begin.’ Hence,

top

and

baş

consist of the Roots √T

OP and √B

AŞ plus a null

n

, while the verbs consist of the same Roots suffixed with –

lA

.

The syllabic distribution of derivational suffixes can also provide a clue to the identity of the Root. Turkish has a large number of verbal suffixes that have a distributional peculiarity: Their left edge always occurs within the second syllable of a derived verb stem. In the case of suffixes that consist of a single syllable, this simply means that the suffix forms the second syllable. For example, the causative allomorphs –

Ir

, –

Ar

, and –

It

are found exclusively with this syllabic distribution, e.g.,

kaç-ır-

‘make escape,’

bat-ır-

‘make sink,’

çık-ar-

‘remove,’

kop-ar-

‘detach,’

kork-ut-

‘frighten,’

ak-ıt-

‘make flow.’ In contrast, the causative allomorph –

DIr

occurs freely in the second, third, and fourth syllables:

yap-tır-

‘cause to do,’

konuş-tur-

‘cause to speak,’

seyrekleş-tir-

‘cause to become sparse.’ A superficial explanation of this difference is that the stems to which the second-syllable suffixes attach is always monosyllabic. As is shown in the chapter, Turkish monosyllables are virtually always simplex—that is, a monosyllabic stem does not contain overt suffixes (apart from exceptionally rare cases). It follows from this that the second-syllable suffixes attach only to bases that have no overt suffixes. The most natural account of this is that they attach only to Roots, while –

DIr

attaches freely to verbs (as well as Roots). This account fits well with the frequencies of these suffixes.

19

In Nakipoğlu & Üntak’s (2008) dataset, –

Ir

, –

Ar

, and –

It

causativize 20, 4, and 9 verb stems, respectively, while –

DIr

causativizes several hundred. I therefore conclude that suffixes with a strict second-syllable distribution can only appear adjacent to particular

Roots.

3

The final tool I use to identify Roots is allomorphy. Root-conditioned allomorphy requires that the suffix be linearly adjacent to the Root (Marantz 2010), or, to put it another way, that there can be no overt suffixes between the Root and the allomorph. The causative allomorphs mentioned above have this requirement. Similarly, the aorist allomorph –

Ar

, which is conditioned by most monosyllabic verbal Roots (Hankamer

2013), can only occur when there are no intervening suffixes; otherwise, the aorist form –

Ir

occurs (or –

r

after a vowel). Hence I use causative and aorist allomorphy as positive evidence that a stem contains no phonologically overt suffix. This leads to the conclusion, for example, that the verbs

doy-

‘become sated’ and

duy-

‘hear, sense’ do not contain a suffix in –

y

(contra Nakipoğlu & Üntak 2008), as they causativize with –

Ir

(

doy-ur-

,

duy-ur-

), and have aorist forms in –

Ar

(

doy-ar

,

duy-ar

).

Chapter 3, ‘Root Causatives,’ focuses on the merger of Root phrases with the terminal node having the feature bundle [v,

CAUS

]. The causative allomorphy discussed above is used to identify Root causatives, which are seen to belong to certain welldefined categories: change-of-state (COS) verbs, motion verbs, psychological predicates, ingestives, and evaluative causatives. (The last category is a novel proposal originating in this study.) In turn, these categories of verbs are used to determine places where the

3

Note, however, that the converse does not automatically follow from this: It is not necessarily the case that polysyllabic stems are always morphologically complex.

20

Elsewhere causative –

DIr

is realizing a Root-adjacent [v,

CAUS

] node. Furthermore, in the chapter an additional type of Root causative exponence is identified, in which the causative VI –

t

is separated from the Root by an underspecified verbalizer such as –

lA

or

A

: e.g.,

ıs-la-t-

‘get wet (tr.),’

kap-a-t-

‘close (tr.).’ Verbs exhibiting such patterns fall into the same categories identified for the more standard causative allomorphy pattern. I argue that in these cases, the Root-adjacent feature bundle undergoes fission, such that the features [v] and [

CAUS

] have separate positions of exponence. This operation is idiosyncratically conditioned by particular Roots.

Chapter 4, ‘The Causative/Inchoative Alternation,’ investigates an area where

Root causatives play a large role: transitivity alternations. These are verb pairs such that one member (the causative) has one more argument than the other member (the inchoative), such as

don-dur-

/

don-

‘freeze (tr./int.)’ In general, the causative is monotransitive and the inchoative intransitive (unaccusative), though ingestives are the exception; the causative may be ditransitive (

ye-dir-

‘feed’) and the inchoative monotransitive (

ye-

‘eat’). Transitivity alternations have been the subject of much research and controversy. One major point of controversy is the derivational relationship between the members of the pairs. It has been argued that the causative verb is derived from the inchoative (‘Causativization,’ Pesetsky 1995), that the inchoative is derived from the causative (‘Anticausativization’ or ‘Decausativization,’ Chierchia 2004,

Reinhart 2002, Horvath & Siloni 2011a), and that each member is derived independently from a common stem (‘Equipollence’ Harley 1995, 2008). This is an important point for any theory, and DM is no exception. For DM, all word formation is syntactic. One

21 consequence of this is that Anticausativization, on the most straightforward understanding, is not a licit operation. If Anticausativization applies to a verb that has a

CAUS

operator and deletes this operator, it cannot be implemented as a syntactic operation. In the syntax, structure can be added, but it cannot be removed. This is the principle of ‘monotonicity’ (Koontz-Garboden 2009). If such an operation existed, it would require an additional generative engine: the lexicon. For proponents of an active lexicon, such as Reinhart (2002) and Horvath & Siloni (2011a),

Anticausativization/Decausativization poses no problem. In fact, these same researchers take the position that Anticausativization is the universal derivational relationship in the causative/inchoative alternation (which they call the ‘unaccusative’ alternation).

Causativization derives a causative verb from an inchoative one, retaining the

BECOME

operator, while Equipollence derives the causative and the inchoative independently. These, therefore, are structure-building operations, and do not inherently pose any problem for syntactic word derivation. But if it can be shown that

Anticausativization exists as a structure-deleting operation, there would be no escape from a generative lexicon. Thus, one of the fundamental underpinnings of DM would be lost.

Koontz-Garboden (2009) takes the position that Anticausativization does exist, but that it does not delete the

CAUS

operator. This, he argues, is achieved through a process of reflexivization. In the transitive verb, there are two arguments, an effector

(either a causer or an agent) and a theme. The reflexivization operation creates an intransitive verb not by deleting the

CAUS

operator and the external argument, but by

22 creating an identity relation between the effector and the theme. Conceived of this way,

Anticausativization could be implemented in the syntax. However, Horvath & Siloni

(2011b) present evidence that casts serious doubt on Koontz-Garboden’s claims. The final sections of the chapter (4.3 and 4.4) suggest an alternative treatment of purported

Anticausatives. I examine Turkish verb pairs ostensibly exhibiting the Anticausative alternation, and I show that in fact these never embed a causative morpheme. Hence, the structure on which these verbs are built never contain a causative structure or

CAUS

operator, and so nothing is deleted.

Chapter 5, ‘Verb-selecting Causatives,’ deals with productive causatives, also known as ‘high-attachment’ or ‘outer’ causatives, which are derived from vPs and not directly from √Ps. Outer causatives have also been the subject of theoretical controversy.

In a study on Japanese causatives, Harley (2008) argues that the only difference between

Root causatives and productive causatives (called ‘lexical’ and ‘syntactic’ in the Japanese linguistic literature) is the attachment site: In Root causatives, the head v.

CAUS

merges with a √P, but in productive causatives, v.

CAUS

merges with a vP. This accounts for the results of various diagnostics that detect two clauses (owing to two vPs) in productive causatives, but only one clause in Root causatives. These diagnostics (not all discussed in

Harley 2008) include agent-oriented adverbials (Shibatani 1972), Binding Condition B

(Miyagawa 1984), coordination (Kuroda 2003), and scope of negation (Hara 1999), among others. Harley (2008) assumes that the verbalizer little-v, which derives the category ‘verb’ (Harley 1995; Chomsky 1995, 2001; Marantz 1997), and Voice, which heads the phrase that introduces external arguments in its specifier (Kratzer 1996), are the

23 same projection, such that each vP-

CAUS

introduces an external argument. On the other side of the controversy, Horvath & Siloni (2011a) counter Harley’s claims with evidence from Hungarian. They accept the evidence that Japanese productive causatives are created in the syntax, but they demonstrate that the same diagnostics detect only one clause in Hungarian productive causatives, and thus fail to distinguish them from simple transitive verbs. From this they conclude that all causatives in Hungarian are generated in the lexicon. As I show in this chapter, the diagnostics yield (mostly) the same results in

Turkish as in Hungarian. The Harley (2008) model is indeed unable to account for such facts. Productive causatives in Hungarian as well as in Turkish do not exhibit idiosyncratic allomorphy or other signs of Root adjacency, while standard Root causatives do show such signs. The only way to account for differences between the two types of causative is with stacked vPs, just as Harley does for Japanese, but this model then predicts that the productive causatives should be biclausal like their Japanese counterparts.

A 21 st

-century innovation to the architecture of argument structure offers a solution to this seemingly intractable problem. Pylkkänen (2002, 2008) proposes that in some languages the external argument-introducing head Voice is distinct from

CAUS

(which in Harleyian (2008) terms is a vP). She also proposes that different types of causative head select for different complements: Those that select for √P are Rootselecting, those that select for vP are verb-selecting, and those that select for VoiceP (as well as High Applicative Phrases) are Phase-selecting. This proposal makes it possible to stack vPs, or causative phrases, without embedding multiple Voice projections. If the

24 clausality tests are sensitive to the presence of an external argument (i.e., a Voice-

ACTIVE projection), then the difference between the productive causatives of Japanese and those of Hungarian and Turkish can be straightforwardly captured: Japanese causatives are phase-selecting, while Hungarian and Turkish causatives are verb-selecting. In addition to demonstrating how this new structure can be implemented, I propose a further innovation: While the Root causative head is v.

CAUS

, as in Harley (2008), the productive causative head is simply

CAUS

. Since the base has already acquired the category

verb

, there is no need for the second verbalizer that appears in the stacked vP model. The fact that the Elsewhere VI appears in both contexts can still be captured under this proposal:

The Elsewhere causative has only the feature [

CAUS

]. By the Subset Principle, it can still realize the Root-adjacent [v,

CAUS

] node, but any VI that has the feature [v] cannot be inserted in the [

CAUS

] node. These facts allow us to account for allomorphy patterns in a more satisfactory way, as discussed in detail in the chapter.

The final sections of Chapter 5 (5.4. and 5.5) compare the split Voice/

CAUS

/v approach to Turkish and Hungarian causatives with a lexicalist approach, that of Horvath

& Siloni (2011a) (Henceforth H&S). H&S employ the theta system of Reinhart (2002) to derive outer causatives in the lexicon via an operation that takes as input a verb with a lexical specified external argument, and adds a causer argument to the verb’s lexical theta grid. This is a radically different approach from the one advocated in the present work, but to a large extent both have the same empirical coverage. Both can account for the fact that the same diagnostics that detect two predicates in Japanese causatives detect only one in Turkish and Hungarian causatives. Because the theoretical differences between the

25 approaches are not trivial, it is necessary to find areas where their empirical coverage differs. In Section 5.5, I take a close look at one such area: the status of the causee.

Although the subject of a transitive or unergative verb may be animate or inanimate, the dative (or, in Hungarian, instrumental) causee of the corresponding causative must be animate. This presents a problem for H&S’s approach. The external argument of the input theta grid may have various feature clusters, including [+c] (‘c’ stands for ‘cause’), or [+c +m] (‘m’ stands for ‘mental state’). Under the lexical causativization operation that H&S propose, if the external argument in the input theta grid has a [+c] feature, it is revaluated to [-c]. Thus, a [+c] argument becomes [-c], and a

[+c +m] argument becomes [-c +m]. Since [+c] is a legitimate input for the external argument, [-c] is a legitimate output for the causee. In Reinhart’s theta system, a [-c] role may be realized by an animate or inanimate argument. The only way to ensure that the argument is animate is for there to be a [+m] feature in the theta grid. However, as I show in the chapter, it is not possible to achieve this without disrupting the integrity of the system as a whole. However, under the Pylkkänen style approach described in the present study, Turkish and Hungarian causatives do not embed the external argument-introducing

Voice phrase, so the causee must be introduced by a different means. Hence, differences between the causee of a causative verb and the external argument of the corresponding non-causative verb present no inherent challenge to the model. I propose the following:

The outer causative head has a [+m] feature, which I borrow from Reinhart’s system. The causee is an adjunct phrase hosted by

CAUS

P. The adjunct phrase itself has no animacy

26 restriction, but an inanimate causee would clash semantically with the [+m] feature of the

CAUS

head.

The proposal (original to this work) that the productive/outer causative is a

CAUS

P rather than a vP makes it possible to explain a puzzling fact about affixal causatives: cross-linguistically, causative recursion is either disallowed, or severely limited. In a comparative study of affixal causatives in six languages (Sámi, Hindi, Nivkh, Amharic,

Kitharaka, Malagasy), Svenonius (2005a,b) observes (among other phenomena) that, at most, the stacking of an outer causative on an inner causative is permitted, and that there is no recursion of outer causatives. This is mysterious if both inner (Root) and outer

(productive) causatives are vPs: The fact that an outer causative can be stacked onto an inner means that v

CAUS

can select for a vP, and so from a syntactic standpoint recursion should be unlimited. On the other hand, if the present proposal is correct and inner and outer causatives are different kinds of functional projections, then the recursion pattern observed in Svenonius simply means that the

CAUS

head can merge with vP but not with

CAUS

P, and so there is no actual recursion at all.

However, there are languages that appear to allow causative recursion. Turkish is one such language (Aissen 1979, Kural 1996, among others), as are Hungarian (Hetzron

1976), Kashmiri (Manetta (to appear)), and Tsez (Kulikov 1993). In Chapter 6, I argue that these languages do not allow causative recursion any more than the Svenonius languages do, and that the supposed recursion is actually an illusion. First, the languages that seem to allow recursion have something else in common as well: the semantically vacuous repetition of causative suffixes. Furthermore, beyond the first outer causative,

27 there is no direct correlation between the number of causative suffixes and the number of causees expressed. Finally, the semantics of indirect causation allow for interpretations with any number of intermediate links that are not syntactically represented, and the agent of one of these intermediate causing events can be expressed as an adjunct phrase.

In a language that allows vacuous repetition of suffixes, the number of causatives can be artificially matched to the number of causees expressed, creating the illusion of syntactic recursion. I argue that this is what is happening in these languages, and not syntactic recursion.

This raises the question of why causative suffixes can iterate. I propose that this iteration is morphological reduplication, and that it realizes focus. I examine two readings—coercion and permission—of Turkish causative iteration, and I show how these readings can be derived from focus. I then model this realization of focus, adapting

Inkelas & Zoll’s (2005) Morphological Doubling Theory to DM.

1.4 Contributions of This Work

In addition to shedding light on a number of Turkish structures, this work makes several theoretical contributions to the field of morphosyntax. First, in Chapters 2 and 3 it provides additional empirical evidence for the category-neutral Root, a central component of the DM model. It also demonstrates that the merger of the first functional head directly with the √P is the locus of a large number of seemingly random, idiosyncratic phenomena, for which no unified account has previously been proposed. I show that the Root, properly recognized, eliminates the need to resort to such vague notions as ‘lexicalization’ to account for such cases.

28

Another contribution is a possible solution to a potential problem for the DM model (Chapter 4) the apparent pattern of Anicausativization in the causative/inchoative alternation (Haspelmath 1993). If an inchoative can truly be derived from a causative, as appears to be the case in Anticausativization alternations, this would entail the deletion of a

CAUS

operator. Such an operation could not be implemented in the syntax, contrary to the DM principle that all word formation is syntactic. While agreeing with Horvath &

Siloni (2011b) that Koontz-Garboden’s (2009) solution to this problem is not tenable, I closely examine apparent Anticausative alternations in Turkish and show that a causative suffix is never embedded under an inchoative suffix. This means that, in Turkish at least, there is no evidence that the structure ever contains a

CAUS

operator, which in turn means that there is no evidence that a

CAUS

operator is ever deleted. Anticausatives are simply built from structures that lack

CAUS

.

There are several contributions involving the Voice projection, argued in

Pyllkänen (2002, 2008) to be distinct from

CAUS

in many languages. I use this theoretical development to respond (in Chapter 5) to H&S’s criticism of Harley (2008) and their claim that Hungarian productive causatives provide conclusive evidence for wordderiving lexical operations. I show how the more recent conceptions of little-v and Voice can account for the phenomena the H&S point out. They all stem from the fact that causatives in Hungarian, as well as in Turkish, cannot embed the Voice projection, being

‘verb-selecting causatives,’ in Pylkkänen’s terminology. Furthermore, I show that this new model of syntactic causative derivation can accommodate the issue of causee animacy, which is an intractable problem in H&S’s model.

29

A novel proposal for syntactic accounts of causatives is that the inner causative head and the outer causative head are of different types: The former is v

CAUS

, while the latter is simply

CAUS

. This provides a simple explanation for different allomorphy patterns found with different types of heads, and allows one to state simply that the condition for contextual allomorphy is linear adjacency, without resorting to elaborate rules involving phase-defining heads, as in Marantz (2010). This innovation also provides a simple account of the lack of causative recursion (on which see the following).

Finally, this study makes an important contribution in Chapter 6 to the literature on Turkish causatives, and on morphological causatives in general. It has frequently been taken for granted that Turkish causatives exhibit syntactic recursion. Whether or not they do is of fundamental importance in the understanding of the construction, and making the wrong assumption could lead one’s analysis far down the wrong path. I demonstrate that there is in fact no recursion of Turkish causatives:

CAUS

P may embed vP-

CAUS

, but

CAUS

P may not embed

CAUS

P. This observation is of great importance for a sound analysis of causatives in Turkish, and in many other languages as well.

1.5 Orthography, Notations, and Glosses

Turkish is written in a modified version of the Latin alphabet. Standard Turkish orthography is employed in Turkish examples and in citations of Roots, affixes, and words, although some modifications are made for notational purposes. First, hyphens are used here to indicate morpheme boundaries, but this of course is not done in written

Turkish. In addition, a capital letter is used here to indicate a variable segment whose precise form is predictable from the phonological environment. The vowels of suffixes

30 are generally underspecified (with a handful of notable exceptions), and their realizations depend on the rules of Turkish vowel harmony. Vowel harmony is not discussed here, but see Göksel & Kerslake (2005) or any standard reference on Turkish for details. A capital

A

represents a low, unrounded vowel, which is realized as

e

(front) or

a

(back). A capital

I

is a high vowel, which is realized as

i

(front, unrounded),

ı

(back, unrounded),

ü

(front, rounded), or

u

(back, rounded). The variability of consonants mostly involves voicing assimilation. The only variable consonant of relevance for causative suffixes is

D

, a denti-alveolar plosive that is realized as

d

when voiced, and as

t

when unvoiced (Göksel

& Kerslake 2005). As an example, the causative suffix –

DIr

has eight realizations: –

dir

,

dır

, –

dür

, –

dur

, –

tir

, –

tır

, –

tur

, –

tür

.

The following abbreviations are used in interlinear glosses: 1 = first person, 2 = second person, 3 = third person, √ = Root, v = little-v,

CAUS

= causative,

INCH

= inchoative,

BECOME

= become,

PAST

= past,

SG

= singular,

PL

= plural,

AOR

= aorist,

ACC

= accusative,

DAT

= dative,

ABL

= ablative,

LOC

= locative,

DEF

= definite,

INDEF

= indefinite,

CONT

= continuous,

PASS

= passive,

MID

= middle voice,

NOM

= nominalizer,

NEG

= negative,

PRVRB

= preverb,

DO

= direct object,

PART

= partitive,

APPL

= applicative

31

CHAPTER 2: THE ROOT IN TURKISH

2.1 Introduction

The focus of this chapter is on Turkish verbs, specifically the Roots and affixes employed in their derivation. The following claim is presented and defended throughout the chapter:

(1) All (native) Turkish verbs are derived from a category-neutral Root and one (or more) overt or null suffixes.

The verbal lexicon of Turkish compiled by Nakipoğlu & Üntak (2008), consisting of more than 4,700 verbs, is the basis for the following analysis. Nakipoğlu & Üntak

(henceforth N&Ü) have painstakingly culled more than 4,700 Turkish verbs from the

Turkish Language Association’s Turkish dictionary and writing guide, and organized them into an appendix based on the character of their derivation. Their analysis is the starting point for my own; however, I diverge from them in important respects. Much of the justification for N&Ü’s morphological analysis is historical. However, the goal of this chapter is not a historically accurate etymology of Turkish verbs, but a synchronic analysis of the verbal derivational system of the modern language.

The empirical basis for defending the above claim primarily comprises native

Turkish verbs, and in large part excludes borrowings from other languages. Loan words do not decompose into overt Roots and affixes as transparently as native words do, and so are impervious to many of the analytical tools employed herein. The morphological status of borrowings is an interesting question that this study does not address. They might be

Roots themselves or, as Arad (2005) suggests with respect to foreign borrowings in

Hebrew, they might be imported from their source languages as full-fledged words, with

32 their functional apparatuses fully intact. The dataset from N&Ü similarly excludes a large number of verbs derived from a foreign base, but for somewhat different reasons: The organization of the reference works from which they gathered their data is such that complex predicates that appear as separate orthographic words do not head their own entries, and so were missed by N&Ü’s collection method. The great majority of such complex predicates are made up of a foreign borrowing and a light verb, and hence there are numerous borrowings (perhaps thousands) that are simply missing from the dataset.

However, two main types of borrowing are represented here: verbs derived with the verbalizing suffix –

lA

and its augmented forms –

lAn

and –

lAş

(N&Ü Appendix 2a: 25,

26, 27, respectively), and complex predicates (relatively small in number) that form a single orthographic unit (N&Ü App. 2b: 3–6).

This chapter is structured as follows: In sections 2.2 and 2.3, I introduce the acategorial Root and motivate its presence in Turkish derivation. In 2.4, I present ways of identifying Roots and affixes in Turkish, including morpheme distribution and allomorphy, with particular attention given to allomorphy of the aorist marker. In 2.5 I examine a large set of onomatopoeic verbs that are derived from disyllabic acategorial stems, and I show that these can be reduced to monosyllabic Roots. Section 2.6 concludes.

2.2 The Acategorial Root

One of the central claims of Distributed Morphology (DM) is that word formation is syntactic. All words are morphologically complex, minimally consisting of a Root and an affix, of which the latter may be overt or null. If words are defined thus, it follows that

33 a Root cannot be a word, but is rather an atomic bound stem. In this work, a distinction is made between

Roots

and

stems

. A Root is an open-class, simplex morpheme that lacks any functional category such as noun, adjective, or verb (Halle & Marantz 1993,

Marantz 1997, Arad 2003, Harley (to appear: b)).

Acategorial Root phrases merge with functional affixes such as

v

,

n

, or

a

to form a word of the categories

verb

,

noun

, or

adjective

. A stem, on the other hand, is any base of affixation. Since Roots admit affixation—indeed must do so in order to enter into a derivation—they are a sub-type of stem. Other types of stem are morphologically complex, comprising a Root and at least one affix, and usually have a functional category.

The syntactic mechanisms employed in DM are essentially the same as those used within the Minimalist Program (MP) (Chomsky 1993, 1995). Items from the numeration are merged to create hierarchical structures. In the MP, a transitive or unaccusative verb

(V) is merged with an internal argument (itself a DP or NP that has previously been built), forming the head and complement, respectively, of a verb phrase (VP). This VP, in turn, merges with a light verb (v) from the numeration. This little-v can be one of several varieties, or ‘flavors,’ including v-

BECOME

and v-

CAUS

(Folli & Harley 2004, 2007). An example of such a derivation is shown below, where ‘break’ is the V, and ‘the glass’ is the internal argument DP.

(2) vP

DP

the glass

VP

V break

vB

ECOME

Ø

34

The derivation continues with subsequent mergers of two types: external and internal.

External merge is the merger of the existing structure with another item from the numeration, such as the heads Voice (on the view that little-v and Voice are distinct projections; see Pylkkänen 2002, 2008, Harley to appear (a), Manetta to appear), Aspect, and Tense. Internal merge (otherwise known as ‘movement’) is the merger of the existing structure with an element that is already present lower in the tree. For example, if the verb is unaccusative, the internal argument, which is the DP complement of the V head, will be internally merged as the subject, in Spec-TP.

DM brings several innovations to the basic MP model. The most basic concern is the nature of the items in the numeration, and consequently of the terminal nodes of the syntactic derivation. In standard MP approaches, the numeration contains words as well as functional heads, but in DM the numeration contains no words. Instead, it contains category-neutral Roots and abstract feature bundles. The feature bundles have no phonological content at this stage; whether the Roots have phonological content is a matter of controversy (see Harley to appear (b) and references therein). The feature bundles (and, on one view, the Roots themselves) receive phonological content on the

Phonological Form (PF) branch of the derivation, a process known as late insertion.

Thus, in the DM derivation of transitive and unaccusative verbs, the head that merges with the DP complement is a Root rather than a V. The structure is not a verb until the Root Phrase (√P) has merged with the functional head little-v.

35

(3)

DP

the glass

√P

vP

B

REAK

v-B

ECOME

At this point, it appears that the only difference between the derivations of standard MP and DM is the label of the phrase that merges with little-v. In the former, it is a VP, and in the latter a √P. However, this difference in labeling has deeper implications. Firstly the

Root, having no category, is not a syntactic word, and thus words do not occupy terminal nodes in the syntax. Secondly, the phrase headed by a Root may in principle merge with any head compatible with Root phrases, including ones that do not select for verbal projections. This accounts for a number of apparent derivational category mismatches (as discussed in 2.3.1 below), which are unexplainable exceptions if the base of the derivation is already of the category V.

At the point of Vocabulary Insertion, the Root and the little-v feature bundle are realized either as independent phonological (but not syntactic) words, or as a stem and affix, depending on whether head movement has taken place. If √ does not head-move to v, then the two will be separate phonological words; if it does head-move, then v will be realized as an affix on the Root. This study addresses Turkish verbs of the latter type, as represented below in (4) with the derivation of the inchoative verb

kır-ıl-

‘break (int.)’ with the internal argument

bardak

‘the glass.’

36

(4)

DP

bardak

√P

vP

K

IR

vB

ECOME

-Il

This model of word formation is partly inspired by the traditional analysis of

Semitic word formation. Indeed, Chomsky has said (p.c.

4

) that DM is based on the premise that ‘all languages are Semitic.’ In Semitic, words can be grouped according to the occurrence of ordered sets of consonants. These sets are traditionally designated

‘consonantal Roots,’ and are typically triliteral (composed of three consonants), although quadriliteral Roots are not uncommon, and biliteral Roots, while rare, do exist. Each set of Root consonants is found in certain predictable distributions in groups of words that usually share some core semantic property but whose membership cuts across categorial lines. In modern work, the existence of a purely consonantal root in Semitic is controversial (Ussishkin 2005). However, whether or not Semitic Roots can actually be reduced to a series of consonants lacking any vocalic content is not relevant to the assumptions of DM. What matters is that Roots, regardless of their precise phonological characterization, lack grammatical category and specific semantics until merged with a functional head.

Turkish represents a particularly interesting testing ground for this premise. Its agglutinating morphology is as typologically removed from the Root-and-pattern system of Semitic as can be imagined, and so perhaps represents one of the greatest challenges

4

During a meeting with graduate students at the University of Arizona, February 2012, in response to a question posed by Deniz Tat and myself.

37 for the hypothesis that Roots underlie words in all languages. On the other hand, agglutinating morphology has a relatively straightforward relationship to derivation and inflection, so if the hypothesis is correct, it is in just such a language that the most transparent evidence should be found. In other words, we expect that acategorial Roots and the affixes with which they merge will be realized as transparent, discrete phonological units. A language like Turkish can potentially provide the best kind of evidence for the Root.

In English, important evidence for the Root comes from conversion-type alternations, where a single phonological string realizes different categories of word without overt affixation. Most famously, English nouns can be used freely as verbs. Bhatt

& Embick (2003), analyzing causatives in Hindi, lament that this language does not present the same kind of evidence for the Root that English does, since nouns and adjectives do not undergo this kind of null conversion, but must combine with an overt auxiliary in order to gain a verbal function, and so to all appearances the verbs are derived from categories rather than from Roots. Turkish, like Hindi, lacks the productive

‘conversion’ that English has. While it is possible to find examples of strings that occur in both nominal and verbal contexts—e.g.,

gerek

‘necessity; necessary (n; a)’ and

gerek-

‘be necessary (v)’—such cases are comparatively rare. However, the paucity of this kind of evidence is easily explained: Turkish, unlike English, lacks a productive null verbalizing affix. As will be seen in the remainder of this chapter, the discrete and, to a large extent, overt morphology of Turkish provides ample evidence for an acategorial

Root. In fact, the relative transparency of Turkish makes it easier to avoid a common

38 mistake pointed out by Marantz (1997). The notion that the word is the domain of special meanings is, according to Marantz, an error due to mistaking the Root for a word. For example, the verb

watch

and the noun

watch

have significantly different meanings—the first is a verb of agentive visual perception, the second a noun denoting a small timepiece. Similarly, the meaning of the derivative

watchman

does not include the full range of the verb’s semantics: while a watchman is a watcher of sorts, the word does not denote watchers of all kinds (it excludes, for example, stalkers, burglars staking out a location, audience members, etc.). The meaning of

watchman

is therefore specialized in an unpredictable way. Since these words do not transparently contain a base that can be distinguished from the word(s)

watch

, it is natural (but incorrect) to conclude that all of these are derived from a word (

watch

) rather than from a Root (√W

ATCH

) and, more generally, that the word is the domain of special meanings. This pitfall is more easily avoided in Turkish, at least among the native vocabulary, where the phonological coincidence of word and Root is less pandemic. Thus, the Root √B

EK

, which never occurs in unsuffixed form, is easily discerned in the verb

bek-le-

‘wait, await’ and the noun

bek-çi

‘watchman,’ as are the verbalizer –

lA

and the agentive nominalizer –

CI

, both of which are extremely productive derivational suffixes.

Care must be taken, however, to distinguish forms that represent synchronically active processes from forms that are merely fossilized remnants of a historical stage of the language. Especially pertinent are cases where strings of phones that once realized morphologically complex units now realize simple morphemes. For example, while the

39 causative –

DIr

historically derives from the Old Turkic causatives –(

X

)

t

and –

Ur

5

(Ramstedt 1912, cited in Erdal 1991), any attempt at a bimorphemic analysis of the modern suffix would probably be unwise. This observation can, and must, be extended to

Roots. A piece of phonology that was once an affix may be reanalyzed as part of the

Root. I refer to this process as

radical reanalysis

. There is evidence of radical reanalysis occurring sometime between Old Turkic and Old Ottoman. For example, the /y/ of

doy-

‘become satiated’ is the phonological reflex of what was very likely a derivational morpheme in Old Turkic (Erdal 1991), but I argue in 2.4.3 that it is now merely a piece of the Root’s phonology.

2.3 Motivating the Root in Turkish

Outside of Semitic linguistics, the notion of the acategorial Root has not traditionally played a role in morphological analysis. The Turkish tradition is no exception. Turcological literature employs a notation of plus and minus signs to indicate derivational categories. A suffix with a hyphen on the left edge is deverbal, and one with a plus-sign is denominal or deadjectival (Erdal 1991, Tekin 2001, inter alia). Below is a sample of Turkish suffixes that derive verbs, shown in the standard notation based on standard assumptions about selection.

(5) suffixes deriving verbs from nouns and adjectives:

+A, +(A)l, +Ar, +At,+DA, +I, +lA, +lAn, +lAş,

suffixes deriving verbs from verbs:

–DIr

,

–Il

,

–In

,

–Ir

,

–IştIr

,

–It

5

In Erdal’s notation for underspecified Old Turkic vowels,

X

represents any of the high vowels

i, ı, ü,

or

u

(i.e., the equivalent of the notation

I

for modern Turkish), while

U

represents either of the high rounded vowels

ü

or

u

(no modern Turkish equivalent).

40

This notational convention reflects a system that does not recognize the Root as an acategorial entity. In the following, I show that the acategorial Root is necessary to a coherent analysis of derivational morphology in the Turkish verbal system.

One type of evidence for the Root comes from bound stems. Bound stems are found in combination with one or more identifiable affixes, but they do not occur without overt affixation. On the Acategorial Root Hypothesis, all Roots are bound stems, since they cannot appear in a derivation without first merging with a functional head. In cases where the phonology of the word coincides with the phonology of the Root, the affix is null. In the analysis developed in this chapter, almost all monosyllabic verbs

(approximately 200 in number) are Roots that merge with a null verbalizer. Given the existence of overt functional heads, it is expected that there will also be Roots that merge only with overt heads—i.e., bound Roots in the traditional sense. A number of verbs built from such Roots are shown in Table 1. In each case, the base to which the verbalizer is affixed does not occur independently,

6

but is found as the base of other words. The first column contains verbs formed with bound stems and overt verbalizing suffixes (Word 1).

These suffixes are presented separately in the second column. The third column contains other words derived from the same bound stem (Word 2) (the suffixes are not listed separately), and the last column contains a third word for some stems (Word 3).

6

At least not straightforwardly in the modern language.

kap

is a noun meaning ‘container,’ which is arguably based on the same root as

kap-a-

‘close,’ but with greater semantic drift than that between

kap-a-

and the noun

kap-ı

‘door.’

uz

is an archaic word almost exclusively found in fairytale formulas, with roughly the same meaning as

uzak

. In other cases, the semantics suggest that the coincidence in form is nothing more than chance homophony:

yas

‘mourning,’

yak-

‘burn (tr.),’

öd

‘bile,’

kurt

‘worm; wolf,’

sar-

‘wrap (tr.),’

kız

‘girl,’

kur-

‘establish,’

dağ

‘mountain,’

yan-

‘burn (int.).’ In a few instances, the semantics are arguably close enough to indicate root identity, though this is far from certain:

kap

‘container’/

kap-

‘seize,’

kız-

‘become angry,’

uy-

‘conform.’

41

Table 1.

Verbs Based on Bound Stems

Verb (Word 1)

bes-le-

‘nourish’

yas-la-

‘lean, recline’

sak-la-

‘hide (tr.)’

bek-le-

‘wait’

bu-lan-

‘become turbid’

yak-laş-

‘approach’

kap-a-

‘close (tr.)’

uz-a-

‘extend (int.)’

öd-e-

‘pay’

kurt-ar-

‘rescue’

sar-ar-

‘yellow (int.)’

kız-ar-

‘redden, blush’

yeş-er-

‘become green

(vegetation)’

kur-u-

‘dry (int.)’

uy-u-

‘sleep’

dağ-ıt-

‘scatter (tr.)’ suffix

–lA

–lA

–lA

–lA

–lAn

–lAş

–A

–A

–A

–Ar

–Ar

–Ar

–Ar

–I

–I

–It

Word 2

bes-ili

‘plump, well-fed’

yas-tık sak-lı kur-u uy-andağ-ıl-

‘pillow’

‘hidden (adj.)’

bek-çi

‘watchman’

bu-ğu

‘condensation’

yak-ın

‘near’

kap-ı-

‘door’

uz-un

‘long’

öd-ünç

‘loan’

kurt-ul-

‘get free’

sar-ı

‘yellow’

kız-ıl

‘red’

yeş-il

‘green’

‘dry (adj.)’

‘wake up (int.)’

‘scatter (int.)’

Word 3

sak-ın-

‘avoid’

uz-ak

‘far’

kur-ak

‘arid’

uy-ku

‘sleep

(n.)’

dağ-ın-ık

‘in disarray’

yan-sı-

‘reflect (int.)’

çıl-dır-

‘go crazy’

küç-ül-

‘shrink’

–sI

–DIr

–Il yan-kı

çıl-gın küç-ük

‘echo’

‘crazy’ (adj.)

‘small’

Based on this, the following bound Roots plausibly exist: √B

ES

,

√Y

AS

,

√S

AK

,

√B

EK

,

√B

U

,

√Y

AK

,

√K

AP

,

√U

Z

,

√Ö

D

,

√K

URT

,

√S

AR

,

√K

IZ

,

√Y

EŞ ,

√K

UR

,

√U

Y

,

√D

AĞ ,

√Y

AN

,

Â

IL

,

√K

ÜÇ

. Yet evidence of this type proves merely that they are bound Roots, not that they lack grammatical category. Often, bound Roots are consistent in the type of affixes they take. For example,

B

EK

combines with

–lA

and –

CI

, both denominal, while

√D

combines with the deverbal suffixes –

It

and –

Il

(and the compound deverbal

–In-

Ik

). This being the case, it might be claimed that these Roots are categorial in spite of being bound. Pesetsky (1995: 73), though he posits the existence of Roots such as

42

amuse

and √

annoy

, concludes that these Roots must be of the category V, since they admit only deverbal morphology. However, such morpheme distributions are not conclusive evidence of the category of the Root. Radical reanalysis predicts a tendency for such distributions: if Roots have been reanalyzed from words that were historically categorial, they would be fairly consistent in taking deverbal or denominal suffixes, as the case may be, as an artifact of the earlier stage of the language when the etymon of the

Root in question was a verb or a noun. Furthermore, this consistency does not hold in all cases: √Ö

D

combines with both denominal –

A

and deverbal

–Inç

, while

S

AK

combines with denominal –

lA

and

–lI

as well as deverbal

–In

. This is consistent with the principle that derivational affixes may, according to the specifications of each, select for categories as well as for Roots.

These cases of selectional contradictions point to a serious problem of morphological analysis. Without the acategorial Root, a coherent analysis of Turkish morphology is impossible, unless one assumes large amounts of allophony and coincidence. This point is succinctly illustrated with the suffixes –(

A

)

l

and –(

I

)

k

.

(6) verb

alçal

- ‘descend’

küçül

- ‘shrink’

seyrel ufal

- ‘thin out’

- ‘dwindle, shrink’

yüksel

- ‘ascend’

tümsel-

‘become convex’ adjective/noun

alçak

‘low’

küçük

‘small’ base

*

alç-

/*

alça-

*

küç-

/*

küçü- seyrek

‘sparse’ *

seyr-

/*

seyre- tümsek

‘mound (n.); *

tüms-

/*

tümse-

convex (adj.)’

ufak

‘tiny’ *

uf-

/*

ufayüksek

‘high’ *

yüks-

/*

yükse-

The verbs ending in /l/ correspond to adjectives (and some nouns) ending in /k/, but there is no free stem to which the suffixes attach. In this way, these words resemble the words in Table 1. However, the standard analysis has it that the adjectives are in fact

43 the base of the alternation, and that the verbs are derived with the suffix –(

A

)

l

, or, more properly, +(

A

)

l

. This suffix, the story goes, deletes a final /k/ when one is present

(Banguoğlu 1986, Hengirmen 1995, Kornfilt 1997, Göksel & Kerslake 2005). Otherwise, it leaves the base unmolested, as in (7).

(7) a. b. adjective

ayrı

‘separate’

diri

‘living’

doğru

‘straight, correct’

eğri

‘bent, crooked’

ince

‘slender, fine’

kısa

‘short’

koca

‘large; old’

sivri

‘pointed’

yüce

‘lofty, exalted’

az

‘few, little’

boş

‘empty’

çok

‘much, many’

dar

‘narrow’

düz

‘straight’

genç

‘young’

kör

‘blind; blunt’

sağ

‘sound, healthy’

sert

‘hard, severe’

şen

‘cheerful’

yön

‘direction’ intransitive verb

ayrı-l-

‘separate’

diri-l-

‘be revived, come to life’

doğru-l-

‘straighten’

eğri-l-

‘bend’

ince-l-

‘become slender’

kısa-l-

‘shorten’

koca-l-

‘become old (not *‘become large’)

sivri-l-

‘become pointed’

yüce-l-

‘become lofty’

az-al-

‘decrease’

boş-al-

‘empty (int.)’

çoğ-al-

‘increase’

dar-al-

‘become narrow’

düz-el-

‘ameliorate’

genc-el-

‘become youthful’

kör-el-

‘become blunt

sağ-al-

‘recover’

sert-el-

‘become violent’

şen-el-

‘become cheerful’

yön-el-

‘head for a particular direction’

This resembles the pattern seen with some English suffixes such as –

ee

:

evacu-ate

evacu-ee

, but

address

address-ee

. This suffix, too, is traditionally analyzed as truncating some words, while leaving others intact (e.g., Aronoff 1976). In DM, however, there is no truncation; both –

ate

and –

ee

attach independently to the Root √E

VACU

(Marantz 2001). As we will see, the same must be true in the case of –(

A

)

l

: it does not delete the phonological segment /k/, but rather combines with some of the same Roots that the suffix –(

I

)

k

combines with independently.

44

As it happens, there is a /k/-final suffix that derives adjectives (as well as nouns), apparently from verbs.

(8) a. b. verb

büyü

- ‘grow’ adjective

büyü-k

‘large’

çürü

- ‘rot’

soğu ılı

-

ürkat-

‘throw’

- ‘get cold’

çürü-k soğu-k

‘rotten’

‘cold’

‘get luke-warm’

ılı-k

‘luke-warm’

‘start (with fear), shy’

ürk-ek

‘timid’

at-ak

‘bold, daring’

aç as

-

-

‘open (tr.)’

‘hang’

bat

- ‘sink’

boğ

- ‘strangle, choke’

boz

- ‘destroy, spoil’

bük

- ‘bend’

bas-

‘tread; press’

çek

- ‘pull, draw’

çık

-

‘emerge’

del

-

‘pierce’

don

- ‘freeze’

düş

- ‘fall’

-

ez

-

‘bend’

‘crush’

kes

- ‘cut’

kır

-

‘break’

sap

- ‘deviate’

sol sön

-

-

‘fade’

‘be extinguished’

yen

- ‘win/overcome’

yit

-

‘become lost’

aç-ık as-ık

‘open’

‘hanging, hung’

bat-ık

‘sunken’

boğ-uk

‘choked (voice), muffled (sound)’

boz-uk

‘spoiled, broken’

bük-ük

‘bent’

bas-ık

‘low (ceiling); low in stature, squat’

çek-ik

‘pulled back; slanted (eyes)’

çık-ık

‘protruding’

del-ik

‘pierced; (n.) hole’

don-uk

‘frozen; expressionless (face)’

düş-ük

‘low’

eğ-ik

‘bent’

ez-ik

‘crushed’

kes-ik

‘cut (adj.)

kır-ık

‘broken’

sap-ık

‘perverted’

sol-uk

‘faded’

sön-ük

‘extinguished’

yen-ik

‘defeated’

yit-ik

‘lost’

In Kornfilt 1997, such formations are attributed to two suffixes, -

Ik

and –(

A

)

k

, both deverbal. Here I conflate them as –(

I

)

k

, but it may well be that the different vowels are an allomorphic variation. Regardless, what is clear is that there are one or more /k/-final suffixes found in adjectives (and nouns) that correspond to change-of-state verbs.

45

This being the case, the pairs in (6) (repeated below for ease of reference) present a puzzle: The /

l

/-final verbs correspond to /

k

/-final adjectives. Given the prevalence of at least one /

k

/

-

final suffix, a simplex analysis of the adjectives in (6) is not well-motivated.

Not only do they resemble the derived /

k

/-final words in form, they also have the same kinds of semantic relationships with their corresponding verbs. The relationship of

küçük

‘small’ to

küçül-

‘shrink’ is not different from that of

büyük

‘large’ to

büyü-

‘grow,’ nor is the relationship of

tümsek

‘mound (n.)/convex (adj.)’ to

tümsel-

‘become convex’ different from that of

delik

‘hole (n.)/pierced (adj.)’ to

del-

‘pierce.’

(6) verb

alçal

- ‘descend’

küçül

- ‘shrink’

seyrel

- ‘thin out’ adjective

alçak

‘low’

küçük

‘small’

seyrek

‘sparse’ base

*

alç-

/*

alça-

*

küç-

/*

küçü-

*

seyr-

/*

seyre- tümsel-

‘become protuberant’

tümsek

‘mound (n.); *

tüms-

/*

tümse-

convex (adj.)’

ufal

- ‘dwindle, shrink’

ufak

‘tiny’ *

uf-

/*

ufayüksel

- ‘ascend’

yüksek

‘high’ *

yüks-

/*

yükse-

If, as seems plausible, these adjectives are morphemically complex, then it appears equally possible that –(

I

)

k

deletes the /l/ of the verb as the other way around. After all, there are no –(

I

)

k

formations that preserve a final /l/, of the type, e.g.,

düzel-

*düzelk

/

*düzelik

. The simplest solution is that the two suffixes combine independently with a bound stem. But here’s the catch: if –(

A

)

l

is deadjectival and –(

I

)

k

is deverbal, there is no common category that each could combine with. In all likelihood, it is this paradox that has led grammarians to opt for the truncation analysis, which forces entirely different derivations for two groups of /k/-final adjectives, even though they all have the same semantic relationship to corresponding verbs.

46

Once one admits the category-neutral Root into the system, such paradoxes resolve themselves with the greatest of ease. The suffixes –(

A

)

l

and –(

I

)

k

are eligible for insertion adjacent to adjective and verb phrases, respectively, as well as to Root phrases, which have no category. The verbs and adjectives in (6), then, are alternations of suffixes on a bound stem, and this stem is a Root.

2.4 Identifying the Roots of Turkish Verbs

Now that the existence of category-neutral Roots in Turkish has been motivated, the next step is to identify them. A priori, there are several possibilities. A given verb stem might be phonologically coextensive with the underlying Root, in which case the verbalizing morpheme is null. A stem might also comprise a Root and an overt suffix, in which case the question of demarcation between Root and affix arises. In addition, the existence of the Root does not preclude derivation from categorial stems, and indeed, just as a coherent analysis requires Root derivation, so it cannot dispense with categorial derivation.

2.4.1 Allomorph Distribution

N&Ü identify 34 verb classes based on derivational morphemes. (These classes do not include ‘augmented’ verbs derived with diathesis markers including causative, passive, reflexive, and reciprocal.) In N&Ü’s analysis, these 34 classes correspond to the same number of suffixes, listed below.

7

7

One might take issue with some of the specifics of N&Ü’s morphological analysis, as in fact I do regarding several points later in this chapter and in Chapter 4. However, the distributional patterns outlined below remain essentially unchanged.

47

(9)

–A, –(A)l, –AlA, –An, –Ar, –ArlA, –At,–DA, –DAr, –I, –Ik, –I/AklA, –IksA, –Il,

–ImsA, –In, –Ir, –IrgA, –IşlA, –IştIr, –It, –KI, –KIn, –KIr, –lA, –lAn, –lAş, –nA,

–rA, –sA, –s(I), –sIn, –y

Of these 34 suffixes, 24 have a distributional peculiarity: Their left edge always occurs in the second syllable of a derived verb stem. Thus, for example, –

KIn

forms the entire second syllable of a verb stem derived with it (as in

yut-kun-, öy-kün-

), while –(

A

)

l

either forms the second syllable (as in

boş-al-

) or is contained within it (as in

ayrı-l-

). One consequence of this distribution is that when the suffix is monosyllabic, the resulting verb stem is disyllabic (10a), and when the suffix is disyllabic, the verb stem is trisyllabic

(10b).

(10) a.

–(A)l

(N&Ü App. 2a:2)

An

(2a:4)

–Ar

(2a:5)

–At

(2a:7)

–DAr

(2a:9)

–I

(2a:10)

–Ik

(2a:11)

–Il

(2a:14)

8

–In

(2a:16)

–Ir

(2a:17)

–It

(2a:21)

–KI

(2a:22)

–KIn

(2a:23)

–KIr

(2a:24)

–nA

(2a:28)

–rA

(2a:29) b.

–AlA

–ArlA

–I/AklA

–IksA

(2a:3):

(2a:6):

(2a:12)

(2a:13)

alçal-, ayrıl-, azal-, boşal-, bunal-... aban-, aldan-, beğen-, bezen-, diren-

...

ağar-, bozar-, çalar-, göver-, karar-...

gözet-, ilet-, ornat-, yarat-, yönet-

aktar-, gönder-, göster-, kotar-... berki-, cıvı-, eri-, farı-, ılı-, ışı-, kaşı-... acık-, ayık-, birik-, gecik-, gözükasıl-, atıl-, ayıl-, bayıl-, boğul-, bozul-...

acın-, açın-, ağın-, alın-, arın-, aşın-, avun-... aksır-, anır- böğür-, çağır-, hapşır-, osur-, öğür-... azıt-, dağıt-, damıt-, eğit-, ısıt-, kırıt-, öğüt-... oku-, şakı- yutkun-, öykün- bağır-, çemkir-, haykır-, hıçkır-, pavkır-.. ağna-, çiğne-, kayna-, kişne-

çokra-, doğra-, kıpra-, kıvra-, südre-, şakra-...

çabala-, çalkala-, çisele-, deşele-... içerle-, toparla-, tekerle-, uyarla-... açıkla-, ayıkla-, didikle-, durakla-, dürtükle-... kanıksa-

8

The verbs listed in 2a:14 of N&Ü are identified as “reflexive.” In Chapter 4, Section 4.3.2.1, I argue that these are inchoative, or anticausative, rather than reflexive. While the anticausative is syncretic with the passive (see 4.3.2.2), the distributional pattern described here applies to anticausative –

Il

and not to passive –

Il

(the latter of which is not included in N&Ü).

48

–ImsA

(2a:15)

–IrgA

(2a:18)

IşlA

(2a:19)

IştIr

(2a:20)

anımsa-, ayrımsa-, azımsa-, benimse-... basırga-, esirge-, indirge-, yadırga- dürtüşle- anıştır-, araştır, çağrıştır-, eleştir-, geçiştir-...

Since these suffixes always begin in the second syllable of a verb stem, I will refer to them as second-syllable suffixes. In contrast, the suffixes

-lA, -lAn,

and

–lAş

occur in the second, third, and fourth syllable positions.

(11)

–lA

(2a:25)

–lAn

(2a:26)

–lAş

(2a:27)

başla-, baltala-, ihtiyarla- adlan-, akıllan-, bereketlen- adlaş-, ağaçlaş-, asalaklaş-

The total number of verbs formed with second-syllable suffixes exceeds 400

(437), and so a principled explanation of this distributional generalization is desirable.

There are two very different types of possible approach. The first is prosodic: These suffixes may require that their base be a monosyllable. The second is morphological:

These suffixes may be compatible only with a category that has a tendency to be a monosyllable. I argue in 2.4.3 below that monosyllabic stems are always Roots

9

; although this does not entail the converse, that polysyllabic stems are nearly always complex, it does mean that the second-syllable suffixes always appear adjacent to Roots.

I suggest here that Roots do tend to be monosyllables, though this is an empirical question that is currently unresolved. The first hypothesis, based on prosody, makes the syllabicity of the derived verbs a property of the suffixes themselves, while the second makes it an artifact of the properties of the category they merge with. The prosodic hypothesis, as a rule or constraint, ought to be exceptionless, while the morphological

9

With the exception of monosyllables that contain the passive morpheme –

n

. There are only three of these:

de-n-

‘be said,’

ye-n-

‘be eaten,’ and

ko-n-

‘be put’ (see 2.4.3).

49 hypothesis predicts that we will find exceptions to the extent that the category of the complement has exceptions. As already noted, there are no exceptions among verbs formed with the suffixes in (10). However, among the remainder of the suffixes in (9), there are two whose derivations have strong but not exceptionless tendencies. The suffix

A

is found in 52 verbs, of which 51 are disyllabic, the sole exception being

yumuşa-

‘soften (int.),’ which is trisyllabic. Thus, but for a single word, it patterns with the suffixes in (10). (The second is –

y

; in 2.4.3, I argue that this is not a suffix.) As for

–lA

,

– lAn

, and –

lAş

, which occur in any non-initial syllabic position, they derive verbs from nouns and adjectives (N&Ü). This too is consistent with the second hypothesis: if the second-syllable suffixes’ distribution is a result of the fact that they merge with phrases headed by Roots only, and these Roots tend to be monosyllabic, then we would expect denominal and deadjectival suffixes to have no such syllabic restriction.

Nevertheless, the possibility that this distribution is the fossil of an earlier stage of the language cannot be discounted. The second-syllable suffixes are all unproductive, which means that the verbs containing them are all old. It may be that when these verbs were coined, there were few or no polysyllabic Roots, but that since that time many complex polysyllabic stems have undergone radical reanalysis (proposed in 2.2 above), and hence the generalization that Roots tend to be monosyllables no longer obtains. In fact, I argue in Chapter 4 (Section 4.3.2.2) that a set of disyllabic stems ending in –

Ir

are in fact simplex Roots in the modern language, most likely owing to radical reanalysis.

However, this possibility does not pose a challenge to the hypothesis that the stem to which the second-syllable suffixes attach is a Root.

50

2.4.2 Competition and Blocking

If monosyllables are usually simplex Roots, and polysyllables are frequently complex stems, then the syllabic distributions of verbal derivational suffixes are very naturally captured by the DM vocabulary insertion model. In this model, syntactic structure is built first, via the merger of complements with terminal nodes of two basic types: Roots and abstract feature bundles (Marantz 1997, Harley (to appear)).The feature bundles have no phonological content. After the syntactic structure has been built, vocabulary items (VIs) compete for insertion at terminal nodes. At a given node, the most highly specified VI that has no conflicting features is inserted, blocking insertion of the less specific VIs

(Embick 2003, Harley 2008).

2.4.2.1 The Distribution of Causative Allomorphs

Turkish has the following causative allomorphs:

–DIr

,

–t

,

–Ir

,

–Ar

,

–It

,

–Art

(Göksel & Kerslake 2005). In a competition-and-blocking model, each of these must be compatible both with the category of its complement, and with the terminal node [v,

CAUS

]. A given VI is compatible (or not) for insertion outside of a particular category. If the node is linearly adjacent to the Root, then only VIs compatible with Root phrases are eligible for insertion; if the node is adjacent to little-v, then only VIs compatible with vPs are eligible. This being the case, it is of interest that different subsets of the causative VIs appear to be competing for insertion next to monosyllabic stems as opposed to disyllabic stems. The majority of monosyllabic stems take the VI -

DIr

as the causative exponent

(12) ( N&Ü 3.2.1.), while between 20 and 30 monosyllabic stems (depending on how

51

(12)

(13) b. c. d. a. one counts them) take one of the VIs –

Ir

(13a), –

Ar

(13b), –

It

(13c), or –

Art

10

(13d)

(N&Ü 3.1., 3.3.1.). The VI –

t

never attaches to a monosyllabic stem (14).

aç- bak-

çizdol- gül- koşyebat- doğ- düşyit-

çık- kop- ak- korksark-

ürk-

çök- aç-tır- bak-tır-

çiz-dir- dol-dur- gül-dürkoş-tur- ye-dirbat-ır- doğ-urdüş-ür- yit-ir-

çık-arkop-ar- ak-ıtkork-utsark-ıt-

ürk-üt-

çök-ert-

(14)

ye-

öl- ye-dir-

öl-dür-

*

*

ye-t-

öl-t-

When the stem is polysyllabic, the allomorphy is quite different in nature: If the stem ends in a vowel or a liquid, the causative is realized by –

t

(15); otherwise, it is realized by

DIr

(16).

(15) a. b. c.

parlaalçal- karar- parla-talçal-tkarar-t-

(16) a. b. c. d.

bırak- kullan- unut- değiş- bırak-tır- kullan-dırunut-turdeğiş-tir-

10

The allomorph –

Art

, given in Göksel & Kerslake, is not recognized in N&Ü. In Chapter 3, it is argued that this is not a single VI but a compound of –

Ar

and –

t

.

52

Thus, the distribution of causative allomorphs can most obviously be described in phonological terms: –

t

occurs with polysyllabic vowel- and liquid-final stems, and –

DIr

occurs with all other polysyllabic stems as well as with monosyllabic stems apart from about two dozen that take idiosyncratic allomorphs. However, the present proposal invites a fresh look at this distribution. In terms of stem syllabicity, –

t

on the one hand and the idiosyncratic forms –

Ir

, –

Ar

, –

It

, and –

Art

on the other hand have a complementary distribution: –

t

attaches only to polysyllabic stems, while the other attach only to monosyllabic ones. According to the present proposal, verbal Roots are frequently monosyllables. The syllabic distribution of allomorphs may then be an artifact of selection: –

t

selects for vPs and the others select for √Ps. –

DIr

, a good Elsewhere form, may attach to either. These combinatorial properties are indicated below. The members of the sets in (17) represent the heads of phrases that provides the conditioning environment for the insertion of each of the VIs. Because Turkish is a strictly suffixing, head-final languages, the VIs will always be linearly adjacent to the head of the phrasal complement.

(17) v v

CAUS

CAUS v

CAUS v

CAUS v

CAUS

–Ir

–Ar

/_____{√B

/____{Â

–DIr

AT

IK

,

,

√D

√K

OP

,

√D

,...}

ÜŞ ,

√Y

IT

,...}

–It

/_____{√A

K

,

√K

ORK

,

√S

ARK

,

Æ

RK

,...}

–t

/______{√,v}

11

/____{√, v}

According to this, the Root-conditioned allomorphs have lists of particular Roots. In contrast, –

t

and –

DIr

merge in the environment of

v

as well as √, though they have no list

11

As presented here, –

t

has the same environment as –

DIr

. However, –

t

also has phonological specifications: it can appear adjacent only to a polysyllabic base ending in a vowel or liquid. This matter is addressed, but not resolved, in Chapter 5, Section 5.3.4.

53 of individual Roots; –

t

additionally has phonological specifications that prevent it from appearing adjacent to a monosyllabic base of any kind. When a Root merges with the feature bundle v.

CAUS

, the Root-conditioned VIs –

Ir

, –

Ar

, –

It

, and –

DIr

compete for insertion. During Vocabulary Insertion, the VIs’ lists are consulted, and if the Root in question is not on any of those lists, then the Elsewhere form –

DIr

wins the competition.

(18) a. vP-

CAUS

DP

√P

D

OL v

CAUS

-

DIr

b.

dol-dur-

‘fill’

If, on the other hand, the Root is on the list of one of the VIs, that VI is inserted, blocking the insertion of the Elsewhere form.

(19) a.

DP

√P vP-

CAUS

√D

ÜŞ v-

CAUS

-

Ir

b.

düş-ür-

‘drop; knock down’

What determines the distribution of causative allomorphs is the specifications of the VIs. The syllabic properties of this distribution fall out from the fact that all of the

Roots involved are monosyllables. There is no need to invoke prosodic constraints.

If the √P has already merged with a little-v (v-

CAUS

or some other variety), then

Vocabulary Insertion for a v-

CAUS

node outside this vP structure looks quite a bit

54 different. In this case, only the vP-compatible allomorphs, -

t

and –

DIr

, are eligible for insertion.

(20) vP vP v-

CAUS

DP

√P

√ v

Note that –

DIr

occurs in both environments, which means that it occurs with monosyllabic and polysyllabic bases. It is therefore in competition with a number of unpredictable idiosyncratic forms when it occurs in the second syllable, but not when it occurs in later syllables.

12

2.4.2.2 Underspecified Verbalizing Allomorphs

The causative is only one type of verbalizer. The functional head little-v comes in a variety of ‘flavors,’ including

CAUS

,

BECOME

, and

DO

(Folli & Harley 2005, 2007; Folli,

Harley & Karimi 2005). The causative morphemes discussed in the preceding section all realize the feature bundle [v,

CAUS

]. They must therefore be specified for a subset of that bundle. Any of these may be specified for [v,

CAUS

], [v], or [

CAUS

]. If a given VI is [v,

CAUS

], it will appear only in a causative verbal context. A VI that is [v], on the other hand, is capable of realizing any flavor of little-v, provided that it isn’t blocked by a more specific VI. For example, a VI that is compatible with Root phrases and that is specified for [v] could realize a terminal node with the bundle [v,

CAUS

] unless there is another VI

12

As mentioned in the previous footnote, beyond the second syllable –

DIr

and –

t

are in competition in a way that is not idiosyncratic, but fully predictable from the final segment of the stem.

55 compatible with Root phrases that is specified for [v,

CAUS

], in which case the latter VI would win the competition for insertion.

We might see such an underspecified VI not only in causative contexts, but in inchoative (

BECOME

) contexts as well. It turns out that one of the causative allomorphs mentioned above, –

Ar

, is indeed found with inchoative verbs (N&Ü App. 2a: 5.i., ii.).

(21) Causative:

Inchoative: a. b. c. a. b. c.

çık-

√-

‘come out’

çık-ar-

√-

CAUS

-

‘bring out’

kop-

√-

kop-ar-

√-

CAUS

-

‘detach (int.)’ ‘detach (tr.)’

git-

√-

‘go’

gid-er-

√-

CAUS

‘get rid of’

ak

ağ-ar-

√-

BECOME

-

‘white (adj.)’ ‘become white or pale’

*sar-

sar-ar-

√-

BECOME

-

(

sar-ı

‘yellow’)‘yellow (int.); turn pale’

yaş

yaş-ar-

√-

BECOME

-

‘tear (drop)’ ‘mist up, water (eyes)’

It is reasonable to posit that –

Ar

is an underspecified verbalizer. Being [v], it is eligible to realize the bundles [v,

CAUS

] and [v,

BECOME

].

13

But now a question arises: How could such a VI ever win the competition to realize [v,

CAUS

] given the existence of VIs that are fully specified for this feature bundle? I propose that Root-conditioned exponence takes precedence over featural specification, so long as there are no incompatible features. A

VI that has a Feature

f

and is specified to occur in the context of a Root

r

will be inserted

13

As well as [v, D

O

]. Possible instances of this:

baş-ar-

‘succeed,’

bec-er-

‘carry out successfully,’

yak-ar-

‘implore,’

yalv-ar-

‘beseech.’

56 at a node

n

containing

f

that is adjacent to

r

, so long as the VI has no features incompatible with

n

. Thus, -

Ar

, which is [v], realizes [v,

CAUS

] in the context of the

Roots Â

IK

, √K

OP

, and √G

İT

, blocking –

DIr

, which, if it has [v,

CAUS

], is more fully specified featurally but is not specified for those Roots. On the other hand, an exponent can contain any subset of the features to be realized. Thus, –

DIr

might be specified for

[

CAUS

] and not for [v,

CAUS

], in which case it would not be more specified than a VI with

[v]. In fact, in Chapter 5 I argue that this is the case.

Here I propose that, of the ‘causative’ suffixes discussed above,

–Ar

is the only underspecified verbalizer. However, it should be noted here that other causative suffixes have intransitive uses that might be construed as inchoative (v-B

ECOME

) or unergative (v-

D

O

):

del-ir-

‘go crazy,’

çıl-dır-

‘go crazy,’

sap-ıt-

‘go off one’s head,’

sır-ıt-

‘grin smugly,’

kır-ıt-

‘behave coquettishly,’

dire-t-

‘be obstinate.’ I argue in Chapter 3 (Section

3.3.5) that these verbs, which I term evaluative causatives, are in fact causative in spite of being intransitive.

There are other suffixes that, like –

Ar

, are found in both causative and inchoative verbs, including -

A

(N&Ü App. 2a: 1.) and –

lA

(N&Ü App. 2a: 25). These are not traditionally classified as causative suffixes, very likely because the stem used without any overt suffix is rarely a verb, and even when it is, the suffixed and unsuffixed verbs do not stand in a causative/inchoative relationship.

(22) Verbs in –

A

(not exhaustive) bare stem verb

ad

‘name (n.)’

beniz

(

benz-

) ‘facial skin, complexion’

bez

‘cloth’

ad-abez-e

‘dedicate’

benz-e

‘resemble’

- ‘adorn’

57

boş

‘empty’ (adj.)

dol-

‘fill (int.)’

harç

‘expenditure’

kan

‘blood’

kap

‘container’

tür

‘kind, species’

yaş

‘age (n.)’

boş-a

- ‘divorce (tr.)’

dol-a-

‘wind around (tr.)’

harc-a-

‘spend’

kan-a-

‘bleed’

kap-a-

‘close (tr.)’

tür-e

- ‘spring up, derive’

yaş-a

- ‘live’

(23) Verbs in

–lA

(not exhaustive) bare stem

at av

‘horse’

‘hunt (n.)’ verb

at-laav-labelge

‘document (n.)’

belge-lehareke

‘vowel mark

(in Arabic script)’

hareke-le-

‘business; occupation’

iş-leak kok-

‘white’

‘smell (give off a fragrance or odor)’

temiz

‘clean (adj.)’

hafif

‘light (weight; adj.)’

ak-lakok-latemiz-lehafif-leihtiyar

‘aged (adj.); old man’

ihtiyar-laiz

‘(foot)print, trace’

iz-le-

öz sap

‘essence; self’

‘handle; stem’

öz-lesap-latop

‘ball’

top-la-

‘jump’

‘hunt’

‘document (v.)’

‘vocalize (insert vowel marks into Arabic script’

‘penetrate; carve; embroider; function (machine)'

‘acquit; prove innocent (tr.)’

‘smell, sniff (inhale so as to detect fragrance or odor)’

‘clean (tr.)’

‘become light’

‘grow old’

‘follow; watch (e.g., a movie)’

‘miss, long for’

‘thrust (something into something else)’

‘gather, collect (tr.)’

Although N&Ü do not discuss the suffixes in terms of their syllabic distribution, it is striking to note that, while –

A

(22) always appears as the second syllable, –

lA

(23) may be the second (

av-la-

), third (

belge-le-

), or fourth (

hareke-le-

) syllable. But these lists are not exhaustive. As mentioned before, in N&Ü’s dataset, there are 52 verbs formed with –

A

, in 51 of which the suffix forms the second syllable. The sole exception is

yumuş-a-

‘soften (int.),’ where it is the third syllable. The suffix –

lA

, on the other hand, forms 953 verbs in the dataset, and frequently occurs as the second and third syllables, less frequently as the fourth. Indeed, the syllabic position of –

lA

appears to be constrained

58 only by the length of the stem to which it attaches. Both the higher productivity and the freer syllabic distribution of –

lA

are very naturally captured in the present proposal: –

A

is specified for phrases headed by a limited number of specific Roots only, while –

lA

is specified for adjective and noun phrases, and is the default verbalizer of Roots not on the conditioning list of another suffix. Thus, –

A

is found with a limited number of Roots, all but one of which are monosyllabic, while –

lA

is free to combine with unclaimed Roots as well as nouns and adjectives. This pattern mirrors that found between the idiosyncratic causative suffixes, which occur only with a small number of stems and always in the second syllable position, and –

DIr

, which occurs as the second, third, and fourth syllable, and is highly productive.

There is substantial semantic drift between some of the verbs and the corresponding bare stems. Note that when the base is a disyllabic word such as

belge

‘document,’

temiz

‘clean,’ or

ihtiyar

‘aged,’ the semantics of the verb follow straightforwardly, but when it is monosyllabic, the semantics of the verb may either be straightforward (e.g.

av

‘hunt’ /

av-la-

‘hunt’;

kan

‘blood’ /

kan-a-

‘bleed’) or radically drifted (e.g.

öz

‘essence’

öz-le-

‘long for’;

bez

‘cloth’ /

bez-e-

‘adorn’). This too is consistent with the claim that most Roots are monosyllabic, since merger with Roots is a domain of special or ‘negotiated’ meanings (Marantz 1997). These lists are, however, not exhaustive, and there are a few disyllabic bases that exhibit similar drift:

ince

‘thin, fine’ /

ince-le-

‘examine, scrutinize’;

balık

‘fish (n.)’ /

balık-la-

‘dive head first.’ These may be instances of disyllabic Roots. Alternatively, they may be categories, since it is possible to find special meanings within domains larger than the Root, as in verb-object idioms and

59 sentential idioms. Reliably passing judgment in every case would not be a viable goal for this study, and analyses might even vary between speakers.

Finally, note the apparent category mismatch between

kok-

and

kok-la-

, both of which are verbs. –

lA

is widely attested as forming verbs from nouns and adjectives (N&Ü

App. 2a: 25.1., 25.2.), not from other verbs, so the existence of

kok-la-

is a mystery if we insist that stems must be categories. Once we admit category-neutral Roots into the system, this ceases to be a mystery, or even an exception. √K

OK

is a Root, not a verb.

2.4.2.3 Fully Specified and Underspecified Morphemes in Competition.

A catalogue of the VIs realizing little-v heads is presented below. As in (15), the items listed under “Selectional Specifications” are the heads of the phrasal complements

(24) of the feature bundle.

VI

–Ir

–It

–Art

14

–DIr

–t

–Ar

–A

–lA

Features

[v,

[v,

CAUS

CAUS

]

]

[v,

CAUS

]

[

CAUS

]

[

CAUS

]

15

[v]

[v]

[v]

Selection

√B

√A v

Â

√A

AT

K

√, v

IK

D

,

,

,

, √B

√K

√K

√B

İT

OK

OP

ENZ

,

,

,

Â

ÖK

,

√G

ÖÇ

,

√D

√K

√G

√B

İT

√, noun, adjective

,

EZ

,

,

ORK

,

√D

√S

√A

K

√K

OY

,

AN

,

,

√D

ARK

√K

,

IZ

√Y

,

UY

Æ

√S

,

√D

RK

AR

√Y

,

ÜŞ

√Y

...

UMUŞ ,...

When a Root phrase merges with the bundle [v,

CAUS

], the realization of this bundle depends on the specific Root. If it is on the list of one of the VIs, then that VI can be inserted. For example, if the Root is √B

AT

, then the exponent will be –

Ir

:

bat-ır-

‘sink

(tr.).’ This is so even if the VI is underspecified. For example, if the Root is √K

OP

, then underspecified –

Ar

, which has this Root on its list, wins over the fully specified –

DIr

. If

14

In Chapter 3, this suffix is decomposed as –

Ar

[v] plus –

t

[C

AUS

].

15

Justification for this underspecification is found in Chapter 3.

60 the bundle [v,

CAUS

] merges with a category, then the Root-specific VIs will not be inserted. If the category is a verb, then –

DIr

or –

t

will be inserted, according to the phonology of the stem, while if the category is a noun or adjective, -

lA

will be inserted.

2.4.3 Complex Monosyllables

Of the 4,700+ verb stems in N&Ü’s dataset, 211 are monosyllabic. In this section,

I address the question of whether any of these is (overtly) morphologically complex. This is undoubtedly so in a small number of transparent cases where the passive morpheme –

n

is added to a vowel-final stem, producing an overt stem-and-affix combination that is still monosyllabic. In the data set, there are only two vowel-final monosyllabic stems:

de-

‘say’ and

ye-

‘eat.’ The passives of these verbs are

de-n-

‘be said’ and

ye-n-

‘be eaten.’

Missing from the dataset is

ko-

, a colloquial variant of

koy-

‘put’ (N&Ü exclude colloquial and regional verbs from their catalogue), which has the passive

ko-n-

. Thus in principle it is possible for monosyllables to be morphologically complex (in terms of their overt morphology). However, in the following I present evidence that there are no complex monosyllabic stems beyond the three passive forms just mentioned.

N&Ü identify an affix –

y

at the end of 15 monosyllabic verbs (App. 2a: 33). Their list is reproduced below:

(25)

caydoyduygiygöykaykıykoyoysay-

‘swerve; deviate from a purpose’

‘be satiated’

‘hear’

‘put on’

‘burn’

‘slip; slide’

‘mince, slaughter’

‘put’

‘engrave, carve’

‘count; respect’

61

siytüyuyyay

-

‘(of a dog) urinate’

‘slip away’

‘conform; agree; fit’

‘spread; scatter’

The reasons they give for analyzing –

y

as an affix in these verb stems are historical:

–y

is the phonological reflex of a suffix –(

X

)

d

found in Old Turkic (attested 7 th –13 th

centuries

(Erdal 1998)), but which was very rare even at that stage. The features of the purported morpheme are unclear. Although Erdal (1991) suggests that the OT affix –(

X

)

d

formed medial verbs, the modern

y

-verbs have no unifying feature: Their small set includes transitive, intransitive, agentive, and non-agentive verbs. Still, –

y

could be a completely underspecified verbal affix.

In a synchronic analysis, etymology does not provide conclusive evidence. How are we then to determine whether a purported affix is in fact an independent morpheme in the modern language, or merely part of the phonology of the Root? There are several criteria we can use: synchronic attestation of the purported affix in other environments; synchronic attestation of the base to which such an affix would attach (i.e., occurrence of the verb stem without the purported affix); and the absence of Root-conditioned allomorphy in affixes following the verb stem. The first two criteria are self-evident. The third is based on the principle that a Root may trigger allomorphy in a linearly adjacent morpheme. Thus, the presence of such allomorphy indicates that the base is a simplex

Root.

There is little or no synchronic attestation of a suffix in –

y

beyond the 15 monosyllabic verbs ending in this sound. The fact that 15 out of 211 verbs end in a

62 particular sound is not, on its own, persuasive. For example, there are 20 verbs ending in

/l/, 20 in /r/, and 19 in /z/, yet this does not require an affixal analysis of these segments.

Direct evidence for –

y

is weak, but evidence for the stems to which the purported affix attaches is somewhat better. As mentioned above,

koy-

‘put’ has a colloquial variant

ko-

. In addition, there are the verb pairs

doy-

‘be satiated’ /

dol-

‘be full,’ and

say-

‘count, respect’ /

san-

‘suppose,’ suggestive of Roots in √D

O

- and √S

A

-, respectively. The verb

siy-

‘(of a dog) urinate’ is claimed to share a Root with the nouns

sidik

‘urine’ and

sik

‘penis,’ and with the verb

sik-

‘fuck,’ at least from an etymological standpoint (N&Ü,

Erdal 1991). Finally, although not mentioned in N&Ü, the relationship between

kıy

-

‘mince’ and

kıs

- ‘cut down’ is suggestive. Thus, of the 15

y

-final verbs, there is potential evidence for an indepent Root in 5 cases: √D

O

, √K

I

,

√K

O

,

√S

A

,

√S

İ . Also note that, barring some deletion rule, such bases provide evidence not only for –

y

, but also for suffixes in –

l

, –

n

, and –

k

(as well as –

s

), making the verbs

dol-

,

san-

, and

sik-

(as well as

kıs-

) complex stems as well. Indeed, the frequency evidence for these is at least as robust as that for –

y

. There are 20 monosyllabic verbs ending in –

l

, 18 in –

k

, 14 in –

s

, and 17 in

n

(not counting the passives

de-n-

and

ye-n-

). As for the last of these, affixal –

n

is widely attested in Turkish verbal morphology as an allomorph of the passive, the reflexive, and the inchoative. It furthermore appears to alternate with a suffix –

k

in the transitive/intransitive pair

yak-

/

yan-

‘burn (tr./int.)’. If this is so, then the bound stem

*ya-

would in fact be the Root √Y

A

.

Hence, in purely distributional terms, there may be a basis for recognizing Roots

√D

O

, √K

I

,

√K

O

,

√S

A

,

√S

İ , and

√Y

A

, which combine with suffixes –

y

, –

l

, –

n

, and –

k

(and

63 perhaps –

s

). On the other hand, if the historical process of radical reanalysis proposed above is correct, another possibility is that these Roots and affixes, if they ever existed as such, have fused, resulting synchronically in the atomic Roots √D

OY

, √D

OL

,

√K

OY

,

√S

AY

,

√S

İY ,

√Y

AK

, and √Y

AN

. In this case, the distributional facts of the final segments and the remainder of the stem would be nothing more than fossils of historical morphemes.

Clearly, etymological evidence of the type adduced in N&Ü will not serve to distinguish fossils from synchronically active morphology. What is needed is a diagnostic that is sensitive to morphological complexity. Allomorphy provides just this sort of diagnostic. Recall that an idiosyncratic allomorph is conditioned by a Root, and must therefore be linearly adjacent to it, while a regular or ‘Elsewhere’ form may be separated from the Root by one or more affixes. Therefore, although the appearance of an

Elsewhere form is not particularly informative, the appearance of an idiosyncratic allomorph constitutes positive evidence that no morpheme intervenes between the Root and this allomorph. For example, if

doy-

takes an Elsewhere suffix, it may be complex

(

do-y-

) but is not necessarily so, while if it takes an idiosyncratic suffix, it must be simplex (√D

OY

), because it could not be the case that a morpheme –

y

intervenes between a Root √D

O

and the allomorph. Turkish verbal morphology is ‘regular’ to a great extent, but the causative and the aorist exhibit the relevant kind of allomorphy.

2.4.3.1 Causative Allomorphy

As discussed in 2.4.2.1 above, causative allomorphy is determined by the Root.

The idiosyncratic allomorphs are Root-conditioned, and hence can occur only in a Root-

64 adjacent position. Note that the presence of the Root alone is not sufficient to license an allomorph. If the causative node is linearly separated from the Root by another head— e.g., another causative—then either –

DIr

or –

t

occurs (according to the phonological environment).

(26)

ak-

ak-ıt-

√-v.

CAUS

-

‘flow’ ‘make flow’

(*

ak-tır-

)

ak-ıt-tır

-

√-v.

CAUS

-v.

CAUS

-

‘cause to make flow’

(*

ak-ıt-ıt-

)

This follows directly from the fact that the idiosyncratic allomorphs are not compatible with categories. Any stem that contains a suffix is not a simplex Root. It therefore makes sense to look at the causative versions of

y

-final stems. If an idiosyncratic allomorph appears, there can be no affix intervening between the causative and the stem, and the stem must be simplex. Note, however, that the occurrence of the Elsewhere form –

DIr

tells us nothing, since this may combined with √Ps as well as vPs. It turns out that two of the 15 y-final verbs take the causative allomorph –

Ir

:

doy-

and

duy-

.

(27)

caydoyduygiygöykaykıykoyoysaysiytüyuyyay

-

cay-dır-

doy-urduy-ur-

giy-dir-

[

göy-dür-

16

]

kay-dırkıy-dırkoy-duroy-dursay-dır-

---

tüy-düruy-duryay-dır-

16

This causative does not occur in the dataset, but is found in dictionaries, e.g.,

Redhouse

Turkish/Ottoman-English Dictionary

(1997).

65

Both of these verbs are specifically argued in N&Ü to be formed with –

y

(based on historical evidence). The causative form

doy-ur-

is particularly meaningful, since

doy-

is one of the verbs where /y/ appears to alternate with another segment:

doy-

‘become satiated’ /

dol-

‘become full.’ We are forced to conclude that

doy-

does not have any overt suffix. As for

dol-

¸ its causative has the Elsewhere form:

dol-dur-

‘fill.’ While this tells us nothing, we will see evidence below that

dol-

too is a simplex stem. Despite the obvious similarities in form and semantics between

doy-

and

dol-

, and their historical relationship, these two stems do not currently share a base in

do

- to which monosegmental affixes are attached. Synchronically, they may be distinct Roots, or variant (allomorphic) stems of a single Root.

2.4.3.2 Aorist Allomorphy

Causative allomorphy gives us positive evidence for only two of the

y

-final verbs.

Fortunately, Turkish has another morphological category exhibiting a type of allomorphy that is informative for present purposes: the marker of the aorist tense.

17

The aorist suffix has the form –

r

after a vowel, and –

Ir

or –

Ar

after a consonant. All polysyllabic consonant stems take –

Ir

(28a). Most monosyllabic consonant stems take –

Ar

(28b), while about a dozen take –

Ir

(26c). All vowel-final stems take –

r

(28d).

(28) a. b. bare stem

bırak-

‘leave’

konuş-

‘speak’

sev-

‘love’ aorist

bırak-ır

‘leaves’

konuş-ur

‘speaks’

sev-er

‘loves’

17

The function of the aorist is a complicated matter that is not immediately relevant. Herein, it is glossed with the English simple present, to which it corresponds in some usages.

66 c. d.

döv-

‘beat’

al-

‘take’

de-

‘say’

dene-

‘try’

döv-er

‘beats’

al-ır

‘takes’

de-r

‘says’

dene-r

‘tries’

Ito & Hankamer (1989) conflate –

Ir

with –

r

, such that –

r

is the basic form, and a high vowel is epenthesized following a consonant. This leaves us with two aorist allomorphs,

Ar

and

–(I)r

, and simplifies the generalization to be made about distribution: –

Ar

attaches to most but not all monosyllabic stems, while –

(I)r

attaches to all polysyllabic stems, as well as a minority of monosyllabic ones.

The aorist is also sensitive to the morphological complexity of the stem. All stems with a consonant-final derivational affixes take –

(I)r

, even those whose bare form takes

Ar

:

(29) bare stem

sev-

‘love’

döv-

‘beat’ aorist

sev-

er

‘loves’

döv-

er

‘beats’ increased stem

sev-dir-

(causative)

‘cause to love’ aorist

sev-dir-

ir

‘causes to love’

döv-üş-

(reciprocal)

döv-üş-

ür

.

‘fight each other’ ‘fight each other’

Note that the complex (increased) stems above are also polysyllabic, such that the allomorphy could still be explained in purely phonological terms without invoking morphology. However, the allomorphy is sensitive to morphological complexity even when the complex stem is a monosyllable. Monosyllabic stems containing passive –

n

, which are clear-cut cases of complex monosyllables, take not –

Ar

, but –

(I)r

:

de-n-

‘be said,’

den-ir

;

ye-n-

‘be eaten,’

ye-n-ir

.

Particularly revealing is the fact that the apparently simplex stem

yen-

‘defeat,’ which is homophonous with

ye-n-

‘be eaten,’ takes

67 the –

Ar

variant:

yen-er

‘defeats.’ Furthermore, the colloquial variant

ko-

‘put’ (standard

koy-

) has its passive in

ko-n-

, which is homophonous with simplex

kon-

‘alight.’ Again, the complex stem takes the aorist in –

(I)r

, and the simplex stem in –

Ar

:

ko-n-ur

‘is put,’

kon-ar

‘alights.’ Thus, in all cases where the base is demonstrably complex, the aorist is –

(I)r

, irrespective of the syllable count. Also, an overwhelming majority of monosyllabic verbs take –

Ar

(196 out of 209 consonantal-final stems). Ito & Hankamer (1989) nevertheless advanced a purely phonological account of aorist allomorphy, but Hankamer

(2013) includes sensitivity to morphological complexity in his account.

As it turns out, the 15

y

-final verbs uniformly take their aorist in –

Ar

, which disqualifies them from being overtly complex, and thus rules out the possibility that they contain a suffix in –

y

. Similarly, both members of the apparently equipollent pair discussed above,

yak-

/

yan-

‘burn (tr./int.),’ also fail this test for morphological complexity, taking–

Ar

:

yak-ar

‘burns (tr.)’,

yan-ar

‘burns (int.).’ This particular alternation is better characterized as Root allomorphy than as overt suffixation.

Apart from the demonstrably complex stems containing the passive morpheme

n

, there are about a dozen monosyllables that take –

Ir

rather than –

Ar

in the aorist. All but one of these ends in –

l

or –

r

, while

san-

‘suppose’ ends in –

n

. Since there are synchronically attested suffixes ending in these three consonants, the possibility that they are complex stems remains. Furthermore, none of the verbs taking the aorist in –

Ir

takes an irregular causative allomorph. Thus, there is no positive evidence that these stems are simplex. All but one of them ends in –

l

or –

r

, but it is by no means the case that all stems ending in these segments take –

Ir

, and so this cannot be attributed to a simple

68 phonological rule. These facts are suggestive of morphological complexity, especially since the only other monosyllable taking the –

Ir

aorist ends in –

n

(

san-

‘suppose’), which is widely attested as an affix (being a conditioned variant not only of the passive, but also of the reflexive and the inchoative). However, another possibility is that the stems are simplex, and that their Roots are simply not present on the selectional list of the –

Ar

allomorph. In the absence of clear diagnostics in these cases, their morphological status is considered to be unresolved.

2.4.3.3 Historical Forms of the Aorist

If –

y

is the phonological, but not the morphological, reflex of an archaic derivational affix, and if the aorist allomorph –

Ar

is indicative of the absence of any overt derivational morphology, there is no conflict in a synchronic analysis in which

y

-stems are simplex. Nevertheless, questions arise as to the historical development of this situation. Either the aorist of these formerly complex stems has normalized following their reanalysis as simplex Roots, or at an earlier stage of the language –

Ar

was not Rootexclusive. The truth may well be a combination of these two possibilities.

In Old Turkic, as today, –

Ar

was the usual form with simple stems. There were major differences, however. Even apparently simple polysyllabic stems took –

Ar

, as did intransitive stems containing one of a particular set of derivational affixes (Marcel 2004).

However, by the Old Ottoman period (13 th

–15 th

centuries), the earliest stage at which a distinct ancestor of modern Turkish is identifiable, the distributions of –

Ar

versus –

Ur

(the archaic form of –

Ir

) were much as they are today: -

Ar

on most simple monosyllables;

Ur

on all complex stems and polysyllabic stems, and on a small minority of simple

69 monosyllables, which include virtually all of the monosyllables taking –

Ir

today

18

(Mundy 1954). There are, however, some monosyllables whose aorist in modern Turkish does not coincide with that of Old Ottoman. Interestingly, one example that Mundy cites is the

y

-stem

gey-

, the archaic form of

giy-

‘put on (clothing),’ whose aorist is recorded as both

gey-ür

(with –

Ur

) and

gey-er

(with –

Ar

, as in modern Turkish). This suggests a reanalysis of

gey-

as a simplex stem, followed by a transitional period during which it may have retained the –

Ur

aorist before this was replaced with the Root-adjacent –

Ar

morpheme. A parallel case is seen with the verb

san-

‘suppose.’ In Old Turkic, this was derived from the stem

sa-

‘count’ (Mundy 1954, Erdal 1991), and so was complex.

Today, this verb shows variation in its aorist similar to that of

gey-

in Old Ottoman.

Although

san-ır

(minus the hyphen) is the standard written form, sanctioned by dictionaries, the colloquial form

san-ar

is often found in speech and on-line. Google searches for forms such as

sanırsın

and

sanarsın

‘you (sg.) suppose’ yield among their top results discussions about which is the correct form. This too is consistent with a model where a formerly complex stem has been reanalyzed as simplex, though the normalization process of its aorist has taken considerably longer than that of

giy-

. Indeed, it is not necessary for reanalyzed verbs to normalize at all: By the Elsewhere Principle, the Elsewhere form –

Ir

can occur with simple stems as well as complex ones. Reanalyzed stems may normalize or not.

18

Mundy includes 12 of the 13 simple monosyllables taking –

Ir

today. Missing is

var

- ‘arrive,’ but since he does not discuss this verb, it is unclear whether this is an oversight, or whether it in fact took –

Ar

in Old Ottoman .

70

The previously complex

sa-n-

has become the synchronically simplex

san-

.

Although it is historically derived from the same Root as the y-stem

say-

‘count,’ the reanalysis renders the apparent –

y

/-

n

alternation spurious, further diminishing any evidence for a synchronic affix in –

y

.

2.4.3.4 Acquisition of the Aorist

The basis of using the aorist as a diagnostic for stem complexity is based on the hypothesis that the allomorphy is sensitive to stem complexity rather than prosodic factors. This idea finds interesting support in an acquisition study—somewhat surprisingly so, since the authors of the study themselves reach different conclusions.

Nakipoğlu & Ketrez (2006) conducted an aorist elicitation study in 135 children aged from 2 years 9 months to 8 years 2 months, recording error rates with different types of verb stems in different age groups. Although this is an acquisition study, their results are potentially important to our understanding of adult grammar. Their assumptions and their interpretation of the results are at considerable variance from the proposal advanced in this study. Nonetheless, I argue that the present proposal is a better fit for N&K’s results than the explanations given in their paper.

71

Table 2.

Results from Nakipoğlu and Ketrez (2006) for error rates (%) in aorist formation with consonant-final verbs.

Verb stem type

Direction of error polysyllabic -Ir>*-Ar monosyllabic (varies)

-Ar>*-Ir nonsonorant– final mono-σ sonorantfinal mono-σ

(varies)

-Ir>*-Ar

-Ar>*-Ir

G1 (mean age: 3;5)

0.6%

26%

5%

34%

39%

29%

G2 (mean age: 4;5)

9%

24%

12%

28%

18%

38%

G3 (mean age: 5;5)

8%

15%

3%

21%

13%

28%

G4 (mean age: 7;0)

0.4%

8%

1%

18%

7%

27%

G5 (mean age: 7;9)

0%

6%

2%

8%

3%

11%

The table shows the error rates in aorist formation in consonant-final verb stems according to stem type and age group. There are many striking facts about these results.

Among these are increases in certain types of error between G1 and G2: errors with polysyllabic stems, virtually non-existent in G1 (0.6%), explode to 9% in G2, and errors with monosyllabic stems not ending in a sonorant more than double, from 5% to 12%.

Although the overall trend in G1–G5 is a reduction in errors of all types, these sharp increases in two specific types of errors from G1 to G2 (which are not addressed in

Nakipoğlu & Ketrez’s study) suggest a paradigm shift between these two age groups.

This surprising change, which looks like a case of the U-shaped learning curve (Marcus et al. 1992), can be accounted for under the present proposal. This proposal is reformulated as the adult grammar in (30) below.

(30)

Adult grammar

: The choice of aorist allomorph with consonant-final stems is determined by the Elsewhere Principle:

1. The Root-conditioned morpheme –

Ar

occurs with overtly simplex stems

(Roots) which are on this morpheme’s list.

2. The variant –

r

occurs with vowel-final stems.

72

3. The Elsewhere morpheme –

Ir

occurs in all other environments, including complex stems, as well as simplex stems not listed for –

Ar

.

It is safe to say that G1 subjects do not yet have the adult grammar. In fact, it looks like their criteria for allomorph choice are purely phonological. These are formulated in (31).

(31)

G1 grammar

: The choice of aorist allomorph is determined by the stem’s phonology:

1. Non-sonorant–final monosyllabic stems take –

Ar

.

2. Sonorant-final monosyllabic stems take either –

Ir

or

–Ar

.

3. Vowel-final stems take –

r

.

4. Polysyllabic stems take –

Ir

.

The generalization that polysyllabic consonant stems take –

Ir

is in fact exceptionless in

Turkish. This explains the virtual absence of mistakes made in this category by G1

(0.6%–there was only 1 such error in 154 elicitations with polysyllabic stems). The generalization that non-sonorant–final monosyllabic stems take –

Ar

is also exceptionless, and hence we have a very low error rate in this category (5%), which is significantly higher than that for polysyllables plausibly because monosyllables as a whole have a much higher rate of –

Ar

than of -

Ir

, and hence performance errors are more likely. The areas of high error rates (29% and above) are those where this simple, phonology-based grammar does not provide sufficient guidance: sonorant-final monosyllabic stems.

Between G1 and G2 a radical shift occurs. The subjects suddenly perform dramatically worse in precisely the environments where the phonology-based grammar provides reliable guidance: polysyllabic stems, and non-sonorant–final monosyllabic stems. This strongly suggests that the subjects have abandoned the phonological account.

In fact, it stands to reason that they have actually settled on the adult grammar as stated in

(30), but that their morphological analysis of the actual verb stems is still in its infancy.

73

That is, they still have to complete what is one of the major tasks of this chapter: demarcating Root material from affixal material. The increase in the ungrammatical use of the Elsewhere form –

Ir

with sonorant-final monosyllabic stems (from 29% to 38%) may be the result of an analysis of stem-final sonorants as derivational morphemes.

From this point on, the morphological analysis gradually approaches that of adult speakers, and error rates in all categories gradually decrease from age group to age group, with no further cataclysmic upheavals. There is, however, an interesting development between G3 and G4: The error rate with polysyllabic stems drops sharply from 8% to

0.4%, very close to the G1 rate of 0.6%. This suggests that G4 subjects have re-adopted the abandoned phonological hypothesis, but only for polysyllabic verbs. In that case, polysyllables aren’t necessarily complex, but we would have two distinct rules governing aorist allomorphy, a morphological one and a phonological one, and the vast majority of complex stems would be redundantly required to take –

Ir

, by virtue both of being complex, and of being polysyllabic.

Nakipoğlu & Ketrez do not consider a morphologically based account of aorist morphology, and they have a significantly different interpretation of the results. They explain their data in terms of the verb vocabulary and verbs occurring in child-directed speech at each of the age groups. Of the verbs that children of the G1 age group are habitually exposed to, 52% are monosyllabic, and of these 70% take –

Ar

. This would explain the high rate of –

Ir

>*–

Ar

errors in monosyllabic verbs in this group. With age, children are increasingly exposed to more verbs taking –

Ir

, whether polysyllabic or morphologically complex, which would explain the gradual decrease in errors in this

74 same category of verb. However, although the vocabulary of different age groups is doubtless a factor in aorist production, it cannot explain why, in G2, learners begin erroneously using –

Ar

with polysyllabic stems at a rate of 9%, an error type that is virtually non-existent in G1 (0.6%–there was in fact only 1 error out of 154 elicitations in this category). Indeed, the verb vocabulary account would predict the exact opposite: G1 ought to be using –

Ar

with polysyllables as well as monosyllables, while G2 ought to be doing so at a lower rate. The current proposal provides a satisfactory account: Between

G1 and G2, learners abandon their initial phonology-based hypothesis, which, for polysyllabic verbs, was an infallible guide.

2.5 Polysyllabic Stems

We now turn to polysyllabic verb stems. If the overwhelming majority of monosyllabic stems are (overtly) simplex, the next question that arises is the morphological status of polysyllabic stems. Are they too mostly simplex? Or are they mostly, or even entirely, complex? Slicing up these stems in various ways, or not slicing them up, would result in numerous and varied potential Roots and affixes. The most plausible of these myriad possibilities are addressed in some detail in the following sections. With regard to the allomorph diagnostic, there is a sweeping generalization to be made: Not a single polysyllabic stem takes an irregular causative suffix, and none take the Root-conditioned –

Ar

form of the aorist (other than complex predicates formed with the auxiliary

et

- ‘do,’ itself a monosyllabic verb whose aorist is

ed-er

). Of course, this amounts merely to the absence of evidence that the stems are simplex Roots, and does not necessarily mean that all of them contain an overt suffix. Nevertheless, it is striking that

75 of 211 monosyllabic verbs, 28 (13%) take an idiosyncratic causative allomorph and 196 take the Root-conditioned –

Ar

aorist (93%), while out of more than 4,000 single-word polysyllabic stems, not one takes either type of Root-conditioned allomorph. We have already seen that the distribution of aorist morphemes is sensitive to the presence of overt suffixes independently of syllable count. A case could be made that most polysyllabic stems do indeed contain suffixes.

19

The facts about second-syllable suffixes discussed in 2.4.1 suggest that polysyllabic stems are, generally, morphologically complex. Verifying that this is indeed the case would require many detailed studies that are beyond the scope of the present work. However, in the following, I take a preliminary step in this direction. First, I present several dozen verbs derived from disyllabic stems that appear at first glance to be

Roots. I then show that, despite appearances, these stems are complex, and that the Roots from which they are formed are monosyllabic.

The verbs in question are onomatopoeic verbs derived with the suffix –

DA

(N&Ü

Appendix 2a:8.i). The stems to which –

DA

attaches are disyllabic, and are also categoryneutral. Nonetheless, close inspection reveals that these category-neutral disyllables can also be reduced to monosyllabic Roots.

19

Key (2012b) observes that polysyllabic verb stems in Turkish end either in a vowel or in one of six consonants: –

k

, –

l

, –

n

, –

r

, –

ş

, or –

t

. Note that these same six consonants are a perfect match for the consonants found at the end of the verbal derivational suffixes listed in N&Ü in (9) above (with the exception of –

y

, which, as I argue in 2.4.3, is not truly a suffix). Monosyllabic stems, on the other hand, can end in any of 15 consonants (Key 2012b erroneously states that there are only 14). Of the 21 (orthographic) consonants of Turkish, the occurrence of 5 can be ruled out for independent reasons (historical and phonological), such that 15 of the 16 expected consonants do occur at the end of monosyllabic stems. These facts are consistent with a generalization that monosyllables are Roots, and polysyllables contain overt

(verbal) suffixes. This is a matter that merits further investigation.

The disyllabic bases to which –

DA

attaches are category-neutral (Table 4); in general, they do not occur alone as nouns, adjective, adverbs, or verbs (though there are sporadic exceptions). As a rule, they appear in combination with derivational morphology:

–DA

(deriving verbs),

–tI

(deriving nouns), or full reduplication (deriving adverbs).

Table 3.

Onomatopoeic verbs, nouns, and reduplicated adverbs based on disyllabic bound stems. base

*bangır

*bıngıl

*cayır

-

DA

(verb)

bangır-da-

‘yell loudly’

bın-gıl-da-

‘quiver like jelly’

cayır-da-

‘crackle’

-

tI

(noun)

cay-ır-tı

‘crackling’

*cazır

*cırıl

*cıvıl

*cıyır

*cızıl,

*cızır

*cumbul,

*cumbur

*çağıl

*çakıl

*çangır cırıl-da-

‘chatter, screech’

cıv-ıl-da-

‘chirp, tweet’

cıy-ır-da-

‘make a ripping sound (paper, cloth)’

cız-ıl-da-

,

cız-ır-da-

‘sputter’

cum-bul-da-

‘splash’

cum-bur-da-

‘plop’

çağ-ıl-da-

‘burble

(water)’

çakıl-da-

‘clatter’

çangır-da-

‘jangle’

cırıl-tı

‘creaking noise’

cıv-ıl-tı

,

cıv-ır-tı

‘chirping sound’

cıy-ır-tı

‘sound as of cloth tearing’

cız-ıl-tı

,

cız-ır-tı

‘sizzling noise’

cum-bul-tu

‘splashing sound’

çağ-ıl-tı

‘burbling’

çakıl-tı

‘clattering’

çangır-tı

‘jangling’

*çatır

*çıtır cazır-da-

‘crackle’

çatır-da-

‘crackle’

çıtır-da-

‘crackle’

çatır-tı

çıtır-tı

‘crackling’

‘crackling’ reduplication

ban-gır ban-gır

‘sobbingly’

cayır cayır

‘(burning) fiercely’

cazır cazır

‘with a crackling noise’

cıv-ıl cıv-ıl

‘with a soft, gurgling sound’

çağ-ıl çağ-ıl

‘with a burbling noise’

çangıl çungul

‘with a clanging noise’

çatır çatır

‘with a crackling noise’

çıtır çıtır

‘with a crackling noise’

76

*dangır dingil

*fıkır

*fısıl

*fışır

*fokur dangır-da-

boorishly’

fısır-da-

‘speak

dingil-de-

‘rattle’

fıkır-da-

‘make a bubbling noise’

fısıl-da-

‘whisper’

‘make a light puffing noise’

fokur-da-

‘gurgle’

fıkır-tı fısıl-tı fısır-tı

noise’

‘bubbling noise’

‘whisper’

‘light puffing

fokur-tu

‘gurgling’

dangıl dungul

‘(talking) boorishly’

fıkır fıkır

‘with a bubbling noise’

fısıl fısıl

‘with a whispering voice’

fısır fısır

‘with a light puffing noise’

fokur fokur

‘with a gurgling noise’

fosur fosur

‘in puffs’

*fosur

*foşur

*gacır

*gıcır

*gurul fosur-da-

‘breathe noisily’

foşur-dagacır-dagıcır-da gurul-da gürül-de harıl-da-

‘make loud and continuous noise’

haşır-da-

‘make a rough scraping sound’

hırıl-da

‘growl; wheeze’

hışıl-da-

wheezing/rustling noise

hışır-da

‘rustle’

horul-da

gurgle’

‘make a

‘snore,

hüngür-de

‘sob’

harıl-tı

‘loud and continuous noise’

haşır-tı

sound’

hırıl-tı

‘harsh crashing

‘sound of snoring, snarling, or growling’

hışıl-tı

‘repeated wheezing or rustling sound.’

hışır-tı

‘a rustling, grating’

horul-tu

‘snore, a snoring’

ışıl-da

‘plash’

‘creak’

- ‘creak’

- ‘rumble’

*gümbür gümbür-de

‘boom;

*gürül

*harıl

*hırıl

* hışıl

*hışır

*ışıl

* haşır

*horul

*hüngür

thunder’

‘gurgle’

‘glimmer’

gacır-tı

‘creaking noise’

gıcır-tı

‘creaking noise’

gacır gucur

‘with a creaking noise’

gıcır gıcır

‘squeaky clean, brand new’

gurul-tu

‘rumbling noise’

gümbür-tü

‘a booming’

gürül-tü

‘noise’

gümbür gümbür

‘a booming noise’

gürül gürül

‘in a loud, throaty voice’

harıl harıl

‘incessantly’

ışıl-tı

‘flash, twinkle’

haşır haşır huşur hırılhırıl

snort’

‘crunch, crash’

hışıl hışıl

;

haşır

‘repeated

‘with a rustling or wheezing noise’

hışır hışır

‘harsh grating noise’

horul horul

‘with a snoring sound’

hüngür hüngür

‘(sobbing) violently, bitterly’

ışıl ışıl

‘sparkling,

77

*inil

*kakır kıkır

*kımıl

*kıpır kıtır kütür lıkır

*mırıl

*

mışıl

* parıl patır

*pırıl

* pıtır

*pofur

*sakır

*şakır

*şangır

*şapır inil-de

‘moan, groan’

kıpır-da

‘move slightly’

kıtır-da-

‘make a crunching noise’

kütür-de-

‘make a crunching sound’

lıkır-da-

‘gurgle’

kakır-da

‘rattle; rustle’

kıkır-da

‘giggle; rustle or crackle’

kımıl-da

‘budge’

inil-ti

‘moan, a moaning’

kıkır-tı

‘giggling, snicker’

kımıl-tı

‘movement, agitation’

kıpır-tı

‘slight and quick movement’

kıtır-tı

‘crunch, crack

(sound’

kütür-tü

‘sound of cracking’

mırıl-damışıl-da-

soundly’

‘murmur’

‘sleep

parıl-da-

‘gleam, glitter’

patır-da-

‘make a knocking’

pırıl-dapıtır-daşakır-da-

‘sparkle’

‘make a

‘make a tapping sound; patter’

pofur-da-

beat; sing’

şapır-da-

‘make a popping noise

sakır-da-

‘shiver due to fear/cold

şangır-da-

‘(of rain) to

‘make a sound of crashing

mırıl-tı

murmuring’

parıl-tı

flash’

patır-tı pırıl-tı pıtır-tı

‘light tapping or crackling’

sakır-tı

‘glitter, gleam,

‘noise, tumult’

‘a shivering with fear or cold’

şakır-tı şangır-tı

‘the noise of breaking glass’

şapır-tı

‘a muttering,

=

‘repeated clatter or rattle’

parıl-tı

‘smacking noise shining brightly’

kıkır kıkır

‘giggling(ly)’

kımıl kımıl

‘wiggling, fidgeting’

kıpır kıpır

‘wiggling, fidgeting’

kıtır kıtır

‘(eat) with a cracking noise’

kütür kütür

‘a crunching sound’

lıkır lıkır

‘(drink) going glug glug’

mırıl mırıl

‘muttering, grumbling’

mışıl mışıl

‘sound of deep breathing in sleep’

parıl parıl

‘gleamingly’

patır patır

‘sound of footfalls’

pırıl pırıl

=

parıl parıl pıtır pıtır

‘lightly, softly (footsteps)’

pofur pofur

‘puffing regularly’

sakır sakır

‘shivering, trembling’

şakır şakır

;

şakır şukur

‘a rattling noise’

şangır şıngır

;

şangır şungur

(imitates the noise of breaking glass)

şapır şapır

;

şapır

78

*şarıl

*şıkır şıngır

*şıpır

*şırıl

*takır

*tangır

*tapır tıkır

*tıngıl

*tıngır

*tıpır

*tiril tokur

*uğul

*vıcır

*vırıl

slurping noise’

şarıl-da-

‘flow with a splashing noise’ of the lips’

şarıl-tı

‘a gurgling, splashing noise’

şıkır-da

‘rattle; jingle’

şıkır-tı

‘a jingling noise’

şupur

(imitate the smacking of lips)

şarıl şarıl

(imitating the sound of running water)

şıkır şıkır

‘dazzling, shiny’

şıngır şıngır

‘crash

(of breaking glass’

şıngır-da

‘crash; make the noise of breaking glass’

şıpır-da

‘(of water) to make a lapping noise’

şırıl-da

‘burble’

şıngır-tı

=

şangır-tı şıpır-tı

‘splash’

şırıl-tı

‘splashing, gurgling’

takır-da

‘make a tapping or knocking noise’

takır-tı

‘a repeated tapping or knocking noise’

tangır-da

‘clatter; clang’

tapır-da

‘make the noise of footsteps’

tangır-tı

‘a repeated clanging’

tıkır-da

‘rattle lightly’

tıkır-tı

‘a rattling’

şıpır şıpır

‘with a dripping sound’

şırıl şırıl

(imitates the noise of running water)

takır takır

;

tukur tukur

(the noise of horses’ hooves, etc.)

tangır tungur

‘noisily, clangingly’

tıkır tıkır

‘with a rattling noise’

tıngıl-da

‘tinkle; clink’

tıngır-da

‘clink; clang’

tıngır-tı

‘clinking, clanging’

tıpır-da tiril-de

‘make a light tapping noise’

‘shiver’

tokur-da

‘make a bubbling noise with a hookah’

uğul-da

‘hum; buzz’

tıpır-tı

‘sound of drops falling’

tokur-tu

‘bubbling noise of a hookah’

tıngır tıngır

,

tıngır mıngır

(imitates the sound of metalllic things knocking together)

tıpır tıpır

(imitates the sound of drops falling)

tiril tiril

‘shivering, trembling’

vıcır-da vırıl-da

‘chirp’

‘talk

uğul-tu

‘humming, buzzing’

vırıl-tı

‘a whirring,

79

80

*vızıl

*yelp

*zangır

*zımbır

*zıngır

*zırıl

incessantly’

vızıl-da

‘buzz, hum’

yelpir-de

‘move slightly’

zangır-da

‘tremble with teeth chattering’ buzzing noise’

vızıl-tı

‘buzzing or whirring noise’

zımbır-da

strum’

zırıl-da zıngır-da

blubber’

‘twang;

‘rattle’

‘bitch;

zangır-tı

‘a clanging, rattling noise’

zangır zangır

‘trembling; with the teeth chattering’

zımbır-tı

‘a twanging or strumming noise’

zıngır-tı

‘a rattling noise’

zırıl-tı

‘continuous chatter’

zıngır zıngır

(imitates the sound of violent trembling)

However, despite this acategorial behavior, there is evidence that the disyllabic base of –

DA

is never a Root. These bases demonstrably consist of a monosyllabic Root plus a suffix. The first piece of evidence that they contain a suffix is that they all end in the vowel

I

(with realizations

i

,

ı

,

ü

,

u

) plus a liquid (r or l), such that there appear to be suffixes of the forms –

Ir

and –

Il

. The generalization can even be made more specific: The majority of these disyllabic bases consist of the string (C)VC plus –

IL

(where L represents a liquid with realizations /l/ and /r/):

fıs-ıl

,

hor-ul-

,

küt-ür

,

şıp-ır

. In the minority of cases that have the base CVCC before the proposed suffix, the consonant cluster always consists of a nasal plus a homorganic voiced stop:

20

ban-gır

,

bın-gıl

,

cumbul

,

güm-bür

. I propose that these are all different phonological variants of a single morpheme, -

IL

. When the base ends in /r/, the suffix is always realized as –

Il

, but otherwise is idiosyncratically –

Ir

or –

Il

. Likewise, when the base ends in a nasal, the

20

This characterization is a bit of a short-cut. While the resultant nasal-stop combination is indeed homorganic, a stem-final /n/ is paired not with a /d/ but with a /g/, to which it assimilates: /ŋg/.

81 suffix may idiosyncratically be preceded by the excrescence of a homorganic voiced stop. These phonological variations are doubtlessly the result of historical processes, and they do not submit to a synchronic analysis with predictive power, apart from the observation that bases ending in /r/ always take the –

Il

variant. This suffix can be notated as –(□)

IL

, where the symbol □ represents the position of the homorganic stop following a nasal. The suffix –(□)

IL

derives a category-neutral base from which verbs in –

DA

, nouns in –

tI

, and reduplicated adverbs can be derived.

The monosyllabic Roots of many of these formations occur independently of

–(□)

IL

in two morphological contexts: with the verbalizer –

lA

, and in reduplication

(Table 4).

Table 4.

Monosyllabic Roots of Onomatopoeic Stems

√-(□)

IL ban-gır bın-gıl cay-ır caz-ır cır-ıl cıv-ıl cıy-ır cız-ıl

,

cız-ır cum-bul

,

cumbur

çağ-ıl

Root

√B

AN

√B

IN

√C

AY

√C

AZ

√C

√C

√C

√C

√C

Â

IR

IV

IY

IZ

UM

√-

lA

-

ban-la-

‘cry out’

cır-la-

‘creak, chirp’

cız-la-

‘make a sharp, sizzling noise’

çak-ıl

çağ-la-

‘to burble, purl...(falling water)’

çan-gıl

çat-ır

Â

AK

Â

AN

Â

AT

çat-la-

‘crack’

√-R

çakçak

ED

cır-cır

(n.) ‘creaking sound’

‘sound of repeated blows’

har-ıl haş-ır hır-ıl hış-ıl hış-ır hor-ul hün-gür ış-ıl in-il kak-ır kık-ır kım-ıl kıp-ır kıt-ır küt-ür

çıt-ır dan-gır din-gil fık-ır fıs-ır fış-ıl fış-ır fok-ur fos-ur foş-ur gac-ır gıc-ır gur-ul güm-bür gür-ül lık-ır mır-ıl mış-ıl par-ıl

√G

ÜR

√H

AR

√H

√H

IR

√H

√H

√H

OR

√H

ÜN

√I

Ş

√İ

N

√K

AK

√K

IK

√K

IM

√K

IP

√K

IT

√K

ÜT

Â

IT

√D

AN

√D

İN

√F

IK

√F

IS

√F

√F

√F

OK

√F

OS

√F

√G

AC

√G

IC

√G

UR

√G

ÜM

√L

IK

√M

IR

√M

√P

AR

çıt-lafıs-la-

‘crackle’

‘whisper’

güm-le-

‘emit a booming sound’

gür-lehar-la-

‘flare up’

haş-la-

‘boil; scald’

hır-lahış-la-

=

‘thunder, roar’

‘growl’

hışıl-dadan dan

‘bang bang’

fıs fıs

‘whispering sound; spray’

güm güm

‘boomingly’

gür gür et-

‘make a gurgling noise’

hırhır

‘continuous snarling or growling’

hışhış

‘a repeated rustling sound’

“”

hış-la-

=

hışır-dahor-la-

‘snore’

in-le-

‘moan, groan, whine’

kak-la-

‘dry (fruit)’

küt-le-

‘thud, knock’

mır-la-

‘purr (a cat)’

par-la-

‘shine’

küt küt

‘repeatedly, violently (beating of the heart)’

lık lık

‘(drink) going glug glug’

mır mır

‘grumbling’

par par

‘(shine)

82

şar-ıl şık-ır şın-gır şıp-ır şır-ıl tak-ır tan-gır tap-ır tık-ır tın-gıl tın-gır tıp-ır tir-il tok-ur uğ-ul vıc-ır vır-ıl vız-ıl yelp-ir zan-gır zım-bır pat-ır pır-ıl pıt-ır pof-ur sak-ır şak-ır şan-gır şap-ır

√Ş

AR

√Ş

IK

√Ş

IN

√Ş

IP

√T

IN

√T

IP

√T

İR

√T

OK

√U

Ğ

√V

IC

√V

IR

√Ş

IR

√T

AK

√T

AN

√T

AP

√T

IK

√T

IN

√P

√P

AT

IR

√P

IT

√P

OF

√S

AK

√Ş

AK

√Ş

√Ş

AN

AP

√V

IZ

√Y

ELP

√Z

AN

√Z

IM

pat-la-

‘explode’

pır-la-

‘flutter

(young bird)’

pof-la-

‘puff, snort’

şak-la-

‘make a loud, cracking noise’

şap-la-

‘make a smacking noise

(with the lips or hand)’

şar-la-

=

şarıl-da-

brightly’

pat pat şak şak şap şap

‘thuds; footfalls’

‘(knock) repeatedly and noisily’

(imitates the sound of repeated kissing)

şar şar

=

şarıl şarıl şıpşıp

‘slipper without any back’

şır-la-

‘rain or flow in torrents’

tık tık

‘with a ticking sound’

tın-la-

‘tinkle, ring, clink’

“”

tir tir

‘(tremble) violently’

uğ-la-

=

uğul-davır vır

(imitates continuous and exasperating talk)

vız-la-

‘whizz’

83

84

zın-gır zır-ıl

√Z

√Z

IN

IR

zır-la-

‘weep

(contemptuous)

zır zır

‘(crying) incessantly’

Thus, there is ample evidence that the first (C)VC sequence is itself a Root, and that the remainder of the stem is an affix, and hence that the disyllabic base of the–

DA

verbs is a derived form. But this derived form also has no syntactic category. The inescapable conclusion is that there are derived forms that, like Roots, lack syntactic category. In these cases, the affix derives a new stem that is the input for further derivation, but does not introduce features that would make the stem a functional category such as verb or noun. Onomatopoeic stems in –□

IL

belong to this class.

2.6 Conclusion

In the foregoing, I have argued that Turkish words are derived from acategorial

Roots, and I have suggested as a preliminary observation that these Roots tend to be monosyllables. I have also made that case that disyllabic Roots, though less common, do occur. I make these arguments only with regard to native Turkish vocabulary. Turkish also has a large number of loan words, mostly from Arabic, Persian, and French. The observations of this chapter do not generalize to these, and so it appears that loan words are morphologically opaque. (Note, however, that some of the verbs cited in this chapter are based on loan words, since the suffixes –

lA

, –

lAn

, and –

lAş

can form verbs from the categorial phrases nP and adjP as well as from √P.)

Turkish also has a large number of complex predicates consisting of a non-verbal element and a light verb, notably

et

- ‘do’ and

ol

- ‘be/become.’ In the vast majority of cases, the non-verbal element is a loan word. The category, or perhaps the lack thereof, of

85 the non-verbal element is an important question, which might well take up an entire dissertation in itself. The present work focuses exclusively on causatives in verbs derived from a single stem, rather than a separate non-verbal element and light verb, and so the nature of the complement of the light verb in complex predicates is regrettably not addressed herein.

Having established what is meant by the Root, in the following chapter I investigate verbs formed by the merger of a Root with a [v,

CAUS

] head.

86

CHAPTER 3: ROOT CAUSATIVES

3.1 Introduction

In the preceding chapter, the category-neutral Root underlying Turkish verbs was identified. This Root heads a phrase that forms a verb by merging with a functional head of the category

v

, which comes in flavors such as

CAUS

,

BECOME

, and

DO

(Folli & Harley

2005, 2007; Folli, Harley & Karimi 2005). The aim of the present chapter is to explore

Root mergers with v

CAUS

, and their relation to non-causative verbs with common Roots.

Root causatives go by various names, depending to some extent on one’s theoretical orientation: lexical causatives (Miyagawa 1984), inner causatives (Svenonius 2005a, b), and low-attachment causatives (Harley 2008). Cross-linguistically, they often exhibit unpredictable allomorphy, as they do in Turkish, and they generally denote direct causation.

This chapter further illustrates the advantages of recognizing the acategorial Root in the derivational system. Previous works on Turkish causatives make little more than passing mention of the allomorphy of idiosyncratic –

Ir

, –

It

, and –

Ar

versus predictable

DIr

and –

t

(Aissen 1979, Dede 1986, Özkaragöz 1986, Kural 1996). This is not surprising, given that the frameworks available in the past did not have the tools to capture this allomorphy in a principled way. Since the Root is the locus of idiosyncrasy

(Marantz 1997, Arad 2003), Root derivation provides an explanation for these forms, as well as several other idiosyncratic phenomena that have either gone unmentioned or been briefly noted as ‘irregular.’ These include –

t

/–

n

alternating pairs such as

öğre-t-

‘teach’/

öğre-n-

‘learn’; the occasional failure of the causative –

t

to change a verb’s

87 valency or semantics, as in

kapa-

‘close (tr.)’/

kapa-t-

‘close (tr.),’ and

kopar-

‘break off

(tr.)’/

kopar-t-

‘break off (tr.)’; cases of semantic drift such as

duy-

‘hear, feel, smell’/

duyur-

‘announce, inform (NOT ‘cause to feel or smell,’ or ‘cause to hear’ in the usual sense); and a special class of causatives that are intransitive and have dramatic semantic drift, such as

sap-

‘turn, deviate (int.)’/

sap-ıt-

‘go off one’s head (int.),’

kır-

‘break

(tr.)’/

kır-ıt-

‘behave coquettishly (int.),’ not to mention

çıl-dır-

‘go crazy (int.),’ whose stem is not independently used as a verb:

*çıl-

. In the past, there has been no way to accommodate such oddities, other than to consign them to the Lexicon, that dark room in the basement where one locks up freaks and monsters. In DM, these all find a comfortable home above ground. While their occurrence is not predictable in any absolute sense—this is what makes them idiosyncratic—it is possible to state the domain where such phenomena can occur: at the merger of the Root phrase with the first functional head.

This chapter is organized as follows. In section 3.2, I discuss the types of realization of Root-merged causatives, and in 3.3 I identify the categories of verbs in which they occur. In 3.4, I identify another kind of exponence of Root causatives, and propose that this is due to the process of causative fission, a Root-conditioned

Morphological Form operation. Section 3.5 concludes.

3.2 Root Causative Allomorphy

Linear adjacency to the Root triggers allomorphic realization (Arad 2003,

Marantz 2010). This configuration is shown in (1), with the verb

bat-ır-

‘sink (tr.)’ as an

88

(1) example. (Naturally, the full sentence involves more structure to be built on top of this, which under the analysis adopted in this study include Voice as well as Tense.) a.

Çocuk gemi-yi bat-ır-dı

. child ship-

ACC

√-v

CAUS

-

PST

‘The child sank the ship.’ b.

DP

gemi

√P vP-

B

CAUS

AT v

CAUS

–Ir

Vocabulary Items (VIs) realizing a Root-adjacent terminal node comprise unpredictable allomorphs as well as Elsewhere forms. What this means for Turkish causatives is that the Root-adjacent feature bundle [v,

CAUS

] may be realized, depending on the Root, by the idiosyncratic [v,

CAUS

] allomorphs –

Ir

, –

It

, or –

Ø

(the null form); by the Elsewhere [v,

CAUS

] form –

DIr

; or by the idiosyncratic underspecified [v] allomorphs

A

or –

Ar

. I do not claim that these lists of allomorphs are exhaustive. The morphemes so far identified that compete to realize the Root-adjacent terminal node [v,

CAUS

] are summarized below. (Note that the following merely indicates the VIs that realize this node; it does not indicate that each VI is fully specified for the entire feature bundle.)

(2) -Ø

-Ir

-Ar

-It

-DIr

[v, C

AUS

] / {√A

Ç

,

√A

S

,

√B

,

Â

ÖZ

,

√E

Z

,

√K

IR

,

√S

IK

,...}

[v, C

AUS

] / {√A

RT

,

√B

AT

,

√B

İT

,

√D

,

√P

İŞ

,

√Ş

İŞ

,

√T

,

√Y

AT

,...}

[v, C

AUS

] / {√K

OP

,

√O

N

,

√K

URT

,...}

[v, C

AUS

] / {√A

K

,

√S

ARK

,

√K

ORK

,

Æ

RK

,

√D

AĞ }

[v, C

AUS

] / Elsewhere

21

21

The allomorph –

t

also belongs in this list. It is unusual in that its distribution can be described in entirely phonological terms: It is found with polysyllabic stems ending in a vowel or liquid. This is problematic in terms of Late Insertion, since the phonology of the stem would have to be present for –

t

to

89

If the Root adjacent to the [v,

CAUS

] node is on the list of one of the VIs, then that VI will be inserted. By the Elsewhere Principle, if the Root is not listed with any of the causative

VIs, then the Elsewhere form –

DIr

will be inserted.

That being the case, the most easily recognized Root causatives are those with idiosyncratic allomorphs. Hence these are a good starting point for identifying Root causatives. Verbs bearing these suffixes group into a small number of fairly well-defined semantic classes: change-of-state verbs, motion verbs, psychological predicates, ingestives, and evaluative causatives. Conspicuously absent from this list are unergatives and standard (i.e., non-ingestive) transitives. These do not take idiosyncratic allomorphs, but rather the predictable forms –

DIr

or –

t

(according to phonological environment).

Below,

koş-

‘run’ and

konuş-

‘speak’ are given as examples of unergatives:

22

(3)

koş-

‘run’

konuş-

‘speak’

koş-tur-

‘make run’

konuş-tur-

‘make speak’

The same is true of most transitives.

(4)

çiz-

‘scratch; draw’

öp-

‘kiss’

çiz-dir-

‘make scratch; make draw’

öp-tür-

‘cause to kiss’

Japanese causatives exhibit a similar allomorphy pattern, which Harley (2008) attributes to the fact that the causatives of unergatives and transitives attach to the category vP, and hence are separated from the Root. While this is undoubtedly so, it is be inserted. I do not at present see how this can be straightforwardly captured in this model, though see

Chapter 5, Section 5.3.2, for some discussion of the distribution of –

t

.

22

These verbs test as unergative by various Turkish-internal diagnostics:

koş-

allows impersonal passivization (Nakipoğlu-Demiralp 2001) and disallows double causativization (Özkaragöz 1986);

konuş-

yields the same results on these tests (among others) (Acartürk & Zeyrek 2010).

90 not at all clear that null morphemes block allomorphy. See Chapter 5, Section 5.3.3, where I address this issue, and propose that the head that causativizes unergative and transitive stems is

CAUS

rather than v

CAUS

. Thus, any VI having the feature [v]—which includes all of the idiosyncratic causative VIs—is ineligible for insertion at this node.

What is important for present purposes is that idiosyncratic allomorphy is indicative of a

Root causative.

In the competition and blocking model, we can also expect to find an Elsewhere form in Root causatives. Identifying Root causatives containing –

DIr

is less straightforward than identifying those with idiosyncratic forms, but we can use certain heuristics in this endeavor. Causative verbs that belong to the same semantic classes as the idiosyncratic forms are likely suspects for Root causatives—so long, that is, as no morpheme intervenes between the Root and the causative affix (Pylkkänen 2002, 2008).

Furthermore, if these semantically compatible forms in –

DIr

are formed with a monosyllabic base, they are especially likely to be Root causatives, since monosyllabic stems generally do not contain overt affixes.

3.3 Verb Classes

In the following sections, I identify Root causatives and describe the properties of the syntactico-semantic classes of verbs formed with them.

3.3.1 Change-of-State Verbs

Perhaps the largest semantic class displaying causative allomorphy is change of state (COS), including change of body position. Causative verbs of this class have intransitive counterparts that also indicate a change of state and are unaccusative. These

91 are called inchoative verbs. The pairs taken together form the causative/inchoative alternation (Haspelmath 1993). This alternation is a linguistic universal, and is discussed in detail in Chapter 4 of this study. The morphological relationship between the causative

(transitive) and inchoative (intransitive) members of COS verbs varies both across and within languages. If the causative member appears to be formed by the addition of a morpheme to the inchoative member, the pair is termed Causative. If the inchoative member appears to be derived from the causative, the pair is termed Anticausative. If both members appear to be derived separately by the addition of morphemes to a common stem, the pair is termed Equipollent. If the members appear to be derived from distinct, unrelated Roots, the pair is termed Suppletive. Finally, if both members are phonologically identical, the pair is termed Labile (as in most English pairs, e.g.,

melt/melt

,

shrink/shrink

). Turkish pairs with idiosyncratic causative morphology are usually Causative alternations, as the causative member has an overt suffix that the inchoative member lacks. However, given the premise that the stem to which idiosyncratic allomorphs attach is a Root rather than a verb, it follows that the causative suffix does not embed an inchoative verb, but in fact alternates with an inchoative morpheme, which is null in such cases. (For the precise characterization of the this

(5) morpheme, see Chapter 4, Section 4.4, as well as 3.4.2 below.)

–Ir causative

art-ır-

‘increase (tr.)’

bat-ır-

‘sink (tr.)’

bit-ir-

‘finish (tr.)’

doğ-ur-

–Ø inchoative

art-

‘increase (intr.)’

bat- bit-

‘sink (int.)’

‘finish (int.)’

doğ-

92

(6)

–Ar

–It

‘give birth to’

piş-ir-

‘cook (tr.)’

şiş-ir-

‘swell, inflate (tr.)’

taş-ır-

‘make overflow’

yat-ır-

‘lay (tr.)’

yit-ir-

‘lose’

kop-ar-

‘detach (tr.)’

on-ar-

‘repair (tr.)’

sark-ıt-

‘dangle (tr.)’

‘be born’

piş-

‘cook (int.)’

şiş-

‘swell, inflate (int.)’

taş-

‘overflow’

yat-

‘lie’

yit-

‘become lost’

kop-

‘detach (int.)’

on-

‘heal (int.)’

sark-

‘dangle (int.)’

To these can be added the null VI that realizes [v,

CAUS

]. This alternates with the overt inchoative morpheme –

Il

, resulting in an apparent Anticausative alternation. (See Chapter

4, Sections 4.3.2.1 and 4.4 for a detailed discuss of this and other inchoative morphemes.)

–Ø causative

aç-

‘open (tr.)’

as-

‘hang (tr.)’

boğ-

‘strangle; drown (tr.)’

–Il

çöz-

‘untie; solve’

ez-

‘crush’

kır-

‘break (tr.)’

sık-

‘squeeze; vex’

tık-

‘squeeze into (tr.)’ inchoative

aç-ıl-

‘open (int.)’

as-ıl-

‘hang (int.)’

boğ-ul-

‘choke; drown (int.)’

çöz-ül-

‘come undone’

ez-il-

‘get crushed’

kır-ıl-

‘break (int.)’

sık-ıl-

‘get squeezed; become bored or vexed’

tık-ıl-

‘squeeze into (int.)’

93

Unlike the other VIs, the null causative morpheme does not alternate with zero, but with the overt morpheme –

Il

. This appears to be a fundamental difference between null and overt causatives. It may be that this pattern is the fossilized distribution of an earlier stage of the language when –

Il

attached to a transitive verbal base. The alternation is examined in detail in Chapter 4. Despite this robust pattern, in principle an overt Root causative could alternate with an overt inchoative. This happens in precisely two cases.

(7)

–Ar

–It causative

kurt-ar-

‘rescue’

dağ-ıt-

‘scatter (tr.)’

–Il inchoative

kurt-ul-

‘get free’

dağ-ıl-

‘scatter (int.)’

This suggests that the analysis of the zero causative as simply one more causative VI is on the right track.

Now we turn to change-of-state causatives realized by the Elsewhere from. The following alternations have –

DIr

attaching to a monosyllabic base for the causative variant, and hence the transitive alternant is very likely a Root causative. Note the irregular stem alternation between

kal-dır-

and

kalk-

. This happens with no other verb, and is likely an allomorphic alternation of this particular Root.

(8)

–DIr causative

din-dir-

‘make subside’

dol-dur-

‘fill (tr.)’

don-dur-

‘freeze (tr.)’

kal-dır

-

‘lift/raise’

öl-dür-

‘kill’

sön-dür-

–Ø inchoative

din-

‘subside’

dol-

‘fill (in.)’

don-

‘freeze (int.)’

kalk

-

‘rise’

öl-

‘die’

sön-

94

‘extinguish’

sus-tur

-

‘silence’

ol-dur-

‘bring into being’

‘go out, become extinguished’

sus

-

‘become silent’

ol

-

‘become’

Change-of-position verbs can also be included here. These tend to indicate body position, although their theme need not always be a body. Furthermore, the non-causative stem does not always indicate a

change

in position, although the causativized stem necessarily does so.

(9)

–Ir

–It causative

yat-ır-

‘lay (tr)’

sark-ıt-

‘hang down (tr.)’

–Ø inchoative

yat

-

‘lie (int.)’

sark

-

‘hang down (int.)’

3.3.2 Motion Verbs

Motion verbs indicate a change in location. The following motion pairs have transitive alternants with irregular causative allomorphs.

(10)

–Ir

–Ar

–It transitive

düş-ür-

‘drop; knock down’

geç-ir-

‘pass (tr.)’

kaç-ır-

‘miss (fail to catch); kidnap’

uç- ur-

‘make fly; blow up (tr.)’

çık-ar-

‘take out’

gid-er-

‘get rid of’

ak-

‘flow’ intransitive

düş-

‘fall’

geç-

‘pass (int.)’

kaç-

‘flee, escape’

uç-

‘fly’

çık-

‘come out’

git-

‘go’

ak-ıt-

‘make flow’

95

There are also several cases where the functional equivalent of a causative of a verb of motion differs radically from the intransitive form. These may be cases of suppletion, though all except

sok-

/

gir-

might alternatively be cases of stem allomorphy.

(11)

–Ir?

–Ar?

Supp. transitive

gönd-er-

‘send’

sokgöt-ür-

‘take (to a place)’

get-ir-

‘bring’

‘insert; allow in’ intransitive

git-

‘go’

gel-

‘come’

git-

‘go’

gir-

‘enter’

The following verbs of motion have their transitive in –

DIr

.

(12)

–DIr transitive

dön-dür

-

‘turn (tr)’

dur-dur

-

‘stop (tr.)’

sap-tır

-

‘turn (tr.)’ intransitive

dön

-

‘turn (int.)’

dur

-

‘stop (int.)’

sap

-

‘turn (int.)’

3.3.3 Psychological Predicates

There are a handful of psychological predicates that display causative allomorphy.

(13)

–Ir

–It causative

doy-ur-

‘satiate’

kork-ut-

‘frighten’

ürk-üt-

‘startle’ non-causative

doy-

‘become sated’

kork-

‘fear’

ürk-

‘start (with fear)’

Likewise, there are monosyllabic psych verbs that have causative variants in –

DIr

.

(14)

–DIr causative

bez-dir-

‘exasperate’

bık-tır-

non-causative

bez-

‘tire of, weary of’

bık-

96

‘weary (tr.)’

kız-dır-

‘anger’

küs-türyıl-dır-

‘offend; hurt’

‘daunt, intimidate’

‘weary (int.)’

kız-

‘get angry’

küs-

‘become offended’

yıl-

‘become daunted’

The non-causative alternants of psych verbs have varying argument structures.

Some (

doy-

,

kız-

,

küs-

) take dative complements (15), while others (

kork-

,

ürk-

,

bez-

,

bık-

,

yıl-

) take ablative complements (16).

(15) Ben san-a kız-dı-m/küs-tü-m

1

SG

2

SG

-

DAT get.angry-

PAST

-1

SG

/get offended-

PAST

-1

SG

‘I’m angry at/offended by you.’ 23

(16) Ben sen-den kork-tu-m/bık-tı-m.

1

SG

2

SG

-

ABL fear-

PAST

-1

SG

/weary-

PAST

-1

SG

‘I’m afraid of/sick of you.’

In thematic terms, the subject of the non-causative alternant is the Experiencer, while the dative or ablative complement is the Stimulus. In the causative variant, the Stimulus is the subject, and the Experiencer is uniformly the direct object (accusative).

(17) Ben-i kız-dır-dı-n.

1

SG

-

ACC

√-

CAUS

-

PAST

-2

SG

‘You’ve angered me.’

(18) Ben-i

1

SG

-

ACC kork-ut-tu-n.

√-

‘You scared me.’

CAUS

-

PAST

-2

SG

Because of the uniform accusative case-marking of the Experiencer in these causatives, and because they match the argument structure of the other Root causatives

23

With some verbs, the simple past in Turkish is used to express present states, which is why the

Turkish past tense is rendered by the English present tense in these examples.

97 discussed thus far, I assume that causative psych predicates have the same syntactic structure as causative COS verbs.

(19)

DP

ben

√P vP-C

AUS

K

ORK vC

AUS

–It

3.3.4 Ingestive Verbs

Ingestives are verbs whose theme argument (the ingestum) is transferred to an animate being (the ingestive location). Ingesta have a very small semantic range, comprising 1. food and drink, 2. information, and 3. clothing and accessories. In the noncausative variant, the location argument is also an agent who executes the transfer of the theme. For example, in ‘John ate applesauce,’ John agentively transfers the applesauce to the ingestive location—namely, himself. In the causative variant, a third argument is presented as the agent of the transfer of location. In ‘John’s mother fed him applesauce,’ it is John’s mother who is responsible for transferring the applesauce to John, whether she put the spoon in his mouth (direction causation) or merely set a bowl of applesauce in front of him (less direct). Bhatt & Embick (2003) identify the following verbs as ingestoreflexive in Hindi:

98 causative variant may be transitive, and the causative distransitive, these show the same causative allomorphy as COS verbs. Verbs answering to these glosses in Hindi have a causative form with the suffix –

aa

, the same suffix found on the causative counterparts of unaccusatives. In Turkish we see the familiar allomorphy.

(21)

–Ir

–zIr

24 causative

em-zir-

‘suckle, 25

nurse (a baby)’ transitive

duy-ur-

‘announce, inform’

duy-

‘hear; feel’

iç-iriç-

‘make drink, give to drink’ ‘drink’

em-

‘suck’

Note in particular that, while the non-causative

duy-

has a variety of semantic functions

(‘hear information (ingestive), hear a sound (non-ingestive), feel (non-ingestive)), the causative

duy-ur-

is strictly ingestive (‘announce, inform’), denoting the transfer of information.

There is also one possible case of a suppletive or stem-allomorphic ingestive causative:

24

em-zir-

is the sole instance of this causative allomorph.

25

Is it possible that –

le

is a causative suffix in English? Note the same alternation in

startle

/

start

.

99

(22)

–Ar? causative

göst-er-

‘show’ transitive

gör-

‘see’

Again, ingestive causatives with –

DIr

are most likely Root-causatives as well.

(23)

–DIr causative

ye-dir-

‘feed’

giy-dir-

‘dress (someone, in something)’ transitive

ye-

‘eat’

giy-

‘put on (clothing)’

Cross-linguistically, the argument structure of ingestives is unstable. This is true of both the causative and non-causative variants. In English, the non-causative ingestives

eat

,

read

, and

learn

can be used transitively or intransitively (by dropping the object) with equal ease.

Drink

, although normally transitive, has an intransitive use where the consumption of alcohol is implied. Non-causative

dress

is obligatorily intransitive, while

wear

is transitive. The causative ingestives

feed

and

teach

may be mono- or ditransitive.

In Turkish, too, there is considerable variation.

Ye-

‘eat’ may be transitive or intransitive, and

iç-

‘drink’ is normally transitive, but intransitively implies alcoholic consumption, parallel to English.

Giy-

‘put on’ is obligatorily transitive.

The theme of the transitive variant is the direct object (marked accusative or unmarked according to Turkish differential object marking—see Enç 1991, von

Heusinger & Kornfilt 2005, Nakipoğlu 2009, and Key 2012a), and if it is retained in the causative counterpart, it remains so, while the causee—the ingestive location—is marked with dative case.

(24) a. Bebek mama ye-di. baby mush eat-

PAST

‘The baby ate baby food.’

100 b. Anne bebeğ-in-e mama ye-dir-di. mother baby-1

SG

-

DAT

mush √-

CAUS

-

PAST

‘The mother fed her baby baby food.’

When the theme is not retained in the causative variant, the causee is the direct object, normally accusative-marked. Hence the following ambiguity:

(25)

Kuş-lar-ı bird-

PL

-

ACC ye-dir-di-m.

√-

CAUS

-

PAST a. ‘I made the birds eat.’

-1

SG b. ‘I made (someone) eat the birds.’ (Kornfilt 1997: 335)

In the interpretation paraphrased in (25a), the ingestum is absent from the argument structure, while in (25b) the birds

are

the ingestum, and the causee (the ingestive location) has been omitted.

When the causee of an ingestive is the direct object, I assume the same structure as with the other causatives already discussed: the causee is the complement of the Root.

When the ingestum is the direct object, it is still the Root complement, but the dative causee is an adjunct hosted by the vP.

(33) Voice-P

Voice’

DP anne vP v’

Voice-

ACTIVE

–Ø v’

Dative-P bebeğine

√P

DP mama

Y

E v

CAUS

–DIr

The status of the dative causee is discussed in detail in Chapter 5, Section 5.3.1.

101

3.3.5 Evaluative Causatives

What I am calling evaluative causatives are a new proposal in this work. These are Root-causatives that indicate a judgment, generally negative, on the part of the speaker towards the behavior of the agent of which they are predicated. They are exceptional in that, though they are morphologically caustive, they are intransitive. The non-causative form may be transitive or intransitive, and there is usually a run-of-the-mill causative version with –

DIr

.

(27)

–It

–Ir

Evaluative causative standard causative

kır-ıt-

‘act coquettishly’

kır-dır-

‘cause to break’

sap-ıt-

‘go too far’

del-ir-

26

‘go crazy’

‘go off one’s head’

az-ıtsap-tır-

‘cause to turn’

az-dır-

‘cause to go wild’

del-dir-

‘cause to bore a hole in’ non-causative

kır-

‘break, bend (tr.)’

sap-

‘turn (int.)’

az-

‘go wild’

del-

‘bore a hole in (tr.)’

There are several verbs of this class with bound stems, another indication that the stems are Roots.

(28)

–It

–Ir

Neg-Eval causative

sır-ıt-

‘grin smugly or annoyingly’ stem

*sırkud-ur-

‘become rabid (with rage); become infected with rabies’

*

kud-

27

A couple of evaluative causatives are found with –

DIr

as well. One of these is a bound stem (

*çıl

).

26

Standard analyses would derive

delir-

from the adjective

deli

plus +(

A

)

r

. However, as I have argued,

Ar

selects only for Roots, not categories such as adjectives, and the vowel is never elided; cf.

sar-ı

‘yellow

(adj.)’/

sar-ar-

‘yellow (v.),’ not

*sarı-r

. Under my proposal,

delir-

and

deli

(as well as

del-

) are all derived from the Root √D

EL

.

27

But note

kuduz

‘rabies (n.)’.

102

(29)

–DIr

Neg-Eval causative

çıl-dır-

‘go crazy’

sik-tir-

‘fuck off (int.), get the fuck out (int.)’ stem

*çılsik-

(but:

‘fuck (tr.)’

çıl-gın

‘crazy’)

From the standpoint of their morphology, these clearly belong among the Root causatives, but their intransitivity and semantics distinguish them from the others. The non-causative verb may be transitive (

kır-

‘break,’

del-

‘bore,’

sik-

‘fuck’) or intransitive

(

sap-

,

az-

), but the causative verb is, exceptionally,

in

transitive. As for their semantics, we expect to find semantic drift in some Root causatives, but the evaluative causatives’ drift is especially pronounced, and they all drift to the same place: speaker evaluation of the agent’s behavior.

I propose that these causatives have the structure of Root reflexives. In some causative constructions, such as the English

get

-causative, when the theme is coreferential with the agent (30a), the former may be omitted, resulting in an intransitive structure (30b).

(30) a. b.

Don’t get yourself killed.

Don’t get killed.

There is evidence that the evaluative causatives are also missing a reflexive object. At least two of them have a non-causative variant with an inaliebly possessed object.

(31) Neg-Eval causative

kır-ıt-

Non-causative equivalent

gerdan kır-

‘behave coquettishly’ ‘behave coquettishly’ (lit. ‘bend neck’)

dire-tayak dire-

‘insist, be obstinate’ ‘insist; dig in one’s heels’ (

ayak

= ‘foot’)

103

(On

dire-t-

as a Root causative, see Section 3.4.1 below.) Words denoting body parts often serve as reflexive pronouns (see Koontz-Garboden 2009 on this phenomenon in

Ulwa (pp. 95-96); he further cites Faltz (1985) and Schladt (2000) as making the same observation). Not all of the evaluative causatives have corresponding idioms, though plausible body parts can be supplied: for example, the missing object of

del-ir-

‘go crazy’ might very well be the head (

del-

= ‘bore a hole in’). Hence, these look like transitives with omitted reflexive objects. However, note that if this is the right analysis, omission of the object is obligatory, such that inclusion of either the body part or a reflexive pronoun results in ungrammaticality.

(32) *

gerdan

/

*kendin-i

neck /self-

ACC

kır-ıt-

√-v.

CAUS

-

3.4 Another Type of Realization: –

t

/–

n

Alternating Verbs

In addition to the second-syllable causative allomorphs, there is another pattern found in Root-causatives: an alternation with –

t

on the causative and –

n

on the noncausative variant, as shown in (33).

(33) causative

ıs-la-t-

√-v-

CAUS

‘wet (tr.)’

- non-causative

ıs-la-n-

√-v-

INCH

-

28

‘get wet (int.)’ stem

ıs-la-

Causative/inchoative pairs of the type shown in (33) are Equipollent in Haspelmath’s terms (1993). The causative member does not, however, appear to be a Root causative, as an overt verbalizer intervenes between the Root and the causative morpheme.

28

Note that the inchoative morpheme –

n

is not glossed as the little-v flavor

BECOME

, but more as

INCH

, for ‘inchoative.’ In Chapter 4 I argue that this is the realization of a Voice projection and not a little-v flavor, and that the little-v preceding it is flavorless.

104

Nevertheless, the base of the alternation (Root plus verbalizer) is generally a bound stem, although some occur in regional dialects, or in standard Turkish with some difference in meaning from the suffixed versions.

In 3.4.1 below, I identify the verb classes that exhibit this alternation, and propose based on this that the causatives participating in the alternation are Root causatives. In

3.4.2, I explain a problem that these verbs present to the phenomenon of blocking as it is implemented in DM, and in 3.4.3, as a solution to this problem, I propose the operation of

CAUS

fission, whereby the flavor feature of little-v is separated from its original terminal node for independent exponence.

3.4.1 Verb Classes

Remarkably, the –

t

/–

n

alternation occurs in the same verbal classes as the more canonical allomorphy shown in the previous section. An asterisk (*) indicates that the stem is listed in Püsküllüoğlu 2007 either as a non-standard variant of the suffixed form, or with a different definition.

Change of State:

(34) causative

arı-t-

‘purify’

aydınla-t-

‘illuminate’

ısı-t-

‘heat (tr.)’

ısla-t-

‘wet (tr.)’

kirle-t-

‘dirty (tr.)’

29

There is a noun

ısı

‘heat.’ non-causative

arı-n-

‘become pure’

aydınla-n-

‘become illuminated’

ısı-n-

‘heat up (int.)’

ısla-n-

‘get wet (int.)’

kirle-n-

‘get dirty (int.)’ stem

*

arı-

*

aydınla-

*

ısı-

29

(*)

ısla-

*

kirle-

105

pisle-t-

‘make filthy’

tüke-t-

‘use up, exhaust’

yıpra-t-

‘wear out (tr.)’

Psychological:

pisle-n-

‘get filthy (int.)’

yıpra-n-

‘wear out (int.)’

tüke-n-

‘run out, become exhausted’

(*)

*

pisle- tüke-

(*)

yıpra-

(35) causative

alda-t-

‘deceive’ non-causative

alda-n-

‘become deceived’

avu-t-

‘console’

avu-n-

‘become consoled, take consolation’

inci-tinci-n-

‘hurt (emotionally)’ ‘get hurt (emotionally)’

Ingestive: stem

*

*

alda- avu-

(*)

inci-

(36) causative

öğre-t-

non-causative

öğre-n-

stem

*

öğre-

‘teach’ ‘learn’

kuşa-tkuşa-n-

*

kuşa-

‘gird someone; gird’ ‘gird on (tr.); gird oneself (int.)’

dona-t-

‘deck out’

dona-n-

‘deck oneself out’

(*)

dona-

Evaluative:

(37) causative

dire-t-

‘be obstinate; insist on having one’s own way’ (int.) non-causative stem

dire-n-

(*)

‘resist; show fortitude’

dire-

30

A few words should be said regarding the evaluative causative in (37):

diret-

is perhaps less pejorative than other evaluative causatives, but in comparison to the –

n

variant

diren-

it certainly connotes a less positive evaluation. The verb

diren-

is appropriate in the context of people resisting in the face of adversity or oppression, and

30

Non-occurring except in the idiom

ayak dire-

‘dig in (one’s) heel (lit. ‘foot’).’ See 3.3.5 above

106 suggests that they are justified in their struggle. This is the word used in hashtags related to the Gezi Park protests of 2013: #direngeziparki, #direnankara, #direndikmen, etc.

(‘resist’ followed by place names: Gezi Park, Ankara, Dikmen (a neighborhood in

Ankara)). The noun derived from this verb stem,

diren-iş

, is used in the protest movement’s ubiquitous slogan:

Her yer Gezi, her yer direniş!

(‘Everywhere is Gezi, everywhere is resistance!’) In contrast,

diret-

merely indicates that the subject insists on having his own way. Thus it is possible to say the following:

(38)

Halk diren-iyor

,

başbakan diret-iyor

. people resist-

CONT prime.minister be.obsinate-

CONT

‘The people are resisting, [and] the prime minister is digging in his heels.’

Like the other evaluative causatives, discussed in the last section, the verb

diret-

is intransitive despite having a causative morpheme.

3.4.2 The Problem with Blocking

The –

t

/–

n

alternation does not conform to the characterization of causative VI distribution presented thus far. The idiosyncratic VIs and –

DIr

appear on monosyllabic bases (always Roots in the former case), while –

t

and –

DIr

appear on polysyllabic bases according to their phonological properties: –

t

follows a vowel or a liquid, and –

DIr

appears elsewhere. By this description, a verb stem ending in /n/ should have its causative in –

DIr

, and it does in many cases.

(39)

–DIr causative

bulan-dır

‘make turbid’ non-causative

bulan-

‘become turbid’

t

causative

*bulat-

31

31

This does exist as the causative of

bula-

‘smear’:

bula-t-

‘cause to smear.’

107

The inchoative alternants of the –

t

/–

n

alternation are polysyllabic stems ending in /n/, and hence meet the description for causativization by –

DIr

. In most cases, however, this is

(40) ungrammatical.

n

inchoative

aydınlan-

DIr

causative

*aydınlandır-

‘become illuminated’

ıslan- *ıslandır-

‘get wet’

kirlen-

‘get dirty’

*kirlendir- tüken-

‘run out, become exhausted’

yıpran-

‘wear out’

*tükendir-

*yıprandır- aldan-

‘become deceived’

*aldandır- incin- *incindir-

‘get hurt (emotionally)’

öğren-

‘learn’

kuşan-

‘gird oneself’

*öğrendir-

*kuşandır- donan-

‘become decked out’

*donandır-

t causative aydınlat-

‘illuminate’

ıslat-

‘wet’

kirlet-

‘dirty’

tüket-

‘wear out’

aldat-

‘deceive’

incit-

‘hurt (emotionally)’

öğret-

‘teach’

kuşat-

‘gird’

donat-

‘deck out’

‘use up, exhuast’

yıprat-

The phenomenon in (41), where –

t

shows up instead of the expected –

DIr

, looks like a classic case of blocking: the Elsewhere form in –

DIr

is blocked by the idiosyncratic occurrence of –

t

. However, a blocking analysis of this phenomenon is highly problematic within the framework employed here. First of all, it assumes that the stem to which –

t

is suffixed is a Root. In that case –

t

could block the Elsewhere form in the context of Roots that it has on its list.

(41)

t

[v,

CAUS

] / [√A

RI

,

√A

YDINLA

,

√I

SI

,

√I

SLA

,

√K

İRLE

,

√P

İSLE

,...]

108

The problem with this is that several of these bases transparently comprise two morphemes. For example, the base of the alternation

kirle-t-

/

kirle-n-

‘get dirty (tr./int.)’ consists of

kir

and –

lA

. The first occurs without overt suffixation as the noun

kir

‘filth,’ while the second is a highly productive verbalizer (the simple and augmented forms of which are found in nearly 2,000 of the 4,700 verbs in N&Ü’s dataset).

Even if we could find a way around this, there remains the more serious problem that these verbs do not present a structural environment for blocking. In DM, blocking occurs strictly at terminal nodes (Embick & Marantz 2008). In the following configuration, VIs can compete to realize [v,

CAUS

], which is a single terminal node.

(42)

DP

√P vP-

CAUS

√K

İRLE v

CAUS

In the environment of √K

İRLE , the insertion of –

t

would fall out naturally. However, recall that the verb stem that idiosyncratically rejects causativization with –DIr is

kirlen-

:

*kirle-n-dir-

. If –

n

is an inchoative morpheme, then the node for insertion of a causative

VI is adjacent not to the Root, but to the inchoative.

32

32

The labels

INCH

(inchoative) and

INCH

P (inchoative phrase) are used here and in subsequent examples as a way of remaining neutral, for the time being, about the identity of phrase that is responsible for forming the inchoative verb.

109

(43)

vP vP

DP

√P

√K

İRLE

v v

CAUS

BECOME

The node at which the insertion of –

DIr

is blocked is not the Root-adjacent node at which

t

is inserted, but a higher node above

INCH

. In this configuration, it is simply not possible for –

t

to block –

DIr

.

Furthermore, the existence of a causative in –

t

does not always mean that the –

DIr

form is ungrammatical. There are cases where both forms are found, with a difference in meaning. the one where –

DIr

is competing, but also that it does not block –

DIr

when the two forms have distinct meanings. If this is blocking, it appears to be a strictly semantic phenomenon, the view taken in Aronoff (1976).

The answer to this problem is that it is not blocking. Given the structure in (43), there is no reason why the appropriate causative Elsewhere form –

DIr

should not be

110 inserted into the v

CAUS

node. As already noted, it is in fact inserted in the cases of

bulan-

‘become turbid,’

arın-

when it means ‘become pure’ in the spiritual sense, and

ısın-

when it means ‘warm up to’ in the emotional sense. This indicates that (43) is the correct structure for these verbs. The tree below is populated with the morphemes in

bulan-dır-

‘make turbid,’ but the same structure holds for all of these verbs.

(45)

DP

√P

vP

√B

U vP v

CAUS

-DIr

v

BECOME

-lAn

On the other hand, the n-final inchoative verbs whose causative counterpart ends in –

t

must have a different structure. Furthermore, the failure of –

DIr

to causativize these verbs cannot be a matter of blocking, but must instead fall out naturally from the structure itself. I propose that in these verbs, the suffix –

n

realizes a projection higher than little-v: namely, Voice (middle). The verb

ıslan-

‘become wet’ is represented in the tree below as an example.

(46)

DP

√P

VoiceP

vP

√I

S

v

–lA

Voice

MID

–n

Argumentation justifying this structure is presented in Chapter 4, Sections 3 and 4. The failure of –

DIr

to be suffixed to a verb stem built of the terminal nodes of this structure is

111 due not to blocking, but to the fact that the Turkish causative cannot embed Voice (see

Chapter 5). No VI is blocked; rather, no causative head can merge with VoiceP.

This means the surface strings

arın-

‘become pure’ and

ısın-

‘warm up’ are ambiguous between the structures in (45) and (46). They have the former structure when the sense is physical, and the latter when the sense is spiritual/emotional. This also raises the question of the structure of the causative counterparts in –

t

. That is the subject of the next section.

3.4.3

CAUS

Fission

In this section, I argue that the feature bundle [v,

CAUS

] may undergo fission postsyntactically, creating two positions of exponence, one with [v] and the second with

[

CAUS

]. A VI with only the feature [v] (see. Chapter 2, Section 2.4.3.2) is inserted in the first position, and a VI with only the feature [

CAUS

] in the second position. This analysis forces the conclusion that the VI –

t

has the feature set [

CAUS

] and not [v,

CAUS

].

DM adopts Logical Form (LF) and Phonological Form (PF), the post-syntactic levels of representation employed in the Minimalist Program (which in turn inherited these from the Y-model of Government & Binding Theory (Haegeman 1991)). LF is the level where remaining movement operations (such as quantifier-raising) that are required to create the necessary scope relations occur. PF is the level where phonological rules apply. In DM, this model is augmented with an intermediate stage, Morphological Form

(MF), where adjustments are made to the feature bundles and VIs are inserted (Halle &

Marantz 1993). This level is necessary, as the operations that apply here must be postsyntactic but pre-Vocabulary Insertion.

112

(47)

Morphological Form

Phonological Form Logical Form

Fission is an adjustment that occurs at MF prior to Vocabulary Insertion (Noyer 1997).

The features composing a bundle at a terminal syntactic node are separated, creating new

(post-syntactic) positions of exponence. Thus, if a feature bundle [x, y] undergoes fission, there will be two positions of exponence, one with [x] and one with [y], and hence two

VIs will be inserted, each discharging one of the features. Herein I propose that the [v,

CAUS

] node of Root causatives (as in 48a) may undergo fission (as conditioned by certain

Roots idiosyncratically) (48b). vP (48) a. b.

DP

DP

√P

√P

√ vP

[v,

CAUS

[v ] v

]

[C

AUS

]

In the configuration in (48b), no VI with the features [v,

CAUS

] can realize the head of the vP, as this would violate the Subset Principle: It can not be inserted in the [v] node, since it has [

CAUS

], a feature that does not belong to that node, nor can it be inserted in the

[

CAUS

] node, since it has [v]. Instead, a VI with the feature [v] but no flavor feature must

113 be inserted in the [v] node, and similary, a VI with [

CAUS

] only must be inserted into the

[

CAUS

] node.

This structure not only accounts for the causative side of the –

t

/–

n

alternation, it is also consistent with other occurences of the VI –

t

that are otherwise mysterious. These are the subject of the following section.

3.4.4

CAUS

fission beyond the –

t

/–

n

alternation

Given the standard assumption that –

t

is a phonologically conditioned variant of the productive causative, it shows up in some surprising places, where it does not causativize the stem in any obvious way. The first of these is in the supposed causative suffix –

Art

(Dede 1986, Göksel & Kerslake 2005). This is used with the stems

çök-

and

göç-

.

(49) causative

çök-ert-

‘collapse (tr.)’

göç-ert-

‘collapse (tr.)’ non-causative

çök-

‘collapse (int.); squat’

göç-

‘collapse (int.); migrate’

Transparently, this ‘suffix’ is composed of –

Ar

and –

t

. Of course, there is no

a priori

necessity that these be separate morphemes in a synchronic analysis; sequences of morphemes can be reanalyzed as single morphemes, much as the suffix –

DIr

was a composite of two causatives in Old Turkic (Erdal 1991). In the case of –

Art

, however, there is synchronic evidence that this string comprises two suffixes. There are two verbs that have causatives in –

Ar

, optionally followed by –

t

with little or no difference in meaning. In other words, they too may causativize with –

Art

(more accurately –

Ar-t

) as well as with –

Ar

alone.

114

(50) causative

çık-ar-

çık-ar-t-

‘take out, remove’

kop-arkop-ar-t-

‘break off, detach (tr.)’ non-causative

çık-

‘come out’

kop-

‘break off, detach (int.)’

Recall that –

Ar

has already been identified as an unspecified verbalizer, having the feature bundle [v]. If a terminal node has the bundle [v,

CAUS

], as in (48a), –

Ar

may be inserted into the node thanks to the Subset Principle (Halle 1997) provided that the

Root is specified for this VI. On the other hand, if this node has undergone fission, as in

(48b), the same VI is also suited to realize the [v] node, while a separate VI (–

t

) realizes

[

CAUS

].

I would like to argue that the extraneous –

t

in the above causative stems and the –

t

in the –

t

/–

n

alternating pairs are one and the same. In support of this claim is the verb stem

kapa-

‘close (tr.),’ which exhibits both types of pattern. As shown in Chapter 2, this verb comprises the Root √K

AP

(also seen in

kap-ı

‘door’) and the underspecified verbalizer –

A

. It optionally occurs with

t

-suffixation, and has its intransitive counterpart in –

n

.

(51) simple transitive

kapa-

‘close (tr.)’

t

-suffixed transitive intransitive

kapa-t-

‘close (tr.)’

kapa-n-

‘close (int.)’

When the simple and

t-

suffixed transitives are considered alone, then

kapa-

/

kapa-t-

looks identical to

kopar-

/

kopar-t-

, with the vacuous addition of –

t

. On the other hand, when the

t

-suffixed transitive and the intransitive are considered alone, then

kapa-t-

/

kapa-n-

looks just like

ısla-t-

/

ısla-n-

. In fact,

ısla-

is the transitive form found in some regional

115 dialects.

33

As it clearly exhibits both patterns,

kapa-

is the ‘missing link’ between them in modern standard Turkish.

We thus have causatives in –

Ar

where an additional –

t

has no morphosyntactic effect on the verb; the verb

kapa-

for which the same is true; and causatives in –

t

than alternate with –

n

on the non-causative counterpart. If all of these are instances of the same structure, then these stems must all comprise a Root VI followed by an underspecified verbalizer followed by –

t

.

First, it needs to be demonstrated that the bases to which the verbalizers attach are indeed all Roots. This is clearest in a case like

ıs-la-t-

‘wet (tr.),’ as the base is monosyllabic and does not occur independently: *

ıs

. This analysis is also fairly straightforward in a case like

kop-ar-t-

‘break off (tr.),’ where a monosyllabic base combines with a Root-conditioned verbalizer, and it is quite plausible that forms such as

tük-e-t-

‘exhaust’ and

öğr-e-t-

‘teach’ comprise Roots and the Root-selecting verbalizer –

A

. But the form

aydın-la-t-

‘illuminate’ appears to present a challenge to the Rootconditioning analysis. The base,

aydın

, independently occurs as an adjective, and combines with –

lA

, the Elsewhere verbalizer, which can form verbs out of adjectives.

These facts seem to suggest that

aydın

is a word rather than a Root. However, as I explain below, the semantics of this base in different environments are suggestive of Root behavior. Furthermore, historical dictionaries indicate that these semantics are a relatively recent innovation, and that this innovation correlates with the development of the –

t

/–

n

33

The form

ısla-

appears in Redhouse (1997) with the note

prov.

(‘provincial’). Several native speakers I have consulted confirm that this form occurs in some dialects. One informs me that it is used by many speakers in his native city of Samsun (on the Black Sea coast).

116 alternation. This evidence suggests that this stem has recently undergone

radical reanalysis

.

Although modern dictionaries include ‘bright’ among the definitions of

aydın

, this is at best an antiquated use. Outside of stock phrases such as

günaydın

‘good morning’

(literaly: ‘day is bright’), the modern meaning of

aydın

is ‘intellectual,’ in both the nominal and adjectival senses. The word is not appropriate to characterize the result state of the verb

aydınlat-

‘illuminate.’ Rather, this requires a suffixed form of the Root,

aydınlık

. The suffix –

lIk

generally derives abstract nouns, but

aydınlık

is an adjective meaning ‘bright’ as well as a noun meaning ‘brightness.’

(52) a. b.

#

Burası

this.place

Burası

çok

very

çok aydın!

aydın

aydın-lık

! this.place very aydın-lIk

‘This place is very bright!’

This semantic oddity appears to be relatively recent. Two late-nineteenth-century dictionaries I consulted, the Redhouse

Lexicon

(1890) and the

Kamus-ı Türkî

(1899), define

aydın

only as ‘bright,’ in the physical sense. Furthermore, both lack the form

aydınlat-

. The Lexicon lists

aydın-lan-dır-

as the causative of

aydın-lan-

, while the

Kamus-ı Türkî

lists the form

aydın-la-

(without –

t

) as an intransitive verb. Neither dictionary lists

aydın-la-t-

. Furthermore, the forms

aydın-lan-dır-

and

aydın-la-

are nonexistent in modern Turkish. This constitutes evidence that the emergence of the

CAUS

fissioned form

aydın-la-t-

roughly corresponds to the emergence of the Root-like semantic behavior of the base.

117

Next, it needs to be shown that all of the relevant stems contain a suffix realizing an unflavored little-v. For several stems, this falls out automatically from the analysis already presented, as they contain the suffixes –

lA

and –

A

(Chapter 2.4.3.2; N&Ü App.

2a: 25 and 1, respectively).

(53)

–lA

–A alternating causative

aydın-la-t-

‘illuminate’

ıs-la-t- kir-le-t-

‘wet’

‘dirty’

pis-le-ttük-e-tkuş-a-tdon-a-tdir-e-t-

‘make filthy’

‘consume, exhaust’

‘gird’

‘deck out’

‘be obstinate’

The stems

yıpra-

and

öğre-

could also be analyzed as containing –

A

, in which case their respective Roots would be √Y

IPR

and √Ö ĞR . Alternatively, they might contain the unproductive verbalizer –

rA

(N&Ü App. 2a: 29), with the Roots √Y

IP and√Ö

Ğ . Since the

Roots are not attested independently of the verbalizer, there is no obvious way to decide between the two analyses, but either will work.

In the other cases, it is necessary to include –

I

and –

DA

in the inventory of underspecified verbalizers.

(54)

–I

–DA alternating causative

ar-ı-t-

‘purify’

ıs-ı-tav-u-t-

‘heat’

‘console’

inc-i-t-

al-da-t-

‘hurt, offend’

‘deceive’

N&Ü classify several verbs under a derivational suffix they notate as –

I

/

U

according to rounding environment (App. 2a: 10). They list both transitive and intransitive verbs under this suffix:

118

(55) a. transitive verbs in –

I/U kaş-ı-

‘scratch (tr.)’

kaz-ı-

‘erase, scrape (tr.)’

tan-ı-

‘recognize (tr.)’

taş-ı-

‘carry (tr.)’

bür-ü-

‘cover up, enfold (tr)’

dok-u-

‘weave (tr.)’

kor-u-

‘protect (tr.)’

sür-ü-

‘drag along the ground (tr.)’ b. intransitive verbs in –

I/U er-iış-ı-

‘melt (int.)’

‘shine, sparkle (int.)’

büy-ü-

‘grow (int.)’

çür-ü-

‘rot (int.)’

kur-u-

‘dry (int.)’

soğ-u-

‘become cold’

sol-u-

‘pant (breathing)’

ul-u-

‘howl’

uy-u-

‘sleep’

ür-ü-

‘howl’

üş-ü-

‘feel cold’

yür-ü-

‘walk (int.)’

Not only is modern –

I

found with both transitives and intransitives, but the intransitives include unaccusatives as well as unergatives. This is fairly good evidence that –

I

is an underspecified verbalizer.

As for –

DA

(N&Ü App. 2a: 8), although it is quite productive with onomatopoeic stems formed with the suffix –(□)

IL

(Chapter 2.5), there are precious few verbs that could be analyzed as Root-DA-:

iste-

‘want,’

bağda-

‘tangle,’ and

sapta-

‘determine, establish.’

The third of these is a product of the language reform (Lewis 1999), and so it cannot be

119 considered as evidence until the status of reform words is determined.

34

It is also possible that

alda-

comprises a Root √A

LD

and the suffix –

A

.

An alternative analysis is that both –

Ar

and –

Art

involve independent exponence, but that in the first instance [C

AUS

] has a null exponent.

(56) a. b.

kop-ar-Ø-

√-v-

CAUS

-

kop-ar-t-

√-v-

CAUS

-

In this configuration, the alternation is separated from the Root by an overt morpheme, and therefore must be independent of it. However, the independent exponence of [

CAUS

] is sensitive to the Root. For example, it is obligatory with the Root Â

ÖK

(57a), optional with √K

OP

(57b), and disallowed with √K

URT

(57c).

(57) c. a. b.

çök-er-t-

√-v-

CAUS

-

‘collapse (tr.)’

kop-ar-t-

√-v-

CAUS

-

‘break off (tr.)’

*kurt-ar-t-

√-v-

CAUS

-

* çök-er-

√-v

CAUS

-

kop-ar-

√-v.

CAUS

-

‘break off (tr.)’

kurt-ar-

√-v.

CAUS

-

‘rescue’

Thus, the analysis represented in (56a) is not tenable. For the same reason,

CAUS

cannot be a phrasal projection distinct from vP, as was originally proposed in Key 2012b.

34

The Turkish Language Reform of the 20 th

century was a state-sponsored effort to replace loan words

(mainly from Arabic and Persian) with words of Turkish origin. Many of the Turkish replacements were invented by members of the Turkish Language Committee, among others (such as

sapta-

‘determine,’ replacing Arabic

tespit et-

). Many, but by no means all, of the proposed replacements have been adopted by

Turkish speakers. How closely these coinages (‘neologisms’) adhere to the morphological system of words occurring naturally in the language is a fascinating question that merits detailed study. As far as the present study is concerned, the status of neologisms is indeterminate. For a detailed history of the Turkish

Language Reform, see Lewis (1999).

120

In the cases where fission is optional, such as

kap-a-

/

kap-a-t-

‘close,’

kop-ar-

/

kop-ar-t-

‘detach,’ and

çık-ar-

/

çık-ar-t-

‘remove,’ the two forms feed nominalization independently.

35

In the following, the non-fissioned and the fissioned causative forms are combined with the productive deverbal nominalizer –

mA

. It should be pointed out that all of these forms, with and without –

t

, also have straightforward meanings deriving directly from the verb (i.e., ‘(the act of) closing,’ etc.). It is in their special meanings that they differ.

(58) Root causative

kapa-

/

kapat-

‘close’

çıkar-

/

çıkart-

‘take out’

kopar-

/

kopart-

‘detach’

Non-fissioned nominal

kapama

‘lamb stew’

çıkarma

‘military debarkation’

koparma

‘clean and jerk

(weightlifting)’

Fissioned nominal

kapatma

‘kept woman’

çıkartma

‘decal, sticker’

kopartma

(no special meaning)

Although the causatives are synonymous, the nominalizations show a very sharp distinction between the fissioned and non-fissioned forms. It will be noted that the nominalizer is not adjacent to the Root. The question arises as to how these special meanings are created. Marantz (2010) suggests that the locality domains for allomorphy and allosemy are different. Phonological adjacency is the relevant relationship for allomorphy, such that it is not blocked by intervening heads if they are null. This is seen in English past tense forms such as

went

and

took

, as well as in the aorist allomorph –

Ar

discussed in 2.4.3.2. For allosemy, Marantz suggests that semantic adjacency is the relevant relationship. In other words, intervening heads that are phonologically overt but semantically null do not prevent allosemy. The precise characterization of ‘semantically

35

Thanks to Nurcan Abacı for pointing this out.

121 null’ and how it applies to [v,

CAUS

] is unclear (at least to me). Still, the parallels should be obvious between the Turkish nominalizations and the Japanese nominalizations that

Marantz cites.

(59) Japanese nominalizations (Volpe 1995; cited in Marantz 2010) verb nominalization a. b. c. chir-asu

‘scatter’ d-asu

‘expel’ nag-asu

‘wash away’ chir-as-i

‘a leaflet’ d-as-i

‘soup stock’ nag-as-i

‘a sink’

3.5 Conclusion

In this chapter, I have shown that a number of idiosyncratic phenomena related to

Turkish causatives occur in a common domain: the merger of the √P with v

CAUS

. These include standard causative allomorphy, the –

t

/–

n

alternation, and intransitive causatives with idiom-like meanings (evaluative causatives). I have proposed that the –

t

/–

n

alternation is the result of

CAUS fission, a Morphological Form operation that applies to a v

CAUS

node adjacent to certain Roots. The –

n

on the non-causative alternant is a realization of a (middle) Voice head; this, and not blocking, is the reason for the failure of some verbs in –

n

to be embedded under a causative. (The middle Voice analysis of inchoatives is explored further in Sections 4.3 and 4.4 of the next chapter.) Root causatives also participate in transitivity alternations with unaccusative (inchoative verbs). These alternations are the subject of Chapter 4.

122

CHAPTER 4: THE CAUSATIVE/INCHOATIVE ALTERNATION

4.1 Introduction

As mentioned in the previous chapter, Root causatives participate in what is known as the causative/inchoative alternation (Haspelmath 1993). In the present chapter,

I take a close look at the alternation itself, other verbal structures participating in it, and the theoretical controversy surrounding the alternation. The greatest controversy is over the directionality of the derivation, namely, whether the transitive member is derived from the intransitive member (causativization); the intransitive is derived from the transitive (anticausativization); or each is derived independently (equipollence). This issue has significant implications for a syntactic theory of word formation, since an anticausative derivation would ostensibly involve the deletion of a causer argument and the

CAUS

operator. In syntactic formations, structure can only be added, not removed.

This chapter is organized as follows. In section 4.2, the alternation is defined, and the cross-linguistic morphological patterns are presented. In 4.2.1, 4.2.2, and 4.2.3, the three competing hypotheses of universal causativization, anticausativization, and equipollence are discussed. In 4.3, the problematic pattern of anticausativization is treated in detail. Koontz-Garboden’s (2009) proposal of anticausativization as reflexivization, a structure-preserving operation, is presented in 4.3.1. Morpheme distribution in the

Turkish anticausative is examined in 4.3.2, and it is concluded that it does not support the reflexivization analysis, and furthermore that there is in fact no causative embedded under the anticausative, and hence no deletion of structure. In 4.4, the Turkish alternation as a whole is evaluated. 4.5 concludes.

123

4.2 The Causative/Inchoative Alternation

The alternation consists of pairs of verbs, one transitive and one intransitive, in which the object of the transitive member (causative) has the same thematic role (Theme) as the subject of the intransitive member (inchoative). The subject of the transitive verb has the role of causer.

(1) a. b.

(inchoative)

(causative)

The snowman melted.

The sun melted the snowman.

(Haspelmath 1993: 90)

While (1a) and (1b) differ in their argument structure, the argument ‘the snowman’ has the same relation to the event in both sentences. The alternation has several other properties. Neither the causer nor the theme has an animacy restriction; either or both may be inanimate (as in 1) or animate. Next, the transitive member has the semantics of

direct causation

: the causer was the immediate cause of the event, with no intermediate links in the causal chain.

Another important fact about the alternation is that it is universal. All languages have a system of such transitive/intransitive verb pairs with the properties mentioned above. There is, however, considerable cross-linguistic and intra-linguistic variation in the morphological marking of the transitive and intransitive members. These involve the issue of directionality. Haspelmath (1993) identifies and names three basic types of alternation: causative, where the transitive member is morphologically more complex than the intransitive; anticausative, where the intransitive member is more complex; and non-directed, where the two members do not differ in morphological complexity.

Examples of the causative and anticausative, or the ‘directed’ alternations, are given below.

124

(2) Causative alteration:

Persian

suz-ân-d-an sux-t-an

burn-

CAUS

-

PAST

-

INF

burn-

PAST

-

INF

‘burn (tr.)’ ‘burn (int.)’

(3) Anticausative alternation

Italian

romp-ere

break-

INF

‘break (tr.)’

romp-er-si

break-

INF

-

ANTI

‘break (int.)’

In (2), the transitive Persian verb meaning ‘to burn’ has a morpheme that the intransitive verb lacks. In (3), the opposite relationship holds: The intransitive Italian verb meaning

‘to break’ has a morpheme not found in the transitive verb. These alternations are called

‘directed’ because the morphological patterns appear to indicate a direction of derivation:

In (2), the intransitive is the ‘basic’ form, while the transitive is derived, and in (3) the transitive verb is ‘basic.’

In contrast, non-directed alternations do not suggest a direction of derivation in any obvious way. Haspelmath divides these into three subtypes: equipollent, suppletive, and labile. In an equipollent alternation, the members share a common base, but each contains a derivational morpheme not present in the other member. Thus, both forms are complex, and neither appears to be basic.

(4) Equipollent alternation a. b.

Japanese:

Persian:

nukum-e-ru

‘warm (tr.)’

âb kar-d-an

water do-

PST

-

INF

‘melt (tr.)’

nukum-or-u

‘warm (int.)’

âb šo-d-an

water become-

PST

-

INF

‘melt (int.)’

In (4a), both Japanese verbs share the base

nukum-

, but the transitive and intransitive have the morphemes –

e

and –

or

, respectively. Similarly, the Persian verbs in (4b) have a

125 common base,

âb

‘water,’ but each has a different derivational morpheme (a light verb in this case).

In the suppletive alternation, the members do not share a common base. From a phonological standpoint, they are entirely different words.

(5) Russian

žeč’

‘burn (tr.)’

goret’

‘burn (int.)’ (Haspelmath 1993: 92)

The third type of non-directional alternation is

labile

. In the labile alternation, the two members are phonologically identical.

(6) Labile alternation a. English b. Persian

break

šekas-tan

break-

(tr.)

INF

‘break (tr.)’

break

break-

(int.)

šekas-tan

INF

‘break (int.)’

Languages differ dramatically in the types of alternations used and the preponderance of various types. English has an overwhelming majority of labile pairs, though there is a handful of cases where there the transitive has a different vowel from the intransitive, indicating a morphological change:

raise

(tr.)/

rise

(int.),

set

(tr.)/

sit

(int.),

fell

(tr.)/

fall

(int.). English also has the suppletive pairs

kill

/

die

,

teach

/

learn

, and

feed

/

eat

.

In his study of 31 causative/inchoative verbs across 21 languages, Haspelmath (1993) found that, for example, Russian has the highest incidence of the anticausative (23 pairs),

Turkish of the causative (17.5 pairs), Japanese of the equipollent (20.5 pairs), and English of the labile (25 pairs). None of the languages has an especially high proportion of suppletive alternations. Despite this wide variation, he found that certain verb pairs have a marked tendency to occur in a particular alternation type across languages. Thus,

‘break,’ ‘open,’ ‘spread,’ ‘lose,’ ‘raise,’ ‘rock,’ ‘connect,’ ‘gather,’ and ‘split’ exhibit the

126 anticausative alternation in more than half of the languages in his sample, while ‘boil’ and ‘freeze’ exhibit the causative alternation in more than half. None of the verbs has any cross-linguistic tendency to appear in a non-directional alternation.

These facts raise several related questions. First, can the alternation types be taken at face value? If one member of a pair shares a base with but is more complex than the other member, does this mean the complex verb is derived from the other? Are there truly three different directionalities in the causative/inchoative alternation, or is there a true, uniform directionality? While there are vastly differing views among theorists, virtually none takes the position that the superficial pattern is the final arbiter of the derivation.

4.2.1 Causativization

Pesetsky (1995) argues that, in English at least, the direction underlying the alternation is causativization. He posits a zero morpheme

CAUS

, for which he makes extensive arguments (most of these go beyond the causative/inchoative alternation itself).

He relates this zero morpheme to Chomsky’s (1972) famous observation that derived nominalizations of alternating verbs have only the argument structure of the inchoative alternant (7c) and not that of the causative (7d).

(7) a. b. c. d.

Tomatoes grow.

Bill grows tomatoes. the growth of tomatoes

*Bill’s growth of tomatoes (Chomsky 1972: 25)

According to Pesetsky, the basic form of an alternating verb like

grow

is intransitive. The transitive contains the null

CAUS

morpheme.

(8) [[

grow

]

INT

-Ø]

CAUS

The nominalizing morpheme (-

th

in this case) attaches to the intransitive base.

127

(9) [[

grow

]

INT

-

th

]

NOM

Since this derivation does not include

CAUS

, the causative interpretation of the nominal is not available. A derivation in which the nominalizer embeds

CAUS

, as in (10), is impossible.

(10) *[[[

grow

]

INT

-Ø]

CAUS

-

th

]

NOM

The impossibility of this structure is an instance of Myers’ Generalization (Myers 1984):

(11)

Myers’ Generalization

Zero-derived words do not permit the affixation of further derivational morphemes.

Pesetsky attempts to analyze Myers’ Generalization as belonging to a more general class of phenomena observed by Fabb (1988) regarding restrictions on affixation: some suffixes never attach to a suffixed word (call these Type I), others attach outside only one

(or more) particular suffix(es) (call these Type II), and still others can attach to any word of the appropriate category (noun, verb, etc.) regardless of prior suffixing (call these Type

III). In Pesetsky’s proposal, suffixes of the second type require certain properties in the suffixes they attach to, and these properties presuppose phonological realization, which disqualifies null morphemes from satisfying this requirement. The upshot is that Type I and Type II suffixes can never attach outside a null morpheme. This leaves Type III suffixes, which attach to a word of a given category without consideration of any affixes already present. Type III membership appears to be quite exclusive. As a case in point, there are only two deverbal suffixes that answer to this description in English, –

er

and

able

. These two suffixes, but apparently no others, can be added to null-derived verbs, e.g.,

grower

,

growable

.

128

For Pesetsky then, the unavailability of the causative reading in derived nominalizations of alternating verbs is proof that the transitive alternant contains a null

CAUS

morpheme. He also assumes that the Root is of the category Verb, and that it has an inherently inchoative meaning. This explains the availability of the inchoative meaning in the Root nominalization.

4.2.2 Anticausativization

Levin & Rappaport Hovav (1995) (henceforth L&R) take the position that the universal directionality is anticausativization. They point out that, despite initial appearances, the possible objects of the transitive alternant do not coincide with the possible subjects of the intransitive. Importantly, the latter is a proper subset of the former.

(12) a. b. b’.

He broke the vase/the window/the toaster/his promise/the contract/the world record.

The vase/The window/The toaster broke.

*His promise/*The contract/*The world record broke.

(adapted from L&R)

(13) a. a’. b. b’.

The wind cleared the sky.

The sky cleared.

The waiter cleared the table.

*The table cleared. (L&R)

Since the single argument of the intransitive is only a subset of the possible objects of the transitive verb, the transitive cannot be derived from the intransitive.

Another piece of evidence cited by L&R for universal anticausativization is the fact that, cross-linguistically, unaccusatives with no causative counterpart are

‘unstable’—in other words, they are prone to turn up with transitive argument structure, either as a nonce usage in a language variety that normally lacks the transitive variant, or

129 as a systematically available form in a different dialect. An example of the former type given is the transitive use of

deteriorate

: ‘The pine needles were deteriorating the roof.’

An example of the latter type is Italian

crescere

‘grow,’ strictly intransitive in standard

Italian but having transitive uses in some dialects. They point out that this contrasts with the stable intranstivity of unergatives such as ‘cry’ and ‘sweat.’

L&R also adduce cross-linguistic directionality patterns found in the causative counterparts of unaccusative verbs as opposed to those of unergative verbs. The directionality of the verb ‘break’ is variable, but there is a tendency for it to display the anticausative pattern. Of the 21 languages in Haspelmath (1993), ‘break’ is anticausative in 12.5 cases, non-directional in 7.5 cases, and causative in only one case.

36

In

Nedjalkov’s (1969) study of 60 languages, ‘break’ is anticausative in 22 cases, labile in

19, and causative in 9. Thus, while all three directionalities are observed, the anticausative type for ‘break’ appears to have a frequency of one third to one half crosslinguistically. Nedjalkov also looks at patterns for ‘laugh.’ In 54 out of 60 languages, the causative pattern is observed with this verb; there are no anticausative alternations. L&R consider this as evidence not only that the unaccusative/causative alternation is distinct from the unergative/causative alternation, but also that the former is universally anticausative and the latter universally causative.

Their final piece of evidence (from Chierchia 1989) is that languages have adverbial modifiers that detect the presence of a cause argument, and that these are

36

The .5 is due to Modern Greek, which has both an anticausative and a labile alternation glossed as

‘break.’

130 felicitous with unaccusative verbs. The English adverbial is

by itself

, on the reading ‘with no outside help’ as opposed to ‘alone.’

(14) a. b. c.

The plate broke by itself.

The door opened by itself.

Molly laughed by herself. (L&R: 88–89)

With unaccusative verbs in sentences (14a) and (b),

by itself

readily has the interpretation

‘with no outside help. This contrasts with (14c), where the meaning ‘alone’ is far more salient. L&R make the following point: ‘the most natural interpretation of

Molly laughed by herself

is that Molly laughed unaccompanied rather than without outside help.’

4.2.2.1 Decausativization

One group of generative linguists seeks to capture the generalizations about the causative/inchoative alternation through a word-forming operation they call decausativization (Reinhart 2002; Reinhart & Siloni 2004; Horvath & Siloni 2011).

According to them, the transitive member is universally the basic form, from which the intransitive member is derived through decausativization, an operation that eliminates the

Causer role from a verb’s lexical entry. Their model assumes that words can be formed in the lexicon or in the syntax. However, since thematic roles, arguments, and operators cannot be deleted from the syntax (see 4.3 below), decausativization necessarily applies in the lexicon.

Decausativization is one of a variety of so-called arity operations, which alter the

θ

-grid of a verb’s lexical entry. In this model, thematic roles are not primitives, but decompose into feature clusters. There are two types of feature: c, which stands for

‘cause’ and indicates whether the argument associated with the role is responsible for

131 causing an event; and m, which stands for ‘mental state’ and indicates whether the argument’s mental state is relevant to the event. Each of these can be positively or negatively valued, and a feature cluster may also be underspecified for one of the two features. This yields eight possible distinct feature clusters, which more or less correspond to the thematic roles generally recognized, though the traditional roles are merely labels in this framework and have no formal status.

(15) Thematic feature clusters and their associated roles (Reinhart 2002)

[+c+m]

[+c-m]

[-c+m]

[-c-m]

[+c]

[+m]

agent instrument experiencer theme / patient cause

(…)

(Unspecified for / m)

[subject of non-alternating Experiencer verbs such as

love

,

know

,

believe

]

[-m]

[-c]

(Unspecified for / c):

benefactor

subject matter / locative source

(Unspecified for / m): Internal roles like

goal,

An agent causes the event in question, and the agent’s mental state is relevant to the event, hence both features are positively valued: [+c+m]. This contrasts with instrument, which is negatively valued for mental state—[+c-m]—and with cause, which is underspecified for mental state: [+c]. On the other hand, a theme or patient has a cluster with both features negatively valued: [-c-m]. Note that a positive value for mental state must necessarily be linked to an animate argument, but a negative value for this state simply indicates that the mental state is not relevant, and so the linked argument may be animate or inanimate.

132

Transitive verbs participating in the alternation have a lexical entry of ([+c], [-c

-m]). The cause role is underspecified for mental state, such that the argument fulfilling this role may be an agent [+c+m], an instrument [+c-m], or some other cause [+c].

(16) The wind / Max / the key opened the door. (Reinhart 2002: 233)

Non-alternating transitive verbs, on the other hand, have the cluster [+c+m] in their lexical entry, which cannot be realized by a non-agentive cause. So-called ‘manner verbs’ such as

drill

and

peel

(Levin & Rappaport Hovav 1995) have the θ-grid ([+c+m], [-c-m],

[+c-m]), with an agent, patient, and instrument. When there are two +c roles, only one need be realized. Thus, such verbs may have either an agent or an instrument subject, but not another cause.

(17) a. b. c.

Max peeled the apple (with a knife).

The knife peeled the apple.

*The heat peeled the apple. (Reinhart 2002: 235)

The intransitive member of the causative/inchoative alternation is derived by

Decausativization, an arity operation that deletes a [+c] cluster from a lexical entry.

(18) V (

θ

[+c]

,

θ

j

) → V (

θ

j

)

(Horvath & Siloni 2011)

The underspecified [+c] cluster is therefore the defining property of transitive verbs that participate in the operation. Verbs whose cause roles are specified for mental state are not valid input for Decausativization.

(19) a. b.

The door opened.

*The apple peeled. (Reinhart 2002: 235)

Unaccusative verbs are universally derived via reduction of a [+c] role, and hence, by definition, are verbs that have a transitive alternant with a [+c] role. This

133 straightforwardly predicts that unaccusative verbs should always have a transitive alternant. Horvath & Siloni (2011) acknowledge that some languages have idiosyncratic gaps where the transitive member is missing. They account for such cases with so-called

‘frozen lexical entries’: lexical entries that are not linked to actual verbs in the output, but that are nevertheless available for lexical operations.

4.2.3 Equipollence

A third possibility is that the alternation is underlyingly equipollent, the position taken by Harley (1995, 2008) and by Alexiadou et al. (2006). On this view, neither member of the alternation is derived from the other, but rather, each member is derived independently from a common base. This avoids the pitfalls of the directional approaches. As in Pesetsky’s causativization, the transitive member of the alternation is derived with a causative morpheme. Thus, both approaches are able to capture the unavailability of a causative reading and transitive argument structure in derived nominalizations, since the

CAUS

operator is not part of the Root to which the nominalizer attaches. Where the approaches diverge is the characterization of this Root. In Pesetsky, the Root is not only of the category Verb, it also has inchoative semantics. Under universal equipollence, the Root has no category, and it is no more inherently inchoative than it is inherently causative. The intransitive reading of the nominalization is not due to the presence of a

BECOME operator, but is a consequence of the argument-taking property of the Root. Unlike the external argument, the internal argument is an argument of the

Root, and thus its realization is not dependent on the presence of any functional head.

The complement of the Root is the theme of the change of state, independent of whether

134 it is ultimately realized as the subject or object of a finite verb (20), or as the ‘subject’ or

‘object’ of a nominalization (21).

(20) a. The tomatoes grew. b. John grew the tomatoes.

(21) a. the tomatoes’ growth b. the growth of the tomatoes

Universal equipollence also avoids the problem noted by L&R that the possible objects of the intransitive member are a subset of the possible objects of the transitive member, since, unlike in causativization, the latter is not derived from the former. Still, there are issues that need to be addressed. If the unaccusative and causative alternants are derived independently, there is no a priori reason why the possible theme arguments of one should exactly coincide with those of the other. Of course, since both derivations begin with the merger of the internal argument and the Root, the discrepancy cannot be accounted for at this level, but it is in principle possible that certain arguments are incompatible with the semantics contributed by one functional head or the other. Thus, since a table has no means of clearing without the intervention of agents or natural forces, it is infelicitous when the phrase formed by the merger of the DP

the table

and the Root

clear

merges with a functional head that precludes the introduction such an external argument. As this infelicity is semantic rather than syntactic, (13b’) would be more accurately marked with a pound sign (#) than with an asterisk (*). As for the cases in

(12), they are all idioms. The merger of a Root phrase with a causative head is a valid domain for idiom formation, to the exclusion of the merger of the same Root phrase with a different functional head.

135

Where universal equipollence has a clear advantage over either directional approach is in capturing the widely attested variation in morphological patterns already mentioned. Proponents of anticausativization point to the fact that this morphological pattern is the most common, a fact that is difficult to explain if one advocates universal causativization. This is not the only morphological problem. As Laks (2011) points out, any model in which causativization is the universal operation faces the challenge of explaining why causatives of transitives and unergatives (productive causatives) do not exhibit the same array, but consistently make use of the overt morphological causativization pattern. I give examples from Turkish.

(22) non-causative causative

koş-

‘run’

çalış- koş-tur-

‘make run’

çalış-tır-

‘work’

zıpla-

‘hop’

‘make work’

zıpla-t-

‘make hop’

öp-

‘kiss’

döv-

‘beat’

öp-tür-

‘make kiss’

döv-dür-

‘make beat’

Further examples could be given ad nauseum, but there is no need. The pattern has no exception in Turkish, and is quite robust in other languages that have this kind of causative. Laks provides the results of a dictionary search for Hebrew: 84% of such alternations are causative, 7% are anticausative, and the rest are non-directional types. If both transitivity alternations are causativization, why do we see such variety in one and such uniformity in the other?

136

On the other hand, the equipollent analysis predicts exactly this pattern. The transitive and intransitive members of the causative/inchoative alternation are derived independently from a Root. If null morphemes are included along with overt ones, then, in the most basic form of the model, we get four out the five surface alternation types.

This is shown in abstract fashion below. Italicized forms in brackets represent null morphemes.

(23) The causative/inchoative alternation: underlying and surface forms transitive

√-

CAUS

-

√-[

CAUS

]-

√-

CAUS

-

√-[

CAUS

]- intransitive surface alternation type

√-[

BECOME

]- Causative

√-

BECOME

-

√-

BECOME

-

Anticausative

Equipollent

√-[

BECOME

]- Labile

In contrast, causatives of unergatives and transitives are formed by a causative morpheme attaching to a verbal stem. Only two surface patterns are predicted: causative, and labile.

(24) The productive causative alternation: underlying and surface forms non-causative causative

√-v-

√-v-

√-v-

√-v-[

CAUS

CAUS

-

]- surface alternation type

Causative

Labile

Thus, all of the occurring basic patterns except suppletive are immediately predicted by the model, even before it is augmented with post-syntactic operations such as fission.

Furthermore, the suppletive pattern is easily captured with Root allomorphy. Although anticausative is the most common pattern, the other patterns are common enough that they cannot be ignored. Only the equipollent analysis satisfactorily captures the full tableau.

137

Directionality is also of importance for the locus of derivation. If word formation is syntactic, then causative and equipollent derivations are permissible, but a true anticausative would be problematic. This problem is taken up in the following section.

4.3 Anticausativization and Monotonicity

From an a priori standpoint, the members of a causative/inchoative alternating pair have, on the assumption that they are derivationally related, three possible basic types of relationship: causativization, anticausativization, and equipollence. In a framework where word formation is modeled as function application, as it must be in syntactic derivation, the first and the third are admissable derivations, but the second, anticausativization, is not. Syntactic word formation proceeds from the Root upwards.

Functional heads introducing operators and arguments are added one after another. Each new head/projection complex can only add structure; it cannot delete existing structure.

In causativization, a

BECOME

operator is added to a base, and a

CAUS

operator is built on top of that. In equipollence, the inchoative is derived via a

BECOME

operator added to a base, while the causative is independently derived via a

CAUS

operator added to that same base without the addition of the

BECOME

operator. In both types of derivation, structure is added, and none is removed. However, on the most literal understanding of anticausativization, structure must be deleted. This is because the anticausative is added to a base that has a

CAUS

operator, and the result of this operation is a non-causative verb.

Frameworks that distinguish between derivation in the lexicon and derivation in the syntax (or that have all-lexical derivation) may permit lexical operations that delete operators and arguments (Grimshaw (1982), Reinhart 2002, Reinhart & Siloni (2005),

138

Horvath & Siloni 2011a), so if anticausativization can be shown to exist on its most literal interpretation, there would be no escape from a generative lexicon.

4.3.1 Anticausativization as Reflexivization

Koontz-Garboden (2009) frames the problem in terms of the Monotonicity

Hypothesis.

(25) The Monotonicity Hypothesis

Word formation operations do not remove operators from lexical semantic representations. (Koontz-Garboden 2009: 80)

He claims that anticausativization does exist, but that it does not delete the

CAUS

operator from the base verb. He focuses on Spanish, where the anticausative is realized by a reflexive clitic on a causative base verb.

(26) causative anticausative

romper romperse

‘break (tr.)’ ‘break (int.)’

Koontz-Garboden uses neo-Davidsonian semantics along the lines of Parsons (1990). The variable

e

represents events,

s

represents states, and the two are subsumed under

v

, eventualities. He also makes use of the category EFFECTOR (Van Valin and Wilkins

1996) as an underspecified argument that brings about an event, subsuming Agents and inanimate Causers. He formulates the lexical semantic representation of a causative verb such as

romper

as in (27). The symbol

ϕ

is a random stative predicate (‘not whole’ in the case of

break

).

(27)

λ

x

λ

y

λ

s

λ

e

[

v

[

CAUSE

(

v

,

e

)

EFFECTOR

(

v

,

y

)

BECOME

(

e

,

s

)

THEME

(

s

,

x

)

ϕ

(

s

)]] (Koontz-Garboden 2009: 82)

An anticausative form such as

romperse

retains the CAUSE operator and the EFFECTOR role. It therefore explicitly does

not

have a representation such as that in (28).

139

(28)

λ

x

λ

s

λ

e

[

BECOME

(

e

,

s

)

THEME

(

s

,

x

)

ϕ

(

s

)]

Koontz-Garboden proposes instead that the anticausative is built onto the causative representation in (27), deleting nothing. Anticausativization, he claims, is none other than reflexivization, an operation that identifies two arguments with each other.

(29) The reflexivization operator

λ

λ

x

[

(

x

,

x

)] (Koontz-Garboden 2009: 83)

Applied to (27), this will ensure that the THEME and the EFFECTOR are identical.

Thus, an anticausative appears to have fewer arguments than the corresponding causative, but this is because two roles have been collapsed into a single argument.

A crucial piece of evidence that the anticausative retains the CAUSE operator comes from negation. Koontz-Garboden claims that it is possible to negate an anticausative while affirming the causative. This is surprising, since the anticausative entails the same result state as the corresponding causative. Thus,

el vaso se rompió

‘the glass broke’ has the same result state entailment as

rompiste el vaso

‘you broke the glass’: namely, that the glass is broken. If this were the whole story, then negating the anticausative while affirming the causative should result in a contradiction. But apparently it does not.

(30) Father: ¿Que pasó, hijo?

‘What happened, child?’

Son: El vaso se rompió.

‘The glass broke.’

Father:

No se rompió

sino que tú lo rompiste!

‘The glass didn’t

break

—you broke it!’

(Koontz-Garboden 2009: 112)

This makes sense if the lexical semantic representation of the anticausative includes both a BECOME operator and a CAUSE operator. Negation could, in principle, target either

140 of these. In (30), the verb

se rompió

‘(it) broke’ has a CAUSE operator the EFFECTOR of which has been identified with the THEME through reflexivization. It is this reflexivized CAUSE that the father is negating, not the result state associated with

BECOME.

Koontz-Garboden further argues that (30) is not a case of meta-linguistic negation, in which a speaker is not negating a truth value, but is merely objecting to the choice of words. Metalinguistic negation differs from logical negation in that it does not license negative polarity items (Horn 1985).

(31) a. b.

John didn’t manage to solve

SOME

of the problems—he managed to solve

ALL

of them. (Horn 1985: 135, (24))

*John didn’t manage to solve

ANY

of the problems—he managed to solve

ALL

of them. (Horn 1985: 135, (24))

(cited in Koontz-Garboden 2009: 116).

Koontz-Garboden therefore uses this to test whether the CAUSE-negation in anticausatives is ordinary (logical) or metalinguistic. He claims that CAUSE-negation does license negative polarity items, and concludes that it is ordinary negation.

(32) itself;

No se rompió ningún vaso; los rompió Andrés. no se broke any.neg(NPI) glass them broke Andrew

‘Any glass didn’t break; Andrew broke them.’ (lit: Any glass didn’t break by

Andrew broke them all.) (Koontz-Garboden 2009: 117)

As we will see shortly, this crucial piece of data may be inaccurate, calling the entire analysis into question. For the moment, however, let us proceed with Koontz-

Garboden’s argumentation. Allegedly in contrast to the Spanish sentence, the English translation of (32) is a contradiction. This too is predicted, as this approach does not assume that the anticausativization is the universal direction of derivation, and Koontz-

141

Garboden suggests that it is not the correct analysis for English. According to him, the same is true of Spanish alternating verbs whose intransitive member lacks the reflexive clitic. Since these lack a CAUSE operator, the kind of negation seen in (32) should be impossible, as it is for English alternating verbs. He gives the example of

empeorar

‘to worsen (tr./int.),’ which exhibits a labile alternation.

(33) #No empeoró ningún paciente. Los empeoró el tratamiento. no worsened any.neg patient them worsened the treatment

‘Any patient didn’t worsen. The treatment worsened them.’

(Koontz-Garboden 2009: 117)

Anticausativization-as-reflexivization, if correct, is significant in that it allows anticausatives to be built through functional application, and is a legitimate operation in syntactic word derivation. However, Horvath & Siloni (2011b) cast doubt on Koontz-

Garboden’s evidence, and hence on his claim. They report that the Spanish speakers they consulted found the sentence in (32) to be just as contradictory as the sentence in (33).

They tested such sentences with other NPIs in Spanish as well, with the same result in all cases.

Horvath & Siloni also provide evidence from Hebrew and Hungarian that the negation of an anticausative paired with the assertion of the result state is coherent only with metalinguistic negation and not with logical negation. My own efforts corroborate their results. I attempted to replicate Koontz-Garboden’s data in Turkish, writing out a number of father-son dialogues with verbs of anticausative, causative, and equipollent alternations, in order to see if the test detected any difference between alternation types. I showed the sentences to three speakers. They judged some sentences to be good and others bad, but there was no consistency among speakers, nor any coherent pattern in the

142 judgments of any given speaker. When I added negative polarity items to the sentences, they were uniformly judged to be terrible.

If anticausativization is an operation that deletes a CAUSE operator, it cannot be implemented in syntactic derivation. Anticausativization-as-reflexivization does not appear to be a viable option. Is there another analysis of anticausativization that does not involve operator deletion? Turkish morphology suggests that there is, at the same time providing further evidence against the anticausativization-as-reflexivization hypothesis.

4.3.2.1 There is no Causative in Anticausative

For Koontz-Garboden, anticausativization

is

reflexivization. He provides further evidence for this from morphology. He presents a table of syncretism patterns in 13 languages taken from Haspelmath 1990. In all but four of these, the anticausative morpheme is syncretic with the reflexive morpheme. This is impressive on the face of it; however, the same table shows that anticausative is syncretic with the passive in all but one language. In fact, the languages with reflexive-anticausative syncretism are a subset of those with passive-anticausative syncretism. He addresses this in a lengthy footnote in response to a reviewer’s observation of this fact. He suggests that a closer investigation of the languages in Haspelmath’s table would reveal that many of them have either passive or anticausative but not both, and that the supposed syncretism is actually a case of mistaken identity. He does allow that such syncretism exists, but suggests that its typological prevalence is ‘overstated’ (p. 92, n. 11). His point about the need to take a closer look at the individual languages in broad typological studies is a good one. In this

143 section, I take a close look at anticausatives and syncretism in Turkish (although this language is not included in Haspelmath 1990).

One pattern would be especially antithetical to the anticausativization-asreflexivization hypothesis: anticausative and passive are both morphologically marked, and they are marked differently. Koontz-Garboden says as much:

Perhaps the most curious kind of language from the perspective of this paper would be one in which anticausativization and reflexivization are marked differently, and where the language either does not have a passive or has a passive morphologically marked in a different way, so that it is clear the anticausativization and passive functions of the same morphology are not simply being confused. (Koontz-Garboden 2009: 93)

Turkish is just such a curious language. This might initially be surprising, given that the reflexive, passive and anticausative look alike in most environments. On closer investigation, however, we find contrastive environments where they can be systematically distinguished from one another. These environments reveal not only that the anticausative morpheme is distinct from the reflexive, but also that, unlike the passive, it never attaches outside of a causative morpheme. This is an especially serious problem for a proposal that (a) anticausativization derives an inchoative/anticausative verb from a causative one, and that (b) it leaves the causative operator intact.

In many cases, the passive and the reflexive are phonologically identical. Indeed,

Nakipoğlu (1998) claims that they are always syncretic, characterizing the realization for both as follows: ‘-n after a vowel, -In after /l/ and –Il elsewhere’ (32 n. 9). This is not quite right, however. The description she gives is accurate for the passive, but not the

144 reflexive, which is never realized as –

Il

. Göksel & Kerslake (2005) identify the reflexive suffix as –(

I

)

n

, which, as we will see below, is the correct characterization.

As Nakipoğlu (1998) says, both the passive and the reflexive morphemes are –

n

following a vowel (34a) and –

In

following the liquid /l/ (34b).

(34) a. b. bare stem

boya-

‘paint (tr.)’

yıka-

‘wash (tr.)’

bul-

‘find’ passive reflexive

boya-nboya-n-

‘be painted’ ‘put on make-up’

yıka-nyıka-n-

‘be washed’ ‘wash oneself’

bul-un-

‘be found’

bul-un-

‘find onself (i.e., be located)’

However, they are distinct following a consonant other than /l/.

(35) bare stem

döv-

‘beat’

giy-

‘wear’

ört-

‘cover’ passive

döv-ül-

‘be beaten’

giy-il-

‘be worn’

ört-ül-

reflexive

döv-ün

-

‘beat oneself/one’s chest’

giy-in-

‘get dressed’

ört-ün-

‘be covered’ ‘cover oneself’

The mistaken assumption that –

n

/–

In

/–

Il

is both passive and reflexive appears to be behind N&Ü’s organization of their lexicon, whereby several verbs ending in –

Il

are classified as reflexive (App. 2a: 14.), and reflexive verbs (including those in (35)) are grouped under –

n

(App. 2c: 1.). Given the systematic contrast between passive and reflexive following any consonant except /l/, the conflation of the two morphemic categories is not warranted. As for the supposed reflexives in –

Il

, these are actually anticausatives. Indeed, wherever a verb providing a contrastive environment has both a reflexive and a passive form, it is the passive that is syncretic with the anticausative, and

145 not the reflexive. (Note also that both the reflexive and the inchoative exhibit semantic drift in several cases, while the passive does not.)

(36) bare stem

dök-

‘pour (tr.)’

sık-

‘squeeze’

soy-

‘peel’

tut-

‘hold’ passive

dök-ül-

‘be poured’

soy-ul- sık-ıl- sık-ın-

‘be squeezed’ ‘restrain

‘be peeled’

tut-ul-

‘be held’ reflexive

dök-ün-

‘pour water on oneself’ oneself’

soy-un-

‘get undressed’

tut-un-

‘hold on tight’ anticausative

dök-ül-

‘pour out (int.)’

sık-ıl-

‘get bored’

soy-ul-

‘peel (int.)(skin, etc.)’

tut-ul-

‘get stiff (neck, back); fall in love’

Semantic drift may occasionally blur the distinction between reflexive and anticausative

(and indeed verbs of the anticausative category are realized with reflexive morphology in other languages). However, the anticausative always has a straightforward inchoative meaning.

Since passives and reflexives syncretize in other environments, anticausatives in –

n

and –

In

are in principle ambiguous between the passive and the reflexive morpheme.

(37) bare stem

böl-

‘divide (tr.)’

kapa-

‘close (tr.)’ inchoative

böl-ün-

‘divide (int.)’

kapa-n-

‘close (int.)’

However, given positive evidence such as that in (36) that the passive morpheme has an inchoative function, and the lack of any positive evidence that the reflexive morpheme

146 does, it is reasonable to conclude that the anticausative is syncretic with the former and not the latter.

The facts presented thus far do not, on their own, present a serious threat to the

Anticausativization-as-reflexivization hypothesis. Indeed, the formal identity of passive and anticausative suggests a way out. The apparent syncretism could be a case of mistaken identity; the so-called ‘inchoatives’ with passive morphology could in fact simply be variant readings of the passive. If this is so, then Turkish lacks the anticausative, and hence poses no threat. In the next section, I show that this avenue is not a viable option. I introduce two diagnostics for distinguishing between the passive and the anticausative. I then show how these diagnostics correlate with the distributional differences between passive and anticausative suffixes.

4.3.2.2 Telling Apart the Look-Alikes

Koontz-Garboden makes the point that distinguishing between passives and anticausatives is extremely difficult, and he suggests that many languages’ verbal systems have been misanalyzed as having syncretic passive and anticausative, when in reality they have only one or the other.’ In Spanish, he says, there is only one reliable diagnostic: the phrase

por sí solo

‘by her/him/itself.’ The Italian equivalent was used by Chierchia

(2004) for the same purpose. Because this phrase indicates that the event had no external causer, it is compatible with an anticausative (inchoative) (38a) but not with a passive (b).

(38) a.

Garboden b.

El barco se hundió por sí solo.

the boat

REFL

sank by

SELF only

‘The boat sank by itself.’ (Mendikoetxea 1999: 1594; cited in Koontz-

200?: 97)

*

El senador se asesinó por sí solo

.

147 the senator

REFL

assassinated by

SELF

only

‘*The senator assassinated/was assassinated by herself.’ [

sic

]

(Mendikoetxea 1999: 1592; cited in Koontz-Garboden 2009: 89.)

The

by-self

phrase distinguishes forms that are ambiguous between passive and anticausative readings from those that are unambiguously passive. Thus, its felicity with a certain verb does not mean that that verb is only anticausative; rather, the verb may be ambiguous, and the phrase picks out the anticausative reading to the exclusion of the passive, which would otherwise be available. Its value as a diagnostic is therefore limited, in a language like Spanish at any rate. In Turkish, as will become clear, it can be used with greater precision.

The Turkish equivalent of ‘by itself’ is

kendi kendine

. Turkish also has a

by-

phrase for expressing the agent of a passive, with the postposition

tarafından

. Since this is only available with the passive, it is a reliable diagnostic for that construction. If verbs with –

Il

are truly syncretic, then both expressions should be compatible with them. This appears to be the case. The following sentences contain the verb

boğ-ul-

, which is ambiguous between ‘be drowned’ and ‘drown (int.).’ (The morpheme under consideration is glossed simply as –

Il

to maintain a semblance of analytical neutrality.)

(39) a. b.

Patron, göl-de hizmetli-si tarafından boğ-ul-du.

boss lake-

LOC employee-3

SG

by drown-Il-

PAST

‘The boss was drowned in the lake by his employee.’

Patron göl-de kendi kendin-e boğ-ul-du.

boss lake-

LOC self self-

DAT

drown-Il-

PAST

‘The boss drowned by himself 37

in the lake.’

37

The English phrase

by himself

has two meanings: 1. with no external instigation, and 2. with no company. Only the first is relevant here, as Turkish

kendi kendine

lacks any such ambiguity.

148

The

by

-phrase in (39a) forces a passive reading, and the

by-himself

phrase in

(39b) forces an inchoative (anticausative) reading, indicating that the form

boğ-ul-

is actually ambiguous. The possibility that –

Il

is a marker of the passive only begins to look less likely.

But can we be sure that these phrases are reliable diagnostics in Turkish? Might there be some factor about these phrases in Turkish that is crucially different from their translation-equivalents in other languages, for example, something that would make

kendi kendine

felicitous with the passive? If passive verbs and anticausative verbs were always identical in Turkish, this question might remain forever unanswered. However, although the syncretism of the exponents is perfect, there are differences between the constructions in terms of the distribution of these exponents. Namely, when a transitive verb has an overt causative morpheme, the form augmented with –

Il

is compatible with the

by

-phrase

(40a), but not with the

by-himself

phrase (40b), which is good only with the bare intransitive stem (40c).

(40) a. b. c.

Patron, hizmetli-si tarafından

öl-dür-ül-dü.

boss employee-3

SG

by die-C

AUS

-Il-

PAST

‘The boss was killed by his employee.’

*

Patron

boss

kendi kendin-e

self self-

DAT

‘The boss was killed by himself.’

öl-dür-ül-dü.

die-C

AUS

-Il-

PAST

Patron kendi kendin-e

boss self self-

DAT

‘The boss died by himself.’

öl-dü.

die-

PAST

This shows that the diagnostics truly do make a distinction. Note also that the form where

Il

attaches outside of an overt causative morpheme tests as a passive, not an anticausative. This yields a contrast in verb pairs where the transitive alternant has a

149 causative suffix, such as

öl

-

dür

-/

öl-

‘kill/die’ and those where it does not, such as

boğ-

/

boğ-ul-

‘drown (tr.)/drown (int.).’ The next step is to test this distributional difference on a single verb stem, so as to have a minimal pair. This is not possible with the verb

öl

-

‘die’ and a multitude of others like it, since –

Il

does not yield an inchoative when attached directly to this stem: *

öl-ün-

.

38

Fortunately, Turkish has verbs with more finely articulated realization than this. Consider the verb

kap-a-

‘close (tr.)’ from Section 3.4.3.

The intransitive alternant takes the anticausative in –

n

. (Note that in (41a), the VI –

A

realizes v.

CAUS

, while in (41b) the same VI realizes an unflavored little-v).

(41) a. b.

Hizmetli

employee

kapı-yı

door-

ACC

kap-a-dı

.

√-v.

‘The employee closed the door.’

CAUS

-

PAST

Kapı kap-a-n-dı

. door

√-v-

IL

-

PAST

‘The door closed.’

Recall that an alternative form of the transitive is

kap-a-t-

, where realization of

CAUS

is overt and discrete.

(42)

Hizmetli

employee

kapı-yı

door-

ACC

kap-a-t-tı

.

√-v-

‘The employee closed the door.’

CAUS

-

PAST

The form with the overt causative morpheme can be passivized.

(43)

Kapı kap-a-t-ıl-dı

. door

√-v-

CAUS

-

IL

-

PAST

‘The door was closed.’

39

38

Forms such as these are grammatical on an impersonal passive reading:

öl-ün-ür

‘one dies/dying is done.’ This is clearly not an inchoative/anticausative, since the subject cannot be expressed:

İnsan öl-ür

‘a person dies,’ but

*İnsan öl-ün-ür

.

39

In the eventive sense: The door was the theme of a closing event. The adjectival passive reading

(The door was in a closed state) is not available in the Turkish sentence.

150

This array of affixation possibilities gives us two positions for the passive/anticausative morpheme: one that embeds an overt causative, and one that does not. The passive/anticausative diagnostics distinguish between these two configurations. When it attaches outside of the causative, the

by

-phrase but not the

by-itself

phrase is licensed.

(44) a. b.

Kapı hizmetli

door employee

tarafından

by

kap-a-t-ıl-dı

.

√-v-

‘The door was closed by the employee.’

CAUS

-

IL

-

PAST

*

Kapı kendi kendin-e kap-a-t-ıl-dı.

door self self-

DAT

√-v-C

AUS

-

IL

-

PAST

‘*The door was closed by itself.’

Conversely, when passive/anticausative does not embed the causative, the

by

-phrase is infelicitous (or perhaps merely degraded), but the

by-itself

phrase is licensed.

(45) a. b.

*/?

Kapı hizmetli tarafından

door employee by

‘*The door closed by the employee.’

Kapı kendi kendin-e

door self self-

DAT

‘The door closed by itself.’

kap-a-n-dı

.

√-v-

IL

-

PAST

kap-a-n-dı

.

√-v-

IL

-

PAST

(45b) shows a clear difference from (44b) with respect to the

by-itself

phrase. Judgments for (45a) are somewhat less sharp, but note that, since

kap-a-

is a transitive verb, in principle

kap-a-n-

should have a passive reading available. The degradation of the acceptability of the

by

-phrase, then, is what is surprising. The judgments I obtained ranged from ‘good’ to ‘bad’ to ‘I wouldn’t say it but others would.’ This may be a matter of idiolect variation. In standard Turkish,

kap-a-

is exceptional in having two optional forms of the transitive. We can get a sharper judgment with the –

t

/–

n

alternating stems that do not have this optionality.

151

(46) a. b.

Halk doğal kaynak-lar-ı

people natural resource-

PL

-

ACC

tük-e-t-ti

.

√-v-

CAUS

‘The people used up the natural resources.’

-

PAST

*Halk doğal kaynak-lar-ı

people natural resource-

PL

-

ACC

tük-e-di

.

√-v-

PAST

The verb

tüket

- is often translated as ‘exhaust’ or ‘consume,’ but since these are nonalternating verbs in English, I translate it here as ‘use up,’ and the intransitive

tüken-

as

‘run out.’ Unlike

kap-a-

, the –

t

/–

n

alternating stem *

tük-e-

never occurs independently.

This means we can avoid the confound in (45a). The form in (47) where –

Il

does not embed the causative morpheme is sharply ungrammatical with a

by

-phrase (a), while it is fine with a

by-self

phrase (b).

(47) a. *

Doğal kaynak-lar halk tarafından

natural resource-

PL

-

ACC people by

‘*The natural resources ran out by the people.’ b.

tük-e-n-di

√-v-

IL

-

.

PAST

Doğal kaynak-lar kendi kendin-e

natural resource-

PL

-

ACC self self-

DAT

‘The natural resources ran out all by themselves.’

tük-e-n-di

.

√-v-

IL

-

PAST

Conversely, the form where –

Il

embeds the causative is fine with a

by

-phrase (48a), but bad with a

by-self

phrase (48b).

(48) a. b.

Doğal kaynak-lar halk tarafından

natural resource-

PL

-

ACC people by

‘The natural resources were used up by the people.’

tük-e-t-il-di

.

√-v-

CAUS

-

IL

-

PAST

*

Doğal kaynak-lar

natural resource-

PL

-

ACC

kendi kendin-e

self self-

DAT

tük-e-t-il-di

.

√-v-

‘The natural resources were used up all by themselves.’

CAUS

-

IL

-

PAST

At this point, there should be no doubt that when –

Il

attaches outside of an overt causative morpheme, the result is a passive verb and not an anticausative/inchoative. But can we be equally certain that the verb in (47b), where there is no causative, contains this

152 same morpheme, which I have uniformly glossed as –

Il

? After all, in these cases, it has the form –

n

, which is an allomorph not only of passive –

Il

, but also of reflexive –(

I

)

n

.

Thus, might the differences in the a. and b. sentences be due not to different attachment sites of the same morpheme, but to different properties of two distinct morphemes? To test this, we need a verb stem where the same allomorph, –

Il

, is found in both positions.

To my knowledge, there are precisely two verb stems where this is the case:

kurt-

and

dağ-

(or, more precisely, the Roots √K

URT

and √D AĞ ). The causative is realized on the former stem as

–Ar

, and on the latter as –

It

.

(49) causative -Il attached to Root

kurt-ar-

√-

CAUS

-

‘rescue’

dağ-ıt- kurt-ul-

√-

IL

-

‘get free (int.)’

dağ-ıl-

√-

CAUS

-

√-

IL

-

‘scatter (tr.)’ ‘scatter (int.)’

-Il attached outside causative

kurt-ar-ıl-

√-

CAUS

-

IL

-

‘be rescued’

dağ-ıt-ıl-

√-

CAUS

-

IL

-

‘be scattered’

These verbs exhibit the same pattern we have seen repeatedly: When –

Il

attaches outside the causative, the result is passive (50), and when it does not, the result is

(51) anticausative/inchoative (51).

(50) a. b.

Rehine polis tarafından

hostage police.officer by

kurt-ar-ıl-dı.

√-

CAUS

-

IL

-

‘The hostage was rescued by the police officer.’

PAST

*Rehine

hostage

kendi kendin-e

self self-

DAT

kurt-ar-ıl-dı.

√-

CAUS

‘The hostage was rescued all by himself.’

-

IL

-

PAST a. b.

*Rehine

hostage

polis tarafından

police.officer by

kurt-ul-du.

√-

IL

‘The hostage was rescued by the police officer.’

-

PAST

Rehine kendi kendin-e

hostage self self-

DAT

‘The hostage got free all by himself.’

kurt-ul-du.

√-

IL

-

PAST

153

This leaves little room for doubt that the morpheme –

Il

, with the phonologically conditioned variants –

n

(after a vowel) and –

In

(after /l/), has both passive and inchoative functions, and that there are minimal pairs where these two functions correspond to the presence versus the absence of a causative suffix between the stem and –

Il

. This raises the question of whether an inchoative verb

ever

includes a causative morpheme. The answer is clearly no for the great majority of verbs, but there is one class of verbs where the answer is not immediately clear. There are certain transitive verbs ending in the sequence /

Ir

/ whose intransitive counterparts have –

Il

. Although the vowel in the /

Ir

/ sequence is deleted, the /r/ remains.

40

(52) transitive

ayır-

‘separate’ intransitive

ayr-ıl-

‘separate’

devir-

‘knock over’

eğir- devr-il-

‘fall over, capsize’

eğr-il-

‘spin (e.g., cotton)’ ‘bend’

kavur-

‘broil, char’

kıvır-

‘coil (tr.)’

sıyır-

‘scrape’

kavr-ul-

‘become scorched’

kıvr-ıl-

‘coil (int.)’

sıyr-ıl-

‘sneak away; become scraped (skin)’

These look as though they might be composed of a Root and the causative suffix –

Ir

. And yet the intransitive variants are for the most part compatible with a

by-self

phrase.

(53) a.

Sık-ış-an

√-

RECIP

-

REL

dişli-ler

gear-

PL

kendi kendin-e

self self-

DAT

‘The jammed gears separated by themselves.’

ayrıl-dı

. separate-

PAST

40

The intransitive alternants in (52) are not all grouped together in N&Ü. While several are found under App. 2a: 14. ‘-Il verbs (reflexive verbs),’ others appear under App. 2a: 2., identified with –(

A

)

l

, which supposedly derives verbs from adjectives, such that

ayrıl-

would be derived from

ayrı

‘separate

(adj.)’ and

eğril-

from

eğri

‘crooked.’

154 b.

Kule kendi kendin-e devril-di.

tower self self-

DAT fall.over-

PAST

‘The tower fell over by itself.’

This suggests that these are inchoatives, so if the /r/ is the causative suffix, they constitute an exception to the generalization that inchoatives never include a causative morpheme.

If –(

I

)

r

is a suffix, the bases to which it attaches do not occur independently as verbs (at least not with relevant meanings):

*çağ-

,

*dev-

, etc. Of course, this in itself by no means precludes a stem-suffix analysis under the proposal about Roots that is advanced in this dissertation.

Note that the segment preceding /(

I

)

r

/ is always

y

,

v

, or

ğ

. Could it be that the causative –

Ir

undergoes vowel deletion following these segments, or that the causative is really –

r

and it undergoes vowel epenthesis in other environments? The answer is no.

Verbs that transparently contain this suffix and whose stems end in /y/ and /ğ/ (there are none in /v/) do not exhibit vowel deletion.

(54) non-causative causative

doğ- doğ-ur-

‘be born’ causative + -

doğ-ur-ul-

‘give birth to’ ‘be birthed’

Il

(= passive)

doy-

sated’

‘become

doy-ur-

‘satiate’

doy-ur-ul-

‘be sated’

duy-

‘hear; feel’

duy-ur-

‘announce’

duy-ur-ul-

‘be announced’ deletion

*doğ-r-ul-

*doy-r-ul-

*duy-r-ul-

Clearly, then, there is no phonological process by which –

Ir

becomes –

r

following certain segments. I propose that in (52), the overt base to which –

Il

attaches is a Root.

41

The fact that these Roots all end in /r/ is probably due to their historical development. It is

41

This is not to say that there is no intervening null morphology. In 4.4.1, I propose that there is a null little-v in these cases.

155 plausible that at some stage of the language, this was affixal material, but is now part of the Root. The Roots are √A

YR

,

√D

EVR

,

√E

ĞR ,

√K

AVR

, etc. The transitive verbs must include causative morphology. There are several possible analyses of how this is realized.

I will outline a few here. First, it is possible that the causative affix is null, and that in the environment of this null causative the Roots in question exhibit a stem alternation with vowel epenthesis. Thus, √A

YR

ayır-

/_

CAUS

. Another, rather ethereal, possibility is that the causative morpheme is realized by abstract syllabicity, a sort of disyllabic template that the Root expands in order to fill. A third possibility is that the appended /

I

/ is itself the exponent of the causative, which is idiosyncratically realized as an infix. I will not decide on a particular analysis here (though I will note that the latter two would be typologically anomalous for Turkish). The important point is that the epenthetic vowel is indicative of the presence of a causative morpheme. If this is correct, then it should be possible to passivize these verbs without deleting the vowel, and the diagnostics should yield different results than when the vowel is deleted. Judgments on this point are delicate; in normal speech, the distinction between the two is likely to be lost.

Furthermore, speakers appear not to have a strict mapping of the different orthographies to the different pronunciations, as they may see a form with /I/ but pronounce it without.

This may be because the vowel-retaining variants, not being prescriptively sanctioned or listed in dictionaries, are infrequent in writing. Nevertheless, when a form such as

ayırıl-

is clearly and deliberately pronounced as three syllables, the speakers I have consulted have a clear intuition that it is passive.

156

(55)

Dövüş-en adam-lar polis tarafından ayır-ıl-dı.

fight-

REL man-

PL police.officer by separate-

IL

-

PAST

‘The fighting men were separated by the police officer.’

As for the carefully pronounced disyllabic stem

ayr-ıl-

, one speaker said it would be acceptable with the

by

-phrase if it were spoken ‘quickly,’ while another rejected it outright.

(56) */?

Dövüş-en adam-lar polis tarafından ayr-ıl-dı.

fight-

REL man-

PL police.officer by

‘*The fighting men separated by the police officer.’ separate-

IL

-

PAST

The remark about saying it ‘quickly’ suggests that this is felicitous only if interpreted as a rapid-speech version of

ayırıldı

, with a trisyllabic stem, and that indeed

ayrıl-

is an inchoative while

ayırıl-

is a passive.

The pattern exhibited by these verb types is summarized in Table 5. Null morphemes are not glossed.

157

Table 5.

Variable morphological marking of causative, anticausative, and passive.

Causative

boğ-

√-

‘drown (tr.)’

öl-dür-

√-v.

CAUS

-

‘kill’

kap-a-

√-v.

kap-a-t-

CAUS

tük-e-t-

-

√-v-

CAUS

-

‘close (tr.)’

√-v-

CAUS

-

‘use up’

kurt-ar-

√-v.

CAUS

-

‘rescue’

ayır-

√.(v.

CAUS

)

‘separate (tr.)’

Anticausative

boğ-ul-

√-

IL

-

‘drown (int.)’

öl-

√-

‘die’

kap-a-n-

√-v-

IL

-

‘close (int.)’

tük-e-n-

√-v-

IL

-

√-

IL

-

‘run out’

kurt-ul-

√-

IL

-

‘get free’

ayr-ıl-

‘separate (int.)’

Passive

42

boğ-ul-

√-

IL

-

‘be drowned’

öl-dür-ül-

√-v.

CAUS

-

IL

-

‘be killed’

kap-a-t-ıl-

√-v-

CAUS

-

IL

-

(

kap-a-n-

)

(√-v.

CAUS

-

IL

-)

‘be closed’

tük-e-t-il-

√-v-

CAUS

-

IL

-

‘be used up’

kurt-ar-ıl-

√-v.

CAUS

-

IL

-

‘be rescued’

ayır-ıl-

√.(v.

CAUS

)-

IL

-

‘be separated’

This should suffice to show that, even though their exponents are syncretic, Turkish has both an anticausative and a passive, which are distinguishable from one another.

Furthermore, whenever this syncretic morpheme, –

Il

, attaches outside an overt causative morpheme, the verb is passive, never inchoative. From this it can reasonably be inferred that, in forms without overt causative morphology such as

boğ-

/

boğ-ul-

, the antipassive does not embed a null causative. The alternative would be to claim that the antipassive can embed a causative

just in case

that causative is null. This would be the exact inverse

42

Again, although many of the English glosses have adjectival passive readings available, the Turkish forms are unambiguously eventive passives.

158 of Myers’ Generalization. Conversely, it can be inferred that in such forms, the passive does embed a null causative morpheme.

4.4 The Alternation in Turkish

We are finally in a position to analyze the causative/inchoative alternation in

Turkish. The empirical basis for this analysis is from Key (2012b), which presents a dataset comprising 146 alternating pairs, or 292 verbs in total. Key (2012b) groups the verb pairs according to surface alternation type, and subgroups them according to the actual suffixes employed to effect the alternation. In the present study, I evaluate these pairs more closely to see which morphemes are embedded under the alternating suffixes, and attempt to establish their underlying structure. Hence this is a more fine-grained analysis than Haspelmath (1993) carries out for any of his 21 languages, as he evaluated them only with respect to morphemes that are added or that alternate to produce the alternation, and not with respect to which morphemes are embedded under them.

In the following, I show that there are two structurally distinct types of inchoative verb stems: middle Voice inchoatives (4.4.1), and v

BECOME

inchoatives (4.4.2). The causative alternant of a middle Voice inchoative is a Root causatives, while the causative of a v

BECOME inchoative actually embeds the vP headed by v

BECOME

.

4.4.1 Middle Voice Inchoatives

In the preceding section, it was shown that the ‘anticausative’ suffix –

IL

(with realizations

–Il

,

–In

, and

–n

) never embeds a causative suffix. When this VI (with the same realizations) does embed a causative suffix, the result is passive rather than inchoative. Examples of the anticausative alternation from Key (2012b) are given below.

159

(57) a. b. c. causative

aç- boz- bur- dök- eğ- göm- kır-

üz- yor- böl- kap-a- giz-le- sak-la- bağ-la- top-la- parça-la- kilit-le- yara-la-

inchoative

aç-ılboz-ulgöm-ülburk-uldök-üleğ-ilkır-ıl-

üz-ülböl-ün yor-ul-

-

kap-a-ngiz-le-nsak-la-nbağ-la-ntop-la-n-

‘open’

‘ruin’

‘sprain’

‘pour, spill’

‘bend’

‘bury’

‘break’

‘sadden’

‘tire’

‘divide’

‘close’

‘hide’

‘hide’

‘attach’

‘gather’

parça-la-n-

‘shred/shatter’

kilit-le-n-

‘lock’

yara-la-n-

‘injure’

Note that when the phonologically conditioned variant –

n

appears, a verbalizing suffix intervenes between it and the Root. In (57c), the suffix is usually –

lA

, and in one case –

A

(

kap-a-

), both of which were shown in 2.4.3.2 to be underspecified verbalizers having the feature [v]. This structure makes sense if the anticausative suffix is really a realization of middle Voice.

43

Voice is a functional category in a verbal projection. In other words, it merges with a vP. It is not itself a verbalizer. The derivation of the verb

kap-a-n-

up to the Voice projection is shown in (58) (‘middle Voice’ is glossed as

MID

).

(58) a.

Kapı kap-a-n-dı.

door

√-v-

MID

-

PST

‘The door closed.’

43

Note that this is the inverse of Alexiadou et al.’s (2006) proposal that anticausatives have v

CAUS

but no Voice projection. The Turkish evidence indicates that anticausatives have a (middle) Voice projection but no v

CAUS

.

160 b.

DP

kapı

√P

VoiceP

vP

K

AP

v

–A

Voice

MID

–n

Note that the little-v in (58) is notated not as v

BECOME

, but as plain (unflavored) little-v.

This is intentional. Of course, the fact that a little-v node is realized by a VI not specified for flavor does not mean that the node lacks a flavor feature. By the Subset Principle, an underspecified VI could very well realize v

BECOME

. However, this is a pattern that is never seen overtly. Although Turkish has a healthy inventory of VIs with the features [v,

BECOME

] (see 4.4.2 following), only the underspecified verbalizers intervene between the

Root and the middle Voice head.

In the cases where middle Voice has the variant –

Il

(–

In

after /l/), there is no overt verbalizer, yet a null one can be assumed to be present. First of all, Voice merges with vPs, not with Roots. Secondly, cases such as

kap-a-n-

provide overt evidence of this structure. The fact that we only see it with –

n

is an artifact of the phonological environment: The most common underspecified verbalizer in Turkish in –

lA

(forming

957 verbs in N&Ü), and the second most common is –

A

(forming 52 verbs). Both of these end in vowels, providing the environment for the –

n

variant. There is one underspecified verbalizer that ends in a consonant, –

Ar

, so in principle it should be possible to find an anticausative in –

Ar-Il

-, but there seem not to be any. This is perhaps not so surprising, given that this suffix forms a total of only 26 verbs in N&Ü.

161

We can therefore conclude that an inchoative verb such as

aç-ıl-

‘open’ has the same structure as

kap-a-n-

:

(59) a.

Kapı aç-Ø-ıl-dı.

door

√-v-

MID

-

PST

‘The door opened.’ b.

DP

kapı

√P

VoiceP

vP

v

–Ø

A

Ç

Voice

MID

–Il

This structure is also found in the inchoative member of superficially equipollent alternations. Most of these are –

t

/–

n

alternating verbs (see Chapter 3, Section 3.4).

(60) causative

ıs-la-t- kir-le-t- pis-le-t- ıs-ı-t44 tük-e-t- yıpr-a-t-

inchoative

ıs-la-nkir-le-n-

‘get wet’

‘get dirty’

pis-le-nıs-ı-n-

‘get filthy’

‘heat’

tük-e-nyıpr-a-n-

‘exhaust’

‘wear out’

In Chapter 3, the causative members were argued to exhibit

CAUS fission. As for the inchoative member, it is natural to conclude that they have the same structure as

kap-a-n-

(58) and the other verbs discussed herein. First of all, we see the same array of morphemes overtly in these verbs, where a null verbalizer (–

lA

in the first three cases) is followed by –

n

. Secondly, all of these inchoative verbs fail to causativize, regardless of whether they are part of an overtly anticausative or equipollent alternation.

44

An alternative analysis is that the Root is √I

SI

and that the verbalizer is null. Similar alternatives are possible for the stems

tüke-

and

yıpra-

.

162

(61) inchoative

aç-ılkap-a-nıs-la-n-

‘open’

‘close’

‘get wet’ causativized form

*aç-ıl-t-

*kap-a-n-dır-

*ıs-la-n-dır-

In Chapter 5, I argue that the Turkish causative cannot embed Voice. If the inchoative suffixes in (60) as well as those in (57) are all realizations of Voice, their failure to causativize falls out naturally.

We now turn to inchoative verb stems with no overt suffixes. In most cases, these alternate with Root causatives, with one of the idiosyncratic VIs –

Ir

, –

It

(not shown), or –

Ar(t)

(62a) or with the Elsewhere form –

DIr

(62b).

(62) a. b. causative

art-ır- bat-ır- bit-ir- doğ-ur- doy-ur- düş-ür- kaç-ır- piş-ir- şiş-ir- taş-ır- yat-ır-

çık-ar- kop-ar-

çök-ert- göç-ert- din-dir- dol-dur- don-dur- kal-dır-

öl-dür- sön-dür- sus-tur- dön-dür- dur-dur- ol-dur- sap-tır- yat-

çıkkop-

çökgöç- dindoldonkalk-

öl-

inchoative

artbat- bitdoğdoy- düş- kaç- piş- şiş- taşsön- susdöndurolsap-

‘increase’

‘sink’

‘finish’

‘be born’

‘satiate’

‘fall’

‘flee’

‘cook’

‘swell’

‘overflow’

‘lie/lay’

‘emerge’

‘break off’

‘collapse’

‘cave in’

‘subside’

‘fill’

‘freeze’

‘rise’

‘die’

‘extinguish’

‘silence’

‘turn’

‘stop’

‘become’

‘turn’

163

In these cases, there is less positive evidence, but here too it is reasonable to infer that these ‘simple’ inchoatives have the same functional structure as

kap-a-n-

: an unflavored verbalizer, and middle Voice (both null in these cases). There are several reasons for this inference. First, they are found in alternation with Root causatives. In

Chapter 3, it was argued that –

t

/–

n

alternating causatives were also Root causatives, and the inchoative alternants of these verbs transparently manifest this structure. Furthermore, there are two cases where a standard (un-fissioned) Root causative alternates with an overt exponent of middle Voice.

(63)

–It

–Ar causative

dağ-ıt- kurt-ar-

inchoative

dağ-ılkurt-ul

-

‘scatter’

‘get free’

Thus, any time a Root causative is seen to alternate with overt inchoative morphology, it is always one of the configurations seen in the realization of middle Voice inchoatives.

Harley (1995, 2008) argued that Root-adjacent v

CAUS

in Japanese alternates with

Root-adjacent v

BECOME

, and that when the inchoative verb has no overt suffix, there is a null v

BECOME

. In Turkish there is no evidence of such an alternation. All of the overt evidence is indicative of middle Voice, and not v

BECOME

. Also, as predicted by this analysis, the inchoative member cannot be causativized.

(64) inchoative

dağ-ılkurt-ul-

causativized form

*dağ-ıl-t-

*kurt-ul-t-

Note further that a causative morpheme never alternates with any of the fully specified v

BECOME

morphemes, such as –

, –

Al

, or –

Ik

(which are the topic of the following section).

‘scatter’

‘get free’

164

4.4.2 v

BECOME

Inchoatives

Turkish also has an inventory of suffixes that realize v

BECOME

: –

, –

Ik

, –

Al

, –

lAş

, and –

lAn

45

(this list may not be exhaustive). The causative alternant always has a causative suffix added to the v

BECOME

suffix. This immediately distinguishes them from the middle Voice suffix.

(65)

–Iş causative

değ-iş-tir- dön-üş-tür- ol-uş-tur- sık-ış-tır- bur-uş-tur- bar-ış-tır-

inchoative

değ-işdön-üş- ol-uşsık-ışbur-uşbar-ış-

‘change’

‘metamorphose’

‘form’

‘jam’

‘crumple’

‘make peace’

–Ik

–Al

ac-ık-tır- bir-ik-tir- alç-al-t- az-al-t- boş-al-t-

çoğ-al-t- düz-el-t- kıs-al-t- yüks-el-t- ac-ıkbir-ikalç-al-

46 az-al- boş-al-

çoğ-aldüz-el- kıs-al- yüks-el-

‘get hungry’

‘accumulate’

‘lower’

‘decrease’

‘empty’

‘increase’

‘straighten’

‘shorten’

‘rise’

–lAş

47

buhar-laş-tır- yer-leş-tir- zor-laş-tır- yak-laş-tır- bir-leş-tir- fena-laş-tır- gerçek-leş-tir- iyi-leş-tir- buhar-laş- yer-leş- zor-laşiyi-leş- yak-laş- bir-leşfena-laş- gerçek-leş-

‘vaporize’

‘settle’

‘get difficult’

‘approach’

‘unite’

‘deteriorate’

‘realize’

‘improve’

45

Of these, –

and –

lAş

are homophonous with reciprocal suffixes, while –

lAn

is homophonous with reflexive (as well as passive). Whether these are the same VIs or merely homophones is an interesting question that is not addressed here.

46

Another possible morpheme division is

alça-l-

. A similar point holds for

kısal-

and

yüksel-

.

47

As –

lAş

and –

lAn

are highly productive suffixes (forming 615 and 362 verbs, respectively, in N&Ü), there is the possibility that some of the bases are categorial words (nouns and adjectives) rather than Roots.

Furthermore, it is possible that these suffixes decompose as –

lA-ş

and –

lA-n

.

165

kolay-laş-tır- kötü-leş-tir- uzak-laş-tır-

–lAn

bu-lan-dır- ayak-lan-dır- ev-len-dir- hasta-lan-dır- umut-lan-dır- yaş-lan-dır- neşe-len-dir- onur-lan-dır-

öfke-len-dir- hüzün-len-dir- kaygı-lan-dır- kolay-laşkötü-leşuzak-laşbu-lan- ayak-lanev-len- hasta-lan- umut-lan- yaş-lan- neşe-len- onur-lan-

öfke-len- hüzün-len- kaygı-lan-

‘facilitate’

‘deteriorate’

‘distance’

‘become turbid’

‘revolt’

‘marry’

‘get sick’

‘get hopeful’

‘age’

‘cheer up’

‘honor’

‘enrage’

‘sadden’

‘get anxious’

As already mentioned, these suffixes never participate in the superficial anticausative alternation. In other words, there are no verb pairs such that the inchoative member has one of these suffixes and the causative member overtly differs only in that in lacks this suffix. Instead, the causative member always has a causative suffix that embeds the inchoative. There are rare instances where the same Root appears in another formation, but the semantics indicate that the two verbs do not belong to the same pair.

For example, the Root of

bur-uş-

‘become wrinkled’ is seen with null morphology as a transitive verb:

bur-

‘wring, twist.’ However, rather than forming an anticausative alternation, these belong to two separate pairs derived from the same Root.

(66) a. b. transitive

bur-

‘wring, twist’

bur-uş-tur-

‘wrinkle’ intransitive

bur-ul-

‘become twisted’

bur-uş-

‘become wrinkled’

Anticausative

Causative

It is apparent here that the combination of the Root √B

UR

with the inchoative –

yields the semantics of ‘wrinkle,’ which are preserved when the causative is added to this stem

166

(cf. Arad 2003). In the absence of this suffix, different semantics are picked out from the same Root (66a).

There are also cases where the VI –

Il

behaves like v

BECOME

, admitting causativization with –

t

.

(67) causative

ay-ıl-t- bay-ıl-t- dar-ıl-t- yan-ıl-t-

inchoative

ay-ıl- bay-ıldar-ılyan-ıl-

‘bring/come to’

‘faint’

‘offend/be offended’

‘lead/go astray’

In these cases, the base to which –

Il

is affixed shows signs of being a Root. When it occurs without overt suffixation, the meaning is substantially different:

ay-

is an intransitive verb meaning ‘come to one’s senses’ rather than ‘come to (i.e., regain consciousness);

bay-

means ‘cloy, bore to tears’;

dar

is an adjective (not a verb!) meaning ‘narrow’; and

yan-

means ‘burn (int.).’ The last of these is very likely not the same Root, and hence the Root of

yan-ıl-

‘go astray’ does not occur with null morphology. Thus it appears that the same VI that realizes the middle Voice head can also realize v

BECOME

in a Root-adjacent position. For this to be so, by the Subset

Principle the VI –

Il

cannot have any features that are not common to both positions. I propose that the common feature shared by these positions, for which –

Il

is specified, is

BECOME

.

The differences between the two groups of suffixes are summarized below.

(68) High-attachment inchoatives

Are realized by the same VI as the passive: –

Il

and its phonologically conditioned variants –

n

(after a vowel) and –

In

(after /l/).

Cannot be causativized

Form ‘anticausatives’

167

Low-attachment inchoatives

Are realized by various allomorphs, including –

Il

, –

Al

, –

, –

Ik

,–

lAn

, –

lAş

Can be causativized

Do not form ‘anticausatives’

4.5 Conclusion

In this chapter, the causative/inchoative alternation was introduced, and various theoretical perspectives on the derivational relationship between the two members of each pair were presented. Of these, the only kind of derivation that presents a challenge to a syntactic model of word formation is anticausativization (also decausativization), in which the non-causative verb is derived from a causative via an operation that would appear to delete a

CAUS

operator. This would violate the principle of Monotonicity

(Koontz-Garboden 2009), according to which structure may be added to but not removed from an existing structure. Anticausative and equipollent alternations in Turkish were analyzed, and it was shown that anticausative/inchoative morphemes never embed causative morphemes. While this constitutes evidence against Koontz-Garboden’s proposal is that anticausativization is reflexivization, it provides another understanding of apparent anticausative alternations that is compatible with syntactic word building.

However, it remains to be seen whether this can be extended to other languages.

The relationship of the causative to Voice is explored in the next chapter.

168

CHAPTER 5: VERB-SELECTING CAUSATIVES

5.1 Introduction

Verb-selecting causatives, as the name straightforwardly indicates, select for a vP, the layer above a √P. This is a rather new characterization, deriving from the work of

Pylkkänen (2002, 2008), who argues that the head that introduces the external argument,

Voice, is distinct from the verbalizing head, little-v. This more fine-grained approach is effective in capturing cross-linguistic variation in productive morphological causatives.

In languages such as Japanese, productive causatives have biclausal properties not found in other languages, such as Turkish and Hungarian. In this chapter, I apply this recent theoretical development to causatives of the Turkish/Hungarian type.

In section 5.2.1, I present data on Japanese causatives, and show how such facts are captured by Harley (2008) using a version of vP that introduces an external argument in its specifier. In 5.2.2, I show how this analysis fails to make the right predictions for

Turkish causatives. In 5.2.3, I discuss Pylkkänen’s proposal of separating Voice from little-v, and in 5.3 I present my own analysis of verb-selecting causatives, making use of this development. In 5.4, I present an alternative, the lexical account of such causatives advanced by Horvath & Siloni (2011a). The syntactic and lexical approaches, though radically different from a theoretical standpoint, to a large extent have the same empirical coverage. Where they differ most significantly is in the status of the causee, which is the topic of 5.5. In that section, I show that Horvath & Siloni’s lexicalist analysis cannot account for animacy restrictions on causees, but that the syntactic analysis advanced herein can. Section 5.6 concludes.

169

5.2 Causatives and the Little-v Hypothesis

Ever since Kuroda (1965), the literature on Japanese causatives has had to address systematic differences between so-called ‘lexical’ and ‘syntactic’ causatives.

Superficially, this looks like the distinction between Turkish Root causatives and productive causatives. Japanese lexical causative suffixes exhibit idiosyncratic allomorphy and occasionally marked semantic drift; syntactic causatives are always realized with the suffix –(

s

)

ase

, which also realizes the lexical causative with certain

Roots, and they are not associated with semantic drift. However, Japanese syntactic causatives also exhibit a range of syntactic phenomena that their Turkish counterparts do not. The existence of such phenomena, which systematically distinguish lexical from syntactic causatives in Japanese, and their absence in Turkish have shaped the debates concerning causatives very differently in the respective languages. In this section, we examine these differences in Japanese, and present the analysis of Harley (1995, 2008), which exploits the little-v hypothesis to account for them.

5.2.1 Japanese Productive Causatives

A range of diagnostics detect two clauses, and in particular two Agents, in

Japanese syntactic causatives, but only one clause in lexical causatives. These include agent-oriented adverbials (Shibatani 1972), Binding Condition B (Miyagawa 1984), coordination (Kuroda 2003), and scope of negation (Hara 1999).

In the Japanese syntactic causative construction, an adverbial phrase that modifies an Agent can modify either the causer subject or the causee (realized with dative marking, as in Turkish) (Shibatani 1972).

170

(1) a. Hanako-wa arui-te it-ta

Hanako-Top walk-

te

‘Hanako, walking, went.’

Taroo-wa goarui-te Hanako-o

PST ik-ase-ta b.

Taroo-Top walk-

te

Hanako-

ACC

go-

sase

-

PST

Readings: ‘Taro made Hanako go, walking’

‘Taro, walking, made Hanako go’ (Harley 2008: 30)

With the syntactic causative in (1b), the agent-oriented –

te

adverbial can be interpreted with either the causer argument or the dative causee. This indicates that both causer and causee are agents, which would incur a Thematic Diversity (Pesetsky 1995: 62) violation if both belonged to the same predicate. The logical conclusion is that there are two predicates, or clauses, and that the two agents belong to distinct predicates: the causer to the

CAUS

predicate, and the causee to the embedded predicate.

This contrasts with the Root or lexical causative in (2b).

(2) a. Hanako-wa nure-te hi-e-ta

Hanako-T wet-

te

cool-

inch

-

PAST

‘Hanako (’s body), getting wet, cooled.’ b. Taroo-wa

Taro-T nure-te wet-

te

Hanako-o

‘Taroo, getting wet, cooled Hanako.’ hi-(y)as-ita

Hanako-

ACC

cool-

caus

-

PAST

Reading:

Impossible: ‘Taroo cooled Hanako, (Hanako) getting wet.’ (Harley 2008: 30)

Despite the fact that it makes much more pragmatic sense for the causee to cool off by getting wet, this reading is unavailable. If the Japanese syntactic causative consists of two predicates, then the lexical causative consists only of one.

A variety of other tests detect two predicates in the productive causative. The productive causative construction also behaves like two binding domains rather than one

(Miyagawa 1984).

171

(3) a. Toru i

-wa Kitahara j

-ni kare* i

/*

j

-o syookai

Toru-

TOP

Kitahara-

DAT

he-

ACC

‘Toru introduced him to Kitahara.’ si-ta. introduction do-

PAST b. Toru i

-wa [Kitahara j

-ni kare i

/* j

-o syookai s]-ase-ta.

Toru-

TOP

Kitahara-

DAT

he-

ACC introduction do-

CAUS

-

PAST

‘Toru made Kitahara introduce him.’(Horvath & Siloni 2011)

Likewise, evidence from coordination (Kuroda 2003) detects two distinct verb phrases. Japanese causatives allow multiple caused events to be coordinated below a single causative suffix. In the following example, -

sase

scopes over two disjunctively coordinated caused events: ‘clean the house’ and ‘pay room rent.’ Hanako will make

Masao do one or the other; Masao may choose either of them.

(4) Hanako-ga

Hanako-

NOM

[[ Masao-ni uti-o

Masao-

DAT haraw]]-aseru kotoni si-ta house-

ACC soozisuru]-ka [ heya-dai-o clean-or room-rent-

ACC pay-

CAUS that to.do-

PAST

‘Hanako decided to make Masao clean the house or pay room rent.’

Reading: -(

s

)

ase

scopes over ‘or’; Masao has a choice. (Kuroda 2003: 455)

The successful coordination of caused events is no surprise if causatives are syntactic:

–(

s

)

ase

scopes over syntactic constituents, so coordination of these should be possible.

Finally, the verbal negation suffix may either follow or precede the causative suffix, with a corresponding difference in scope.

(5) a. Toru-wa Yoko-o ik-ase-nakat-ta.

Toru-

TOP

Yoko-

ACC

‘Toru did not make Yoko go.’ go-

CAUS

-

NEG

-

PAST b. Toru-wa Yoko-o ik-anaku-sase-ta.

Toru-

TOP

Yoko-

ACC go-

NEG

-

CAUS

-

PAST

‘Toru make Yoko not go.’ (Hara 1999, cited in H&S)

When the negative suffix follows the causative, the causing event is negated: Toru did not make Yoko go. When the negative suffix precedes the causative, it negates the caused

172 event: Toru made Yoko not go. These scope alternatives indicate that there are two predicates, since either can be negated.

Thus, various tests show that syntactic causatives in Japanese are biclausal. The same tests detect only a single predicate in lexical causatives. This split behavior requires distinct analyses for the two kinds of causatives. Indeed, the names they traditionally go by reflect such a split derivation: lexical causatives are derived in the lexicon, and syntactic causatives in the syntax. In the lexicon, there is assumed to be no hierarchic structure, which would explain why lexical causatives are a single binding domain, and why coordination and negation cannot target the caused event, but only the causing event.

In the early decades of generative grammar, there was little option for analyzing

Japanese lexical causatives other than to derive them in the lexicon. However, the latter years of the twentieth century saw a revolution in the decomposition of verbal structure.

Marantz (1984) proposed that all predicates decompose into a non-verbal, or noncategorial, part, and a verbalizer. Hale & Keyser (1993, 2002) proposed a functional head v°, whose projection hosts the external argument in its specifier, while Kratzer (1996) argued that the external argument is not part of the verb’s thematic/semantic representation, but is introduced by a distinct functional projection, Voice. These proposals are sometimes conflated into a single projection. Little-v, the functional head that merges with VP (or, in DM, with √P), heads the projection vP, which hosts the external argument in its specifier (see discussion in Horvath & Siloni’s (2002) paper

‘Against the Little-v Hypothesis’).

173

Harley (1995, 2008), building on the proposal of Miyagawa (1994, 1998), exploits this theoretical development to capture the differences between lexical and syntactic causatives without resorting to separate derivational engines. In her work, the differences between the two types of causative all proceed directly from the attachment site of the little-v that introduces the causing event. In the productive causative construction (a.k.a. the syntactic causative), v° merges with a √P, creating a verb. It projects to the phrasal level, with an external argument in its specifier. Then another v°— this one of the variety, or ‘flavor’ (Folli & Harley 2005), ‘

CAUS

’—merges with the existent verbal structure. A second external argument is introduced in the specifier of the causative vP.

(6) a. vP2 ...

DP

Taroo

DP vP1

Hanako

√P

DP hanasi v’ v’ v° sase

√ tutae v°

Ø b. Taroo-wa Hanako-ni hanasi-o

Taroo-

TOP

Hanako-

DAT

story-

ACC

‘Taroo made Hanako convey a story.’ tutae-sase-ta. convey-

CAUS

-

PST

(Harley 2008)

The biclausal properties of the productive causative are immediately captured. The two agent positions explain the ability of agent-oriented adverbials to associate with either argument. Furthermore, if vP is a phase (Chomsky 2001), the presence of two of them

174 explains why there are two binding domains. Finally, there are also two separate vPs that can be targeted by coordination and negation, giving rise to alternative scopal possibilities.

The so-called ‘lexical’ causatives are also built in the syntax. The differences between them and the productive causatives are captured in a different way than a lexicon/syntax dichotomy: They are built out of a single vP, whose head is merged directly to the √P with no intervening functional heads. The following example is a simplified version of the sentence in (2) from Harley. vP (7) a.

DP

Taroo

DP

Hanako

√P v’

√ hi v°

CAUS

-as b. Taroo-wa Hanako-o hi-(y)as-ita.

Taroo-

NOM

Hanako-

ACC

cool

-

CAUS

-

PAST

‘Taroo cooled Hanako.’

There is only one vP and one external argument. The causee here is an internal argument, which is why it cannot control agent-oriented adverbials. Likewise, a single vP means a single phase, hence one rather than two binding domains. Coordination and negation also have only one verbal domain to scope over. Finally, the idiosyncratic realization of the causative morpheme (the allomorph –

as

rather than the Elsewhere form –

sase

) is due to the adjacency of little v-

CAUS

to the Root.

175

All the phenomena of so-called lexical and syntactic causatives are neatly captured in a unified syntactic account. Unfortunately, there are other affixal causative languages that throw a spanner into these elegant works. One such language is Turkish.

5.2.2 Turkish Productive Causatives

The Turkish morphological patterns and their correlation to verb types ostensibly suggest a causative system like that of Japanese. Unaccusatives, among other types, have a causative counterpart that exhibits idiosyncratic allomorphy. These are the Root causatives of Chapter 3. One of the Root-causative allomorphs is –

DIr

, which, like

Japanese –(

s

)

ase

, is also used to causativize unergatives and transitives. Apart from the additional complication of the –

DIr

/–

t

allomorphy, the picture looks very much like

Japanese. The Root-causatives pattern like Japanese lexical causatives, which in Harley

(2008) are also analyzed as Root causatives.

The problem lies with the Turkish productive causatives: On the clausality tests, they pattern with the Root causatives, showing evidence of only one predicate. As in

Japanese, when the non-causative counterpart is transitive, the causee of the causative version is marked with dative case. However, unlike Japanese, the Turkish dative causee cannot control an agent-oriented adverbial.

(8)

(9)

Tarkan Hakan-a Mehmet-i bil-erek

Tarkan Hakan-

DAT

Mehmet-

ACC

know-

PART

‘Tarkan made Hakan beat Mehmet on purpose.’

(‘on purpose’ must refer to Tarkan, not Hakan) döv-dür-dü. beat-

CAUS

-

PAST

Similarly, the Condition B facts indicate a single binding domain. a. Hakan

i on-u* i

Hakan 3

SG

.

ACC

‘Hakan beat him.’ döv-dü beat-

PST

176 b. Tarkan i

Tarkan

Hakan-a j

Hakan-

DAT on-u * i.

/* j

3

SG

‘Tarkan made Hakan beat him.’ döv-dür-dü beat-

CAUS

-

PST

This shows that the subject and the object are clause-mates, and hence that these constructions comprise a single clause.

Coordination of caused events is simply not possible.

(10) *Hakan Mahmut-a ev-i temiz-le-

Hakan

öde-t-me-ye

Mahmut-

DAT

house-

ACC karar ver-di. clean-v- veya kira or rent pay-

CAUS

-

NOM

-

DAT

decision give-

PAST

Intended: ‘Hakan decided to make Mahmut clean the house or pay rent.’

Since the construction is completely ungrammatical, the issue of possible scope readings does not even come into play.

The only test that appears to diagnose two predicates is scope of negation. When a single verb stem has a negative and a causative morpheme, the ordering is strict, such that the negative must follow the causative.

(11) a. b.

çalış-tır-ma- work-

CAUS

-

NEG

-

‘not cause to work’

*çalış-ma-dır-/*çalış-ma-t- work-

NEG

-

CAUS

-

‘cause not to work’

By the Mirror Principle (Baker 1985), the suffix ordering

CAUS

-

NEG

(11a), the only allowable one, should yield only the reading where the negative scopes over the causative. However, contrary to expectation, there is an apparent scope ambiguity, with two readings available.

177

(12) a. b.

Mehmet

Mehmet

Ayşe-yi

Ayşe-

ACC

çalış-tır-ma-dı. work-

CAUS

-

NEG

-

PAST

Mehmet didn’t make Ayşe work.

Mehmet made Ayşe not work (i.e., didn’t let her work).

The reading in (12b) is a surprise. First of all, it appears to be a violation of the Mirror

Principle (Baker 1985), since the causative apparently has scope over negative, contrary to the suffix ordering. Furthermore, it appears that the causing event is negated in (12a) but the caused event is negated in (12b), as with the Japanese syntactic causative in example (5) (but without the change in suffix ordering). Even stranger, this apparent scopal ambiguity does not distinguish productive causatives from Root causatives. We see the same two types of readings with the Root causative

çık-ar-

‘take out.’

(13) Baba-mız dün biz-i hiç dışarı çık-ar-ma-dı. father-1

PL a. yesterday 1

PL

-

ACC at.all outside √-v.

CAUS

-

NEG

-

PAST

‘Our father didn’t take us outside at all yesterday. b.

‘Our father made us not (i.e., didn’t let us) go outside at all yesterday.’

This poses a puzzle for any account of the differences between causative types, be it lexical, syntactic, or mixed. In addition to that, it calls into question the claim about Root causatives, laid out in Chapter 3, that the complement of the causative is a categoryneutral Root phrase rather than a verb phrase. On this approach, the Root does not have a verbal function or specific semantics unless it composes with a verbalizing head, so in

(13b), where the causative scopes over the negative, negation syntactically and semantically separates

CAUS

from the Root, which should not permit any kind of coherent composition. This is more than just a Mirror Principle violation.

The same holds for that rarest of cases in Turkish, the suppletive causative. The causative counterpart of the unaccusative verb

gir-

‘enter’ is

sok-

‘insert, put inside.’

178

(14) Baba-mız dün biz-i hiç içeri sok-Ø-ma-dı. father-1

PL a. b. yesterday 1

PL

-

ACC at.all inside √-v.

CAUS

-

NEG

-

PAST

‘Our father didn’t bring us inside at all yesterday.

‘Our father made us not (i.e., didn’t let us) come inside at all yesterday.’

This perplexing issue is addressed again and resolved in section 5.5. For the time being, it is sufficient to note that, in Turkish, none of the clausality tests distinguish between productive and Root causatives.

In sum, the distribution of allomorphs groups them with Japanese syntactic causatives, while syntactic tests group them with lexical causatives. These facts suggest that they belong somewhere in-between, separated from the Root but comprising less structure than Japanese productive causatives; however, the Harley (2008) model does not leave much room to maneuver between the inner and the outer vP.

5.2.3 Severing the External Argument from its Light Verb

If a latter 20th-century theoretical development made it possible to accommodate so-called lexical/syntactic causatives of Japanese uniformly in the syntax, then an early

21st-century development enables us to reconcile ‘in-between’ causatives of the Turkish type with a fully syntactic approach. Pylkkänen (2002, 2008) proposes that the external argument-introducing head, Voice, is distinct from the verbalizing head, little-v. One type of evidence comes from so-called unaccusative causatives: Root causatives that lack an external argument, as in the Finnish desiderative causative.

(15) Maija-a laula-tta-a.

Maija-

PART sing-

CAUS

-3

SG

‘Maija feels like singing.’ (Pylkkänen 2008: 95)

Although the verb has the causative suffix –

tta

, the event is intransitive. Furthermore, the partitive marking on the subject indicates that it is an internal argument. Direct objects

179 and the subjects of passives are marked with the partitive in Finnish when the event is atelic. External arguments, too, may have partitive marking, but only if they are plural or mass (Pylkkänen 2008). The partitive marking on the singular count subject of (15), then, shows that it is an internal argument. Thus, the verb has a causative suffix, while the sentence has an internal argument but no external argument. Pylkkänen takes this to be evidence that the projection that introduces external arguments is distinct from the causative, and from the verbalizer in general.

Harley (2013) also adopts the split between vP and Voice, presenting evidence from the interaction of causative and applicative morphology in Hiaki. The causative and applicative morphemes are both suffixes on the verb stem. In the configuration verb-

CAUS

-

APPL

, then, the causative is more deeply embedded than the applicative. On the assumption that the causative introduces an external argument and the applicative an applied argument, by the Mirror Principle (Baker 1985), the latter should precede the former, as in the schema below.

(16) [appl-arg [ext-arg [verb-obj]-caus]-appl]

However, this prediction is falsified: The external argument precedes the applied argument.

(17) Nee usi-ta avion-ta ni’i-tua-ria-k

I child-

ACC plan-

ACC fly-

CAUS

-

APPL

-

PRF

‘I made the (model) plane fly for the child.’

Harley concludes that the external argument is not introduced by the vP, but by a higher projection, Voice. Manetta (to appear) reaches a similar conclusion based on evidence from Kashmiri causatives.

180

Having severed the external argument from the vP, Pylkkänen lays out a typology of possible causative types that vary in two dimensions: 1. whether they bundle Voice and

CAUS

in a single head, and 2. what type of complement they select for. This more finely articulated structure provides the room needed to distinguish three levels of causative: those that embed only the Root, those that embed little-v but not Voice, and those that embed Voice

48

as well as little-v. The differences between productive causatives in Japanese and Turkish can now be captured: Japanese causatives are phaseselecting (i.e., they embed an external argument), and Turkish causatives are verbselecting (they do not).

Drawing on data from Bantu languages, Pylkkänen shows that these two types of causative show differences in terms of the ability of the causee to control agent-oriented adverbials. Venda and Luganda pattern with Japanese.

(18)

Venda

Muuhambadzi o-reng-is-a salesman 3

SG

.

PAST

-buy-

CAUSE

-

FV

Katonga

Katonga nga dzangalelo. with enthusiasm

‘The salesman made Katonga

buy the car eagerly

.’ modoro car

(19)

Luganda

Omusomesa ya-wantdi-s-a Katonga ne obu teacher 3

SG

.

PAST

-write-

CAUSE

-

FV

Katonga with the

‘The teacher made Katonga write

with dedication

.’

(Pylkkänen 2008: 119) nyikivu. dedication

These contrast with Bemba, which behaves like Turkish.

48

Or a high Applicative. Pylkkänen considers a phase to be defined by an external-argument– introducing head, which on her analysis includes the high Applicative as well as Voice.

181

(20)

Bemba

Naa-mu-fuund-ishya uku-laanda iciBemba

1

SG

.

PAST

-him-learn-

CAUSE a. to-speak Bemba

‘I,

on purpose, made

him learn to speak Bemba. b.

*‘I made him

on purpose learn

to speak Bemba.’

(Givón 1976, 329, (18); cited in Pylkkänen 2008: 115) ku-mufulo on-purpose

This is despite the fact that Bemba causatives allow adverbial modification of the embedded event.

(21) Naa-butwiish-ya Mwape ulubilo.

1

SG

.

PAST

-run-

CAUS a.

Mwape

‘I made Mwape

run quickly

.’ fast b.

*‘I

quickly made

Mwape run.’

(Givón 1976, 343, (120); cited in Pylkkänen 2008: 115)

It seems that the event embedded under the causative is generally available for adverbial modification independently of the causing event, but not for adverbials that modify an external argument. This suggests that no external argument is present beneath the causative. The argument that

is

there must be introduced in a different way. Pylkkänen leaves the details of this point for future work.

In Chapter 4, Section 4.4.1, it was proposed that certain inchoative suffixes in

Turkish are exponents of a middle Voice head. One of the diagnostics for this was their inability to be embedded by the causative. This falls out naturally if the Turkish causative cannot embed Voice. Similarly, the causative of a passive is impossible.

(22)

*Kadın et-i kasap tarafından kes-il-dir-di. woman meat-

ACC butcher by cut-

PASS

-

CAUS

-

PAST

‘The woman had the meat be cut by the butcher.’ (Aissen & Hankamer 1980:239)

In (22), the causative –

DIr

is attached to the passive morpheme –

Il

. Note, however, that following a liquid, –

t

and not –

DIr

is the allomorph used (see section 5.3.4). This turns

182 out to be irrelevant, however, as the form with –

t

is equally bad. Nor does omission of the passive agent by-phrase improve anything.

(23) *Kadın woman et-i meat-

ACC

(kasap butcher tarafından) by kes-il-t-ti. cut-

PASS

-

CAUS

-

PAST

Aissen & Hankamer conclude that causativization is a lexical operation, since it cannot be fed by passivization, a syntactic one. However, with the recent developments in syntactic theory outlined above, this conclusion is no longer forced. If the Turkish causative cannot embed Voice, it follows naturally that passives cannot be causativized.

5.3 The Structure of Verb-Selecting Causatives

With the separation of little-v and Voice, the structures proposed for causatives need to be modified accordingly. Root-selecting causatives, which by definition take only the Root as their complement, will not look any different, except that in non-Voice– bundling languages, vP-

CAUS

will be immediately dominated by a Voice projection.

Verb- and phase-selecting causatives, on the other hand, are more deeply affected by this theoretical development. I take up this matter in the following sections.

5.3.1 Proposal

Japanese productive causatives, being phase-selecting, are essentially the same as in Harley (2008), except that each causative is now dominated by a Voice projection, resulting in two extra layers.

183

(24) a. VoiceP

DP

Taroo

CAUS

P

Voice’

Voice-

ACTIVE

CAUS

’ b.

DP hanasai

√P

DP

VoiceP

Voice’

CAUS

–sase

Hanako vP Voice-

ACTIVE v’

√ tutae v-

CAUS

–Ø

Taroo-wa Hanako-ni hanasi-o

Taroo-

TOP

Hanako-

DAT

story-

ACC

‘Taroo made Hanako convey a story.’ tutae-sase-ta. convey-

CAUS

-

PST

Note that the high-attaching causative position is labeled

CAUS

and not v.

CAUS

, and that the phrase that it heads is labeled

CAUS

-P and not vP. There are several reasons for this.

One is that there is no longer need for verbalizing function at this level of structure. With the understanding of little-v that informs Harley (2008), each external argument, causer and causee, needed to be introduced in the specifier of a separate vP. On the view adopted in the present study, wherein external arguments are introduced by a Voice projection, the only vP that is strictly necessary is the first one, whose head merges with a

√P (or a categorial phrase in the case of denominal/deadjectival formations) to create the category ‘verb.’ While the new architecture does not strictly preclude the recursion of

184 vPs, it does not require it. A second reason has to do with patterns of allomorphy, discussed in detail in 5.3.2 below.

Verb-selecting causatives of the Turkish type should have a similar structure, minus the Voice layer between the causative and the vP.

(25) b.

Voice-P

DP kadın

CAUS

-P

Voice’

Voice-

ACTIVE

CAUS

’ vP v’

CAUS

–DIr

DP et

√P

K

ES v

CAUS

–Ø

Kadın woman et-i meat-

ACC kes-Ø-tir-di.

√-v.

CAUS

-

CAUS

-

PST

‘The woman had the meat cut/had (someone) cut the meat.’

Something is missing, however. With no Voice projection beneath the high-attaching

CAUS

head, the position of the dative causee (when expressed) needs to be addressed anew. Pylkkänen leaves this matter to future research. I propose that the dative causee in

Turkish is an adjunct hosted by

CAUS

-P.

185

(26) b.

Voice-P

DP kadın

CAUS

-P

CAUS

Voice’

Voice-

ACTIVE

Dative-P

Ekrem-e vP v’

√P

K

ES

CAUS

’ v

CAUS

–Ø

CAUS

–DIr

DP et

Kadın woman

Ekrem-e

Ekrem-

DAT et-i meat-

ACC kes-Ø-tir-di.

√-v.

CAUS

-

CAUS

‘The woman had the meat cut/had Ekrem cut the meat.’

-

PST

Introducing the causee as an adjunct rather than in Spec-

CAUS

P accounts for the fact that expression of the causee is optional: compare (25b) to (26b). Causee optionality is a different phenomenon from argument drop. Turkish allows arguments to be dropped if their referents can be identified from the discourse context. In the absence of contextual support, the subject cannot felicitously be dropped.

(27) #Kapı-yı door-

ACC anahtar-la key-with aç-tı. open-

PAST

Omission of the dative causee does not require contextual support. This was named the

‘missing subject construction’ by Aissen (1979).

(28) a. Hasan kutu-yu aç-tır-dı.

Hasan box-

ACC open-

CAUS

-

PAST

‘Hasan had the box opened.’

186 b.

Kadın et-i kes-tir-di. woman meat-

ACC cut-

CAUS

-

PAST

‘The woman had the meat cut.’ (Özkaragöz 1986)

An attempt still might be made to explain causee optionality as a special case of argument drop. In many cases, the missing causee is a professional whose job is to perform the caused action. Hence, if (28b) is uttered out of the blue, the most natural interpretation is that the woman had the meat cut by a butcher. The existence of such professionals can be assumed in many cases, and might be available as an implicit discourse referent. However, there are problems with such an explanation. Not all cases of missing causees involve such professionals. In (28a), there is no obvious professional entity charged with opening boxes for people. The natural interpretation of this sentence is that Hasan had

someone

open the box. (28a) might be paraphrased as in (29).

(29) Hasan bir-in-e kutu-yu

Hasan one-1

SG

-

DAT

box-

ACC

‘Hasan had someone open the box.’ aç-tır-dı.

√-

CAUS

-

PST

In (28a), might it be the vague causee ‘someone’ that is dropped? Does the vagueness of the referent license argument drop? This is unlikely, as it is not a general phenomenon.

When the phrase

biri

‘someone’ is the object of the postposition

ile

‘with,’ it cannot be dropped in an out-of-the-blue context.

(30) a. b.

Bir-i-yle tanış-tı-m. one-3

SG

=with meet-

PST

-1

SG

‘I saw someone.’

#Tanış-tı-m.

√-

PST

-1

SG

From these facts, I conclude that the causee appears as an adjunct dative phrase, and that when it is not expressed, it is absent from the syntactic structure.

187

Another fact about the dative causee is that it must be animate.

(31) a. b.

Müdür (Bekçi-ye) kapı-yı aç-tır-dı. director watchman-

DAT door-

ACC

‘The director made the watchman open the door.’ open-

CAUS

-

PAST

*Bekçi anahtar-a kapı-yı aç-tır-dı. watchman key-

DAT door-

ACC open-

CAUS

-

PAST

Intended: ‘The watchman made the key open the door.’

This contrasts with the subject of the corresponding non-causative verb. The subject can be animate (a) or inanimate (b).

(32) a. b.

Bekçi kapı-yı key door-

ACC

‘The key opened the door.’ anahtar-la watchman door-

ACC key-with

‘The watchman opened the door.’

Anahtar kapı-yı aç-tı. open-

PAST aç-tı. open-

PAST

It would be wrong to attribute this animacy restriction to the causee, because even when no causee is expressed, animacy/agency is implied. Thus, (28b) could not be interpreted as meaning that, for example, the woman made a knife cut the meat. I propose that the head of

CAUS

-P has the feature [m], for ‘mental state.’ I borrow this feature from

Reinhart’s (2002) theta system (discussed in Chapter 4, Section 4.2.2.1), where it indicates a role whose mental state is relevant to the event. It therefore implies animacy.

The fact, then, that an inanimate dative causee is infelicitous is due not to any property of the dative phrase itself, but to semantic incompatibility with the

CAUS

head.

We now return to an issue raised in Chapter 3, Section 3.3.4, regarding the argument structure of ingestive verbs. This is the only class of Root-causatives whose non-causative counterpart is transitive. The causee of these causatives is in the dative

188 case, like the causee of the

CAUS

P, and its expression is also optional (Kornfilt 1997). I therefore propose that it, too, is an adjunct dative phrase, this time hosted by vP-

CAUS

.

(33) a. Voice-P

Voice’

DP anne vP Voice-

ACTIVE

Dative-P bebeğine

√P

DP mama v’ v’

Y

E v

CAUS

–DIr b. Anne bebeğ-in-e mama ye-dir-di. mother baby-1

SG

-

DAT

mush √-

CAUS

-

PAST

‘The mother fed her baby baby food.’

The dative causee of ingestive verbs must be also be animate. This quite naturally derives from the semantics of ingestive verbs, and does not need to be enforced by an [m] feature in the v

CAUS head. The causee, as the ingestive location, must be an entity capable of consuming, learning, or wearing clothing and accessories.

The analysis of

CAUS

P proposed here can fruitfully be extended to the passive, as there are strong parallels. The agent of a passive in Turkish is expressed with the postpositional by-phrase

tarafından.

Although in general the subjects of transitive verbs may be either animate or inanimate, the argument of the by-phrase is obligatorily animate.

(34) a.

Kapı bekçi tarafından aç-ıl-dı. door watchman by open-

PASS

-

PAST

‘The door was opened by the watchman.’

189 b.

*Kapı anahtar tarafından aç-ıl-dı. door key by open-

PASS

-

PAST

Intended: ‘The door was opened by the key.’

As with the dative causee, the animacy requirement of the passive by-phrase cannot be traced to the by-phrase itself. A form such as

öl-dür-ül-

‘be killed’ implies that an agent did the killing, whether there is a by-phrase or not.

(35) a. b.

*Adam man deprem earthquake tarafından by

öl-dür-ül-dü.

√-v

CAUS

Intended: ‘The man was killed by an earthquake.’

-

PASS

-

PST

*/?Adam

man deprem-de earthquake-

LOC

öl-dür-ül-dü.

√-v

CAUS

-

PASS

Intended: ‘The man was killed in an earthquake.’

-

PST

The inanimate by-phrase in (35a) is strictly ungrammatical (unless ‘earthquake’ were somehow personified, as in a fairytale or myth). With

deprem

‘earthquake’ in the locative case in (35b), the sentence is felicitous only on the reading that someone murdered him during an earthquake. This is substantially different from the English passive; not only is the English version of (35a) perfectly grammatical, but the English of (35b) can be readily interpreted to mean that the earthquake was the cause of the man’s death. For that reading in Turkish, the inchoative verb

öldü

‘he died’ is required.

The Turkish passive facts follow naturally if the passive Voice head, like the

CAUS

head, has the [m] feature. The difference between the two is that the passive head, instead of having a [

CAUS

] feature, has a [

BECOME

] feature, which accounts for the syncretism between the passive and the anticausative VIs, which are specified for

[

BECOME

].

When the non-causative counterpart is unergative, the causee of the corresponding causative is marked accusative.

190

(36) a.

b.

Adam çalış-tı. man work-

PAST

‘The man worked.’

Patron adam-ı

çalış-tır-dı. boss man-

ACC work-

CAUS

-

PAST

‘The boss made the man work.’

Unlike the dative causee, the accusative causee is not optional. It also has no animacy restriction.

(37) a. b.

Araba çalış-tı. car work-

PAST

‘The car started.’

Mehmet araba-yı

Mehmet car-

ACC

‘Mehmet started the car.’

çalış-tır-dı. work-

CAUS

-

PAST

This is not merely a case of machines behaving like animate arguments in language

(noted in Reinhart 2002: 235). In the causative counterpart of a transitive verb, a machine is ungrammatical as the dative causee.

(38) a. b.

Araba kutu-yu ez-di. car box-

ACC smash-

PAST

‘The car smashed the box.’

*Mehmet araba-ya kutu-yu ez-dir-di.

Mehmet car-

DAT box-

ACC smash-

CAUS

-

PAST

Intended: ‘Mehmet made the car smash the box.’

Because the accusative causee is not optional and does not have an animacy requirement, it cannot be introduced in the same way as the dative causee. The dative causee is an adjunct, and is hence optional, and it is hosted by

CAUS

P, which has a [+m] feature. These facts suggest that it is an argument position of a different projection, possibly the specificer of vP-

DO

. Note that v

CAUS

cannot be present in the structure at all, or else there would be a semantic clash with an inanimate causee. The obvious alternative

191 is that the causative is a vP rather than a

CAUS

P, as shown in (39) below, which represents a possible structure for (37b).

(39) VoiceP

Voice’

DP

Mehmet vP v’ vP

DP araba

√P 49

Ç

AL

50 v’ v v

CAUS

DO

Voice-

ACTIVE a.

çal-ış-tır-

√-v.

DO

-v.

CAUS

The structure in (39) is an example of a possible stucture for causatives of unergatives. I do not claim that this

is

the actual structure, but instead leave this matter for future research.

5.3.2 Problems of Allomorphy

There are still some unresolved issues concerning allomorphy. The first involves the problem of reconciling causative allomorphy, which appears sensitive to structural adjacency to the √P, with tense allomorphy, which appears sensitive to linear

49

I label the complement of v

DO

as a √P instead of a noun or NP, as is more customary in the Hale &

Keyser (1993) tradition. The reason is that unergative verb stems in Turkish are not nouns; e.g., there is no noun

*

çalış

meaning ‘work.’ In contrast, unergative complex predicates, which are not treated in this study, do often have non-verbal complements headed by nominals.

50

An alternative analysis is that the Root is Â

ALIŞ

, and that a null suffix realizes v

DO

.

192

(phonological) adjacency only. As discussed at length in 5.3.3, this is an intractable problem if high-attaching and low-attaching causatives are the same type of head, but it ceases to be a problem at all under the present proposal that high attaching causatives are distinct from v

CAUS

. The second problem, taken up in 5.3.4, involves the allomorphy of the high-attaching causative: –

t

is inserted when the stem is polysyllabic and ends in a vowel or liquid, and –

DIr

is inserted elsewhere. This poses difficulties for the model adopted here, and its resolution is left to future work.

5.3.3 Attachment Sites and the Elsewhere Principle

In Chapter 2, Section 2.4.4, I presented causative and aorist allomorphies as diagnostics for stem complexity. In Root-adjacent contexts, the v

CAUS

node may be realized by any of the idiosyncratic VIs –

Ir

, –

It

–,

Ar

, or –

Ø

(the null allomorph); by –

t

if the Root is polysyllabic and ends in a vowel or liquid; or by the Elsewhere form –

DIr

. In particular, the insertion of an idiosyncratic VI is conditioned by adjacency to the Root. As an example, the causative derived from the Root √B

AT

is given below. The causative VI conditioned by this Root is –

Ir

.

(40) a. vP v’

DP gemi

√P

B

AT v

CAUS

–Ir

b. çocuk gemi-yi child ship-

ACC bat-ır-dı.

√-v.

‘The child sank the ship.’

CAUS

-

PST

193

It is claimed that the relevant relationship for allomorphy is Root-adjacency. It is important, however to be explicit about what is meant by ‘Root-adjacency.’ There are at least two ways the term could be interpreted in (40): 1. structural adjacency to the √P, or

2. linear (phonological) adjacency to the Root.

Structural adjacency is taken to be the relevant relationship in Harley’s (2008) treatment of Japanese causatives, which show an allomorphy pattern very similar to that of Turkish. Japanese has a rather large inventory of idiosyncratic causative VIs, including

as

, –

s

, –

os

, –

se

, –

akas

and –

Ø

, which are found on the transitive counterpart of certain types of verbs, including unaccusative COS verbs, much the same as idiosyncratic VIs in

Turkish. Japanese also has an Elsewhere causative VI, –

sase

, which is found in the causative versions of unergative and most transitive verbs, as well as with certain Roots.

According to Harley (2008), the reason that only the Elsewhere form occurs with an unergative or transitive base is that there is a functional projection (vP) intervening between the v

CAUS

node and the √P. The point is illustrated in the following example.

51

(41) a. Taroo-ga Hanako-ni piza-o

Taroo-

NOM

Hanako-

DAT

pizza-

ACC

‘Taroo made Hanako eat pizza.’ tabe-sase-ta. eat-

CAUS

-

PST

51

This example, taken directly from Harley (2008), is based on the verb

tabe-

‘eat,’ which, according to Chapter 3, Section 3.3.4, belongs to the ingestive class of verbs, which form Root causatives, unlike what is shown in (41). However, this example will still serve to illustrate the structural analysis proposed for transitive verbs in general, temporarily leaving aside the specific issue of ingestives.

Also, Harley (2008) assumes that external arguments are introduced in Spec-vP, which is reflected in

(41b).

194 b. vP2 ...

DP

Taroo

DP vP1

Hanako-ni

√P

DP piza v’ v’ v° sase

√ tabe v°

Ø

For our discussion, the relevant node for Vocabulary insertion is v° of the higher vP

(vP2). Since the lower v° is null, the position is phonologically adjacent to the Root morpheme:

tabe-sase

-. Harley (2008) therefore claims that what blocks causative allomorphy (in this verb and in transitive verbs in general) is the intervening structure of vP1, even though its exponence is null. The lack of allomorphy in transitive and unergative verbs is due to

high attachment

, as opposed to the

low attachment

(adjacent to

√P) of Root causatives.

Turkish aorist allomorphy runs afoul of this analysis. In Chapter 2, Section

2.4.3.2, it was argued that the aorist allomorph –

Ar

was Root-conditioned, while –

Ir

(or

r

) was an Elsewhere form. This means that aorist –

Ir

can be inserted in both Rootadjacent and non-Root-adjacent positions, but that aorist –

Ar

can be inserted only in

Root-adjacent positions. Because the aorist is of the general category tense/aspect/mood

(TAM) (Nakipoğlu 2001), functional morphemes such as little-v (and Voice) necessarily intervene between the Root and the aorist head (42a). If the structural definition of adjacency is relevant for allomorphy, then unergatives and intransitives should take the

195

Elsewhere aorist –

Ir

. Contrary to this prediction, most monosyllabic verbs, regardless of semanto-syntactic class, take the Root-conditioned allomorph –

Ar

.

(42) a. TP

DP

...

T’

T-

AOR

–Ar b.

DP

DP vP

√P v’

D

ÖV v

CAUS

–Ø

Tarkan

Tarkan

Hakan’ı

Hakan-

ACC boğ-Ø-ar.

√-v

‘Tarkan would strangle Hakan.’

CAUS

-

AOR

Despite the fact that the aorist node is separated from the Root by the null v

CAUS

(as well as other null heads such as Voice and Aspect), the Root-conditioned allomorph –

Ar

is inserted rather than the Elsewhere form –

Ir

. A similar point can be made with regard to irregular past tense forms in English, e.g.,

rise

/

rose

,

fall

/

fell

. Given such facts, it is hard to maintain that structural adjacency is the relevant relationship. We must therefore turn to linear (phonological) adjacency. In cases where Turkish verbs take the Rootconditioned aorist VI –

Ar

, this VI is always phonologically adjacent to the Root, e.g.,

boğ-ar

. This approach looks more promising, but it in turn brings up a different problem:

If linear rather than structural adjacency is responsible for allomorphy, how can causative

196 allomorphy be blocked in high-attachment causatives when the intervening head is null, as in (41)?

The example in (41) is from Japanese, but the same problem applies to Turkish.

Monosyllabic transitives and unergatives, while frequently taking the Root-conditioned aorist form –

Ar

, resolutely take –

DIr

and never any of the Root-conditioned allomorphs.

(43) unergative/transitive stem

atdövgülkoş-

‘throw’

‘beat’

‘laugh’

‘run’ aorist

at-ar döv-er gül-er koş-ar

causative

at-tırdöv-dürgül-dürkoş-tur-

The same stems that take the Root-conditioned aorist allomorph never take a Rootconditioned causative allomorph. These facts appear to contradict each other.

Marantz (2010), comparing the Japanese causative to the English past tense, addresses this very problem. According to him, the locality domain for contextual allomorphy is linear (phonological) adjacency, but he suggests a further factor to account for the lack of allomorphy in high-attachment causatives:

(44) A phase boundary blocks contextual allomorphy, when the context is another phase head.

(Marantz 2010)

For Marantz (as well as Chomsky 2001, among others), little-v is a phase head—in other words, a head that defines the boundary of a phase domain. Thus, in a configuration such as the one that Harley (2008) proposes, VI insertion for the high-attachment causative node meets Marantz’s condition in (44). The tree for the Japanese example from (41) is shown again, this time with the purported phase boundaries indicated.

197

(45)

DP piza

√P vP2 ...

DP

Taroo vP1 v’ v’ sase v°

√ tabe v°

Ø

Both v° nodes are phase heads in this analysis. Thus Vocabulary insertion in the higher node is subject to the stipulation in (44): the context for allomorphy is the higher v°, a phase head, and allomorphy is therefore blocked by the lower v°, another phase head, even though the lower head has null realization.

On the other hand, when the context for allomorphy is the tense node, (44) does not apply, because T° is not a phase head. Hence, we may see Root-conditioned allomorphy here so long as all intervening heads have null realization. This point is illustrated for the Turkish aorist in the tree below, where the phase boundaries are shown.

198

(46)

DP

... vP

√P

TP v’

D

ÖV

CP

C’

T’

T°v°-

AOR

–Ar

CAUS

DP

DP

This solution works only if little-v is a phase boundary. However, it runs into complications with the separation of the external argument-introducing head Voice from the verbalizer little-v. Now Voice, and not the little-v beneath it, is the phase boundary.

In the case of Japanese causatives, which are phase-selecting, a Voice head does intervene, but the context of allomorphy is v-

CAUS

, which is not a phase boundary.

199

(47)

DP piza

√P

DP

VoiceP

Voice’ v-

CAUS

–sase

Hanako vP Voice-

ACTIVE

VoiceP

Voice’

DP

Taroo vP v’

Voice-

ACTIVE v’

T

ABE v-

CAUS

–Ø

Turkish causatives also do not meet the condition stated in (44). The v-

CAUS

node is not a phase head, nor does it embed Voice. The proposal in Marantz (2010) (modified with the new architecture) would therefore predict that allomorphy should not be blocked in high-attachment causatives, and yet allomorphy never occurs in that position.

(48) a. Hakan Tarkan-ı

Hakan Tarkan-

ACC döv-dür-dü.

√-v

CAUS

‘Hakan had Tarkan beaten.’

-

PST

200 b.

DP

VoiceP

Voice’ vP v’

Voice-

ACTIVE

DP

√P vP v’

D

ÖV v

CAUS

-Ø v-

CAUS

–DIr

The analysis presented in this chapter provides a different avenue to account for the lack of allomorphy. I have claimed that the outer causative, unlike the inner, is not a flavor of little-v, but the head of a different type of phrase,

CAUS

P, which is not a verbalizer. The head has the feature [

CAUS

] but lacks [v]. This requires that –

DIr

, like –

t

, have the underspecified feature bundle [

CAUS

].

52

This has no effect on the analysis presented thus far. Underspecification allows it to compete with other VIs for the position

[v,

CAUS

]. As argued in Chapters 2 and 3, these competing idiosyncratic allomorphs are either fully specified for [v,

CAUS

] or underspecified as [v]. In both cases, they contain the feature [v], which is incompatible with the outer causative node [

CAUS

]. This renders them ineligible for competition in the outer domain, even when the node is linearly adjacent to the Root, as shown below.

52

Note to Committee: This is at odds with the analysis of earlier chapters, where –

DIr

is [v,

CAUS

].

This change should not affect the results of competition for vocabulary insertion.

201

(49)

DP

DP

VoiceP

Voice’ vP v’

Voice-

ACTIVE

√P vP v’

D

ÖV v

CAUS

-Ø v-

CAUS

–DIr

5.3.4 Outer Causative Allomorphy

We have identified the outer or verb-selecting causative as a

CAUS

phrase that is distinct from the verbalizing vP-

CAUS

. The

CAUS

head thus has only the feature [

CAUS

] and does not contain [v]. We have proposed that both –

DIr

and –

t

are specified for the feature [

CAUS

] only, which makes them compatible with this node, and the other causative allomorphs incompatible, as they all have the feature [v]. The final question concerning allomorphy, then, is what determines the alternation between –

t

and –

DIr

.

The first follows vowels and liquids, and the second occurs in all other environments. A phonological characterization is descriptively adequate, but the motivation cannot be strictly phonological.

The restrictions on -

DIr

do not avoid any phonotactically illicit sequences. This is readily apparent from the fact that the past tense suffix -

DI

, which has the same two initial segments as -

DIr

, and can be added to all verb stems, including those to which -

DIr

cannot.

202

(50) causative -

DIr

*

eri-dir-

melt-

DIr

-

*

sarar-dır-

yellow-

DIr

-

*

çoğal-dır-

increase-

DIr

- past tense -

DI eri-di

“it melted”

sarar-dı

“it yellowed”

çoğal-dı

“it increased”

The distribution of -/DIr/ is therefore not phonotactically motivated. The distribution of –t, on the other hand, does avoid the illicit coda sequence of two stops.

53

(51) *

bırak-t-

leave-

t

- cf. *

vakt

vakit

‘time’ (Arabic loanword)

In other cases, however, the forms prevented by the restriction would be phonotactically licit. Complex codas consisting of a fricative or nasal followed by a stop are allowed in Turkish.

(52) *

kullan-t-

use-

t

- cf.

çent-

‘notch (tr.)’

*

savaş-t-

fight-

t

- cf.

sergüzeşt

‘adventure’ (Persian loanword)

The phoneme /t/ may follow nasals and fricatives, but the suffix may not. Paster

(2006) characterizes the segmental restrictions on –

t

as optimizing syllable structure, preventing a sequence of low-sonority consonants at the end of a causative stem. In any event, it is clear that phonological factors must be taken into account in defining the

53

The illicit combination would arise when -/t/ is followed by a consonant-initial suffix, such as the past tense -

DI

, or is not followed by any overt suffix, as in the singular imperative: *

Bırakt-tı

“He made

[someone] leave [it],’ *

Bırakt

! “Make [him] leave [it]!” If followed by a vowel-initial suffix such as the future -

AcAk

, -

t

would resyllabify, and the phonotactically illicit combination would be resolved: *

Bıraktacak

“He will make [someone] leave [it].” Even with the phonotactic violation resolved, however, the form is ungrammatical.

203 distribution of this VI. This presents problems in terms of the Late Insertion component of DM. At the stage when –

t

is inserted, the phonology of the node that linearly precedes it must be available, which would require that Vocabulary Insertion be cyclic, along the lines of Bobaljik (2000). I leave the precise formulation to future work.

5.4 Causativization in the Lexicon

The separation of Voice from little-v can accommodate the behavior of productive causatives in Turkish and other languages, which are like Japanese syntactic causatives in distribution and form, but test as monoclausal, like lexical (Root) causatives. Nevertheless, causativization as a lexical operation still has its adherents.

Horvath & Siloni (2011a) provide evidence from Hungarian causatives very similar to what we have already seen in Turkish. They fail the biclausality tests.

Binding Condition B

(53) a. Kati i

Kati.

NOM le-fotóz-ta down-photograph-

PST

.

DEF

.

DO

‘Kati has photographed her.’ őt* she.

i

.

ACC b. Kati i le-fotóz-

tat

-ta őt* i

/* j

Mari-val j

.

Kati.

NOM down-photograph-

CAUS

-

PST

.

DEF

.

DO

she.

ACC

Mari-

INSTR

‘Kati made Mari photograph her.’ (Horvath & Siloni 2011)

Scope of Negation

(54) Nem énekel-tet-t-em a gyerek-ek-et. not sing-

CAUS

-

PAST

-1

SG

the kid-

PL

-

ACC

‘I didn’t make the kids sing.’

Narrow scope impossible: ‘I made the kids not sing.’

Horvath & Siloni (2011a) (henceforth H&S) propose that the Hungarian productive causatives are lexically derived. They assume an active lexicon. As far as verbs are concerned, this means that lexical entries contain specifications for the complete

204 argument structure, including an external argument where relevant. Arity operations can change argument structure within the lexicon, adding, removing, or modifying arguments. They make use of the theta system of Reinhart (2002) (see Chapter 4.2.2.1), where thematic roles are not primitives, but are composed of the features [c] ‘cause’ and

[m] ‘mental state.’ Agents have the feature bundle [+c+m], because they are responsible for causing the event, and because their mental state is relevant to the event. The thematic role ‘Cause,’ in contrast, is [+c], since it causes the event, but is underspecified for mental state. For example, in ‘Marvin melted the ice,’ Marvin is an agent, because he caused the event, and, in the most natural interpretation, he consciously decided to melt the ice, but in ‘The rising temperature melted the ice,’ the rising temperature is merely a cause, since no mental state can be attributed to it. Themes are [-c-m], since neither do they cause the event, nor is their mental state relevant to the event. Experiencers may be

[-c+m] or [+m]—although they do not cause the event, their mental state is relevant.

H&S formulate the lexical arity operation of causativization as follows:

(55) Causativization in the lexicon

V<α> →

CAUS

-V<[+c+m], α’>, where α includes a role specified as external; if this role includes a [+c] feature, the feature is revaluated to [-c] (otherwise α equals α’)

(Horvath & Siloni 2011a)

The Greek letter α is the notation for argument structure, and V<α> is the argument structure of the lexical entry of a given verb. The notation to the left side of the arrow indicates that the arity operation adds the meaning of causation, as well as a [+c+m], or

Agent, argument. It also may modify the arguments of the original entry, which is why the new argument structure includes α’ rather than the original α. The stipulation ‘where

205

α includes a role specified as external’ defines the input to lexical causativization: any verb that has a lexically specified external argument. Theta roles with uniformly positive values ([+c+m], [+c], [+m]) are always mapped externally in the lexicon, and those with uniformly negative values ([-c-m]) are always mapped internally. Roles with mixed values ([-c+m]) are not lexically specified as internal or external, and may be realized as either. The input to causativization is therefore a verb with a [+c+m], [+c], or [+m] argument in its theta grid. This includes transitive verbs, unergative verbs, and nonalternating subject-experiencer verbs such as ‘love’ and ‘hate.’

If the external argument of the input verb has a [+c] feature, causativization revalues it to [-c]. Thus, [+c+m] becomes [-c+m], a cluster that corresponds to experiencer or benefactor, and [+c] becomes [-c], which corresponds to goal or recipient

(Reinhart 2002). As an example, take the verb

fel-olvas

‘read.’ Its theta grid includes an agent, the reader, and a theme, the reading material:

fel-olvas

<[+c+m], [-c-m]>. The lexically specified external argument makes it legitimate input to causativization, after which it looks like this:

fel-olvas-tat

<[+c+m], [-c+m], [-c-m]>. In the new grid, the

[+c+m] argument is the one added by causativiation, the cause, while [-c+m] is the reader, whose initial [+c] has been revalued to [-c].

Since there is no hierarchical structure available in the lexicon, a lexical derivation would explain why Hungarian causatives consist of one predicate rather than two. As we have seen, separating Voice from little-v also explains this. Thus it would seem that the two approaches have the same empirical coverage in terms of their ability

206 to explain differences in clausality. In order to decide between them, we must find areas where there coverage differs. One very significant area is the causee.

5.5 Causee Animacy

In comparing the lexical and syntactic accounts of monoclausal productive causatives, the causee is of special interest. Under both approaches, the internal argument is part of the input to the causative verb (though in quite different ways), but the causee, as the external argument of the corresponding non-causative verb, has a very different place in the two derivations. In Causativization in the Lexicon, the external argument is part of the θ-grid of the base verb’s lexical entry, and indeed is the sole factor that makes the verb licit input to the causativization operation. If the external argument has the feature +c, this is revalued to –c, but otherwise it is the same argument. In contrast, in the syntactic derivation, monoclausal productive causatives select for the verbalizing head little-v, but not for the external argument-introducing head Voice, such that the external argument of the non-causative counterpart is not part of the input. The causee, when expressed, must be introduced another way, either by a separate head, or in the specifier of vP-

CAUS

, as proposed here. What this means is that in the lexical approach, the causee should not differ from the subject of the non-causative except in ways directly attributable to the revaluation of +c to –c, while in the syntactic approach, the causee might differ in any number of ways, depending on the details of how it is introduced. In the following sections, it is shown that the causees of productive causatives differ from the corresponding external argument in ways that cannot be captured by H&S’s

Causativization-in-the Lexicon.

207

As already discussed, the causee of productive causatives in Turkish has animacy restrictions not found in the external argument corresponding non-causative. The same holds for Hungarian. When the non-causative verb is transitive, the subject may be an animate agent, but it may also be an inanimate argument, such as an instrument.

(56) a.

Az őr ki-nyit-ott-a the watchman

PRVRB

-open-

PAST

.

SG

3-

DEF

.

DO kulccsal. az ajtó-t a the door-

ACC

the b. key.inst

‘The watchman opened the door.’

A kulcs ki-nyit-ott-a the key

PRVRB

-open-

‘The key opened the door.’

PAST

.

SG

3-

DEF

.

DO az ajtó-t. the door-

ACC

In the causative variant, however, the causee—marked instrumental in Hungarian—must be animate. Instruments are not acceptable causees.

54

(57) a. b.

A főnök the director az őr-rel the watchman-inst ki-nyit-tat-t-a

PRVRB

-open-

CAUS

-

PAST

.

SG

3- az ajtó-t.

DEF

.

DO

the door-

ACC

‘The director made the watchman open the door.’

*Az őr ki-nyit-tat-t-a az ajtó-t the watchman

PRVRB

-open-

CAUS

-

PAST

.

SG

3-

DEF

.

DO

the door-

ACC

the a kulccsal.

55 key.

INST

Intended: ‘The watchman made the key open the door.’

This restriction runs counter to H&S’s Causativization-in-the-Lexicon. The lexical entry for alternating verbs such as

open

has a [+c] argument in its θ-grid, underspecified for the feature m. The subject may therefore have any +c cluster: [+c+m]

(Agent), [+c-m] (Instrument...), or [+c] (Cause). This is what allows the subject to be any

54

Thanks to Eszter Ótott-Kovács for all following Hungarian examples.

55

Irrelevantly, this is grammatical on the reading where the instrumental is interpreted not as the causee, but as an instrument: ‘The watchman made (someone else) open the door with the key.’

208 type of cause. As a uniformly positively valued role, [+c] is specified as external, and therefore alternating transitive verbs are also legitimate input to Causativization in the

Lexicon. When such a verb is causativizatized, the c feature is revalued, such that [+c] becomes [-c]. Still underspecified for mental state, the [-c] cluster is compatible with [c+m] (Experiencer), [-c-m] (Theme/Patient), and [-c] [Goal, etc.] (Reinhart 2000). There is thus no basis on which to rule out an instrument as a causee.

Note that arity operations are not operations on arguments, but on the θ-grids of lexical entries. When an alternating transitive verb undergoes Causativization, the output is a θ-grid with a [-c] role. There is no way to stipulate that it must be realized by an argument that is [+m], since [-c] is underspecified for m. In principle, there are only two ways to ensure that the causee will be animate: the feature +m must be included either as part of the external argument’s feature cluster in the input θ-grid, or as part of the causee’s feature cluster in the output θ-grid. The first option is not really an option at all.

Recall that transitive verbs undergoing Decausativization have an external causer role that is underspecified for mental state: [+c]. This is what permits a range of causer arguments from agents to instruments, and it is what defines the class of verbs undergoing the alternation in the first place. It is perhaps not an overstatement to say this single point is the linchpin of the entire enterprise of arity operations. It is therefore nonnegotiable.

The second option, building the [+m] feature into the arity operation, would be far less radical. Causativization in the lexicon could be revised as follows (my modification is in bold):

209

(58) Causativization in the lexicon

V<α> →

CAUS

-V<[+c+m], α’>, where α includes a role specified as external; if this role includes a [+c] feature, the feature is revaluated to [-c]

and the feature

[+m] is added

(otherwise α equals α’)

This might have the desired result, except that it cannot capture the facts associated with variable case marking of the causee. In Hungarian, the causative counterpart of an unergative verb can have an instrumental causee (59b) or an accusative one (59c).

(59) a. b. c.

A gyerek-ek the child-

PL

‘The children run.’ fut-nak run-3

PL

.

IND

A tanár fut-tat-ott a the teacher run-

CAUS

-

PAST

.

INDEF

the

‘The teacher made the children run.’

A tanár fut-tat-ta a the teacher run-

CAUS

-

PAST

.

DEF

.

DO

‘The teacher made the children run.’ gyerek-ek-kel. child-

PL

-

INST gyerek-ek-et. the child-

PL

-

ACC

In Hungarian, a definite direct object triggers agreement on the verb. Note that when the causee is marked accusative, the verb shows object agreement (59b), but not when the causee is marked instrumental (59c). (The non-object-agreeing conjugations are known as the ‘indefinite paradigm’—hence the

IND gloss). This is indicative of distinct structural positions. But verbal agreement is not the only difference between the two types of marking. It turns out that causee case marking correlates with the animacy restriction. An instrumental causee must be animate, as we have already seen, but an accusative causee need not be.

210

(60) a. b. c.

A program fut a gép-en. the program run.3

SG

.

IND

‘The program runs on the machine.’ the machine-on

*A diák fut-tat-ott a program-mal the student

A diák run-

CAUS fut-tat-ta

-3

SG

.

INDEF

the program-

INST a program-ot. the student run-

CAUS

-3

SG

.

DEF

.

DO

the program-

ACC

‘The student ran the program.’

The construction in (60c), with an accusative-marked inanimate causee, is exactly like the pattern seen in Turkish. The construction in (60b), on the other hand, is like the causative of a transitive verb, where the instrumental causee must be animate. The syntactic account proposed here can capture these facts. When the causee is in the instrumental case, it is hosted by a

CAUS

P,

56

a phrase whose head has the feature [+m]. When it is accusative, the causative is a vP, with no [+m] feature; the causee may be in Spec-vP

DO

, as in (39), or in Spec-vP

CAUS

. I do not see how the distinction between the instrumental and accusative causees could be captured by modifying theta features in the lexicon.

Horvath & Siloni address the matter of causee animacy in a footnote:

It

may be worth noting

here that there seem to be some semantic factors that influence the acceptability of inanimate causees

in certain cases

[emphasis mine].

But according to our preliminary exploration, these effects involve independent properties of the semantics of causation (possibly, obligation vs. permission), and may not be specific to lexically derived causatives (a prima facie similar constraint is discussed regarding Italian periphrastic causatives by Folli and

Harley 2007). (H&S 2011a: 691–692, n. 36)

They downplay the phenomenon, suggesting that it is only in effect ‘in certain cases.’

This is an understatement: the animacy restriction is in fact quite general. To support their case, however, they provide the following example:

56

With causatives of unergatives in Hungarian, the instrumental causee is probably an argument in

Spec-

CAUS

P rather than an adjunct, because its expression is not optional.

211

(61) Mari meg-szár-ít-tat-ja a nap-pal

Mari.

NOM

a

PERF haj-á-t.

.-dry-

TRANS

-

CAUS

-

PRES

.

DEF

.

DO

the sun-

INSTR the hair- a

POSS

.3

SG

-

ACC

‘Mari has/lets the sun dry his/her hair.’

I have several points to make in this regard. First, the causee in (61) is the sun, which is a natural force. It has been noted that, as a class, natural force arguments are permitted in many positions that normally only permit animate agent arguments (Folli & Harley 2007,

Sichel 2010). Second, the judgments of (61) from my Hungarian consultants range from awkward to ungrammatical. The Turkish equivalent is strictly ungrammatical. Finally, the suggestion that the restriction can be attributed to the semantics of causation is not promising. I will explain why.

First, the analysis in Folli & Harley (2007) is of little use to H&S. Doubtlessly, their motivation for citing this work is not the analysis contained therein, but the simple empirical fact that a similar restriction is found in periphrastic causatives (which they implicitly assume to be formed in the syntax), and therefore that the animacy requirement is an artifact of the semantics of causation, and, as such, that it is not a phenomenon that needs to be addressed by their lexical analysis. Nevertheless, it is worth pointing out that

Folli & Harley’s (2007) analysis would not derive the animacy restriction that H&S seek.

Folli & Harley address the sense of obligation inherent in Italian causatives of the FI

(

faire infinitif

) variety. They argue that the light verb

fare

takes a vP headed by v

DO

as its complement in that construction. They also posit that such a vP always has an Agent in its specifier. Agents are typically animate. The obligation effect is a result of the interaction of the semantics of causation with the presence of an Agent causee.

212

Thus, Folli & Harley derive the obligation reading from the agency/animacy of the causee, whereas H&S propose the reverse: deriving an animacy effect from the semantics of causation. This is a challenging task, because there is nothing inherent in the semantics of causation as such that would rule out an inanimate causee. An instrument such as

the key

is a good causee in the English

make

causative.

(62) I made the key open the door.

Note that, with an animate causee, there is an obligation effect that is absent with an inanimate causee.

(63) I made John open the door.

This lends support to Folli & Harley’s proposal that the obligation effect is the consequence of an agentive (animate) causee, and challenges the idea that a causee animacy effect could be derived from the semantics of obligation.

As for the semantics of permission, this is a non-starter with respect to Hungarian causatives. This is one way in which Hungarian differs from Turkish. Turkish causatives are systematically ambiguous between basic causation and permission.

(64)

(65)

Öğretmen öğrenci-ler-i konuş-tur-ma-dı. teacher

Basic: student-

PL

-

ACC speak-

CAUS

-

NEG

-

PAST

‘The teacher didn’t make the students talk.’

Permissive: ‘The teacher didn’t let the students talk.’

Baba kız-ın-a araba-sı-nı kullan-dır-ıyor. father daughter-3

SG

-

DAT

Basic: car-3

SG

-

ACC

use-

CAUS

-

PRES

‘The father makes his daughter drive his car.’

Permissive:

‘The father lets his daughter drive his car.’

Just as systematically, Hungarian causatives lack a permissive reading.

213

(66) A the tanár teacher nem beszél-tet-te

NEG speak-

CAUS

-

PAST

.

DEF

.

DO

Basic: ‘The teacher didn’t make the students talk.’

Permissive: Unavailable! a diák-ok-at. the student-

(67) Az apa vezet-tet-i

the father a lány-á-val. drive-

CAUS

-

PRES

.

DEF

.

DO a kocsi-já-t the car-3

SG

-

ACC the

Basic: daughter-3

SG

-

INSTR

‘The father makes his daughter drive his car.’

Permissive: Unavailable!

PL

-

ACC

There appears to be no basis, then, for attributing the animacy restriction on the causee to semantics of permission in Hungarian.

The structural analysis presented herein captures the facts much more adequately.

The verb-selecting

CAUS

P projects a specifier with the feature [+m], which may be realized by a dative phrase in Turkish, and by an instrumental phrase in Hungarian.

Incidentally, the permissive reading of causatives in Turkish provides an answer to why Turkish causatives seem to test as biclausal on the scope-of-negation test. Kural

(1996, 2000) addresses the supposed scope ambiguity in (68).

(68) Emine-yi bugün çalış-tır-ma-dı-m.

Emine-

ACC today work-

CAUS

-

NEG

-

PAST i. ‘I didn’t make Emine work today.’ ii. ‘I made Emine not work today.’ (Kural 1996)

As already noted, the Hungarian equivalent lacks this ambiguity.

(69) Nem énekel-tet-t-em a gyerek-ek-et. not sing-

CAUS

-

PAST

-1

SG

the

‘I didn’t make the kids sing.’ kid-

PL

-

ACC

Narrow scope impossible: ‘I made the kids not sing.’

Kural (1996, 2000) explains that there is no scope ambiguity in Turkish. Rather, the ambiguity is between the Basic (“make”) and the Permissive (“let”) causative. He

214 defines the relationship between the two as a type of modality in which the

make

causative is a universal modal and the

let

-causative an existential one.

(70)

(71) a. b. a. b.

A

let

P = 1 iff

∃ w

C

, w

C

a causally possible world,

A cause P = 1 in w

C

A

make

P = 1 iff

∀ w

C

, w

C

a causally possible world,

A cause P = 1 in w

C

let

P = cause

C

P

make

P = cause ◻

C

P

Universal and existential quantifier pairs are duals—that is, the universal quantifier under the scope of negation has the same truth value as the existential quantifier scoping over negation.

(72) a. b.

¬

=

¬

¬

=

¬

(73)

It is not the case that every boy had fun = There is some boy who didn’t have fun.

It is not the case that there is some boy who had fun = Every boy didn’t have fun.

This, and not the scope of negation, is the source of the ambiguity in (67).

(74) i

I didn’t make Emine work today.

¬

∀ ii.

I didn’t let Emine work today.

¬ =

¬ , hence is truth-value equivalent to

“I made Emine not work today.”

This has interesting consequences for locating the source of different causative semantics. Lexical or Root causatives are traditionally characterized as indicating direct causation. Yet, as we have already seen, Root causatives allow a permissive reading.

(75) Baba-mız dün biz-i hiç dışarı çık-ar-ma-dı. father-1

PL a. b. yesterday 1

PL

-

ACC at.all outside √-

CAUS

-

NEG

-

PAST

‘Our father didn’t take us outside at all yesterday.

‘Our father didn’t let us go outside at all yesterday.’ = ‘Our father made us not go outside at all yesterday.’

215

(76) Baba-mız dün biz-i hiç içeri sok-Ø-ma-dı. father-1

PL a. b. yesterday 1

PL

-

ACC at.all inside √-v.

CAUS

-

NEG

-

PAST

‘Our father didn’t bring us inside at all yesterday.

‘Our father didn’t let us come inside at all yesterday.’ = ‘Our father made us not come inside at all yesterday.’

In Hungarian, on the other hand, not even verb-selecting causatives allow a permissive reading, regardless of causee case. This suggests that the causative readings of coercion versus permission are not associated with particular syntactic structures, but are abstract features.

5.6 Conclusion

In this chapter, I have proposed a structure for Turkish causatives that is like

Harley’s (2008) analysis of Japanese causatives, except that there is no external argument introduced by the vP. Furthermore, I have argued that, while the Root-selecting causative is v-

CAUS

, the verb-selecting causative is not a flavor of little-v, but a distinct head,

CAUS

.

This has syntactic consequences beyond those discussed in this chapter. If outer causatives are vPs, then the causative head is a little-v that selects for little-v, and therefore recursion should be possible. On the other hand, if outer causatives are a different syntactic category, then recursion might be possible, depending on selectional factors, but it is not automatically predicted. I take up this matter in the next chapter, wherein I present evidence that, despite appearances to the contrary, Turkish and various other languages do not in fact have causative recursion.

216

CHAPTER 6: CAUSATIVE ITERATION

6.1 Introduction

In the previous chapter, I proposed that the outer (productive) causatives realize a different type of syntactic head than inner (Root) causatives do. The inner causative is the

CAUS

flavor of the verbalizer little-v, while the productive/high-attaching causative is a

CAUS

head that lacks any verbalizing properties. The Turkish outer causative is verbselecting and the Japanese outer causative is phase-selecting, in the sense of Pylkkänen b.

(2002, 2008). The structures for Turkish and Japanese outer causatives under the new proposal are given in (1) and (2), respectively (repeated from Chapter 5).

Turkish

(1) a.

DP kadın

Voice-P

CAUS

-P

CAUS

Voice’

Voice-

–Ø

ACTIVE vP v’

CAUS

–DIr

DP et

√P

K

ES v

CAUS

–Ø

Kadın woman et-i meat-

ACC kes-Ø-tir-di.

√-v.

CAUS

-

CAUS

-

PST

‘The woman had the meat cut/had (someone) cut the meat.’

217

Japanese

(2) a. VoiceP

Voice’

DP

Taroo vP v’

Voice-

ACTIVE

–Ø

DP

Hanako

VoiceP

Voice’ v-

CAUS

–sase vP v’

Voice-

ACTIVE

–Ø

DP hanasai

√P

√ tutae v-

CAUS

–Ø b. Taroo-wa

Taroo-

TOP

Hanako-ni hanasi-o

Hanako-

DAT

story-

ACC

‘Taroo made Hanako convey a story.’ tutae-sase-ta. convey-

CAUS

-

PST

This is in contrast to Harley (2008)’s analysis of Japanese causatives, where both

‘lexical’ (inner) and ‘syntactic’ (outer) causatives are of the category v

CAUS

, and all differences between them proceed from the attachment site: inner causatives are adjacent to √P, and outer causatives are adjacent to vP. The latter structure is shown in (3).

218

(3) a. b. vP2 ...

DP

Taroo

DP vP1

Hanako

√P

DP hanasi v’ v’ v° sase

√ tutae v°

–Ø

Taroo-wa Hanako-ni hanasi-o

Taroo-

TOP

Hanako-

DAT

story-

ACC

‘Taroo made Hanako convey a story.’ tutae-sase-ta. convey-

CAUS

-

PST

(Harley 2008)

While the most salient difference between the present proposal (2) and Harley’s analysis

(3) is the introduction of the external argument via a Voice projection, there is a further difference (proposed in Chapter 5): The outer causative is not of the category vP, but

CAUS

P. This difference is not trivial. If the outer causative is of the same syntactic category as the inner, then Turkish outer causatives, which do not embed Voice, would still look very much like Harley’s original proposal for Japanese, with stacked vPs, as shown in (4).

219 a. (4)

DP et

√P

DP kadın

Voice-P vP v’ vP v’

K

ES v

CAUS

–Ø

Voice’ v

CAUS

–DIr

Voice-

–Ø

ACTIVE

This would mean that v

CAUS

is a little-v that selects for a phrase headed by little-v— indeed, by another v

CAUS

. The immediate prediction is that outer causatives should exhibit recursion, in principle without limit, though factors such as performance and pragmatics might restrict this in practice. On the other hand, if inner and outer causatives are distinct syntactic heads, as I propose, then recursion is not automatically predicted.

The structure I propose, shown above in (1), indicates that the

CAUS

head has to be able to merge with vP, but recursion would require that

CAUS

be able to merge with

CAUS

P as well. Outer causative recursion would be compatible with either approach: the stacked vP approach straightforwardly predicts it, while the

CAUS

P approach neither predicts it nor categorically rules it out. On the other hand, the absence of outer causative recursion would be evidence in favor of the

CAUS

P approach, since under the stacked vP approach recursion cannot be ruled out except by stipulation.

As it happens, some languages, including Turkish, apparently allow outer causative recursion, while others, including Japanese, do not. In the present chapter, I

220 examine causative iteration, principally in Turkish but in other languages as well, and reach the conclusion that it is in fact not recursion, but rather morphological reduplication. I propose that this reduplication realizes focus in Turkish.

In section 6.2, I discuss causative iteration cross-linguistically, and find that several languages that appear to have causative recursion also allow the semantically vacuous iteration of causative suffixes. In 6.3, I show that the semantics of indirect causation allow interpretations with multiple causal links that are not individually represented in the syntax. In 6.4, I show how these semantics in combination with the vacuous repetition of suffixes permit the illusion of causative recursion in Turkish. In 6.5,

I analyze causative iteration as focus realized via morphological reduplication. In 6.6, I model this reduplication using Inkelas & Zoll’s (2005) Morphological Doubling Theory, adapting it to DM’s model of derivation. Section 6.6 concludes.

6.2 Causative Iteration Cross-Linguistically

Svenonius (2005a,b) evaluates morphological causative in several languages:

Sámi, Hindi, Nivkh, Amharic, Kitharaka, Malagasy. In all of these, the limit on iteration is two, and even this is restricted in that an inner causative (Root causative) may be followed by an outer causative (verb- or phase-selecting causative in the sense of

Pylkkänen), but inner may not be followed by inner, nor outer by outer. In the languages mentioned above, at most an unaccusative base can be causativized twice, by one inner and one outer causative, while unergative and transitive bases do not permit any iteration of the causative. This is fully consistent with the present proposal, in which Root causatives and outer (verb- or phase-selecting) causatives are distinct types of projection.

221

It is perhaps surprising then that Turkish has been claimed variously to permit up to three iterations (Çetinoğlu et al. 2009) or to have no upward limit (Kural 1996). In particular, double causatives of transitive verbs have frequently been cited in the literature (Comrie 1989, Zimmer 1976, Aissen 1979, Kural 1996).

(5) Mektub-u müdür-e kâtip vasıta-si-yle imzala-t-tır-dı-m. letter-

ACC

director-

DAT

secretary means-3

SG

-

INST

sign-

CAUS

-

CAUS

-

PST

-1

SG

‘With the help of the secretary, I got the director to sign the letter.’ (Aissen 1979)

In the above sentence, the transitive base

imzala-

‘sign’ has two causative suffixes, which, given the distributional restriction on inner causatives, must both be outer causatives. Furthermore, in addition to a direct object, the verb apparently has two causees, one with a dative suffix and one with an instrumental postposition. This looks like double causativization of a transitive base, contrary to the otherwise observed crosslinguistic limit.

In the following, I show that this is not a case of recursion, but rather of reduplication. Many causative languages that apparently allow such recursion also have another characteristic in common: They allow the vacuous repetition of causative morphemes. By ‘semantically vacuous,’ I mean causative morphemes that do not correspond to additional causing events. In Turkish, when the base verb is transitive or unergative, a second suffix may be added with no change in meaning.

(6) a. b.

Saç-ım-ı kes-tir-di-m hair-1

SG

-

ACC

cut-

CAUS

-

PST

-1

SG

‘I had my hair cut.’

Saç-ım-ı kes-tir-t-ti-m hair-1

SG

-

ACC

cut-

CAUS

-

CAUS

-

PST

-1

SG

‘I had my hair cut.’ (Göksel and Kerslake 2005)

222

Vacuous causatives are not peculiar to Turkish, but are attested in other causativeiterating languages. Kashmiri (Manetta (to appear), Hungarian (Hetzron 1976), and Tsez

(Kulikov 1993) have all been claimed to allow causative recursion, or the ‘extended causative’ construction.

Kashmiri

(7) Me

I. erg chal-Ina:v-Inǝ:v’

WASH

-

CAUS

-

CAUS

.

MPL ra:mni

Ram.

GEN

zǝriyi by raj-as athi palav

Raj-

DAT INSTR clothes.

NOM

‘I got clothes washed by Raj through Ram.’ (Manetta (forthcoming))

Hungarian

(8) A tanár-ral the teacher-

INST dolgozatot composition-

ACC

ír-at-tat-ta-m write-

CAUS

-

CAUS

-

PAST

-1

SG

.

DEF a diák-ok-kal the pupil-

PL

-

INST

‘I had the teacher have the pupils write a composition.’ (Hetzron 1976)

Tsez

(9) Ala žek’uq užiq magalu bac’-r-er-si

Ali:

ERG old-man:

LOC

child:

LOC scone:

ABS eat-

CAUS

-

CAUS

-

PAST

‘Ali caused the old man to feed a scone to the child.’ (Kulikov 1993)

All of these languages, like Turkish, also allow the vacuous iteration of causatives.

Kashmiri

(10) kar-Ina:v/ kar-Ina:v-Ina:v do-

CAUS

/ do-

CAUS

-

CAUS

‘cause to do’

(11) d‘-a:v / d‘-av-Ina:v give-

CAUS

‘cause to give’

/ give-

CAUS

-

CAUS

/d‘-av-Ina:v-Ina:v

/ give-

CAUS

-

CAUS

-

CAUS

(Emily Manetta, p.c.)

223

Hungarian

57

(12) a. b.

A főnök sokat dolgoz-tat-t-a the boss many work-

CAUS

-

PST

-3

SG

.

DEF

‘The boss made the workers work a lot.’ a munkás-ok-at. the worker-

PL

-

ACC

A főnök sokat dolgoz-tat-tat-t-a a munkás-ok-at. the boss many work-

CAUS

-

CAUS

-

PST

-3

SG

.

DEF

the worker-

PL

-

ACC

‘The boss made the workers work a lot.’

Tsez

(13) žek’a užiq magalu bac’-er-si/bac’-r-er-si old-man:

ERG

child:

LOC scone:

ABS eat-

CAUS

-

PAST

/eat-

CAUS

-

CAUS

-

PAST

‘The old man fed a scone to the child.’ (Kulikov 1993)

Strengthening the connection between the two phenomena of the ‘extended’ causative and vacuous causative iteration is the fact that, in the ‘extended’ causative construction

(14a), it is often possible to omit one of the suffixes (14b).

Turkish

(14) a. Müdür, director yardımcı-sı aracılığıyla yaz-dır-t-tı. işçi-ler assistant-3

SG

by.means.of worker-

PL yeni bir yönetmelik temsilci-si-ne rep.-3sg-

DAT new a by-laws write-

CAUS

-

CAUS

-

PST

‘The director, via his assistant, had the workers’ representative write new bylaws.’ b. Müdür, yardımcı-sı aracılığıyla işçi-ler temsilci-si-ne director assistant-3

SG

by.means.of worker-

PL yeni bir yönetmelik yaz-dır-dı. rep.-3

SG

-

DAT new a by-laws write-

CAUS

-

PST

‘The director, via his assistant, had the workers’ representative write new bylaws.’

57

Thanks to Eszter Ótott-Kovács for this example (as well as all examples in this chapter not attributed to Hetzron 1976), and for obtaining judgments from other native Hungarian speakers. Judgments on these points in Hungarian are variable. Some consultants said that for them the number of causatives had to match the number of causees; however, they also pointed out that they have heard other speakers use vacuous causatives. Googling a form such as

írattatta

(write-

CAUS

-

CAUS

-

PAST

) yields thousands of results, and a scan of the top results suggests that nearly all of them are used in the context of single causation.

Why some consultants should reject such sentences (as well as double causation with a single causative suffix) is an interesting question that will require further investigation. For the moment, three possibilities come to my mind. The first is that these consultants are bringing prescriptivist preconceptions or extra-linguistic logic to the grammaticality judgment task. The second is that the iteration realizes focus

(as argued in Section 6.5 and 6.6 of this chapter), and that for these speakers this focus is closely associated with the multiple-links reading of indirect causatives (see Section 6.4). The third possibility is that these speakers have a different grammar that allows recursion. For this last possibility, these speakers’ grammar would need to allow stacked vPs or stacked

CAUS

Ps. I leave investigation of this issue to future work. For the time being, it is important to note that the vacuous use of causatives is found in Hungarian, as acknowledged even by those who dislike such constructions themselves.

224

Kashmiri

(15) a. b.

Me chal-Ina:v-Inǝ:v’ ra:mni zǝriyi

I.

ERG

wash-

CAUS

-

CAUS

.

MPL

Ram.

GEN

by palav raj-as athi

Raj-

DAT INSTR clothes.

NOM

‘I got clothes washed by Raj through Ram.’

Me chal-Inǝ:v’ ra:mni zǝriyi raj-as athi palav

I.

ERG

wash-

CAUS

.

MPL

Ram.

GEN

by Raj-

DAT INSTR

clothes.

NOM

‘I got clothes washed by Raj through Ram.’ (Manetta (to appear))

Hungarian

(16) a. b.

A tanárral dolgozatot ír-at-tat-t-am the teacher-

INST

composition-

ACC

write-

CAUS

-

CAUS

-

PAST

-1

SG

.

INDEF diák-ok-kal the pupil-

PL

-

INST

‘I had the teacher have the pupils write a composition.’ (Hetzron 1976)

A tanárral dolgozatot ír-at-tam

58 the teachera

INST diák-ok-kal composition-

ACC write-

CAUS

-

PAST

-1

SG

.

INDEF the pupil-

PL

-

INST

‘I had the teacher have the pupils write a composition.’

Tsez

(17) Ala

Ali:

ERG žek’uq old-man:

LOC užiq

child:

LOC magalu scone:

ABS bac’-er-si/bac’-r-er-si eat-

CAUS

-

PAST

/eat-

CAUS

-

CAUS

-

PAST

‘Ali cause the old man to feed a scone to the child.’ (Kulikov 1993)

To sum up these facts: 1. Turkish, Kashmiri, Hungarian, and Tsez all apparently have the ‘extended’ causative construction, with two or more causatives stacked onto a transitive or unergative base that correspond to two or more cause events. 2. All of these languages also have ‘vacuous’ causatives, where suffixes may be iterated without further cause events or arguments being added. 3. In the ‘extended’ causative, a single causative morpheme may be used instead of the expected two. Thus, there is no one-to-one correlation between the number of suffixes and the number of events.

58

See the previous footnote.

225

I claim that the extended causative is not a case of syntactic iteration. Rather, the semantics of indirect causation allow for an indefinite number of causal links that are not syntactically represented. The vacuous use of causative morphemes makes it possible to artificially match the number of suffixes to the number of implied events, but this only creates the illusion of syntactic recursion.

Japanese data strengthen this claim. In this language, it is impossible to stack two outer causative morphemes on the same verb stem, as in the languages reviewed by

Svenonius.

Japanese

(18) *Taroo-ga Hanako-ni Ziroo-ni

Taro-

NOM

Hanako-

DAT

Jiro-

DAT suw-ase-sase-ru. tabako-o cigarette-

ACC smoke-

CAUS

-

CAUS

-

PRS

Intended: ‘Taro will cause Hanako to cause Jiro to smoke.’ (Miyagawa 1999, based on Kuroda 1993)

Nonetheless, an ‘extended’ causative is possible, even without iterating the suffixes.

(19) Taroo-ga Hanako-ni Ziroo-ni tabako-o suw-ase-ru.

Taro-

NOM

Hanako-

DAT

Jiro-

DAT cigarette-

ACC smoke-

CAUS

-

PRS

‘Taro made Hanako make Jiro smoke the cigarette.’ (Miyagawa 1999)

This sentence is identical to the ‘extended’ constructions of the other languages, except that it lacks the doubled causative. The extended causative is grammatical in Japanese; what is ungrammatical is the repetition of the outer causative suffix. Miyagawa (1999) assumes there to be two

CAUS

predicates, the second of which is not pronounced.

However, no such move is necessary. In all of these languages, the semantics of indirect causation allow for interpretations with multiple causal links.

59

The only difference

59

The source of the causee is a question that arises here. In Turkish, this is fairly straightforward: The causee is a freely occurring adjunct (see 6.4, examples (24) and (25), below). However, the same may not

226 between Japanese and the other languages discussed is that Japanese does not allow the vacuous repetition of suffixes, while the others do.

6.3 The Semantics of Indirect Causation

The English construction

have

+

past participle

is a close approximation to the affixal outer causative, to the extent that it denotes indirect causation without embedding a separate TP for the caused event.

(20) a. b.

I had the letter signed.

I had my car repaired.

Although this construction does not allow iteration (*

I had my car had repaired

), it does permit readings with more than one causal link.

(21) a. b.

I had the letter signed by the director by means of a courier.

I had my car repaired by Jim through Tom’s mediation.

In (21a), it is clear that I did not personally have the director sign the letter, but that I got a courier to have him sign it. Similarly, in (21b), it is understood that Jim repaired the car, that Tom arranged the repair, and that I enlisted Tom’s services to make these arrangements. In each case, there is only one

CAUS

predicate, but there are semantically at least two causal events. Thus, we have the illusion of a double causative. How is this achieved?

The trick lies in the semantics of indirect causation. Indirect causation allows any number of intermediate events (Kratzer 2005). Thus, (20b) ‘I had my car repaired’ is equally true if I took the car to the mechanic in person, or if I entrusted it to someone to be true in every language. The additional dative phrase in Japanese, for example, might not be so easily dismissed as the Turkish adjuncts. This matter may need to be addressed on a case-by-case basis in other languages.

227 take it to the mechanic for me, or if I gave it to someone who found someone to take it to the mechanic, and so on. Every caused event has at most one direct cause. As Bittner

(2008) explains, “

The direct

cause of an event

e’

is its < w

-greatest cause -- an event that caused

e’

in

w

and was caused by all the other causes” (30). While the direct cause is necessarily unique, the number of indirect causal links, which comprise “all the other causes,” is potentially limitless.

In the

have

+

past participle

construction, there is a single

CAUSE predicate that semantically allows for one direct cause and an indefinite number of indirect ones. The direct cause may be unambiguously expressed with a

by

phrase, such that in (17b) Jim is the direct cause of the car’s being repaired. Indirect causes, on the other hand, can be picked out by adjuncts such as a

by-means-of

phrase. Given the unambiguous status of the primary causer and the ultimate causer, the adjunct is easily interpreted as being an intermediate link. Note, however, that any attempt to express more than one intermediary puts a strain on the illusion, since intermediate links are not uniquely identified.

(22)

?I had my car fixed by Jim through Tom’s mediation with Joe’s intervention.

Revealingly, in Turkish the illusion breaks at precisely the same point. The following sentence is claimed by Kural (1996) to have five

CAUS

predicates, and yet it shows signs of being a simple case of indirect causation as described above, with

(somewhat excessive) vacuous iteration causative suffixes.

(23) Ahmet Ayşe tarafından soğanlar-ı

A.-

NOM

A. by onions-

ACC

Ali-ye

A.-

DAT doğ-ra-t-tır-t-tır-t-tı

chop-

CAUSE

(x5)-

PAST

-3.

SG i. ‘Ahmet made Ayşe have the onions chopped by Ali.’ ii. ‘Ahmet had Ayşe make Ali chop the onions’

228

The dative phrase unambiguously realizes the immediate causer of the onions’ being chopped, i.e., the one who did the chopping. As to the purported three intermediate causees, Kural notes that only one can be overtly expressed, appearing in a

by

-phrase.

According to Kural, this sentence has two possible interpretations with respect to the intermediate links of the causal chain. The first is that Ayşe is the second link: Ahmet caused Ayşe to cause (through two unnamed intermediaries) Ali to chop the onions. The second interpretation is that Ayşe is the fourth (penultimate) link: Ahmet caused somebody to cause somebody else (both unexpressed) to cause Ayşe to cause Ali to chop the onions. Thus, Ayşe’s exact position in the causal chain is unclear, while the other two purported causees simply cannot be expressed. This array of facts is surprising if there are indeed five

CAUS predicates in the syntax, but fully expected if Turkish multiple causatives are derived semantically rather than syntactically, just as the English examples.

Using English, I have shown that double

CAUS

predicates are not necessary to produce double causative semantics. Furthermore, the limit on expression of causees in

Turkish is in line with the intuition presented for English. In the next section, I will provide further evidence that this is the correct approach to Turkish as well.

6.4 The Illusion in Turkish

The fact that the alleged double-causative sentences in Turkish could be realized with a single

CAUS

predicate has been established, but this on its own does not mean that

Turkish sentences such as (20) are indeed formed this way.

229

(24) Mektub-u müdür-e kâtip vasıta-si-yle imzala-t-tır-dı-m. letter-

ACC

director-

DAT

secretary means-3

SG

-

INST

sign-

CAUS

-?-

PST

-1

SG

‘With the help of the secretary, I got the director to sign the letter.’ (Aissen 1979)

My claim is that the second causative suffix does not realize

CAUSE

. In this section, I present evidence for this claim.

First, the higher causee adjunct, marked by

vasıtasiyle

or the more modern

aracılığıyla

, ‘by means of,’ does not need to be licensed by

CAUS

.

(25)

Şükran kayıp çantasını karakol-da

Şükran lost purse-3

SG

-

ACC

police.station-

LOC aracılığıyla/vasıtasiyle bul-du.

çalış-an work-

PRT bir tanıdık a acquaintance by.means.of find-

PST

‘Şükran found her lost purse by means of an acquaintance who works at the police station.’

Furthermore, as already mentioned, it is possible to construct “double causative” sentences with a single causative morpheme.

(26) Müdür, director yardımcı-sı aracılığıyla/vasıta-si-yle assistant-3

SG

by.means.of işçi-ler worker-

PL temsilci-si-ne yeni bir yönetmelik yaz-dır-dı. representative-3

SG

-

DAT new a by-laws write-

CAUS

-

PST

‘The director, via his assistant, had the workers’ representative write new by-laws.’

If the by-means-of phrase truly introduced a syntactic causee in addition to the dative causee, it should require a second causative suffix. Instead, it looks like this sentence simply has a single outer causative (

CAUS

P) and an extra adjunct phrase. Since the ‘bymeans-of’ adjunct is allowed in the absence of even a single outer causative, it must be hosted by a different projection, such as Voice. This adjunct is labeled ‘OblP’ (oblique phrase) in the tree below.

230

(27) Voice-P

DP müdür

Voice’

OblP

yardımcısı aracılığıyla

Voice’

CAUS

-P

CAUS

Voice-

ACTIVE

–Ø

CAUS

Dative-P işçiler temsilcisine vP v’

CAUS

–DIr

√P

DP yeni bir yönetmelik

Y

AZ v

–Ø

Next, the interpretation of doubled suffixes on an unaccusative base is different from that on an unergative or transitive base (Özkaragöz 1986). Recall that since unaccusatives take inner causatives, and since the only iteration that I claim to be possible is inner plus outer, then only unaccusative bases should unambiguously yield a double causative reading with double suffixes. This is indeed so.

Unaccusatives

(28) a. Can-ı öl-dür-t-tü-m

Can-

ACC die-

CAUS

-

CAUS

-

PST

-1

SG

‘I had Can killed.’

Impossible:‘I killed Can.’ b. Su-yu don-dur-t-tu-m. water-

ACC freeze-

CAUS

-

CAUS

-

PST

-1

SG

‘I had the water frozen.’

Impossible:‘I froze the water.’

An unaccusative base with two causative suffixes is truly a double causative, since it has an outer causative (v

CAUS

) stacked on an inner (vP-

CAUS

). The tree below represents this structure for (28b).

(29) Voice-P

DP

pro

(1

SG

)

CAUS

-P

CAUS

Voice’

Voice-

ACTIVE

–Ø

DP su

√P vP v’

D

ON v

CAUS

–Ø

CAUS

–DIr

However, when the base verb is unergative or transitive, the repetition of causative suffixes does not unambiguously yield a multiple causative interpretation.

Unergatives

(30) a. b.

Can-ı koş-tur-t-tu-m

Can-

ACC run-

CAUS

-

CAUS

-

PST

-1

SG

‘I made Can run.’

‘I had Can made to run.’

Can-ı

çalış-tır-t-tı-m.

Can-

ACC work-

CAUS

-

CAUS

-

PST

-1

SG

‘I made Can work.’

‘I had Can made to work.’

Transitives

(31) a. b.

Yemeğ-i yap-tır-t-tı-m. food-

ACC make-

CAUS

-

CAUS

-

PST

-1

SG

‘I had the food made (by one or more intermediaries).’

Bina-yı yık-tır-t-tı-m building-

ACC

demolish-

CAUS

-

CAUS

-

PST

-1

SG

‘I had the building demolished (by one or more intermediaries).’

231

232

I suggest that the reason these can still be interpreted as single causatives is that there is no syntactic recursion of outer causatives. The possibility of a multiple causation reading simply falls out from the semantics of indirect causation.

If that is the case, then what is the source of the multiple causatives? The DM model of Vocabulary Insertion requires that each VI correspond to a terminal node. I propose that the phenomenon is morphological reduplication, and that it is achieved by the copying of terminal nodes. I further propose that the trigger of this copying is the feature [+focus].

6.5 Causative Iteration as Reduplication

There are certain functions that reduplication tends to realize cross-linguistically.

One of these is the plural.

Warlpiri (noun)

(32) a. b.

kamina kamina kamina

‘girl’

‘girls’ (Inkelas & Zoll 2005)

Carib uses doubled causative suffixes to indicate a plurality of arguments.

Carib (causative)

(33) a.

[...] kaiku:si ?wa kisi:wopoi

‘Do not let him be killed by the jaguar.’ b.

kisi:wopo:poi kaiku:si ?wa

‘Do not let him be killed by all these jaguars.’

wo

- ‘to kill’ -

po:po

- double causative (Kulikov 1993)

Rather than corresponding to two distinct causing events, the doubled Carib causatives appear to have the same function as Warlpiri reduplication. They indicate the plurality of an argument. This is evidence that causative iteration can be a form of reduplication rather than syntactic recursion.

233

Another common function of replication is intensification. We see this with adjectives in Turkish.

(34) a. b. kara kap-kara

‘black’

‘pitch black’

I would like to suggest that the function of causative reduplication in Turkish is focus. This is a novel claim. Göksel & Özsoy (2000) assert that stress is the only indicator of focus in Turkish. A focused constituent tends to appear in the immediately preverbal position (35a), but it need not (35b).

(35) a. b.

Ali-ye yemeğ-i

BEN pişir-di-m.

Ali-

DAT food-

ACC

I

cooked the food for Ali.’

1

SG cook-

PST

-1

SG

BEN Ali-ye yemeğ-i

1

SG

Ali-

DAT food-

I

cooked the food for Ali.’ pişir-di-m.

ACC cook-

PST

-1

SG

The causative, being a suffix, cannot be focused through stress (35a). Primary stress on the verb falls on the last stressable suffix (Göksel & Kerslake 2005), which in (35b) is the past tense morpheme.

(36) a. b.

*Can-ı

Can-

ACC

*Can-ı konuş-TUR-du-m. speak-

CAUS

-

PST

-1

SG konuş-tur-DU-m.

Can-

ACC speak-

CAUS

-

PST

-1

SG

‘I made Can talk.’

The normal word stress rule cannot be violated for the purpose of focus. However, I suggest that there is another option available for the

CAUS

head: It may be reduplicated.

Like stress, reduplication increases the phonological salience of a constituent, but it does so without interfering with normal stress assignment.

Causative reduplication in Turkish may yield an intensified reading.

234

(37) a. b.

Can-ı konuş-tur-t-tu-m.

Can-

ACC speak-

CAUS

-

CAUS

-

PST

-1

SG

‘I forced Can to speak.’

Can-ı

çalış-tır-t-ma-dı-m

Can-

ACC work-

CAUS

-

CAUS

-

NEG

-

PST

-1

SG

‘I didn’t let Can work.’

This construction has much in common with a highly productive form of focus in English analyzed by Ghomeshi et al. (2004), which they name Contrastive Reduplication (CR).

(38) a. b.

I’ll make the tuna salad, and you make the SALAD-salad.

My car isn’t MINE-mine; it’s my parents’. (Ghomeshi et al. 2004)

In CR, an Xº or XP min

is repeated, and the constituent is focused. The semantic effect of CR is to restrict the intended meaning to a subset of the normally possible meanings of the item. Thus, in (38) the contrast is not between salad and things other than salad, but between a narrowly defined salad—the green, leafy kind—and other dishes that fall within the general definition of salad, such as tuna salad, chicken salad, fruit salad, etc. Similarly, the range of items one could call “mine” includes things that one may not be the proper owner of, but CR excludes such things: if something is MINE-mine, then I am its proper owner.

In the following sections, I show how two readings of causative reduplication are derived via CR.

6.5.1 The Coercive Reading

A common interpretation of reduplicated causatives is that of force.

(39) Can-ı

Can-

ACC konuş-tur-t-tu-m. speak-

CAUS

‘I forced Can to speak.’

-

CAUS

-

PST

-1

SG

235

Causation may be conceptualized as a set whose members include different types and degrees, of which the most prototypical is coercion. What more basic way of making someone do something is there than to

force

him? Indeed, this is the interpretation achieved when the English

make

-causative undergoes CR. The following examples are taken from discussion boards.

(40) a. b.

Didn't "Make" make him

do anything, Full doors are more expensive than half doors and that was his offer not mine i didn't mean as in a use of force to

make make her

In the first example, the discussant has traded full jeep doors for half doors and in addition received $1000 from the other party in the trade. He is responding to another discussant who has criticized him for making the other man give him $1000. The sense might be paraphrased thus: “I

did

make him give me the money in the sense that I wouldn’t have traded the doors without reimbursement, but I did not force him to.” In the second example, the discussant has provided such a paraphrase himself.

Thus, if causative reduplication is an instance of CR, the coercive semantics of examples such as (39) are derived straightforwardly. Next, I will show how the permissive semantics can be derived in similar fashion.

6.5.2 The Permissive Reading

Another reading found in causative reduplication is the permissive.

(40) Can-ı

çalış-tır-t-ma-dı-m

Can-

ACC work-

CAUS

-

CAUS

-

NEG

-

PST

-1

SG

‘I didn’t let Can work.’

While the derivation of the coercive reading is straightforward, that of the permissive one is less obviously so. How could simply permitting an event to occur ever

236 constitute a prototypical instance of causation? The first step towards answering this question is understanding the modal relationship between so-called coercive and permissive causatives.

As discussed in the preceding chapter, Kural (1996, 2000) proposes a solution to a long-standing scope ambiguity problem in Turkish. Causative and verbal negation suffixes are subject to strict ordering, the former obligatorily preceding the latter, and yet both scope readings appear to be available.

(41) konuş-tur-m-uyor speak-

CAUS

-

NEG

-

DUR

‘s/he doesn’t make (someone) speak’

‘s/he makes (someone) not speak’

Kural notes that the Turkish causative suffix is ambiguous between coercion and permission.

(42) Hoca teacher

öğrenci-ler-i student-

PL

-

ACC konuş-tur-uyor. speaka. ‘The teacher makes the students speak.’ b. ‘The teacher lets the students speak.’

CAUS

-

DUR

Since the first reading does not actually entail the use of coercive force (though not ruling it out), I will refer to it simply as the

make

-reading. For consistency’s sake, and for another reason that will soon become clear, I will refer to Kural’s permissive causative as the

let

-reading. According to Kural, the

make

-reading is high-involvement, since the causer actively brings about the event, and the

let

-reading is low-involvement, since the “causer” simply fails, through inaction, to prevent the event from occurring.

Kural defines the relationship between the two as a kind of modality in which the

make

causative is a universal modal and the

let

-causative an existential one. The relationship

237 between modals and negation is such that a proposition with an existential modal and negation is truth-value-equivalent to a universal modal with the inverse scope of negation.

(43)

Not every child had fun = some child didn’t have fun

Every child didn’t have fun = there was not some child who had fun

If

make

- and

let

-causatives stand in such a modality relationship, then it follows that they will exhibit this property.

(44)

I didn’t make John sing = I let John not sing.

I didn’t let John sing = I made John not sing.

Therefore, the ambiguity in (42) is not one of scope, but of the modality of the causative. In (42a), the

make

-causative is in the scope of negation. In (42b), the

let

causative is in the scope of negation, which is truth-equivalent to a proposition in which negation is in the scope of the

make

-causative.

There is a second relationship that Kural does not point out: When modals are negated, the entailment is reversed. In the affirmative, the direction of entailment is from the universal to the existential. If every boy had fun, then it is entailed that some boy had fun, but if some boy had fun, this does not entail that every boy had fun. With negation, the direction of entailment is from the existential to the universal: if some boy didn’t have fun, then it is entailed that not every boy had fun.

The same relationship holds with the causative modals:

(45) a. b. c. d.

I made the prisoner escape > I let the prisoner escape

I let the prisoner escape ¬> I made the prisoner escape

I didn’t let the prisoner escape > I didn’t make the prisoner escape

I didn’t make the prisoner escape ¬> I didn’t let the prisoner escape

238

We could express the entailments in terms of subsets: worlds in which I make the prisoner escape compose a subset of worlds in which I let him escape, and worlds in which don’t let him escape are a subset of worlds in which I don’t make him escape.

Furthermore, in each case, the subset is the high-involvement causative.

Let

is lowinvolvement only in the affirmative; in the negative, it is high-involvement. If I don’t let the prisoner escape, then I actively prevent him from escaping, whereas if I don’t make him escape, I am inactive. Therefore, the subset of the affirmative

make

-causative on the one hand, and that of the negative

let

-causative on the other, are in both cases the highinvolvement subset—i.e., the prototypical subset of the type

cause

.

We can therefore predict that CR will restrict possible readings to the

make

causative in the affirmative, and to the

let

-causative in the negative. The following sentences, without CR, are ambiguous between the two readings.

(46) a. b.

Öğretmen öğrenci-ler-i konuş-tur-du. teacher student-

PL

-

ACC speak-

CAUS

-

PST

‘The teacher made the students speak.’

‘The teacher let the students speak.’

Öğretmen öğrenci-ler-i konuş-tur-m-uyor. teacher student-

PL

-

ACC speak-

CAUS

-

NEG

-

PST

‘The teacher doesn’t make the students speak.’

‘The teach doesn’t let the students speak.’

It should be noted that, despite being ambiguous, these sentences are already biased towards one reading or the other. The more salient interpretation of (46a) is that the teacher makes the students speak, but the second interpretation, in which he simply fails to silence them, is also possible. With negation, the inverse holds: the more salient reading of (46b) is that the teacher does not permit the students to speak, although it is possible to understand the sentence as meaning that he does not make them speak. For

239 example, this reading is possible if one is criticizing a foreign-language teacher for not actively motivating his students to talk.

CR biases the sentences towards one reading or the other, but in seemingly opposite directions for the affirmative and the negative. (It should be noted that the readings given below are the most salient ones, but not necessarily the only readings available.)

(47) a. b.

Öğretmen öğrenci-ler-i konuş-tur-t-tu. teacher student-

PL

-

ACC speak-

CAUS

-

CAUS

-

PST

‘The teacher forced the students to speak.’ (

make

-causative)

Öğretmen öğrenci-ler-i konuş-tur-t-madı. teacher student-

PL

-

ACC speak-

CAUS

-

NEG

-

PST

‘The teacher [absolutely] did not let the students speak.’ (

let

-causative)

However, this ambiguity resolution is in fact in a single direction: towards the highinvolvement subset. As we have already seen, the affirmed

make

-causative and the negated

let

-causative each constitute the high-involvement subset of the superset of possible causative events. But CR goes beyond mere ambiguity resolution. The sentences in (47) have a sense of force. The talking of the students is either caused or prevented by means of overwhelming or oppressive tactics. In the affirmative, this can be conveyed by the verb “to force” in English. In the negative, English has no verb that readily expresses this idea, so in the translation the verb “to let” must be reinforced by an adverbial such as

“absolutely” or “categorically.” Thus, CR picks out a subset of a subset: the subset of high-involvement causation in which overwhelming tactics, such as violence or authority, are employed. This is the prototypical type of causation, in which the causer makes absolutely certain that the event occurs, or does not occur, as the case may be.

240

Similarly, when the English causative “make” undergoes CR, it gets the oppressive reading. But there is a puzzle: Why do Turkish and English CR causatives behave differently when negated?

(48) Yap-tır-t-ma-dı-m. make-

CAUS

-

CAUS

-

NEG

-

PST

-1

SG

‘I didn’t let (him) do (it).’ (negative

let

causative)

(49)

I didn’t “make” make him do anything. (negative

make

causative)

As discussed above, negated CR yields an oppressive

let

-causative in Turkish, but apparently not in English. The solution to this puzzle lies in the fact that the Turkish affixal causative is ambiguous between a

make

and a

let

reading, while English “make,” quite obviously, is not. Therefore the only possible interpretation of (49) is, “It is not the case that I forced him.”

6.5.3 The number of repetitions

There is some degree of optionality in the number of iterations of the causative morpheme.

(50)

Yemeğ-i yap-tır-t-(tır)-dı-m food-

ACC make-

CAUS

-

CAUS

-(

CAUS

)-

PST

-1

SG

‘I had the food made.’ (with an implication of several intermediaries)

Based on the analysis presented thus far, this is not expected. If the first causative realizes v-

CAUSE

and the second is a reduplicant realizing the focus head, then what could a third occurrence be, and why would it be optional? In fact, the answers to both of these questions fall out naturally if causative focus reduplication is an instance of CR.

Ghomeshi et al. point out that certain material that is not part of the focused constituent

241 can optionally be reduplicated. Such material includes unstressed object pronouns and inflectional suffixes.

(51) a. b.

Do you LIKE-’EM like-’em?

Do you LIKE like-’em?

Some speakers have a strong preference for (51a) where the object pronoun is repeated, while others find (51b) equally acceptable, in which only the verb is repeated. Ghomeshi et al. conclude that the motivation for including non-focalized material is purely prosodic.

They formulate the constraint as follows:

(52) The reduplicant preferably contains the same number of syllables as the prosodic constituent containing the base of CR. (Ghomeshi et al. 2004)

Recall that the Turkish causative allomorphs -

t

and -

DIr

have complementary phonological environments, and that each provides the environment for the other.

Therefore, if the initial allomorph is -DIr, a single reduplicant will have the form -t.

(53) yap-tır-t- do-

CAUS

-

CAUS

-

The reduplicant is a prosodic mismatch of the base: -

DIr

is a full syllable, while -

t

is nonsyllabic. However, simply repeating the -

DIr

allomorph, in a purely phonological reduplication, results in a morphophonological violation.

(54) *yap-tır-dır-

A solution to this problem is to sandwich the allomorph -t between the two occurrences of -

DIr

. -

t

provides the licit phonological environment for the repetition of -

DIr

, and in effect realizes neither

CAUS

nor focus, but simply phonological support.

(55) yap-tır-t-tır- do-

CAUS

-

t

-

FOC

-

242

Occurrences beyond three, such as the alleged quintuple causative in (19), are not as readily explained. However, it should be pointed out that such occurrences are extremely rare and marginal, while double and triple repetitions are quite common.

Furthermore, to the extent that such marginal uses do arise, they exclusively yield the serial causation reading, and therefore seem to appeal to some kind of iconicity in exaggerated repetition.

6.5.4 The Vacuous Reading

There are also cases, in Turkish and in the other languages discussed in 6.2, where the iteration appears to be truly vacuous. No intensive, high-involvement reading is contributed by the doubled causative marking.

(56) a. b.

Saç-ım-ı kes-tir-di-m hair-1

SG

-

ACC

cut-

CAUS

-

PST

-1

SG

‘I had my hair cut.’

Saç-ım-ı kes-tir-t-ti-m hair-1

SG

-

ACC

cut-

CAUS

-

CAUS

-

PST

-1

SG

‘I had my hair cut.’ (Göksel and Kerslake 2005)

In these cases, I claim that the reduplication still realizes focus. However, in these cases, it is simple information focus (in the sense of É. Kiss 1998) of the verbal complex. Since the verb phrase is the default locus of presentational focus, this does not add any special meaning to the sentence.

6.6 Formalization of Causative Focus Reduplication

Causative focus reduplication in Turkish cannot be phonological in nature, since the two VIs do not share any phonological segments. Although the initial segment of –

DIr

is realized as /t/ in some environments, the vacuous causative following it is always

243

/t/, even though following a liquid is a position where the phoneme /D/ would be realized as /d/.

(57) a. b.

ver-dir-t-ir

give-

CAUS

-

CAUS

-

AOR

‘s/he will make give’

*

ver-dir-d-ir

(57b) is ill-formed, even though the first segment of the ‘base’ is /d/, not /t/, and the

‘reduplicant’ likewise finds itself in an environment where /D/ is realized as /d/. In no way can –

t

be a phonological reduplicant of a base –

DIr

. In cases where –

t

precedes, the first consonant of –

DIr

is always realized /t/, and is hence identical to the ‘base.’ Here, however, a different but equally intractable problem arises: the ‘reduplicant’ is larger than the base. Clearly then, there is no basis for treating causative iteration as phonological reduplication.

The alternative is morphological reduplication. The formalization of such a process is a simple matter in a realizational framework such as DM: the presence of a feature (such as focus) at a terminal node triggers the copying of that node. Since there will now be two copies of the terminal node

CAUS

, at Vocabulary Insertion two separate

VIs will be inserted to realize

CAUS

. The fact that these are allomorphs with no shared phonological segments is not problematic.

Morphological Doubling Theory (MDT) (Inkelas & Zoll 2005) is a theory of reduplication designed specifically to account for cases of reduplication that cannot be explained phonologically. Although formulated within the framework of Sign-Based

Morphology, the model it presents is fully compatible with DM. The position taken in

MDT is that a morphological approach is necessary to account for a large number of

244 cases of reduplication, and that such an approach can also explain all cases analyzed by others as involving a skeletal morpheme R

ED

that copies phonological material from a base, and therefore that the postulation of R

ED

is unwarranted. The only cases of phonologically-based reduplication, according to Inkelas & Zoll, are those where a minimal piece of phonology (generally a single segment) is copied to satisfy a phonological well-formedness constraint, rather than to realize a morpheme. In the MDT formalization, two sister nodes having identical semantic content are dominated by a mother node having the same content plus additional semantics. In the case at hand, the sisters would both have the content [

CAUS

], while the mother would have this in addition to another feature (say, Focus). This is shown in (58):

(58) causative reduplication in an MDT model

Semantics:

CAUS

,

FOCUS

Semantics:

CAUS

Semantics:

CAUS syntax.

In the following derivation, the terminal node [

CAUS

,

FOCUS

] is merged in the

245

(59) a.

DP

√P

DP vP v’

K

ES

Voice-P

Voice’

CAUS

-P

CAUS

’ v

CAUS

[

CAUS

,

Voice-

FOCUS

]

ACTIVE b.

Saç-ım-ı kes-tir-t-ti-m

hair-1

SG

-

ACC

cut-

CAUS

-

CAUS

-

PST

-1

SG

‘I had my hair cut.’

At Morphological Form, the node [

CAUS

,

FOCUS

] is split into two nodes. This process is similar to fission, which also turns a single node into multiple ones (see Chapter 3,

Section 3.4.3). The difference is that in fission, the features of the original node are distributed among the new nodes, such that, for example, the bundle [v,

CAUS

] is realized at two nodes, one with [v] and the other with [

CAUS

]. In focus reduplication, on the other hand, creation of the new node discharges the focus feature so that it does not require further exponence, while each of the new nodes contains a copy of [

CAUS

]. At

Vocabulary Insertion, each receives a separate causative VI.

246

(60) Voice-P

DP

pro

[1

SG

]

CAUS

-P

CAUS

Voice’

Voice-

ACTIVE

–Ø

DP saçımı

√P vP v’

K

ES v

CAUS

–Ø

[

CAUS

]

–DIr

[

CAUS

,

FOCUS

]

[

CAUS

]

–t

6.7 Conclusion

Languages with affixal causatives apparently differ in terms of whether they have causative recursion. Sámi, Hindi, Nivkh, Amharic, Kitharaka, Malagasy (Svenonius

2005a,b) and Japanese (Miyagawa 1999) disallow repeated outer causative morphemes.

In contrast, Turkish, Kashmiri (Manetta (to appear)), Hungarian (Hetzron 1976), and

Tsez (Kulikov 1993) all appear to allow recursion. I have concluded, however, that what really distinguishes the latter group from the former is the ability to iterate causative suffixes without syntactic recursion. The semantics of indirect causation allow for interpretations with multiple causal chains in many, perhaps all, languages, including

English.

The morphological iteration is not recursion, but reduplication, and it has the effect of focusing either the

CAUS

phrase or the entire verb phrase. This focus may, among other things, increase the salience of the indirect causation, facilitating an

247 interpretation with multiple causal links. Since the agents of these implied links can be expressed through adjunct phrases, this effectively creates the illusion of syntactic recursion, and this has been taken at face value by linguists for decades. The lack of a precise correlation between the number of suffixes and the number of causees has either been ignored or attributed to sloppiness. It is important to recognize this phenomenon for what it is, as it has significant implications for syntactic analysis. Failure to see past the apparent recursion could lead one far down the wrong path.

Furthermore, any analysis that assumes recursion in Turkish and Hungarian, among others, has the onus of explaining why Malagasy and various other languages lack such recursion, a very tricky problem from a syntactic point of view. On the proposal advanced herein, these languages do not differ syntactically in terms of recursion. The only difference between them is that some allow causative reduplication and others do not. This, too, is a compelling question, but it is no deeper than, for example, the question of why some languages realize the plural through reduplication but others do not. This is a problem that is already with us. There is no need to increase the explanatory burden by adding a mysterious discrepancy between languages in syntactic recursion, especially when a close examination of the facts show the phenomenon not to be recursion at all.

248

CHAPTER 7: CONCLUSION

An important thematic thread running through this dissertation is the opposition of DM’s all-syntactic derivation to lexicalism, specifically the weak lexicalism of

Reinhart (2002) and Horvath & Siloni (2011a, b), who advocate a split model with both lexical and syntactic derivation. In their model, the lexicon is a generative engine where there are no hierarchic relations, and where idiosyncrasy and irregularity abound, while the syntax is the locus of hierarchic structure and regularity. If we compare the two approaches simply as models attempting to account for empirical data, we see that

Reinhart, Horvath, and Siloni’s lexicon corresponds to small structures in DM, while their syntax corresponds to large structures in DM. Lexical/Root causatives exhibit idiosyncratic morphology and semantics. To the weak lexicalist, these are properties of lexical derivation, but to the Distributed Morphologist, these are properties ascribed to a particular, very small structure: the merger of the √P with the first functional head. As for syntactic/productive causatives, they exhibit biclausal structure in Japanese, but monoclausal structure in Turkish and Hungarian. To the weak lexicalist, this means that

Japanese causatives are built in the syntax, where hierarchic structure is available, and

Turkish and Hungarian causatives are built in the lexicon, where it is not. To the

Distributed Morphologist, it means that the latter type of causative has less structure than the former; more specifically, the latter lacks the embedded Voice projection that the former has.

DM has clear advantages over the dual-engine model of the weak lexicalists. In syntactic derivation, different types and amounts of structure are available to account for

249 linguistics phenomena, and to the extent that the lexicalists make use of syntactic derivation, it is available to them as well. In DM, this available in all derivations, because all derivations are syntactic, but for the weak lexicalists, this avenue is not available for derivation in the lexicon. By consigning Hungarian outer/productive causatives to the lexicon, they have limited means to distinguish between these and inner/Root causatives.

For example, if idiosyncratic morphology is a property of the lexicon, why is it exhibited only by inner causatives and not by outer ones? Also, the arity operation of

Causativization in the Lexicon offers no obvious way of distinguishing between instrumental and accusative causees of unergative verbs. In a syntactic account, the two have distinct structural positions. I have not determined the structures of causatives of unergatives as explicitly as I have the causatives of transitives, but the instrumental causee of an unergative is either an adjunct, or an argument in the specifier position of

CAUS

P. (The latter seems plausible given that expression of the instrumental causee of an unergative is not permitted, unlike the instrumental causee of a transitive verb.) On the other hand, the accusative causee of an unergative is much lower in the structure, probably the complement of the Root. Whatever the specifics of the final analysis, it is clear that they are in structurally distinct positions. According to Horvath & Siloni

(2011a), causatives of unergatives are the output of a lexical arity operation,

Causavization in the Lexicon, the same operation that produces causatives of transitives.

Case marking of the causee would apply in the syntax, to the output of the lexical operation, so the source of the variable case marking is unclear. The difference in case-

250 marking not trivial: an accusative causee can participate in idiom formation (1a), but an instrumental one cannot (1b).

(1) a. A the

Literal:

Idiomatic: b. A the

Literal:

Idiomatic: férfi fut-tat-ta a lány-ok-at man run-

CAUS

-

PAST

.

DEF

.

DO

‘The man made the girls run.’ the girl-

PL

-

ACC

‘The man pimped the girls (managed them as prostitutes).’ férfi fut-tat-ott

Unavailable man run-

CAUS

-

PAST

.

INDEF

.

DO

‘The man made the girls run.’ a the lány-ok-kal girl-

PL

-

ACC

Since idiom formation is idiosyncratic, it is natural that the weak lexicalists would locate the contsruction in (1a) in the lexicon. The natural way to account for the impossibility of idiom formation in (1b) would be to locate this construction in the syntax. This would also provide a natural way of explaining the different case marking. Unfortunately, this particular application of the dual-engine approach is not available, since the analysis in

Horvath & Siloni (2011a) requires that

all

causatives in Hungarian are derived in the lexicon. The distinction between (1a) and (1b) appears to be beyond the power of their model to capture.

On the other hand, this distinction is evidence in favor of a syntactic derivation.

Heads may form idioms with their complements (Marantz 1984), so the possibility of idiom formation with an accusative argument is unsurprising if that argument is merged as a complement. On the other hand, higher structures such as adjuncts and external arguments are not expected to participate in idiom formation with the structures they merge with, and so the unavailability of the idiom with an instrumental causee is also unsurprising.

251

One of the main hypotheses informing this study is the that productive (outer) causatives in Turkish embed vP but not a VoiceP. The primary motivation for this was to implement Pylkkänen’s (2002, 2008) proposal about Voice to account for the monoclausal behavior of Turkish (and Hungarian) causatives. I accomplished this, showing again that phenomena that purportedly force the conclusion of derivation in the lexicon can easily be captured in a syntactic model. Furthermore, I showed that the new syntactic model can capture at least one significant fact that H&S’s causativization in the lexicon cannot: The requirement that the causee be animate.

It is always desirable that a proposal to explain one set of phenomena should have applications in other areas. This is an indication that the proposal has objective reality, and is not merely an ad hoc solution (a ‘hack’ in the vernacular of linguists) invented just to cover up an embarrassment in one’s analysis. The hypothesis that some languages have outer causatives that do not embed a Voice projection is already given support in several domains in Pylkkänen (2002, 2008), so if the clausality facts of Turkish and Hungarian causatives were this dissertation’s only application of this hypothesis, this would still be a valuable result. But this hypothesis also provides an explanation for another set of phenomena herein: The fact that some inchoative verbs can be made transitive by the addition of a causative morpheme, but others cannot. If all inchoative morphemes realize

[v,

BECOME

], then the challenge is to explain why a causative can be added to an inchoative in some cases but not others. A rigorous implementation of blocking within the DM model, where morphemes block morphemes at terminal nodes (Embick &

Marantz 2008), could not account for this. This would appear to be evidence that

252 blocking is semantic (Aronoff 1976) and not structural. One would be compelled either to conclude that words can block other words, or else to make vague claims (‘hand-waving’ in the vernacular) about persistent paradigm gaps (as Key 2012b does). The Voice hypothesis provides a better solution. In a close examination of Turkish causative/inchoative pairs, I show that whenever an inchoative fails to causativize, the inchoative morpheme is syncretic with the passive. Conversely, whenever an inchoative morpheme that is not syncretic with the passive appears, it can be causativized. The conclusion from this is straightforward: There are two inchoative structures. One is like that found in Harley (2008), where a [v,

BECOME

] node is merged with a √P (or nP or adjP). This structure always admits causativization. The other structure is middle Voice, in the strict sense of the term ‘Voice,’ in that a Voice head (of the middle ‘flavor’) merges with a vP. These cannot be causativized in Turkish, for the simple reason that

Turkish causatives do not embed Voice.

The predictions following from this analysis need to be tested on a much larger scale. The conjunction of these two applications of Pylkkänen’s proposal yields an obvious prediction regarding outer causatives: Within a language, there should be a correlation between the results of clausality diagnostics and the possibility of embedding inchoatives. In a language such as Japanese, whose causatives test as biclausal, it should always be possible to causativize inchoatives, whereas in a language such as Hungarian, whose causatives test as monoclausal, there may be inchoatives that fail to causativize

(although there also may not be, if that language lacks middle Voice inchoatives). The

prediction is that these two factors should be correlated in all languages with outer causatives. The testing of this prediction remains for future work.

253

254

REFERENCES

Acartürk, C. & Zeyrek, D., 2010. Unaccusative/Unergative Distintion in Turkish: A

Connectionist Approach.

Proceedings of the 8th Workshop on Asian Language

Resources

, 111–119, Beijing, China, 21–22 August 2010.

Aissen, J., 1979.

The Syntax of Causative Constructions

. New York: Garland Publishing.

Aissen, J. & Hankamer, J., 1980. Lexical Extension and Grammatical Transformations.

In

Proceedings of the Sixth Annual Meeting of the Berkeley Linguistics Society

.

University of California, Berkeley.

Alexiadou, A., Anagnostopoulou, E. & Schäfer, F., 2006. The properties of anticausatives cross-linguistically. In

Phases of Interpretation

, M. Frascarelli (ed), 187–211.

Berlin: Mouton de Gruyter.

Arad, M., 2003. Locality Constraints on the Interpretation of Roots: The Case of Hebrew

Denominal Verbs.

Natural Language and Linguistic Theory

22: 737–778.

Aronoff, M., 1976.

Word Formation in Generative Grammar

. Linguistic Inquiry

Monograph Series. Cambridge: MIT Press.

Baker, M., 1985. The Mirror Principle and Morphosyntactic Explanation.

Linguistic

Inquiry

16: 373-415.

Banguoğlu, T., 1986.

Türkçenin Grameri

. Ankara: Türk Tarih Kurumu Basım Evi.

Bhatt, R. & Embick, D., 2003. Causative derivations in Hindi. Unpublished manuscript,

University of Texas at Austin and University of Pennsylvania.

Bittner, M, 2008. Concealed Causatives.

Natural Language Semantics

7: 1-78.

Bobaljik, J., 2000. The ins and outs of contextual allomorphy. In

University of Maryland

Working Papers in Linguistics

10, K.K. Grohmann & C. Struijke (eds), 35–71.

College Park: University of Maryland.

Çetinoğlu, Ö., Butt, M. & Oflazer, K., 2009. Mono/biclausality of Turkish causatives. IN

Turcologica 79: Essays on Turkish Linguistics

, S. Ay, Ö. Aydın, İ.

Ergenç, S. Gökmen, S. İşsever & D. Peçenek (eds). Wiesbaden: Harrassowitz

Verlag.

Chierchia, G., 1989. A Semantics for Unaccusatives and Its Syntactic Consequences. MS.

Cornell University.

255

Chierchia, G., 2004. A Semantics for Unaccusatives and its Syntactic Consequences. In

The Unaccusativity Puzzle

, A. Alexiadou, E. Anagnostopoulou &

M. Everaert (eds). Oxford/New York: Oxford University Press.

Chomsky, N., 1970. Remarks on Nominalization. In

Readings in Transformational

Grammar

. A. Jacob & P. Rosenbaum (eds). Waltham, MA: Ginn and co.

Chomsky, N., 1995.

The Minimalist Program

. Cambridge: MIT Press.

Chomsky, N., 2001. Derivation by phase. In

Ken Hale: A life in language

, M.

Kenstowicz (ed), 1–52. Cambridge: MIT Press.

Clauson, G. 1972.

An etymological dictionary of pre-thirteenth-century Turkish

. Oxford:

Clarendon.

Comrie, B., 1989.

Language Universals and Linguistic Typology, Second Edition

.

Chicago: The University of Chicago Press.

Dede, M., 1986. Causatives in Turkish. In

Proceedings of the National Conference on

Turkish Linguistics.

A. Aksu-Koç & E. Erguvanlı Taylan (eds), 49–58. Istanbul:

Boğaziçi University Publications.

É Kiss, K., 1998. Identificational focus versus information focus.

Language

74:

245–273.

Embick, D., 2003. Locality, Listedness, and Morphological Identity.

Studia Linguistica

57: 143–169.

Embick, D. & Marantz A., 2008. Architecture and Blocking.

Linguistic Inquiry

39: 1–53.

Erdal, M., 1991.

Old Turkic Word Formation: A Functional Approach to the Lexicon

.

(Vol. 1 & 2). Wiesbaden: Otto Harrassowitz.

Erdal, M., 1998. Old Turkic. In

The Turkic Languages

, L. Johanson & É. Csató (eds),

138–157. London: Routledge.

Fabb, N., 1988. English affixation is constrained only by selectional restrictions.

Natural

Language & Linguistic Theory

6:527–540.

Folli, R. & Harley, H., 2005. Consuming results in Italian and English: Flavors of v. In

Aspectual inquiries

, P. Kempchinsky & R. Slabakova (eds), 95–120. Dordrecht:

Springer.

256

Folli, R. & Harley, H., 2007. Causation, obligation, and arguments structure: On the nature of little v.

Linguistic Inquiry

, 28: 197–238.

Folli, R., Harley, H. & Karimi, S., 2005. Determinants of event type in Persian complex predicates.

Lingua

vol. 115, no. 10, 1365–1401.

Ghomeshi, J., Jackendoff, R., Rosen, N. & Russell, K., 2004. Contrastive Focus

Reduplication in English (The Salad-Salad Paper).

Natural Language &

Linguistic Theory

, 22: 307–357.

Givón, T., 1976. Some constraints on Bantu causativization. In

Grammar of Causative

Constructions: Syntax and Semantics 6

, M. Shibatani (ed), 325–351. New York:

Academic Press.

Göksel, A., 1993.

Levels of Representation and Argument Structure in Turkish

. Doctoral dissertation. University of London.

Göksel, A., 2001. The auxiliary verb

ol

at the morphology–syntax interface. In

The Verb in Turkish

, E. Erguvanlı Taylan (ed), 151–178. Philadelphia: John Benjamins

Publishing Co., North America.

Göksel, A. & Kerslake, C., 2005.

Turkish: A Comprehensive Grammar

. New York:

Routledge.

Göksel, A. &

Ö zsoy, S., 2000. Is there a focus position in Turkish. In

Studies in Turkish and Turkic Languages: Proceedings of the Ninth International Conference on

Turkish Linguistics

, A. Göksel & C. Kerslake (eds.), 219–229.

Grimshaw, J., 1982.

Argument structure

. Cambridge: MIT Press.

Hale, K. & Keyser, S.J., 1993. On argument structure and the lexical expression of syntactic relations. In

The View from Building 20

, K. Hale & S.J. Keyser (eds),

53–109. Cambridge: MIT Press.

Hale, K. & Keyser, S.J., 2002.

Prolegomena to a theory of argument structure

.

Cambridge: MIT Press.

Halle, M. & Marantz, A., 1993. Distributed morphology and the pieces of inflection. In

The View from Building 20

, K. Hale & S.J. Keyser (eds), 111–176. Cambridge:

MIT Press.

Halle, M., 1997. Distributed morphology: Impoverishment and fission. In

MITWPL 30:

Papers at the Interface

, B. Bruening, Y. Kang & M. McGinnis (eds), 425–449.

Cambridge: MITWPL.

257

Hankamer, J., 2013. Turkish Monosyllabism Revisited.

Journal of Linguistics Research

(Dilbilim Araştırmaları Dergisi)

, special issue: “Updates in Turkish Phonology,”

E. Erguvanlı Taylan (ed).

Hara, M., 1999. Causatives in Japanese. Manuscript, Truman State University, Kirksville,

MO.

Harley, H., 1995.

Subject, event, and licensing

. Doctoral dissertation, MIT, Cambridge,

MA.

Harley, H., 2008. On the causative construction. In

The Handbook of Japanese

Linguistics

, S. Miyagawa and Mamoru Saito (eds), 20–53. Oxford: OUP.

Harley, H., 2013. External arguments and the Mirror Principle: On the distinctness of

Voice and v.

Lingua

125: 34–57.

Harley, H., to appear (a). On the identity of roots.

Theoretical Linguistics

.

Harley, H., to appear (b). Semantics in Distributed Morphology.

Semantics: An

International Handbook of Natural Language Meaning: Vol. 3

, C. Maienborn, K. von Heusinger & P. Portner, (eds). Berlin: Mouton de Gruyter.

Harley, H. & Noyer, R., 2000. Formal versus Encyclopedic Properties of Vocabulary:

Evidence from Nominalizations. In

The Lexicon-Encyclopedia Interface

, B.

Peters (ed.), Elsevier Press, August 2000.

Haspelmath, M., 1993. More on the typology of inchoative/causative alternations. In

Causatives and Transitivity

, B. Comrie & M. Polinsky (eds), 87–

Amsterdam/Philadelphia: John Benjamins Publishing Company.

Hengirmen, M., 1995.

Türkçe Dilbilgisi

. Ankara: Engin Yayınevi.

Hetzron, R., 1976. On the Hungarian Causative Verb and its Syntax. In

Syntax and

Semantics: The Grammar of Causative Constructions

, Shibatani, M. (ed). New

York: Academic Press. von Heusinger, K. & Kornfilt, J., 2005. The case of the direct object in Turkish:

Semantics, syntax and morphology.

Turkic Languages

9: 3–44.

Horn, L., 1985. Metalinguistic negation and pragmatic ambiguity.

Language

61: 121–

174.

Horvath, J. & Siloni, T., 2002. Against the Little-v Hypothesis.

Rivista di Grammatica

Generativa

27: 107–122.

258

Horvath, J. & Siloni, T., 2011a. Causatives across components.

Natural Language &

Linguistic Theory

, 29: 657–704.

Horvath, J. & Siloni, T., 2011b. Anticausatives: Against reflexivization.

Lingua

121:

2176–2186.

Inkelas, S. & Orgun, C.O., 1995. Level ordering and economy in the lexical phonology of

Turkish.

Language

, 71.

Inkelas, Sharon & Cheryl Zoll, 2005.

Reduplication

. Cambridge, UK: Cambridge

University Press.

Ito, J. & Hankamer, J., 1989. Notes on monosyllabism in Turkish. In Ito & Runner, eds.,

Phonology at Santa Cruz. Syntax Research Center working papers, UCSC.

Jaeckel, R. & Erciyeş, G.D., 2006.

A Dictionary of Turkish Verbs in Context and by

Theme

. Washington, D.C.: Georgetown University Press.

Kearns, K., 2000.

Semantics

. New York: Palgrave MacMillan.

Key, G., 2008. Differential Object Marking in a Medieval Persian Text. In:

Aspects of

Iranian Linguistics

, S. Karimi, V. Samiian & D. Stilo (eds). Newcastle upon

Tyne, UK: Cambridge Scholars Publishing.

Key, G., 2009.

Differential Object Marking as an Areal Trait in Persian and Turkish

.

Unpublished MA thesis, University of Arizona.

Key, G., 2012a. Differential Object Marking in Turkic and Persian as a Contact

Phenomenon. In

Proceeding of the 38th Berkeley Linguistics Society

.

Key, G., 2012b. The Causative/Inchoative Alternation, and the Decomposition of Little v.

Coyote Papers: Working Papers in Linguistics 19

, D. Brenner (ed). University of

Arizona Linguistics Circle.

Koontz-Garboden, A. 2009. Anticausativization.

Natural Language & Linguistic Theory

.

27: 77–138.

Kornfilt, J., 1997.

Turkish

. London: Routledge.

Kratzer, A., 1996. Severing the external argument from its verb. In

Phrase structure and the lexicon

, J. Rooryck & L. Zaring (eds), 109–137. Dordrecht: Kluwer.

Kratzer, A., 2005. Building Resultatives. In

Event Arguments in Syntax, Semantics, and

Discourse

, C. Maienborn & A. Wöllstein-Leisten, 177–212. Tübingen, Niemeyer.

259

Kulikov, L.I., 1993. The “second causative”: a typological sketch. In:

Causatives and

Transitivity

, B. Comrie & M. Polinksy (eds), 121–154. Amsterdam/Philadelphia:

John Benjamins Publishing Company.

Kural, M., 1996.

Verb Incorporation and Elementary Predicates

. Doctoral dissertation,

UCLA.

Kural, M., 2000. Modality in Causatives. In

Proceedings of the Twenty-Third Annual

Meeting of the Berkeley Linguistics Society

, M.L. Juge & J. L. Mosley (eds), 224–

233. Berkeley, California.

Kuroda, S.-Y., 1965.

Generative grammatical studies in the Japanese language

. Doctoral dissertation, MIT, Cambridge, MA.

Kuroda, S.-Y., 2003. Complex predicates and predicate raising. In

Formal Japanese syntax and Universal Grammar: The past 20 years

, N. Fukui (ed),

Lingua

4: 447–

480.

Laks, L., 2011. How did Hebrew ‘freeze’ (

hikpi

) defrost? Morphological filling of frozen lexical entries. Handout of a talk.

Levin, B. & Rappaport Hovav, M., 1995.

Unaccusativity

. Cambridge: MIT Press.

Lewis, G., 1967.

Turkish Grammar

. Oxford: OUP.

Lewis, G., 1999.

The Turkish Language Reform: A Catastrophic Success

. Oxford/New

York: OUP.

.

Manetta, E., to appear. The shape of the causative verbal domain: evidence from

Kashmiri.

Syntax

.

Marcus, G.F., Pinker, S., Ullman, M., Hollander, M., Rosen, T.J., Xu, F. & Clahsen, H.,

1992. Overregularization in Language Acquisition.

Monographsof the society for research in child development

, 57(4).

Marantz, A., 1984.

On the nature of grammatical relations

. Cambridge: MIT Press.

Marantz, A., 1991. Case and licensing. In

ESCOL ’91: Proceedings of the Eighth Eastern

States Conference on Linguistics

, G.F. Westphal, B. Ao & H.-R. Chae, 234–253.

Columbus: The Ohio State University, Department of Linguistics.

260

Marantz, A., 1997. No escape from syntax: Don’t try morphological analysis in the privacy of your own lexicon. In

Proceedings of the 21st Annual Penn Linguistics

Colloquium

, A. Dimitriadis, L. Siegel, C. Surek-Clark & A. Williams, (eds). 201–

225. Pennsylvania Working Papers in Linguistics 4.2. Philadelphia: University of

Pennsylvania, Penn Linguistics Club.

Marantz, A., 2001. Words and Things. Handout of a talk, MIT, Cambridge, MA.

Marantz, Alex, 2010, Locality Domains for Contextual Allosemy in Words. Handout.

Miyagawa, S., 1984. Blocking and Japanese causatives.

Lingua

64:177–207.

Miyagawa, S., 1994. (S)ase as an Elsewhere Causative. Program of the Conference on

Theoretical Linguistics and Japanese Language Teaching, Tsuda University, 61–

76.

Miyagawa, S., 1998. (S)ase as an Elsewhere Causative and the Syntactic Nature of

Words.

Journal of Japanese Linguistics

16: 67–110.

Miyagawa, S., 1999. Causatives. In

The Handbook of Japanese Linguistics

, N. Tsujimura

(ed). Malden, Massachusetts, and Oxford, UK: Blackwell Publishers

Mundy, C.S., 1954. The -E/-Ü Gerund in Old Ottoman.

Bulletin of the School of Oriental and African Studies

, 16: 298–319.

Myers, S., 1984. Zero-derivation and inflection.

MIT Working Papers in Linguistics 7:

Papers from the January 1984 MIT Workshop in Morphology

. Cambridge: MIT.

Nakipoğlu, M.H., 1998. Split Intransitivity and the Syntax-Semantics Interface in

Turkish. Doctoral dissertation, University of Minnesota.

Nakipoğlu-Demiralp, M., 2001. The referential properties of the implicit arguments of impersonal passive constructions. In

The Verb in Turkish

, E. Erguvanlı Taylan

(ed), 129–150. Philadelphia: John Benjamins Publishing Co., North America.

Nakipoğlu, M. & Üntak, A., 2008. A complete verb lexicon of Turkish based on morphemic anlaysis.’

Turkic Languages

, L. Johanson (ed), 12: 221–280.

Nakipoğlu, M. & Ketrez, N., 2006. Children's overregularizations and irregularizations of the Turkish aorist. In

Proceedings of the Boston University Conference on

Language Development

30:399–410.

261

Nedjalkov, V.P., 1969. Nekotorye verojatnostnye universalii v glagol’nom slovoobrazovanii. In

Jazykovye universalii i lingvističeskaja tipologija

,

Vardul’,

I.F. (ed), 106–114. Moscow: Nauka.

Noyer, R., 1997. Features, Positions and Affixes in Autonomous Morphological

Structure. Revised version of 1992 MIT Doctoral Dissertation. New York:

Garland Publishing.

Özkaragöz, İ.Z., 6891.

The relational structure of Turkish syntax

. Doctoral dissertation,

University of California, San Diego.

Öztürk, B., 2003. Complex predicates in Turkish.

Harvard Working Papers in

Linguistics.

Harvard University.

Parsons, T., 1990.

Events in the semantics of English: A study in subatomic semantics

.

Cambridge: MIT Press.

Paster, M.E., 2006.

Phonological Conditions on Affixation

. Doctoral dissertation, University of California, Berkeley.

Pesetsky, D., 1995.

Zero syntax: Experiencers and cascades

. Cambridge: MIT Press.

Punske, J. Morphological Conspirators: morpho-syntax and the apparent event-structure differences across forms of nominalization. Talk given at University of Arizona

Department of Linguistics Colloquium, Sept. 9 th

, 2011.

Punske, J., 2012.

Aspects of the Internal Structure of Nominalization: Roots,

Morphology, and Derivation

. Doctoral dissertation, University of Arizona.

Püsküllüoğlu, A., 2007.

Büyük Türkçe Sözlüğü

. Istanbul: Can Yayınevi.

Pylkkänen, L., 2002.

Introducing Arguments

. Doctoral dissertation, MIT, Cambridge,

MA.

Pylkkänen, L., 2008.

Introducing Arguments

. Cambridge: MIT Press.

Ramstedt, G.J., 1912. Zur Verbstammbildungslehre der mongolisch-türkischen,

Helsingfors, Druckerei d. Finnischen Literaturgesellschaft.

Redhouse, James, 1890.

A Turkish and English Lexicon shewing the English significations of the Turkish terms.

Constantinople: A.H. Boyajian.

Redhouse Turkish/Ottoman Dictionary, 1997. Istanbul: Redhouse Yayınevi.

262

Reinhart, T., 2002. The theta system: An overview.

Theoretical Linguistics

28: 229–290.

Reinhart, T. & Siloni, T., 2004. Against an Unaccusative Analysis of Reflexives. In

The

Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface

, Alexiadou,

A., Anagnostopoulou, E. & Everaert, M. (eds), 159–180.

Sami, Şemsettin, 1899.

Kamus-ı Türkî

. Istanbul: Akdem Matbaası.

Shibatani, M, 1972. Three reasons for deriving ‘kill’ from ‘cause to die’ in Japanese. In

Syntax and semantics vol. I

, J.P. Kimball (ed), 125–137. New York: Academic

Press.

Shibatani, M. & Pardeshi, P., 2002. The causative continuum. In

The Grammar of

Causation and Interpersonal Manipulation

, M. Shibatani (ed).

Amsterdam/Philadelphia: John Benjamins Publishing Company.

Sichel, Ivy, 2010. Event Structure Constraints in Nominalization. In

The Syntax of

Nominalizations Across Languages and Frameworks

, A. Alexiadou & M. Rathert

(eds). Walter de Gruyter & Co.

Svenonius, P., 2005a. What we think we know about inner and outer causatives. CASTL,

February 5 th

, 2005.

Svenonius, P., 2005b. Two domains of causatives. CASTL, March 10 th

. http://www.hum.uit.no/a/svenonius/papers/Svenonius05TwoDomains.pdf

Tekin, Ş., 2001.

İştikakçının Köşesi: Türk Dilinde Kelimelerin ve Eklerin Hayatı Üzerine

Denemeler

. Istanbul: Simurg.

Tietze, A., 2002.

Tarihi ve etimolojik Türkiye Türkçesi lugatı

. Vol. 1, A-E. İstanbul:

Simurg.

Tubino Blanco, M., 2010.

Contrasting Causatives: A Minimalist Approach

. Doctoral dissertation, University of Arizona.

Ussishkin, A., 2005. A Fixed Prosodic Theory of Nonconcatenative Templatic

Morphology.

Natural Language & Linguistic Theory

, 23: 169–218.

Van Valin, R.D. & Wilkins, D.P., 1996. The case for “effector”: Case roles, agents and agency revisited. In

Grammatical constructions: Their form and meanings

, M.

Shibatani & S.A. Thomposon (eds), 289–322). Oxford: OUP.

Zimmer, K., 1978. Some Constraints on Turkish Causativization. In

Syntax and

Semantics: The Grammar of Causative Constructions

, M. Shibatani (ed), 399–

412. Academic Press, New York.

263

Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement