A SIGN-BASED PHRASE STRUCTURE GRAMMAR FOR TURKISH

A SIGN-BASED PHRASE STRUCTURE GRAMMAR FOR TURKISH
A SIGN-BASED PHRASE
STRUCTURE GRAMMAR
FOR TURKISH
by
Onur Tolga Şehitoğlu
January, 1996
Middle East Technical University
ANKARA
In partial fullfilment of the requirements
for the degree of
Master of Science
in
The Department of Computer
Enginnering
Abstract
A Sign-Based Phrase Structure Grammar for Turkish
Şehitoğlu, Onur Tolga
MS., Department of Computer Engineering
Supervisor: Assist. Prof. Dr. Cem Bozşahin
January 1996, 97 pages
This study analyses Turkish syntax from an informational point of view. Sign based
linguistic representation and principles of HPSG (Head-driven Phrase Structure Grammar) theory are adapted to Turkish. The basic informational elements are nested and
inherently sorted feature structures called signs.
In the implementation, logic programming tool ALE —Attribute Logic Engine—
which is primarily designed for implementing HPSG grammars is used. A type and
structure hierarchy of Turkish language is designed. Syntactic phenomena such as subcategorization, relative clauses, constituent order variation, adjuncts, nominal predicates
and complement-modifier relations in Turkish are analyzed. A parser is designed and
implemented in ALE.
Keywords: syntax, Turkish Grammar, parsing, phrase structure
ii
Öz
Türkçe İçin İm Temelli Öbek Yapısal Sözdizimi
Şehitoğlu, Onur Tolga
Yüksek Lisans, Bilgisayar Mühendisliği Bölümü
Tez yöneticisi: Yrd. Doç. Dr. Cem Bozşahin
Ocak 1996, 97 sayfa
Bu çalışmada, Türkçe sözdizimi bilgiye dayalı bir bakış açısıyla değerlendirilmiştir.
İme dayalı dilbilimsel gösterim ve HPSG ( Baş-sürümlü Öbek Yapısal Dilbilim) kuramı Türkçe’ye uyarlanmıştır. HPSG, dildeki nesnelerin bilgisel içerikleriyle gösterimine
dayanan çağdaş bir sözdizimi ve anlambilim kuramıdır. Temel bilgi öğesi im denilen içiçe
ve kalıtsal türlendirilmiş özellik yapılarıdır.
Uygulamada mantık programlama dili olarak özellikle HPSG uygulamaları için tasarlanmış olan ALE kullanılmıştır. Türkçe’deki dil öğelerinin bir tür ve yapı tanımı yapılmıştır.
Altulamlama, yan cümleler, öbek sıra değişimi, tümleç-niteleyen ilişkileri ve ortaç yapıları
ALE’de çalışan bir ayrıştırıcı ile tasarlanmış ve uygulanmıştır.
Anahtar Kelimeler: sözdizimi, Türkçe Dilbilgisi, ayrıştırma, öbek yapısı
iii
Acknowledgments
I would like to thank NATO TU-LANGUAGE and TÜBİTAK EEEAG-90 projects
for providing development environment and research materials. Hardware and software
resources of the laboratory established by NATO (LcsL) have been used in all stages of
the preperation of the thesis.
I would like to thank Dr. Cem Bozşahin and Elvan Göçmen for their contributions
with corrections, discussions and especially for Turkish syntax chapter.
Thanks are also due to all of the friends and family for their encouragement, support
and friendship during the preperation of the thesis.
iv
Contents
Abstract
ii
Öz
iii
List of Tables
vii
List of Figures
viii
List of Abbreviations
ix
1 INTRODUCTION
1
2 HEAD-DRIVEN PHRASE STRUCTURE GRAMMAR
2.1 Feature Structures . . . . . . . . . . . . . . . . . . . . . . .
2.2 Sign Structure . . . . . . . . . . . . . . . . . . . . . . . . .
2.3 Phrases and Syntactic Structure . . . . . . . . . . . . . . .
2.4 Lexical Organization . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3
. 4
. 8
. 9
. 12
3 TURKISH SYNTAX OVERVIEW
3.1 Noun Phrase . . . . . . . . . . . .
3.1.1 Specifier segment . . . . . .
3.1.2 Modifier segment . . . . . .
3.1.3 The head . . . . . . . . . .
3.2 Postposition group . . . . . . . . .
3.2.1 Postpositions . . . . . . . .
3.2.2 Postposition Attachment .
3.3 Adjective group . . . . . . . . . . .
3.3.1 Comparative adjectives . .
3.3.2 Superlative adjectives . . .
3.4 Adverb group . . . . . . . . . . . .
3.4.1 Reduplications . . . . . . .
3.4.2 Case-marked place adverbs
3.4.3 Temporal adverb groups . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
v
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
14
16
21
25
30
31
32
32
33
33
34
34
34
35
36
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
36
39
39
39
41
43
44
44
45
4 DESIGN
4.1 Sign Structure . . . . . . . . . . . . . . . . .
4.2 Major Categories and Head Features . . . . .
4.3 Complement Selection and Linear Precedence
4.4 Pronoun Drop . . . . . . . . . . . . . . . . . .
4.5 Adjuncts . . . . . . . . . . . . . . . . . . . . .
4.6 Relative Clauses . . . . . . . . . . . . . . . .
4.7 Substantive Predicates . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
48
48
50
52
55
56
59
64
3.5
3.4.4 Verb groups with adverbial use
Verb group . . . . . . . . . . . . . . .
3.5.1 Predicate types . . . . . . . . .
3.5.2 Subcategorization . . . . . . .
3.5.3 Auxiliary verbs . . . . . . . . .
3.5.4 Existential predicates . . . . .
3.5.5 Infinitive form of the verbs . .
3.5.6 Gerundive forms of the verbs .
3.5.7 Syntax of causative verbs . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5 ALE IMPLEMENTATION
67
5.1 Grammar Rules and Principles . . . . . . . . . . . . . . . . . . . . . . . . 68
5.2 Lexicon and Lexical Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
6 CONCLUSION
71
REFERENCES
73
A PARSER SOURCE
A.1 Type Definitions . . . .
A.2 Phrase Structure Rules .
A.3 Constraints and Macros
A.4 Definite Clauses . . . . .
A.5 Lexicon . . . . . . . . .
A.6 Lexical Rules . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
vi
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
75
75
80
81
85
90
93
List of Tables
3.1
3.2
Cases for Turkish nouns . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Segments of a noun group. . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
vii
List of Figures
2.1
2.2
Subtype hierarchy for the type defined for HEAD feature . . . . . . . . . .
Basic Structure of a Lexical Sign (word). . . . . . . . . . . . . . . . . . . .
4.1
4.2
4.3
Sample sign structure for Turkish . . . . . . . . . . . . . . . . . . . . . . . 49
Sort hierarchy of type head . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Projection of the SLASH feature . . . . . . . . . . . . . . . . . . . . . . . 62
5.1
5.2
Sample Source for Head Feature and Subcategorization Principles . . . . . 68
Lexical hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
viii
7
8
List of Abbreviations
1Sg, 2Sg, 3Sg Agreement suffixes first, sec- Past Past Tense (-dH, -mHş)
ond and third person singular
Fut Future Tense (-(y)AcAk)
1Pl, 2Pl, 3Pl Agreement suffixes first, secAsp Aspect markers (-dH, -mHş, -sA)
ond and third person plural
1SP, 2SP, 3SP Possessive suffixes first, sec- Pass Passive Suffix
ond and third person singular
Caus Causative Suffix
1PP, 2PP, 3PP Possessive suffixes first, sec- Neg Negation suffix
ond and third person plural
Ques Question suffix
Abl Ablative Case
Part Complement Participle Suffix (-DHk,
Acc Accusative Case
-(y)AcAk)
Dat Dative Case
Inf Infinitive Suffix (-mAk)
Loc Locative Case
Ger Gerundive Suffix (-mA, -Hş)
Ins Instrumental/commitative Case
Gen Genitive Case
Rel Relative Participle Suffix (-An, -DHk,
-(y)AcAk)
Equ Equitative Case
Cond Conditional Suffix (-(y)sA)
Mun Munitive Case
Adv Adverbial Suffix (-ken)
Rlvz Noun Relativizer
Nec Necessity Suffix (-mAlH)
Cop Copula Suffix
Aux Auxilary Suffix
Pres Present Tense (-Ar)
Prog Progressive Tense (-Hyor)
ix
Chapter 1
INTRODUCTION
This study has two purposes: first, to study Turkish grammar in light of the Headdriven Phrase Structure Grammar (HPSG) formalism, and second, to come up with a
computational model of the languages based on the HPSG principles.
Turkish grammar has been analyzed from the perspective of linguistic theories such
as Transformational Grammar [18], Government-Binding, and Functional Grammar [23].
Lewis [16], Underhill [27], Banguoğlu [1], and Şimşek [26] are also good sources in the
traditional descriptive style. However these studies do not shed any light on how a
computational model can be constructed from the linguistic description.
Recent linguistic theories, such as HPSG [20, 21] and Lexical Functional Grammar
(LFG) [4], differ from the earlier ones in their rigorous definitions and incorporation of
ideas from computer science and artificial intelligence. These ideas range from typetheory in programming languages to unification and knowledge representation. Due to
the formal representations, there are meta-tools for constructing computational models
from formal descriptions, such as Attribute Logic Engine (ALE)[5], CUF [6], and Typed
Feature System (TFS)[15]; Tomita’s parser for LFG [19].
This work is one of the early attempts, together with LFG [10] and Categorial Grammar Models [13, 3] to study Turkish from the perspective of modern linguistic theories.
Our motivation was to design a parser based on the principled account of Turkish syntax in the HPSG framework. It makes use of the ALE formalism to model HPSG-style
definitions.
HPSG makes universal claims about human languages. The main point is that,
although the grammars of languages differ in terms of phrase structure and how grammatical functions are realized, certain principles always hold accross the languages. An
example of such a principle roughly states that the ‘head’ of a phrase plays the most
prominent role in propagating the syntactic and semantic properties of a phrase. Thus,
an HPSG grammar for a language is a collection of specifications for phrase structure,
realization of principles in the language, and the signature of the language in terms of
linguistic features. This division of linguistic description is also reflected in the computational meta-level tools for writing HPSG-style grammars. We hope that these kinds
1
of experiments point out the advantages and disadvantages of such frameworks for underanalyzed language including Turkish.
We aimed to develop a competence grammar rather than a performance grammar
for Turkish. This requires postulating the Turkish realizations of HPSG principles and
their computational counterparts. We chose to provide a breadth of coverage in terms of
lexical types and phrases instead of a comprehensive study with a large lexicon. Moreover, we have implemented some of the morphosyntactic operations (eg. case marking,
possessives, relativization) in the lexicon.
The remainder of the thesis is organized as follows: Chapter 2 introduces the basic
concepts of HPSG. Chapter 3 is an outline of Turkish syntax. Chapter 4 describes HPSG
model of Turkish and Chapter 5 elaborates on the implementation.
2
Chapter 2
HEAD-DRIVEN PHRASE
STRUCTURE GRAMMAR
HPSG (Head-driven Phrase Structure Grammar) was introduced by Pollard and Sag[20]
as an information-based theory of syntax and semantics . HPSG views a human language
as a device used for exchanging information, and tries to explain the relation between
the phonetic form of a word or a phrase, its grammatical structure, and its informational
content. In HPSG, a natural language is defined as a system of correspondences between
certain kind of utterances and certain kinds of objects and situations in the world.
HPSG synthesizes most of the recent (principally nonderivational) syntactic theories
such as Categorial Grammar (CG), Generalized Phrase Structure Grammar (GPSG),
Arc Pair Grammar (APG), and Lexical Functional Grammar (LFG); semantic theories
like Situation Semantics, and some basic concepts of computer science (data type theory,
knowledge representation, unification).
In HPSG, every linguistic component (words, phrases, rules, etc.) is analyzed with
a perspective of the information it provides to the speaker of the language. This information may include not only the syntactic features of the component, but also its
grammatical information, semantic content and its background semantics.
HPSG is a system based on signs. Any structural element (words and phrases), and
principles defining the language are modeled by sorted feature structures (ie. feature
structures with an associated type or sort) and constraints and operations defined on
them. As being one of the most recent examples of the family of the unification based
grammar theories [25], the most fundamental operation of HPSG is unification, which
combines a set of compatible feature structures, and returns a minimum informative
feature structure containing all information present in the operands. Phonetic, syntactic and semantic information coded in the lexicon and information coming from other
resources like lexical rules, universal and language specific principles of well-formedness,
are combined by unification.
Similar to the majority of the contemporary linguistic theories, HPSG defines a lan-
3
guage by a finite set of recursively applicable rules which yields the judgment of grammaticality. Basically, principles are divided into two categories: 1) Universally applicable
basic set of constraints such as head feature principle and subcategorization principle;
and types of phrases available in any human language. 2) Language specific principles
of phrases lexicon itself and further articulation and specification of the principles of the
universal grammar such as constituent order[21].
One of the distinctive aspects of HPSG is that it not only models the language syntactically, but also concerns itself with the interactions between all kinds of information
of a linguistic component. Both syntactic and semantic information content of a sign
is considered. Situation Semantics and Relational Theory of Meaning are chosen for
semantic modeling.
2.1
Feature Structures
HPSG, like other unification-based formalisms, uses recursively embedded feature-value
pairs representing linguistic objects. Feature structures have different names in each
theory: f-structures in LFG, feature bundles, feature matrices or categories in GPSG,
etc. Feature structures are informational objects that consist of feature (attribute)-value
pairs. Usually feature structures are represented by attribute-value matrices (AVM’s).
For example:
(1)


PHON “kedi” % cat
CAT

noun


"
#



PERSON third
AGR

NUMBER singular
In (1), features PHON, CAT and AGR are defined where value of PHON is “kedi”,
CAT is noun and value of AGR is another feature structure with features PERSON and
NUMBER which have values third and singular respectively. As an alternative, feature
structures can be represented in graph notation where nodes are the intermediate feature
structures, vertices are attributes, and values are sink nodes.
(2)
PERSON
third
AGR
singular
NUMBER
noun
CAT
4
As a fundamental property, feature structures can be recursively embedded. Value
of an attribute can be an atomic value or another feature structure. To represent values
embedded in feature structures, “path of attributes” notation is used as a shorthand.
A path is an ordered sequence of attributes separated by ‘|’ to reach the value. In the
example (2), AGR|PERSON has the value third, AGR|NUMBER is singular.
A relation defined on feature structures is the subsumption relation. When a feature
structure A is subsumed by another feature structure B, A is equally or more informative than B. In other words, it contains all of the information provided by B and
possibly more. It is often said A extends B or B subsumes A and written as A B. It
indicates that any object described by B can be appropriately described by A. When a
feature structure has no information, it subsumes every feature structure. It can describe
any object whatsoever. This structure is the root element of the subsumption ordering
(called Top), shown as >:
>=[ ]
Subsumption relation defines a partial ordering between information structures. It
has the properties of reflexivity (∀A, A A), transitivity (A B and B C then
A C), and antisymmetricity (A B and B A then A = B). For subsumption
relation to hold between two feature structures, they should have compatible types
and compatible values in the corresponding attributes. In the example below, feature
structures have incompatible values so, A 6 B and B 6 A.
(3)
h
i
A = PERSON third ,
"
PERSON first
B=
NUMBER singular
#
Another important property of feature structures is structure sharing. Two attribute
paths in a feature structure may describe the same object. This implies the token identity
which should not be confused with the structural identity where only type structure and
feature values are equal. Usually structure shared object are denoted by tags (boxed
numbers).
(4) a.


CAT
verb

"

HD-DTR

PERSON


AGR 1


NUMBER

#
"


CAT
noun
SUBJ-DTR
AGR 1
5
first
plur

#








b.



CAT
verb

"

HD-DTR


AGR PERSON


NUMBER




CAT noun

"


SUBJ-DTR 
AGR PERSON

NUMBER
#

first 


plur 



#

first 


plur
In (4a), HD-DTR|AGR attribute shares the same object with the SUBJ-DTR|AGR
attribute. Although the paths in question have the same value, the source of the values
may be different (i.e., not token identical) contain the same agreement in (4b). Intuitively, it is clear that the structure shared version is more informative than the other;
it is subsumed by the other. Similarly, the effect of structure sharing is reflected in the
formal definition of the subsumption:
(5)
if A and B are atomic, then A B iff A = B.
else, A B holds iff,
(i) for every path in B, same path exists in A and its value is subsumed by
the value in B.
(ii) for every structure sharing path in B, same path is structure sharing in
A.
Perhaps the most important operation on feature structures is the unification, which
constructs a base to a group of linguistic theories. Unification operation builds a new
feature structure which contains all but not more of the information contained in its
operand feature structures. For feature structures to be unified, they should have compatible types. Result of unification is the least informative (the most general) feature
structure which extends all of the operands. Unification is denoted by the symbol ∧ and
if C = A ∧ B then, C A and C B.

(6) CAT noun
h

AGR PERSON
 

CAT noun
V
i 
h
i=
third
AGR NUMBER sing


CAT noun
"
#



AGR PERSON third 
NUMBER sing
6
When operands of the unification have incompatible types or values, the resulting
feature structure does not exist and unification fails. This is indicated by the symbol
‘⊥’ (bottom) which represents inconsistent information. As > is the maximal element
in the subsumption ordering, ⊥ is the minimal element that is subsumed by all feature
structures.
Also, unary negation operator ¬ and disjunction operator ∨ are defined. In negation,
¬a means any value other than a. Similarly, disjunction a ∨ b means the attribute may
be equal to a or b. Attributes may be list-or set-valued. Lists are denoted by comma
separated values enclosed in angle brackets, ha, b, ...i. Sets are enclosed in curly braces
{a, b, ...}. List valued attributes are unified by unifying corresponding elements by order.
Unification of set values is a more complex operation. For detailed information and
formal definition about feature structures, consult Rounds and Kasper[22].
The most significant formal property of HPSG feature structures is that they are
sorted. Every feature structure has a type (sort), and a subtype relation is defined
between these sorts. All defined sort symbols are partially ordered by the subsumption
relation.
head
noun
common
proper
personal
adjective
pronoun
determiner
......
quantificational
verb
qualitative
demonstrative
Figure 2.1: Subtype hierarchy for the type defined for HEAD feature
As the second formal property, HPSG feature structures should be totally well-typed.
For each sort, a set of appropriate features and types is defined, and this set is inherited
by the subsorts of the sort. For example, if the CASE feature of sort case is defined
for the sort noun in Figure 2.1, then it is appropriate for the sorts common, proper,
pronoun, personal and demonstrative sorts. Any other feature which is not introduced
in the sort is not allowed in the feature structure.
Third, HPSG feature structures should be sort-resolved to satisfy the criteria of
completeness as models of the linguistic entities. Sort-resolved means: for every attribute
defined, a sort should be assigned. This sort should be the most specific in the sort
ordering (A leaf node in the subtype hierarchy). For example HEAD feature can be
assigned proper or common but not head or noun which subsume other types in the
ordering and actual sort (value) is not clear.
7
2.2
Sign Structure
Linguistic entities have the general sort sign. Information in all intermediate phrases,
lexical entries, sentences and even multisentence discourses are described by a corresponding sign. The sign sort has two subtypes: word and phrase. word describes lexical
entries, and phrase describes phrasal constructs. phrase has an additional feature DTRS
(daughters) to represent the phrase structure.
HEAD
PHON
CAT
<string,..>
CONT
word
LOCAL
head
SUBCAT
subcat
cont
conx
CONX
SYNSEM
INHERITED
inherited
NONLOCAL
to-bind
TO-BIND
Figure 2.2: Basic Structure of a Lexical Sign (word).
A basic graph briefly describing the structure of a sign can be given as in Figure 2.2.
All signs should have at least two attributes: PHON and SYNSEM. PHON attribute is a
feature representation of the phonetic content of the phrase or word. Usually, it has a list
of strings describing phonological and phonetic structure of the sign. SYNSEM attribute
contains both syntactic and semantic information of the sign. Using SYNSEM instead of
two distinct features SYNTAX and SEMANTICS allows packing all information required
for subcategorization into one attribute.
SYNSEM value is another structured object which has two attributes LOCAL and
NONLOCAL. NONLOCAL represents the information which is not bound to the phrase
described by the sign. This information is used to handle unbounded dependency
constructs like filler-gap dependencies, relative clauses, etc. LOCAL feature describes
the local information which consists of the attributes CATEGORY (CAT), CONTENT
(CONT) and CONTEXT (CONX).
CAT value includes both the syntactic category of the sign and the grammatical
arguments it requires. CONT value is the context independent semantic interpretation
and semantic contribution of the sign.
CONX value contains context-dependent linguistic information such as indexicality,
presupposition and conventional implication. The semantic features are not the object
of this study so we will not go into details of CONT and CONX features.
CAT attribute consists of two attributes HEAD and SUBCAT. HEAD feature is roughly
lexical category (part of speech) of the sign. It describes the information to be passed
8
to phrasal projections of the sign. Contents of HEAD feature varies according to the
category of the sign. It typically contains basic features related with the category of the
sign e.g. case, agreement, verb form, prepositional form, etc.
SUBCAT (subcategorization) describes the valence of the sign which specifies the
group of signs that sign in question requires to become saturated. A saturated sign
means all the subcategorization requirements are met. Group of signs in SUBCAT feature
is described by a list of synsem values. synsem values are used for identifying the
subcategorized objects, so that sign can select not only the syntactic category of the
complement, but also semantic role and even nonlocal attributes.
The order of the synsem values in the SUBCAT list does not correspond to the surface
order of the phrase. However, it may define an obliqueness order which can be used to
describe the constituent order. For example, in English, linear precedence rules defining
surface order is declared by this obliqueness order, as the least oblique element linearly
precedes others for non-verbal heads. When the head is a verb, the first oblique element
is the subject element which precedes the head. For languages having free constituent
order, like Turkish, SUBCAT attribute may have different structure, e.g., unordered list.
2.3
Phrases and Syntactic Structure
HPSG is a constraint-based theory and constraints are defined by partial descriptions
that model linguistic utterances. Descriptions are declarative, order independent and
reversible. Judgment of whether a phrase is well-formed or not is done by a set of
universal principles and language-specific rules. Universal principles are general constraints on universally available phrase types. The most basic principles in HPSG are
Head Feature Principle and Subcategorization Principle. Language specific phenomenon like Linear Precedence (constituent order) is described by a set of language
specific constraints and some kind of specialization of universal principles.
As mentioned in the preceding section, a sign has two subtypes: word and phrase.
phrase has an additional feature DAUGHTERS (DTRS) in which phrase structure is
represented. DTRS feature has a value of constituent-structure (cons-struc) representing
the immediate constituents of the phrase. cons-struct may have several subsorts each
has characterized by different daughter attribute. The most general sort of comp-struc
is headed-structure (head-struct).
(7)


head-struc
HEAD-DTR

sign


D
E

COMP-DTRS sign, ....
Each head-struc has one HEAD-DTR attribute and another attribute which is a list of
signs which are the sisters of the HEAD-DTR. For example tree and DTRS representation
of the sentence “Ahmet kırmızı kitabı aldı.” (Ahmet took the red book.) is:
9
(8) a.
NP
S
H
HH
NP
HH
Adj
N
Ahmet
kırmızı
b.

phrase
SYNSEM






DTRS




HH
HH
V
aldı
kitabı

S[fin]
head-struc
HEAD-DTR V[aldı]
sign
COMP-DTRS

*

SYNSEM NP



head-struc
N[Ahmet],

DTRS HEAD-DTR N[kitabı]

ADJ-DTRS





 

+ 




 

 


 
Adj[kırmızı]
The HEAD value of a phrase is centrally important since it defines the syntactic
properties of the mother phrase. For example, the lexical head of a sentence is of
the sign verb. verb combines with its complement sisters and forms a Verb Phrase (VP)
which takes its syntactic properties from its head daughter (verb). Similarly, verb phrase
combines with the subject complement forming a sentence. In other phrase types like
Noun Phrase (NP), Prepositional Phrase (PP), Adjective Phrase (AP), HEAD feature
is projected —propagated— along the upper phrases until phrase becomes saturated.
The key idea behind this projection is the X-bar theory[14]. HPSG’s Head Feature
Principle describes this syntactic phenomena which is adopted from the Head Feature
Convention of GPSG[8].
(9)
1
NP[nom]
Ahmet
2
LOCAL | CAT |
NP[acc]
"
h
3
HEAD
SUBCAT hi
LOCAL | CAT |
i
HEAD
SUBCAT
kalemi
verb[fin]
3
1
NP[nom],
aldı.
Head Feature Principle(HFP) is defined as follows:
10
2
NP[acc]
#
(10) The HEAD value of a headed phrase is structure-shared with the HEAD value of
the head daughter.
Formally:

phrase


SYNSEM | LOC | CAT | HEAD 1

DTRS | HEAD-DTR | SYNSEM | LOC | CAT | HEAD 1
The other principle which together with HFP describes the basic Immediate Dominance (ID) scheme of HPSG is Subcategorization Principle. Subcategorization
checks the requirements of the phrasal head to be saturated and allows heads to select
its complement sisters by structure sharing the SYNSEM values of the sisters with that
in the SUBCAT list. Subcategorization Principle is defined as follows:
(11) In a headed phrase, SUBCAT value of the head daughter of the phrase is the
concatenation of the SYNSEM values of the complement daughters.
Formally:


phrase
SYNSEM | LOC | CAT | SUBCAT 1







HEAD-DTR | SYNSEM | LOC | CAT | SUBCAT 2 ⊕ .... ⊕ n ⊕ 1 


D
E

DTRS 
SYNSEM 2 , ...., SYNSEM n
COMP-DTRS
Where ⊕ is defined to be list concatenation operation.
The Subcategorization Principle allows all constraints on the arguments of a
phrase to be controlled by an argument. Any kind of argument restriction, complement
structure like sentential complements, unbounded dependencies and other constraints
can be directly controlled and coded into lexicon. In other words, HPSG crucially relies
on the complex descriptions in the lexicon. To deal with the redundancy caused by
the complexity of the lexical entries, lexical rules and multiple inheritance hierarchy
describing relation between lexical entries can be expressed [20].
Phrase structure rules defining tree structure of phrases are described by immediate
dominance (ID) and linear precedence (LP) constraints. There is a general trend in contemporary syntactic theories towards the lexicalization of grammar and elimination of
construction-specific rules in favor of schematic immediate dominance templates. These
schemata may vary for language-specific phrase types and constituent relations. Examples of typical phrase structures are head-complement, specifier-head, and adjunct-head,
conjunct-daughters.
Linear precedence constraints are mostly defined as language-specific rules and constraints on the surface constituent order of the phrases. In English, LP rules are defined
on the obliqueness hierarchy of the SUBCAT list. Subject is the least oblique argument
of a verb. The direct object and the indirect object come next in the obliqueness order.
11
Also, least oblique constituents precede the others. LP rules for English can be defined
as follows:
(12) 1. Any lexical head sign precedes
other signs:
h
i
h i
HEAD-DTR 1 word =⇒ 1 ≤
2. Subject complement precedes the Head
daughter:
h
i
h
i
1 =⇒ 1 ≤ phrase
SUBJ-DTR
3. Least oblique elements linearly precede the others:
E
h
D
i
COMP-DTRS ..., 1 , ...., 2 ,... =⇒ 1 < 2
where ≤ means immediately precede and < means precede.
2.4
Lexical Organization
Lexicalization and the use of meta-rules controlled by a set of universal principles results
in a few number of simple grammar rules. However, the associated information structures
become more complex. Lexical signs in lexicaly-oriented theories should very rich in
information content so it is not always possible to enter and maintain a lexicon without
any organization.
In HPSG —as the other lexicalized formalisms— it is necessary to organize lexicon
such that lexical entries should be represented as compact as possible. Two main devices,
lexical type hierarchy and lexical rules, are the basic solutions to redundancy problem
in the lexicon.
The main idea behind the lexical type hierarchy is the repetition of information in
the lexical items of same category class. Only a small part of a lexical entry carries
exceptional information from the other entries having the same category. For example,
all nouns have noun as the value of the HEAD feature and empty SUBCAT value. All
common nouns have third person in their agreement. So the idea is to create a hierarchy
of types each of that is assigned a set of attribute-value pairs which are inherited along
the hieararchy. Lexical entries can be defined by means of these types plus the special
features.
Lexical hierarchies solve some portion of the redundancy. However, some specific
features of lexical entries may be related to each other by recurrent patterns. These
patterns include some derivational and inflectional phenomena in the language like passivization of verbs, case marking, verb inflections, nominalization etc. The solution that
has been used by most of the unification-based formalisms is to define functions mapping
one class of words to another, called lexical rules.
12
Lexical rules are generally expressed as procedures converting an input form to an
output form. So that all inflections and derivations of a word can be generated from a
base form by application of lexical rules several times. In the example (13a), a simplified
lexical rule for passivization of verbs for English is given where fP SP is the function
mapping verb base to its past-participle form. Also, it is possible to generate different
readings and syntactic behaviours of the same word. Lexical rule in (13b) duplicates
lexical entry of verbs for non-referential objects in preverbal position in Turkish.
(13) a. word
PHON 1





SYNSEM | LOCAL



b.



CAT



"

HEAD


SUBCAT ...
SUBJ
CONT
#
verb
TENSE
PASSIVE
base
−
]2, [
[
]3
2
4

word
PHON


SYNSEM | LOCAL | CAT

HEAD
CONT
1
h

NPacc [ ],
2
(....)



"

HEAD

SYNSEM | LOCAL | CAT
SUBCAT
13
verb
i
]3
4

verb
SUBCAT
word
PHON
7−→

HEAD PASSIVE +
CAT 

SUBCAT PP[BY] 2 , ..., [


3
SUBJ
1







... 



fP SP ( 1 )





SYNSEM | LOCAL


word
PHON




7−→

verb
D 2
, NPnom [ ]
#

E







... 


Chapter 3
TURKISH SYNTAX
OVERVIEW
This chapter is adopted from [9]. Turkish is an agglutinative language where words are
formed by affixation of derivational and inflectional morphemes to root words. So most
of the syntactic properties of a word such as case, agreement, relativization of nouns,
tense, modality, aspect of verbs, and even passivization, negation, causatives, reflexives
and some auxiliaries are marked by suffixes.
(14) a.
b.
ev
-imiz -de
-ki
-nin
house -1PP -Loc -Rlvz -Gen
‘of the one that is in our house’
bak -tır
-a
-mı
-yor
-muş -sun
look -Caus -Able -Neg -Prog -Asp -2Sg
‘you were not able to make look (reported past)’
As a result, Turkish words —especially heads of phrases— have complex and rich
syntactic forms and carry much information.
As another distinct property, Turkish is head-final. Specifiers and modifiers always
precede the specified or modified. Similarly complements and arguments precede the
head in their usual formation. However when head is a verb or predicative noun, complements and objects may follow the head.
(15) a. Benim kapıdaki
kırmızı arabam
ben-Gen door-Loc-Rlvz red
car-1SP
‘my red car at the door’
14
b. Hicabi kitabı
çok çabuk okudu.
Hicabi book-Acc very quick read-Past-3Sg
‘Hicabi read the book very quickly.’
c. Kitapları verdim
Ahmet’e.
book-Plu give-Pass-1Sg Ahmet-Dat
‘I gave Ahmet the books.’
Also in Turkish, constituents have free order. The most usual sentence order is S-OV. However they can scramble causing different readings and interpretations. Sentenceinitial position marks the topic, pre-verbal constituent is the emphasis and post-verbal
position is for the background or afterthought information [7].
(16) a. Onur kalemi
çocuğa verdi.
Onur pencil-Acc child-Dat give-Past-3Sg
‘Onur gave the child the book.’
b. Onur çocuğa kalemi verdi.
‘Onur gave the pencil to the child.’
c. Çocuğa kalemi Onur verdi.
‘It is Onur who gave the child the book.’
d. Kalemi Onur verdi çocuğa.
(c)
e. Onur verdi kalemi çocuğa.
‘Onur did give the child the pencil.’
When the object is non-referential (ie. no case marked or specified), it should immediately precede the verb.
(17) a. Adam bahçede
şiir yazıyordu
man garden-Loc poem write-Prog-Asp-3Sg
‘The man was writing poem in the garden’
b. (*) Şiir bahçede adam yazıyordu.
‘The poem was writing the man in the garden’
c. * Adam şiir bahçede yazıyordu.
Similarly, adverbs and sentential complements may scramble freely (18a–c). Also order variation of constituents is valid for the embedded sentences such as relative clauses,
15
infinitive and gerundive forms, and sentential complements. Relative clauses are strictly
head-final; no constituent belonging to relative clause can follow the head verb (18d–f).
(18) a. Gerçekten onun sınavı
kazanmasını
herkes istemişti.
really
he-Gen exam-Acc pass-Inf-3Sg-Acc everyone want-Past-Asp
‘Everybody realy wanted him to pass the exam.’
b. Onun sınavı kazanmasını herkes gerçekten istemişti.
c. Herkes onun sınavı kazanmasını gerçekten istemişti.
d. Bakkaldan dün
aldığım
kalem kırıldı
store-Abl yesterday buy-Rel-1Sg pencil break-Pass-Past-3Sg
‘The pencil that I bought yesterday from the store was broken’
e. Dün bakkaldan aldığım kalem kırıldı.
f. * Dün aldığım bakkaldan kalem kırıldı.
3.1
Noun Phrase
Phrases with nominal heads are noun phrases. The head noun is the final constituent
of the phrase and determines the syntactic role of the whole phrase. Noun phrases may
act as a subject, object or complement of a sentence or modifier or specifier of another
noun group. A noun —so a noun phrase— can have the cases listed in Table 3.1.1
Table 3.1: Cases for Turkish nouns
case
nominative
accusative
dative/allative
locative
ablative
genitive
comitative/instrumental
suffix
-(y/n)H
-(y/n)A
-(n)DA
-(n)DAn
-(n)Hn
-(y)lA
examples
adam, kedi
adamı, kediyi
adama, kediye
adamda, kedide
adamdan, kediden
adamın, kedinin
adamla, kediyle
Also three suffixes -cA, -lH and -sHz (equative, munitive and privative respectively)
are considered as cases by Banguoğlu [1].
1 We
use A to stand for a or e, H to stand for ı, i, u or ü, and D to stand for d or t.
16
Nominative case is used for marking subjects (19a), indefinite/nonreferential objects
(19b). Also noun with the nominative case can be a classifier for another noun (19c).
(19) a. Köpek kediyi kovaladı.
dog cat-Acc chase-Past-3Sg
‘The dog chased the cat.’
b. Adam kuş avladı.
man bird hunt-Past-3Sg
‘The man hunted a bird.’
c. Güzel bir köpek evi
yaptık.
nice a dog house make-Past-1Pl
‘We made a nice dog house.’
The accusative case is used for marking definite objects. It is obligatory with pronouns and proper nouns in object position.
(20) a. Çocuk kitabı
okumamış.
child book-Acc read-Neg-Past-3Sg
‘The child hasn’t read the book.’
b. Köpek Ayşe’yi ısırdı.
dog
Ayşe-Acc bite-Past-3Sg
‘The dog bit Ayşe.’
c. Herkes onu
suçluyor.
everyone he/she-Acc blame-Prog-3Sg
‘Everyone blames him/her.’
Noun phrases with dative/allative case (-(y/n)A suffix) have three roles: they behave
as prepositional phrases indicating target or aim (21a–b), mark the indirect object (21c),
and they are subcategorized as the oblique object in some verbs (21d).
(21) a. Çocukları
Ankara’ya gönderdik.
child-Plu-Acc Ankara-Dat send-Pass-2Pl
’(We) sent the children to Ankara.’
b. Çiçekleri
sana
aldım.
flower-Plu-Acc you-Dat buy-Past-1Sg
’(I) bought the flowers for you.’
17
c. Mehmet ekmeği
adama verdi.
Mehmet bread-Acc man-Dat give-Past-3Sg
‘Mehmet gave the man the bread.’
d. Kadın bahçeye
baktı.
woman garden-Dat look-Past-3Sg
‘The woman looked at the garden.’
Noun phrases with locative case (-DA suffix) is used to express the location of an
action or object (22).
(22) Kitabın masada duruyor.
book-2SP table-Loc stand-Prog-3Sg
‘Your book lays on the table’
The ablative case (-DAn suffix) indicates the source of an action or object as the
English preposition “from” (23a–b). Also can be subcategorized as direct object by a
group of verbs (23c) .
(23) a. İstanbul’dan yeni gelmiş.
İstanbul-Abl just come-Past-3Sg
‘He has just come from İstanbul’
b. Genelde bu üzümlerden şarap yapılıyor
usually these grape-Plu-Abl wine make-Pass-Prog-3Sg
‘Usually wine has been done from these grapes’
c. Ahmet kedilerden nefret eder.
Ahmet cat-Plu-Abl hate--Pres-3Sg
‘Ahmet hates cats.’
The genitive case is used to mark the possessor in the possesive-possessor relation.
Noun with the genitive case behaves as a specifier of possessed noun which is marked
with the possessive suffix. Person and number information of the noun should agree
with this possesive suffix.
(24) a. Arabanın anahtarını unuttum.
car-Gen key-3SP-Acc forget-Past-1Sg
‘I forgot the key of the car.’
18
b. Senin
kalemini
kullandım.
you-Gen pencil-2SP-Acc use-Past-1Sg
‘I have used your pencil.’
c. İlker’in arabasının motoru
bozuk
İlker-Gen car-3SP-Gen engine-3SP broken
‘The engine of the İlker’s car is broken.’
The -(y)lA suffix is the combined form of the postposition ile with the noun. It
marks the commutative (25a) and instrumental (25b) relationships.
(25) a. Kitabı
Ahmet’le gördük.
book-Acc Ahmet-Ins see-Past-1Pl
‘Ahmet and I saw the book together.’
b. Kuşları
dürbünle
seyrediyoruz.
bird-Plu-Acc binocular-Ins watch-Prog-1Pl
‘We could see the birds with telescope.’
-cA suffix is used for marking subject of a passive sentence. Postposition tarafından
is more commonly used compared to equative case.
(26) a. Kampanya vatandaşlarca destekleniyor.
campaign citizen-Plu-Equ support-Pass-Prog-3Sg
‘The campaign is supported by citizens.’
-lH and -sHz (munitative and privative) suffixes have similar meaning with the
prepositional phrases formed by ‘with’ and ‘without’ in English. Noun phrases with
these suffixes behave as adjective. However, -lH suffix saves some of the properties of
the noun it is attached to. Noun may be still the head of a noun group and can be
modified (27).
(27) a. kırmızı kanatlı
böcek
red
wing-Mun insect
‘the insect with red wings’
b. üç
tekerlekli bisiklet
three wheel-Mun bicycle
‘The bicycle with three wheels’
19
Segments
Specifier
Modifier
Head
Alternatives
Quantifier
Article
Demonstrative Adjective
Genitive noun
Classifier noun
Quantitative Adjective
Qualitative Adjective
Relativized noun
Relative clause
Unit noun
Common noun
Proper noun
Pronoun
Examples
her, bazı, biraz, kimi, herbir, birçok
bir
bu, şu, o, diğer, ilk, sonuncu
bahçenin
mutfak dolabı
dört, yarım, ikişer, üçlü
güzel, zor
evdeki, akşamki
postadan çıkan, yolda gördüğüm
bardak, salkım, tane
ev, kitap
Deniz, Ankara
ben, sen, onlar
Table 3.2: Segments of a noun group.
Another inflection that a noun group may have is the relativizer (-ki suffix). This
suffix is attached to some temporal adverbs and nouns with locative case. A relativized
noun becomes a specifier for another noun group (28).
(28) bahçedeki
çiçekler
garden-Loc-Rlvz flower-Plu
‘The flowers in the garden’
-ki suffix following a genitive noun group behaves as a pronoun meaning ‘one that
belongs to’ and different from the relativizer -ki.
(29) Ayşe’ninkiler
yarın
gelecek
Ayşe-Gen-Pro-Plu tomorrow come-Fut-3Sg
‘Ones that Ayşe owns will come tomorrow.’
A noun group consists of an optional group of specifier and modifiers and a head noun.
Head noun can be a common noun, a pronoun or a proper noun. Order and grammatical
combinations of specifiers and modifiers change according to the type of specifiers and
modifiers. Order and valid combinations of specifiers and modifiers are pragmatically
controlled. Some specifiers/modifiers put some restrictions on the specifier/modifier
types that can further specify/modify the noun.
General structure of a noun group can be viewed as a sequence of segments, head
noun being the last one. These segments are listed in Table 3.2.
20
Specifier and modifier segments are optional:
(30) a. bahçenin
kapısı
garden-Gen gate-3SP
‘the gate of the garden’
b. şu kız
that girl
‘that girl’
c. Ankara
The order of the specifier and modifier segments may vary.
(31)
a.
c.
Her
kırmızı
every
red
‘every red flower’
güzel
bir
beautiful a
‘ a beautiful girl’
çiçek
flower
b.
kız
girl
d.
kırmızı her
red
every
‘every flower that is
bir
güzel
a
beautiful
‘one beautiful girl’
çiçek
flower
red’
kız
girl
Each segment of the noun group are elaborated below.
3.1.1
Specifier segment
Specifiers pick out noun(s) out of a set of possible nouns. In Turkish, specifier segment position is filled by a specifier that can be a quantifier (32a), an article (32b), a
demonstrative adjective (32c), a genitive noun (32d) or a a classifier noun (32e).
(32) a. Yazdıklarımız bazı insanları rahatsız edecek.
write-Part-1PP some people-Acc disturbed make-Fut
‘What we have written will disturb some people.’
b. Yolda
bir kalem buldum.
road-Loc a pencil find-Past-1Sg
‘I found a pencil on the street.’
c. İlk sınavımı geçtim.
first exam-1SP pass-Past-1Sg
‘I passed my first exam.’
21
d. yazarın
her kitabı
author-Gen every book-3SP
‘every book of the author’
e. Onur’un bulduğu
iki caz plağı
Onur-Gen find-Part-3Sg two jazz record-3SP
‘two jazz records that Onur found’
The valid combinations and and order of specifiers are pragmatically controlled. A
noun group may have only one quantifier (33a–b). Also quantifiers cannot be used with
demonstrative adjectives and article (33c,d).
(33) a. *her çoğu kitap
every most book
b. *birçok kimi öğrenciler
many some student-Plu
c.
d.
*her
bu
Quant.
Dem. Adj.
every
this
* kimi
bir
Quant.
Art.
some
a
kitap
book
insan
person
The use of the article with demonstrative adjectives and quantifiers depends on some
selectional restrictions.
(34) a.
b.
*ilk
first
bir
a
bir
a
ilk
first
kitap
book
kitap
book
c. bir üçüncü kitap
a third book
‘yet a third book’
d.
*sonuncu
last
bir
a
kitap
book
22
e. bir şu kitap
a that book
‘only that book’
f. şu bir kitap
that one book
‘that single book’
g. diğer bir kitap
other a book
‘another book’
There are some points to be underlined here, about the different meanings of “bir”
and about some exceptions:
The sequence depicted in (34b) has a limited usage referring to the first book of an
author. In (34c), “bir” is an adverb meaning “yet” or “another”. In (34e), “bir” is an
adverb with a meaning “only”. In (34f), “bir” is not an article but a cardinal number
(a modifier). In (34g), “diğer” acts as an adverb.
Concerning the demonstrative adjectives, there are two subgroups:
i) bu, şu, o
ii) ilk, sonuncu, ordinal numbers, diğer
Only one element from each group can be used within a noun group. The elements
of the first group can sometimes be used in front of the elements of the second group
for emphasizing the demonstration.
(35) a. Bu ikinci kitabı
pek beğenmedim
this second book-Acc much like-Neg-Past-1Sg
‘I didn’t like the second book much’
b. Şu diğer valiz
benimki
that other suitcase mine
‘The other suitcase is mine’
c. *diğer sonuncu kız
other last
girl
Nouns or noun groups with genitive marking also function as specifiers within a
noun group. Genitive nouns can be used in combination with other specifiers (36a–b).
23
The main restriction is that all specifiers and modifiers modifying the possessive marked
noun should follow the genitive noun. Otherwise they specify/modify the genitive noun
(36c):
(36) a. yazarın
bir kitabı
author-Gen a book-3SP
‘a book of the author’
b. kitabın
bu sayfası
book-Gen this page-3SP
‘this page of the book’
c. ilk seminerin
konuşmacısı
first seminar-Gen speaker-3SP
‘the speaker of the first seminar’
Genitive nouns can rather be interpreted as complements of possessive marked nouns
since possessive nouns require a genitive noun which is subject of the owner relation in
the possessive group.
Classifier nouns resemble genitive nouns in that they require a possessive-marked
noun group modified by the classifier noun. However, classifier nouns take no genitive
suffix.2 The difference between a genitive noun and a classifier noun is that the former
provides a definite reading where the latter provides an indefinite or nonreferential one.
(37) a. duvar boyası
wall paint-3SP
‘wall paint’
b. duvarın boyası
wall-Gen paint-3SP
‘the paint of the wall’
Classifier noun groups can act as specifiers of other classifier nouns:
(38)
2 These
kredi kartı
faiz
yüzdesi
credit card-3SP interest percentage-3SP
‘credit card interest rate’
noun groups are called izafet by Lewis [16]
24
A classifier noun is the immediate predecessor of the head noun. Hence, other specifiers and modifiers precede it.
(39) a. her çocuk arabası
every child car-3SP
‘every stroller’
b. o
dere yatağı
that stream bed-3SP
‘that river bed’
c. *çocuk her arabası
d. *dere o yatağı
e. *ev bir kapısı
f. *duvar evin boyası
g. evin duvar boyası
home wall paint-3SP
‘wall paint of the house’
3.1.2
Modifier segment
Modifiers provide information about the properties of the entity or its relations with
other entities. A modifier is either an adjective group, or a noun group. More than one
modifier may exist within a noun group.
(40)
güzel
mavi eteğin
beautiful blue skirt-2SP
‘your beautiful blue skirt’
As a general rule, “whatever precedes modifies” in Turkish. Hence, if a modifier
itself is a noun group or a clause containing a noun, any preceding modifier modifies the
first of the succeeding nouns.
For example, in the phrase below, the modifier “yaşlı” modifies “adam” rather than the
head noun “kadın”.
(41)
yaşlı adamın konuştuğu
kadın
old man-Gen talk-Part-3SP woman
‘the woman to which the old man talked/talks’
25
Certain restrictions apply to the combinations of modifiers. When a noun is modified
by both a qualitative and a quantitative adjective, order of the adjectives may vary but
the quantitative adjective usually precedes the qualitative one.
(42) a. üç
kırmızı kalem
three red
pencil
‘three red pencils’
b. hassas ikili ilişkiler
sensitive dual relationship-Plu
‘sensitive dual relationships’
c. ikişer kalın battaniye
by-two thick blanket
‘two thick blankets for each’
d. rahat
üçlü kanepe
comfortable triple sofa
‘a comfortable triple sofa’
e. yarım çürük elma
half rotten apple
‘a half rotten apple’
f. çürük yarım elma
rotten half apple
‘a half rotten apple’
When used as modifiers, unit nouns are preceded by a cardinal number (43a), a
fractional number (43b), or a distributive adjective (43c):
(43) a. iki bardak süt
two glass milk
‘two glasses of milk’
b. yarım somun ekmek
half loaf
bread
‘half loaf of bread’
c. birer dilim pasta
by-one slice cake
‘a slice of cake (for each)’
26
When the unit noun denotes a container, the word dolusu (“full”-3SP) may optionally
be inserted between the unit noun and the head.
(44)
üç
kaşık dolusu şeker
three spoon full-3SP sugar
‘three spoonful of sugar’
The other group of modifier is the relativized nouns which are inflected by relativizer
suffix -ki. If the head noun is modified by a relativized noun, all other modifiers and
specifiers of the head come after the relativized noun (45a). Otherwise, any modifier
preceding a relativized noun modifies the relativized noun rather than the head (45b).
(45) a. çantamdaki
üç
küçük anahtar
handbag-1SP-Loc-Rlvz three small key
‘three small keys in my handbag’
b. küçük çantamdaki
üç
anahtar
small handbag-1SP-Loc-Rlvz three key
‘three keys in my small handbag’
Noun groups can also be modified by relative clauses. In Turkish, the noun on which
the relativization is performed is placed at the final position of the relative clause.
(46) a. Ağabeyim Ankara’da çalışıyor.
brother-1SP Ankara-Loc work-Prog-3Sg
‘My elder-brother works in Ankara.’
b. Ankara’da çalışan ağabeyim
Ankara-Loc work-Rel brother-1SP
‘My elder-brother who works in Ankara’
As seen above, the main verb of the relative clause is used in participle form. The
example depicts the suffix -en (phonological realization of -(y)An after morphophonemic processes) which is used in producing subject participle (in present). Other subject
suffixes are given below:
27
Suffix
-mHş (olan)
-(y)AcAk (olan)
-Hyor (olan)
Tense
past
future
progressive
The word olan (“being”) can optionally be used with past, future or progressive
participles, but not with present participle.
(47) a. Ankara’da çalışmış olan ağabeyim
Ankara-Loc work-Part be-Rel elder-brother-1SP
‘my elder brother who have worked in Ankara’
b. Ankara’da çalışacak olan ağabeyim
Ankara-Loc work-Part be-Rel elder-brother-1SP
‘my elder brother who will work in Ankara’
c. * Ankara’da çalışan olan ağabeyim
Olan can also be used in forming participle form of the copula.
(48) a. Arkadaşımın
annesi
hasta.
friend-1SP-Gen mother-3SP ill
‘My friend’s mother is ill.’
annesi
hasta olan arkadaşım
mother-3SP ill
be-Rel friend-1SP
‘my friend whose mother is ill’
b. Evin
pencereleri
geniş.
house-Gen window-Plu-Acc wide
‘Windows of the house are wide.’
pencereleri
geniş olan bir ev
window-Plu-3SP large be-Rel a house
‘a house which has large windows’
Apart from the subject participle form, the verb of a relative clause may take complement participle form, which is obtained by attachment of either -DHk or -yAcAk suffixes.
-DHk suffix, as itself, produces adjectives from verbs, although it is not productive:
(49)
28
bilmek → bildik
‘to know’
‘known’
umulmamak
→ umulmadık
‘to be not expected’
‘unexpected’
When used in complement participles, -DHk is always followed by a possessive (marks
the agreement in this case) and participle suffix group becomes DHğ-Agr. The tense of
this participle can be past or present, as examples (50) and (51) depicts, respectively.
Actual tense is usually determiner from the discourse.
(50)
Kitabı
kıza
geri verdim.
book-Acc girl-Dat back give-Past-1Sg
‘I gave back the book to the girl.’
a. kıza
geri verdiğim
kitap
girl-Dat back give-Rel-1Sg book-Acc
‘The book that I gave back to the girl.’
b. kitabı
geri verdiğim
kız
book-Acc back give-Rel-1Sg girl
‘The girl to whom I gave back the book.’
If there is an overt subject noun in the clause, it is marked with the genitive suffix:
(51)
Öğrenci sınıfta şarkı söylüyor.
student class-Loc song sing-Prog-3Sg
‘The student is singing a song in the classroom.’
a. öğrencinin sınıfta söylediği
şarkı
student-Gen class-Loc sing-Rel-3Sg song
‘the song that the student is singing in the classroom’
b. öğrencinin şarkıyı söylediği
sınıf
student-Gen song-Acc sing-Rel-3Sg class
‘the classroom in which the student is singing the song’
Complement participles in future tense are formed by attaching -(y)AcAk suffix to
verb stem. Just like -DHk suffix, -(y)AcAk combines with a possesive suffix to produce
29
-(y)AcAğ-Agr as the future complement participle.
(52) a. öğrencinin söyleyeceği şarkı
student-Gen sing-Rel-3Sg song
‘the song that the student will sing’
b. kitabı
geri vereceğim kız
book-Acc back give-Rel-1Sg girl
‘the girl to whom I will give back the book’
Relative clauses can be embedded as adnominals:
(53)
köyde
yaşayan kızın
yetiştirdiği
ineğin öldüğü
yer
village-Loc live-Rel girl-Gen breed-Rel-3Sg cow-Gen die-Rel-3Sg place
‘the place at which the cow that was breeded by the girl who lives in the village
died ’
3.1.3
The head
The last segment of the noun group is the head, and this position is filled either by a
common noun (54a), a proper noun (54b) or a pronoun (54c).
(54) a. küçük bir elma
small a apple
‘a small apple’
b. güzel
Ayşe
beautiful Ayşe
‘beautiful Ayşe’
c. unutkanlığıyla bilinen
sen
forgetful-3Sg-Ins know-Pass-Rel you
‘you who are known as forgetful’
30
Pronouns
When the head is a pronoun, no determiner or modifier segments are allowed:
(55)
*bazı sen
some you
*sarışın ben
blond I
Proper Nouns
When it is used as the head, a proper noun imposes certain restrictions on the selection
of the preceding segments. For example, particular determiners can be used in front of
a proper noun, while others are not applicable.
(56) a. Bu İstanbul nasıl düzelir?
this İstanbul how get-better-Pres-3Sg
‘How could this İstanbul get better?’
b. Nerede kaldı şu Hasan?
where left that Hasan
‘Where on the earth is Hasan?’
c. Trakya’da birkaç Yeşilköy’e
rastladım
Thrace’-Loc several Yeşilköy’-Dat come-across-Past-1Sg
‘I came across more than one Yeşilköy in Thrace’
d. Ailemizdeki
diğer/ikinci Mehmet dedemdir.
family-1PP-Loc-Rlvz other/second Mehmet grandfather-1SP-Cop
‘The other/second Mehmet in our family is my grandfather’
3.2
Postposition group
By postposition group, we mean a group of elements whose head is a proposition. Postposition group consists of a head and an optional complement noun group. The former
always occupies the final position.
31
.
3.2.1
Postpositions
Postpositions form a closed class of words. They can be viewed in subgroups, with
respect to the case of the complement they subcategorize for. ([16], pp. 85-89) Various
types of postpositions exist which subcategorize for: infinitives or nouns with nominative
case (57a,b), nouns with accusative case (57c), dative case (57d) and ablative case (57e).
(57) a. gelmek üzere
come-Inf for
‘for the purpose of coming’
b. sokak boyunca
street along
‘along the street’
c. Sınavı
müteakiben
exam-Acc following
‘after the exam’
d. şimdiye dek
now-Dat until
‘until now’
e. Dünden
beri
yesterday-Abl since
‘since yesterday’
3.2.2
Postposition Attachment
Attachment of a sequence of postpositions is determined without ambiguity by morphosyntactic cues (e.g., relative suffixes and case marks) and positional cues (head-final
structure). However, if a sentence involves relative clauses and postpositions, ambiguities may arise (58a).3 In “I read the newspaper on the couch”, if on the couch were an
adnominal, it would be relativized in Turkish (cf.,58b-c). Chained postposition groups
are not ambiguous because the predecessor modifies the successor.
(58) a. Bu bilgilere
göre
yazdığımız rapor değişmeyecek.
this data-Plu-Dat according write-Rel-1Pl report change-Neg-Fut-3Sg
‘The report that we wrote according to these data will not change.’
‘According to these data, the report that we wrote will not change’
3 In writing, it this may be disambiguated by seperating the sentential complement with a coma
(before yazdığımız in 58a).
32
b. Kanepedeki
gazeteyi
okudum.
couch-Loc-Rlvz newspaper-Acc read-Past-1Sg
‘I read [ the newspaper on the couch].’
c. Kanepede gazeteyi
okudum.
couch-Loc newspaper-Acc read-Past-1Sg
‘I read the newspaper [ on the couch].’
3.3
Adjective group
Adjective group is a sequence of words last of which is an adjective. Adjective groups
are typically formed by comparative and superlative adjectives.
3.3.1
Comparative adjectives
The head of a comparative adjective group is a qualitative adjective. Three comparatives can precede the head: “daha”, “az” and “çok” meaning “more”, “less” and “very”,
respectively.
(59) a. Elvan daha büyük bir eve
taşındı.
Elvan more big
a house-Dat move-in-Past-3Sg
‘Elvan moved in to a bigger house.’
b. Az şekerli kahve içerdi.
less sweet coffee drink-Pres-Asp
‘(S)he used to drink coffee with a little sugar.’
c. Çok hızlı arabalardan hoşlanmıyorum.
very fast car-Plu-Abl like-Neg-Prog-1Sg
‘I don’t like very fast cars.’
d. Annem
benden çok daha iyi yemek yapar.
mother-1SP I-Abl very more good dish make-Pres-3Sg
‘My mother cooks much better than I do.’
33
3.3.2
Superlative adjectives
The head adjective is qualitative for this group, too. Superlative form is obtained by
preceding the head with “en” (“most”).
(60)
Sınıfın
en çalışkan
öğrencisi
Ali’ydi.
class-Gen most hardworking student-3SP Ali-Cop
‘Ali was the most hardworking student of the class.’
3.4
Adverb group
An adverb group is a segment which has an adverb as its head. Modifiers of an adverbial head may be adverb or adjective groups, including the comparative daha and
the superlative en. Adverbial heads may be classified as manner (alelacele), temporal
(sonra, önce), position (aşağı, beri, ileri), repetition (gene, yeniden, tekrar), sentential
(besbelli, asla, kuşkusuz), frequency (seyrek, sık), possibility (herhalde, belki), definiteness (katiyen, muhakkak), and question (nasıl, hani) adverbs. Basic types of adverb
groups are described below.
3.4.1
Reduplications
Nouns and adjectives can be reduplicated to form an adverb group.
(61) a. Yemeğimizi
çabuk çabuk yedik.
meal-1PP-Acc quick quick eat-Past-1Pl
‘We ate our meal quickly.’
b. Akşam akşam canımızı
sıktı.
evening evening soul-1PP-Acc bother-Past-3Sg
‘It bothered us at this time of the evening.’
c. Geçen yaz
bu sahilleri
koy koy dolaştık.
last summer this shore-Plu-Acc cove cove go-around-Past-1Pl
‘We visited each and every cove of this shore last summer.’
Some of the reduplicated adverbs are onomatopoeic words:
34
(62)
Şırıl şırıl
akan
derenin
sesini
dinledim.
‘splashing’ flow-Part stream-Gen sound-3SP-Acc
‘I listened to the sound of the stream that flows gently.’
Distributive adjectives, when used as adverbs, are reduplicated:
(63)
Merdivenleri üçer
üçer
çıktık.
stairs-Plu-Acc three-Dist three-Dist go-up-Past-1Pl
‘We went upstairs three steps by three steps.’
Adverbs or adjectives can be intensified by phonological reduplication to produce
adverbs as well:
çabuk
hızlı
3.4.2
quick
fast
çarçabuk
hıphızlı
very quickly
very fast
Case-marked place adverbs
Adverbs of place act as the head of an adverb group either by themselves or by taking
a case suffix.
içeri
yukarı
ileri
öte
ön
karşı
inside
up
forward
yonder
front
opposite
dışarı
aşağı
geri
beri
arka
outside
down
backward
hitter
behind
(64) a. Evden
dışarı çıkmadım.
house-Abl outside go-out-Neg-Past-1Sg
‘I didn’t go out from the house.’
b. Yolun
ilerisi
görülmüyor.
road-Gen forward-POSS see-Pass-Neg
‘The forward part of the road is not visible.’
c. Nehirden öteye
nasıl geçilir?
river-Abl yonder-Dat how pass-Pass-Pres-3Sg
‘How can one go beyond the river?’
35
3.4.3
Temporal adverb groups
“sonra” (“after”) and “önce” (“before”) succeed noun groups denoting a time period or
a point in time, and form adverb groups.
(65) a. Dört gün sonra yola çıkacağız.
four day after road go-out-Fut-1Pl
‘We’ll set out on a journey in four days.’
b. Umarım
Perşembeden önce burada olmaz.
hope-Pres-1Sg Thursday-Abl before here
be-Neg
‘I hope he/she won’t be here before Thursday.’
Another type of adverb group denoting time is the one that uses special temporal nouns in head position. These temporal nouns are some time units (gün:“day”,
hafta:“week”, ay:“month”, mevsim:“season”, yıl:“year”, yüzyıl:“century”, dönem:
“semester, age”, çağ:“era, epoch”), days of week, months and year. In such adverb
goups, however, the set of words that may modify the head is rather limited: önceki
(“previous”, “before”), ertesi (“following”, “after”), geçen (“last”), gelecek (“next”),
bu (“this”), o (“that”).
(66) a. Ertesi
gün eski bir arkadaşıma
rastladım.
following day old a friend-1SP-Dat come-across-Past-1Sg
‘The following day, I came across with an old friend of mine.’
b. Gelecek yaz
Paris’e
gideceğim.
next
summer Paris-Dat go-Fut-1Sg
‘I will go to Paris next summer.’
3.4.4
Verb groups with adverbial use
Verb stems may function as adverbs with the addition of certain suffixes. These suffixes
are discussed below.
-(y)A suffix denotes a repeated action that takes place at the same time with the
main verb. Verb groups in this gerundive form consist of two gerunds (either of the
same verb or different verbs).
36
(67) a. Ağacı budaya budaya biçimlendirdi.
tree-Acc prune prune shape-Past-3Sg
‘He/she shaped the tree pruning.’
b. Çocuk düşe kalka büyür.
child fall rise grow-Pres
‘A child grows falling and rising.’
-(y)ArAk suffix denotes a continuous action or a point action which takes place either
at the same time with the main verb or just before it.
(68) a. Öpüşerek ayrıldılar.
kiss-Recp leave-Past-3Pl
‘They kissed each other as they said goodbye.’
b. Koşarak karşıya
geçtik.
run
opposite-Dat pass-Past-1Pl
‘We crossed the street running.’
-(y)Hp suffix is attached to the first of consecutive verb stem pairs and provides a
connection (e.g., temporal sequence) between these stems.
(69) a. Şemsiyemi
işyerinde unutup gelmişim.
umbrella-1SP-Acc office-Loc forget come-Past-1Sg
‘I came, having forgotten my umbrella at the office’
b. Oturup konuşalım
sit-down talk--Wish-1Pl
‘Let’s sit down and talk.’
-(y)HncA suffix marks its stem as the temporal predecessor of the main verb.
(70) a. Eve
varınca seni
ararım.
house-Dat arrive you-Acc call-Pres-1Sg
‘I’ll call you when I arrive home.’
b. Haberleri dinleyince yolculuğumu erteledim.
news-Acc listen
travel-1SP-Acc postpone-Past-1Sg
‘I postponed my travel when I listened to the news.’
37
-DHkçA suffix is a composite one which combines participle suffix -DHk with çA. This
composite suffix has the meaning “so long as” or “the more”.
(71) a. Çalışmadıkça başarılı olamazsın.
study-Neg
successful be-Neg-Pres-2Sg
‘So long as you don’t study, you cannot be successful.’
b. Ankara’ya geldikçe bize
uğrar.
Ankara-Dat come
we-Dat visit-Pres-3Sg
‘Every time he/she comes to Ankara, he/she visits us.’
c. İp atladıkça susuyorum.
rope skip
be-thirsty-Prog-1Sg
‘The more I skip, the more I get thirsty.’
The suffix sequence -(H)r· · ·mAz attach to the same verb stem to produce a verb
group that can be used like an adverb. This construction has a meaning similar to “as
soon as”.
(72) a. İbibikler
öter ötmez oradayım.
hoopoe-Plu sing sing-Neg there-Loc-Cop(1Sg)
‘I will be there as soon as the hoopoes sing.’
b. Otobüsten iner inmez
onu
gördüm.
bus-Abl
get-off get-off-Neg he/she/it see-Past-1Sg
‘I saw her/him/it as soon as I got off the bus.’
-(y)ken suffix is the last one that is to be discussed in this section. It can be
translated to English as “as”. This suffix differs from the previous ones as it attaches
not to a verb stem, but usually to third person singular inflection of the verb in aorist.
It may also attach to narrative past, present and future tense forms for third person
singular. The suffix does not harmonize with the vowels of the verb stem.
(73) a. Çayımı
içerken gazete
başlıklarına
göz atarım
tea-1SP-Acc drink newspaper headline-3PP-Dat eye throw-Pres-1Sg
‘I glance through newspaper headlines as I drink my tea’
b. Düşümde
dövüşmekteyken yanımda
yatanı
tekmelemişim.
dream-1Sg-Loc fight
side-1Sg-Loc lie-Rel-Acc kick-Past-1Sg
‘I had kicked the one lying next to me as I was fighting in my dream.’
38
c. Buraya kadar gelmişken geri dönmek olmaz
here-Dat upto come
back turn-Inf be-Neg-Pres
‘It’s impossible to go back now that we came up to here.’
3.5
Verb group
3.5.1
Predicate types
Predicates in Turkish can be verbal (74a), nominal with an attached auxiliary suffix (74b), nominal with a copula (74c–d), or existential(74e–f).
(74) a. Adam topa
sert vurdu.
man ball-Dat hard hit-Past-3Sg
’The man hit the ball hard.’
b. Kitabın arabadaydı.
book-2SP car-Loc-Aux
’(Your) book was in the car.’
c. Benimki en hızlı arabadır.
I-Gen-Pro most fast car-Cop
’Mine(my car) is the fastest car.’
d. Gökyüzü hep
mavidir.
sky
always blue-Cop
’The sky is always blue.’
e. Ayşe’nin iki çocuğu var.
Ayşe-Gen two child-Acc exist
’Ayşe has two children (there exist two children of Ayşe).’
f. Sokakta kimse yok.
street-Loc nobody not-exist
’There is (there exists) no one on the street.’
3.5.2
Subcategorization
Every verb—except the intransitives—subcategorize for a noun group or a set of noun
groups. These noun groups may be in accusative (75a), dative (75b), locative (75c),
ablative (75d) or instrumental/commitative case (75e).
39
(75) a. Raporu
henüz bitirmedik.
report-Acc yet
finish-Neg-Past-1Pl
‘We haven’t finished the report yet.’
b. Yarın
sinemaya gidelim.
tomorrow cinema-Dat go-Wish-1Pl
‘Let’s go to the cinema tomorrow.’
c. Buzdolabında hiçbir şey kalmamıştı.
refrigerator-Loc no-at-all thing remain-Neg-Past-Asp-3Sg
‘Nothing was left at all in the refrigerator.’
d. Atakule’den dönerken Evrim’i
gördüm.
Atakule-Abl return-Adv Evrim-Acc see-Past-1Sg
‘I saw Evrim as I was coming back from Atakule.’
e. Çocuklar oyuncaklarıyla oynuyorlar.
child-Plu toy-3PP-Ins play-Prog-3Pl
‘The children are playing with their toys.’
Ufuk bir arkadaşıyla
çalışacak.
Ufuk a friend-3SP-Ins work-Fut-3Sg
‘Ufuk will work with a friend of his.’
The number of required noun groups depend on the valency of the verb.
Kitap okuyordu.
book read-Prog-Asp-3Sg
‘He/she was reading a book.’
Ditransitive verb:
Mehmet’e
gitarımı
verdim.
Mehmet-Dat guitar-1SP-Acc give-Past-1Sg
‘I gave my guitar to Mehmet.’
More noun groups may be provided to increase the amount of information provided;
they act as complements.
Transitive verb:
(76)
Sandıkları
İzmir’den Samsun’a gemiyle yolladık
chest-Plu-Acc İzmir-Abl Samsun-Dat ship-Ins send-Past-2Pl
‘We sent the chests from İzmir to Samsun by ship.’
Some verbs subcategorize for a complement clause:
söylemek:“to say”, söz vermek:“to promise”, iddia etmek:“to claim”, inanmak:“to
believe”, zannetmek:“to assume”, tahmin etmek:“to guess”, sanmak:“to suppose”,
ispat etmek:“to prove”, inkar etmek:“to deny”, yemin etmek:“to swear”, düşünmek:
40
“to think”, emin olmak:“to be sure”, kuşkulanmak:“to suspect” etc.
(77) a. Dosyayı bulacağına
söz vermiştin.
file-Acc find-Part-2Sg-Dat promise-Past-Asp-2Sg
‘You had promised that you would find the file.’
b. Randevumuzu
unuttuğumu
iddia ediyor.
Appointment-1PP-Acc forget-Part-1Sg-Acc claim-Prog-3Sg
‘She/he claims that I forgot our appointment.’
Some verbs (yeğlemek:“to prefer”, kabul etmek:“to accept”, çalışmak:“to try”,
çabalamak:“to struggle”, alışmak:“to get accustomed to”, özenmek:“to desire”, karar
vermek:“to decide”, niyetlenmek:“to intend”, bahsetmek:“to mention”, vazgeçmek:“to give up”, anlamak:“to understand” etc.) subcategorize for an infinitive form
of the verb.
(78) a. Bu işi
bitirmeye
söz verdik.
this job-Acc finish-Inf-Dat promise-Past-1Pl
‘We promised to finish this job.’
b. Bugün alışveriş yapmaktan vazgeçtik.
today shopping make-Inf-Abl give-up-Past-1Pl
‘We gave up (the idea of) shopping today.’
3.5.3
Auxiliary verbs
In Turkish, some verbs are composed of a noun and an auxiliary verb. The auxiliary
verbs used in such constructions are etmek:“to do” and yapmak:“to make”, the former
being more frequent.
(79) a.
alay
‘mockery’
b.
kabul
‘acceptance’
c.
alışveriş
‘shopping’
→
alay etmek
‘to mock’
→
→
kabul etmek
‘to accept’
alışveriş yapmak
‘to shop/to do shopping’
41
There is another auxiliary which attaches to nouns to form nominal predicates: olmak (“to be”). This auxiliary differs from etmek and yapmak in two respects. First,
it does not appear as a separate word, but rather a morpheme when the sentence is in
past or present tense. Second, its inflection does not resemble to that of a verb but the
copula.
(80) a. Babam
geçen ay
yurtdışındaydı.
father-1SP last month abroad-Loc-Aux
‘My father was abroad last month.’
b. Üç gündür çok uykusuzdum.
three day
very sleepless-Aux
‘I have been very sleepless for three days.’
c. Kitaplar masanın üstündeymiş.
book-Plu table-Gen top-1SP-Loc-Aux
‘The books were on the table.’
This auxiliary is not present for third person form if the sentence is in the present
tense.
(81)
Bardaklar rafta.
glass-Plu shelf-Loc-(Cop)
‘The glasses are on the shelf.’
For future tense as well as conditional and necessitative forms, olmak succeeds the
noun as a separate word.
(82) a. Babam
geçen ay
yurtdışında olmasaydı...
father-1SP last month abroad-Loc be-Neg-Cond-Asp
‘If my father weren’t abroad last month...’
b. Kitaplar masanın üzerinde
olacak.
book-Plu table-Gen top-1SP-Loc be-Fut-3Sg
‘The books will be on the table.’
Another point to be emphasized about this auxiliary is that it has a different negative
form than the other verbs when the sentence is in past or present tense. Negativization
is performed by introducing the word değil (“not”) just after the nominal. The tense
marker, if exists, attaches to değil.
42
(83)
Cem evde
değildi.
Cem house-Loc not-Aux-3Sg
‘Cem wasn’t at home.’
An ambiguity may arise with negative questions of predicates. This ambiguity is
resolved by stress in speech and by a comma preceding değil in writing.
(84)
Kedi bahçede
değil mi?
cat garden-Loc not Ques
‘The cat is in the garden, isn’t it?’
‘Isn’t the cat in the garden?’
3.5.4
Existential predicates
Existential predicates are formed using var (“existent”) and yok (“non-existent”).
(85) a. Odada
dört koltuk vardı.
room-Loc four armchair exist-Aux
‘There were four armchairs in the room.’
b. Burada kimse
yok.
here
anybody non-existent
‘There isn’t anybody here.’
c. Arabası yokmuş.
car-POSS non-existent
‘She/he doesn’t have a car.’
var and yok cannot be used in future, conditional or necessitative forms. For these
cases, olmak replaces var and yok.
(86) a. Odada
dört koltuk olmalı
room-Loc four armchair be-Nec
‘There should be four armchairs in the room.’
b. Burada kimse olmayacak.
here
nobody be-Fut-3Sg
‘There won’t be anybody here.’
43
3.5.5
Infinitive form of the verbs
Infinitive form of a verb is formed with suffix -mAk attached to the stem.
(87)
Fransa’ya gitmek çok para ister.
France-Dat go-Inf much money require-Pres
‘Going to France costs much’
Infinitive suffix cannot be followed by genitive or possesive suffixes. However case
suffixes are allowed.
(88) a. Koşmaktan yoruldum.
run-Inf-Abl tire-Pass-Past-1Sg
‘I got tired of running.’
b. Kızmakta
haklısın.
get-angry-Inf-Loc right-Cop(2Sg)
‘You are right to be angry.’
3.5.6
Gerundive forms of the verbs
There are a couple of suffixes for producing gerundive forms of a verb: -mA and -(y)Hş.
-mA is used for referring to the action or its result. Genitive and possesive suffixes can
be attached to -mA.
(89) a. Onunla
görüşmenin
bana faydası olmaz.
he/she-Gen-Ins meet-Ger-Gen me use
be-Neg-Pres-3Sg
‘Meeting with him/her is of no use to me.’
b. Okuması
düzeliyor.
read-Ger-3SP improve-Prog-3Sg
‘His/her reading is improving.’
-(y)Hş produces a gerundive which emphasizes the manner the action is performed.
This suffix can also be succeeded by genitive and possesive suffixes.
(90)
Gülüşünü
hatırlıyorum.
smile-Inf-3SP-Acc remember-Prog-1Sg
‘I remember the way you/(s)he smile/s.’
44
3.5.7
Syntax of causative verbs
In Turkish, verbs are causativized by attaching the causative suffixes -DHr, -Hr, -t,
-Ht and -Ar to the stem. Using these suffixes, one can obtain a causative verb almost
from any verb, including the causatives themselves.
(91) a.
inanmak → inandırmak
‘to believe’
‘to persuade’
b.
doğmak
→ doğurmak
‘to be born’
‘to give birth to’
c.
oturmak → oturtmak
‘to sit’
‘to seat’
d.
korkmak → korkutmak
‘to fear’
‘to frighten’
e.
çıkmak
→ çıkartmak
‘to go out/to go up’
‘to remove/to raise’
Appropriate combinations of causative suffixes allow production of multiple causatives:
(92)
ölmek → öldürmek → öldürtmek
‘to die’
‘to kill’
‘to have someone killed’
From syntactical point of view, causativization process has two important results:
increase in the valency of the verb, and the changes in the grammatical functions of the
noun groups.
Any verb form of valency n will require n + 1 noun groups after causativization:
Intransitive verb: uyumak → uyutmak
‘to sleep’
‘ to send to sleep’
Transitive verb: bir şeyi okumak
→ birine birşeyi okutmak
‘to read something’
‘ to make someone read something’
Ditransitive verb:
bir şeyi bir yere koymak
→ birine birşeyi bir yere koydurtmak
‘to put something to somewhere’
‘ to make someone put something to somewhere’
Another effect of causativization is that the noun groups of the original clause change
their grammatical functions:
45
Causativization of an Intransitive Verb: The subject of the intransitive verb
becomes the direct object (accusative-marked noun group) in the causative clause. A
new noun group is introduced for subject position of the causative clause.
(93)
Yiğitcan güldü
→ Işık Yiğitcan’ı güldürdü
NOM
NOM ACC
‘Yiğitcan laughed’
‘Işık caused Yiğitcan to laugh’
Causativization of a Transitive Verb: The subject of the transitive verb becomes
the dative-marked indirect object in the causative clause, whereas the direct object (e.g.,
şiir in 94)) preserves its grammatical function in the causative clause. Also, a new noun
group is introduced for subject position.
(94)
Arzu şiir okudu
→ Öğretmen Arzu’ya şiir okuttu
NOM NOM
NOM DAT NOM
‘Arzu read a poem’
‘The teacher made Arzu read a poem’
The indirect object of the causative clause may sometimes be omitted:
(95)
Öğretmen şiir okuttu.
teacher
poem read-Caus-Past-3Sg
‘The teacher caused a poem to be read.’
If the main verb subcategorizes for a dative noun group, this noun group remains
unaltered in the causative clause. In such a case, the subject of the main verb is marked
as accusative and becomes the direct object of the causative clause.
(96)
Çocuk okula başladı
→ Çocuğu okula başlattık
NOM DAT
ACC NOM
‘The child started school’
‘We made the child start school’
Causativization of a Ditransitive Verb: The subject of the ditransitive verb becomes the dative-marked indirect object in the causative clause, whereas the accusativemarked direct object and the dative-marked object of the main verb preserve their
grammatical functions in the causative clause. Subject position is again filled by a new
noun group.
46
(97)
Hakan kitabı masaya koydu
→ Ali Hakan’a kitabı masaya koydurdu
NOM ACC DAT
NOM DAT ACC DAT
‘Hakan put the book on the table’
‘Ali made Hakan put the book on the table’
Just as for the tansitive verb, the subject of the main verb may be omitted:
(98)
Ali kitabı
masaya koydurdu.
Ali book-Acc table-Dat put-Caus-Past-3Sg
47
Chapter 4
DESIGN
Every language is expected to have a different realization of the language-independent
principles. Some solutions proposed by Pollard and Sag [21] for English have to be modified to model Turkish. The basic features that distinguish Turkish from English are:
importance of morphology in the specification of grammatical functions, overt case and
agreement marking, final head position, free constituent order, pronoun drop, complement drops, and the nature of unbounded dependency constraints. Some of these items
are resolved by extending the sign structure and introducing Turkish-specific versions of
some principles.
4.1
Sign Structure
The template for phrases is given in Figure 4.1. A structure similar to that of English
is used for signs in Turkish, with minor changes. In category (cat) type, an additional
feature SUBJ is added, which is coindexed with the subject complement if it exists in
the subcat list. It is used to distinguish subject complement from other complements
which may have the same case and agreement. It refers to the subject complement
directly. This is necessary for implementing some syntactic phenomena like subject
raising relative clauses and other bindings to subject complement. Example (99) shows
a simplified version of the CAT feature of a lexical entry for the verb “seviyor” (loveProg-3Sg).
48
phrase
PHON








SYNSEM











DTRS

(99)

string, ...
synsem
LOCAL









NONLOCAL

daughters
HD-DTR
SPEC-DTRS

COMP-DTRS




 

cat


 

cat
 

 

head
CAT HEAD
 

 

SUBJ
synsem





SUBCAT subcat-type


CONT sem-obj
i
h
INHERITED | SLASH: null ∨ synsem 

TOBIND | SLASH: null ∨ synsem





sign




sign,... 

sign,...
Figure 4.1: Sample sign structure for Turkish
word
D
E

PHON
“seviyor”
% love-Prog-3Sg




cat

HEAD

head


SYNSEM | CAT SUBJ
1


n


SUBCAT
1
NP[nom], NP[acc]











o

One of the major differences in sign structure is the SUBCAT feature. Since Turkish has free constituent order,1 a type which is a nested combination of ordered and
unordered lists is used instead of lists indicating obliqueness of the complements. Unordered lists are denoted with curly braces similar to sets.
Only the SLASH feature of type null ∨ local is defined as a nonlocal feature. Nonlocal
features are used to process the information coming from arbitrary daughters (not only
head daughter) which will be transmitted to upper phrases and bound to outer structures. Unbounded dependency and other binding constraints are defined by nonlocal
features. HPSG defines three basic nonlocal features: SLASH, REL and QUE which are
used to implement filler-gap dependencies, relative clauses and questions respectively.
In this study, we only used the SLASH feature.
Since adjuncts may exist in any position preceding the head and probably between
any subcategorized constituent, we have chosen to combine adjunct and complements of
the phrase into one “daughters” attribute. Daughters (DTRS) attribute consists of adjunct daughters (ADJ-DTRS) and complement daughters (COMP-DTRS) which are lists
1 For
more information on order-freeness, see Chapter 3 and Section 4.3.
49
of sign. Head daughter (HD-DTR) is of type sign. If phrase has a subject complement,
there is another feature subject complement (SUBJ-COMP) which is coindexed with the
subject in the COMP-DTRS list.
4.2
Major Categories and Head Features
Since derivations involving category change is possible in Turkish, (e.g. relativizer -ki
turning a noun into a specifier), and derived words preserve their syntactic behavior,
head features must be extendable. For example, finite verbs are not nominalized hence
do not carry case. However, sentential complements have inflections which make them
behave as nominals and take case:
(100) Eve
girdiğimizdeki
manzarayı gördünüz.
house-Dat enter-Part-1Pl-Loc-Rlvz view-Acc see-Past-2Pl
‘You saw the view of the house when we entered.’
Basic major categories are: nouns, verbs, adjectives, adverbs and conjunctives. Category information consists of HEAD, SUBJ and SUBCAT features. Each category has its
own set of appropriate features in head attribute. HEAD feature is of type head which
has the following categories defined as subsorts.
head
adjective
noun
common
proper-noun
pronoun
determiner
quantitative
qualitative
questional
adverb
question
direction
temporal
verb
sentential
manner
infinitival
adverbial
finite
quantitative
relative
complement
Figure 4.2: Sort hierarchy of type head
Appropriate features for the head sort noun are case, agreement, relativized, nominalindex and possession (cf. 101). CASE can be nominative, objective, genitive, dative,
50
ablative, genitive or instrumental. In Turkish, there is no gender of objects so agreement(AGR) consists of person and number information. Since possession (if noun is used
in the possessive group) is overtly marked, it is included as a head feature of type either
none or agreement. Also feature nominal index (N-IND) is defined to handle adjuncts
modifying the derived noun (see Section 4.5) and coindexed with the semantic index of
the root noun.
(101) PHON








SYNSEM






“bahçelerimizin”







CAT







| HEAD
CONT | INDEX




# 






third



plural 



#





third

plural 
% “of our garden”
common-noun
CASE "genitive

agr

AGR PERSON

NUMBER

N-IND 1

"

agr

POSS
agr
PERSON
NUMBER
1
As can be seen in Figure 4.2, head sort verb has three subsorts, finite, infinitival
and adverbial. Finite verbs are the heads of the finite sentences. Infinitivals are verbal heads which modify or specify other phrases or subcategorized by other heads as
complements. However, they still have the properties of verbs and construct embedded
sentences. Relative clauses are verbs inflected by suffixes -An, -dHk-Poss, -AcAk+Poss
and contain a gap which is filled by a following noun phrase. Another group of verbal
heads are sentential complements which can be arguments of some verbs. These are
verbs inflected by suffixes -mA(k), -Hş-Poss, -dHk-Poss, and -AcAk-Poss where the
possessive suffix marks agreement rather than possession. Similarly, sentential adverbs
modify the matrix verb.
In addition to the syntactic roles, there are structural differences in these three
sorts. Finite verbs take an secondary tense or aspect marker which is one of none,
past, dubitative or conditional. Infinitivals carry case information so they have the CASE
feature. The following attribute definitions are appropriate for these groups:
51
(102)
verb:
AGR
NEG
TENSE
4.3
null ∨ agr
plus ∨ minus
base ∨ present ∨ continuous ∨ past ∨
dubitative ∨ wish
finite:
AGR
AUX-TENSE
agr
past ∨ dubitative ∨ conditional
infinitival:
CASE
case
Complement Selection and Linear Precedence
Heads select their arguments using the SUBCAT feature. The SUBCAT feature is a
structured type consisting of arguments of sort synsem. Therefore, a head can select any
syntactic property of its arguments like category, case, agreement, nonlocal features and
even semantic content. Any type of category is allowed including sentential complements,
adverbs, adjectives, etc.
As mentioned in the preceding section, scrambling of constituents is handled by
unordered lists. Linear precedence constraints of Turkish can be described generally as:
(103) a.
b.
"
HD-DTR
1
COMP-DTRS
HD-DTR
ADJ-DTRS
1
....,
....,
2
#
SYNSEM | CAT | HEAD
2
sign ,....
sign ,....
=⇒
¬ verb
2
<
=⇒
2
<
1
1
(103a) describes the constraint “complement daughters should precede the head
daughter when head daughter is not a verb” and (103b) describes the constraint “adjunct
daughters should precede the head daughter”.
To handle the cases such as ‘nonreferential object should immediately precede the
verb’, we use a special sort for subcat feature. This sort has a nested mixed structure of
list and set.2 Subcat type can be either a list or a set, and recursively, element of a set
can be either a synsem-arg value or a list. Element of a list can be either a synsem-arg
value or a set (104b). synsem-arg sort is used to enable optional arguments (104a). If
2 Set is used to indicate the property of order-freeness where all permutations of the members of the
set is possible at the surface.
52
OPT attribute of an argument is plus than it is optional and can be omitted. Optional
arguments are generally denoted with enclosing parentheses.
(104) a.


synsem-arg
OPT plus ∨ minus
ARG synsem
b.
subcat-type



list-subcat
HD synsem-arg ∨ set-subcat 
TL e-list ∨ list-subcat

set-subcat
EL synsem-arg ∨ list-subcat 
ELS e-list ∨ set-subcat
In the surface form, lists are ordered and sets are permuted. For example, the
sign in (105a) has the surface forms listed in (105b). It is assumed that adam (man) is
substituted for the subject, çocuğa (child-Dat) is substituted for the dative object, evden
(house-Abl) is for ablative argument and kalem (pencil) is the nonreferential object.
(105) a. PHON



SYNSEM | LOCAL | CAT
‘getirdi’

HEAD
SUBJ

SUBCAT
% bring-Past-3Sg
verb
1
D
1




E

NP[nom], NP[dat], NP[abl] , NP[nom]
b. “Adam çocuğa evden kalem getirdi”
“Adam evden çocuğa kalem getirdi”
“Çocuğa adam evden kalem getirdi”
“Çocuğa evden adam kalem getirdi”
“Evden çocuğa adam kalem getirdi”
“Evden adam çocuğa kalem getirdi”
Sentential complements and any kind of argument-head relation can be declared in
this manner. For example the verb “söyledi” (told) can be defined as:
53
(106) PHON
‘söyledi’
% tell-Past-3Sg


HEAD
finite-verb

1
SYNSEM | LOCAL | CAT SUBJ





NP[nom], NP[dat], S[ inf,acc]
SUBCAT
Where:
S[inf,acc]:


SYNSEM | LOCAL | CAT 
HEAD
SUBCAT
h
infinitival
CASE
i
acc

Verbal categories have the most complex complement structures. All verbal heads
(finite verbs, infinitivals, sentential adverbs) require one or more complements according
to their valence. Turkish is a complement-drop language so complements can be dropped
even if they are obligatory arguments.
Other typical complement-head relationship is in the possessive noun group (107a).
A possessive marked noun subcategorizes for a genitive noun and the part of speech of
the complement should agree with the possessive suffix (107b).
(107)
sarayın
kapısı
palace-Gen door-3SP
‘door of the palace’


‘kapısı’
PHON
% “door-3SP”




common


HEAD



1 agr 
POSS

SYNSEM | LOCAL | CAT 

SUBCAT
NPgen [AGR
1
]
The order of the complements and adjuncts are variable which means adjuncts specifying the head can be in any position. So, instead of generating the surface form from
the subcat list directly by a phrase structure rule, we chose to retrieve the complements
one at a time. This allows the adjunct rule which will be described in following sections
to be applied to the head at any position.
(108) SYNSEM | LOCAL | CAT
SYNSEM
HEAD
SUBCAT
3
1
2
−→
, SYNSEM | LOCAL | CAT
HEAD
SUBCAT
1
4
, selectlast( 3 , 4 , 2 )
Where selectlast selects the last synsem value ( 3 ) from the SUBCAT structure ( 4 ),
and rest is stored in third parameter ( 2 ).
54
This rule applies to the head-final complements. Handling scrambling of verbal head
to pre-complement position is made possible by another schema:
(109) SYNSEM | LOCAL | CAT
HEAD
SUBCAT
1
2
SYNSEM | LOCAL | CAT
−→
HEAD
SUBCAT
1
4
verb
SYNSEM
3
, selectfirst( 3 , 4 , 2 )
Where selectfirst selects the first synsem value ( 3 ) from the SUBCAT structure ( 4 ),
and rest is stored in third parameter ( 2 ).
4.4
Pronoun Drop
One of the distinct properties of Turkish is the pronoun drop; pronoun in the subject
position can be omitted since it is marked by agreement of the head. There are three
constructs where pronouns drop: subject of the verbal heads, substantive predicates
and possessive noun groups. In both cases, including embedded sentences in which the
subject has genitive case, the dropped pronoun has either nominal or genitive case.
(110) a. (biz) Treni
gördük.
We train-Acc see-Past-1Pl
‘We saw the train.’
b. (benim) Güzel bahçem.
I-Gen nice garden-1SP
‘My beautiful garden.’
c. (o) (benim) Eve gittiğimi
gördü.
he I-Gen house go-Part-1Sg-Acc see-Past-3Sg
‘He saw that I went to house.’
d. (o) (benim) En yakın arkadaşımdır.
(he) (I-Gen) most close friend-1SP-Cop(3Sg)
‘He is my best friend.’
A solution to pro-drop is using empty categories, which have null surface forms. A
possible declaration for dropped pronoun as empty category is:
55
(111) PHON



SYNSEM | LOCAL | CAT HEAD
SUBCAT
h
pronoun
CASE
nominative ∨

i

genitive 
This declaration will fill the subject position required by any head feature. However,
empty categories usually cause major problems. In most of the implementations, they
are inserted into any position available in the sentence. This is simply inefficient. More
critically, when the order of the complement filled by the empty category has free order
as it is in Turkish, superfluous parses are generated for each possible position that the
subject can occupy. Therefore more constraints may be necessary to deal with the empty
categories. The same problem also exists for management of the trace in relative clauses
(Section 4.6). For the time being we have chosen the keep dropped pronouns as empty
categories.
4.5
Adjuncts
Adjuncts are optional elements in the phrase structure. Adjuncts cannot be modeled
in the same way as the complements. Their most distinct property is that they do
not change the valence of the phrase they combined with. In other words, a head can
be specified/modified by any number of adjuncts, which may possibly have the same
category.
Another problem about adjuncts is whether the heads should select their adjuncts
or adjuncts should select their heads. One solution proposed by Pollard and Sag [20]
takes the approach where heads select their adjuncts. A new set-typed feature called
adjuncts is added to sort cat, and adjunct is checked by whether it is unified with one of
the elements of the set. The number of elements in the set does not change. However,
adjuncts may come in many different varieties and this set may grow to an unmanageable
size.
In the other approach, adjuncts select their heads [21]. This provides a simpler solution because the heads that an adjunct can modify are more restricted. MOD attribute
of type synsem defined in the lexical entry fot the adjunct is used to select the syntactic
category of the head. MOD is a head feature containing the restrictions for the head to
be modified, and is unified with the SYNSEM value of the head. In examples (112a–b),
the adjunct category of adjective/adverb subcategorizes for the head noun/verb respectively.
(112) a. PHON


SYNSEM | LOCAL | CAT
‘mavi’
"
HEAD
% blue
qualitative-adj
MOD
56
LOCAL | CAT | HEAD
noun

#

b. PHON


SYNSEM | LOCAL | CAT
‘çabuk’
"
HEAD

#

% fast
adverb
MOD
LOCAL | CAT | HEAD
verb
With this model, all adjuncts have similar structure and can be handled by the same
rule. In Turkish, an adjunct with the appropriate MOD attribute can precede the phrase
anywhere. So a preliminary version of adjunct principle can be written as:
(113) 
HD-DTR
DTRS ADJ-DTRS
1
2
⊕
3
SYNSEM | LOCAL | CAT | HEAD
3
4


−→
SYNSEM | LOCAL | CAT | HEAD | MOD
5
,
1
"
SYNSEM
DTRS
5
LOCAL | CAT | HEAD
ADJ-DTRS
2
4
#
Although adjuncts can modify a phrase in any preceding position, there are restrictions on the possible combinations and order of the adjuncts modifying the same head.
Rules defining the grammatical combinations vary; an adjunct modifying the head may
prevent other adjuncts to modify the same head. In (114a) “güzel” modifies “bahçedeki”
and does not modify “çiçek”. Similarly, the quantitative adjective “iki” cannot modify the noun phrase “bu çiçek”. However “iki” does not prevent “bu” from specifying
“çiçek” (114b–c).
(114) a. güzel
bahçedeki
çiçek
beautiful garden-Loc-Rlvz flower
‘The beautiful flower in the garden’
b. * iki bu çiçek
two this flower
c. bu iki çiçek
this two flower
’These two flowers’
In order to control the combinations of adjuncts, we introduce a new feature for all
categories under the CAT feature called ADJUNCTS. This structure consists of a group of
boolean attributes that keep track of the adjuncts that have been applied to the category.
In the adjunct part, the MOD attribute is divided into two attributes: a synsem value
(MODSYN) with the same purpose of MOD in (113), and MODADJ defining the resulting
57
ADJUNCTS structure which will be projected to the mother phrase. Adjunct still selects
the head together with the ADJUNCT value included in the SYNSEM of the head, and
defines which flags will be set and passed to the mother phrase. For example, assume
that ADJUNCTS consist of three flags: RLV indicating that the relativized noun has
been applied, DEM indicating the demonstrative adjective has been applied and QLT
indicating that the qualitative adjective is applied. Simplified lexical entries for each
category could be as in the example (115).
(115) a. PHON




SYNSEM | LOCAL | CAT | HEAD | MOD

b. PHON





SYNSEM | LOCAL | CAT | HEAD | MOD


c.

‘güzel’

% beautiful
MODSYN
LOCAL | CAT | ADJUNCTS

"


RLV
MODADJ DEM
QLT
‘bu’

−
−
+
#
QLT
SYNSEM | LOCAL | CAT
3
1
2
⊕
3
‘bahçedeki’

ADJUNCTS
QLT


4 
2
−→
6
SYNSEM | LOCAL | CAT | HEAD | MOD
"
SYNSEM
DTRS
5
QLT
LOCAL | CAT | HEAD
ADJ-DTRS
2
58
4
MODSYN
MODADJ
#
5
6
,
"
RLV
DEM
QLT

#



1 



−
−
% one that is in the garden
MODSYN LOCAL | CAT | ADJUNCTS


"
#

RLV
+

MODADJ DEM 1

HEAD
RLV
DEM
1
With these definitions, a revised adjunct principle can be written as:
"
MODADJ





SYNSEM | LOCAL | CAT | HEAD | MOD



1
RLV
DEM
% this
MODSYN LOCAL | CAT | ADJUNCTS


"
#

RLV
−

DEM +
PHON
(116) 
HD-DTR
DTRS
ADJ-DTRS



h

i
− 

− 




#


1 


2





−
When relative clauses, quantifiers, article ‘bir’, classifier nouns, and quantitative
adjectives are defined, all noun phrase combinations can be covered. On the other hand,
genitive noun in possessive noun group is not a specifier. It is an argument of the
possessive noun. Thus it requires a special interpretation. Specifiers and modifiers can
specify/modify the possessive marked noun as long as they are between the genitive noun
and the possessive noun. Otherwise they specify/modify the genitive noun. To prevent
adjuncts from passing over the genitive noun, we defined another constraint which can be
informally expressed as: “a noun modifier/specifier modify/specify a possessive marked
noun if it is not saturated”. This constraint can be shown as:
(117) SYNSEM | LOCAL | CAT | HEAD | MOD | LOCAL | CAT
1
4.6
h
SUBCAT
¬
i
1
HEAD
h
noun
POSS
¬ none
i
=⇒
Relative Clauses
Filler-gap dependencies are the contracts in which elements are extracted from their
positions (leaving gaps) and appear in other positions (filler). In Turkish, typical fillergap construction is the relative clauses. Two basic strategies exist for relative clauses
which are called wa and ga by Hankamer and Knecht [12] which are realized respectively
by -(y)An and -DHk-Agr or -(A)cAk-Agr relative participles:
(118)
i. When the gap is the relative clause subject, or a subconstituent of the relative
clause subject, use the wa strategy.
ii. When there is no relative clause subject, use the wa strategy.
iii. When th gap is not a part of relative clause subject, use the ga strategy.
(119a–b) are examples of wa, (119c–d) are examples of ga strategy.
(119) a.
b.
Adama kalemi
veren çocuğu1 gördüm.
man-Dat pencil-Acc give-Rel child-Gen see-Past-1Sg
‘I saw the child who gave man the pencil.’
1
yakınına
köprü yapılan
ev1
near-3SP-Dat bridge build-Pass-Rel house
‘The house1 to which a bridge is built next
1’
1
59
kalemi1
gördüm.
c. Çocuğun adama
1 vereceği
child-Gen man-Dat
give-Rel-3Sg pencil-Acc see-Past-1Sg
‘I saw the pencil1 that the child will give
1 to the man.’
adamı1 gördüm.
d. Çocuğun kalemi
1 verdiği
child-Gen pencil-Acc
give-Rel-3Sg man-Acc see-Past-1Sg
‘I saw the man1 to whom the child gave
1 the pencil.’
(118ii) introduces a special condition where the relative clause has no subject. In
Turkish there are two cases for clauses with no subject [2]: impersonal passives and
verbs with incorporated subject. In these cases, the real agent of the verb does not
exist. The noun in the subject position incorporates to the verb. In the example (119b),
‘köprü’ is an incorporated subject. Similarly, examples below show the relativization of
an adjunct NP—a locative adjunct in this case, with subject incorporation (120a) and
no incorporation (120b).
(120) a.1 Kedi çocuğun yatağında
uyudu.
cat child-Gen bed-3SP-Loc sleep-Past-3Sg
’The cat slept in the child’s bed.’
a.2 yatağında
kedi uyuyan
çocuk
bed-3SP-Loc cat sleep-Rel(wa) child
’the child whose bed cat slept in’
b.1 Ayşe çocuğun yatağında
uyudu.
child-Gen bed-3SP-Loc sleep-Past-3Sg
’Ayşe slept in the child’s bed.’
b.2 yatağ-ın-da Ayşe’nin uyu-duğ-u
çocuk
bed-3SP-Loc Ayşe-Gen sleep-Rel(ga)-3Sg child
’the child whose bed Ayşe slept in’
Gaps in relative clauses may involve dependencies which exist in nested constituents
(121a–b). Infinitival verbs and possessives produce gaps from missing noun phrase constituents and pass them to the upper clause. This gap information is nonlocal to phrase,
and projected until a verb with the relative suffix is reached. The clause headed by the
verb behaves as a modifier and gap is filled (i.e, structure-shared) by the noun phrase
at modified position.
(121) a.
Çocuğu kaybolan kadın1 çok telaşlandı.
child-3SP lost-Rel woman very panic-Past-3Sg
‘The woman whose child is lost has panicked.’
1
60
söylediğim araba1 satılmış.
b. Babama
1 beğendiğimi
father-1SP-Dat
like-Part-1Sg-Acc tell-Rel-1Sg car
sell-Pass-Past-3Sg
is
sold.’
‘The car1 that I told my father that I like
1
Such dependencies and information interaction with the other phrases over the local
phrase boundary are called non-local features by HPSG. These features are ruled by
a principle called Non-local Feature Principle [21] which is adapted from Foot
Feature Principle of GPSG. For filler-gap dependencies, a nonlocal feature called
SLASH is introduced. In English more than one gap is possible in a clause, so set type
is used for SLASH attribute. However in Turkish, a relative clause can contain only one
trace at any intermediate phrase. In case of nested relative clauses, the gap is always
filled and bound to a sister NP. Therefore in our design, SLASH attribute can be null or
of type local. When the trace (empty category) is introduced, non-local feature SLASH
is coindexed with the LOCAL feature of the gapped argument position.
(122)



PHON
SYNSEM

1
LOCAL
NONLOCAL | INHERITED | SLASH
1


Slash feature introduced by the trace is inherited to upper levels. However, in some
position, inheritance should be broken and filler should be searched. In case of Turkish,
this is the level where a relative verb is the head of the phrase. HPSG marks these
positions by dividing NONLOCAL feature into two attributes INHERITED and TO-BIND
of the same structure. TO-BIND|SLASH feature of relative verbs are marked as local (not
null) and coindexed with the INHERITED|SLASH feature. The resulting phrase becomes
a modifier of which the LOCAL feature of the modified phrase is also coindexed with the
slash, so that the filler and its trace are combined (Figure 4.3).
A lexical entry for relativized verbs is given in (123). However, some head features
such as case, relativization, possession and subcategorization are not supposed to be
the same for the filler and the trace. To handle this, selected features CONT|INDEX,
HEAD|AGR are passed to modified structure instead of LOCAL feature.
(123)

PHON







SYNSEM




‘söylediğim’


% tell-Rel-1Sg

LOCAL | CAT






NONLOCAL


obj-rel-verb
HEAD AGR


h
PERSON
NUMBER
first
sing
i
MOD | MODSYN | LOCAL | CONT | INDEX


INHERITED | SLASH
null
TO-BIND | SLASH
61
local
CONT | INDEX
1










1






S
HH
H
HH
NP
V
HH
H
satılmış
HH
HH
HH
S[rel]
NP
INHER | SLASH
TO-BIND | SLASH
NP
HH
1
1
HH
HH
HH
Babama
INHER | SLASH
H
HH
NP
LOCAL
INHER | SLASH
HH
LOCAL
araba
1
H
V
S[inf]
HH
1
H
TO-BIND | SLASH
söylediğim
1
V
1
1
beğendiğimi
Figure 4.3: Projection of the SLASH feature
The problem with the dropped pronouns also exists in relative clauses. When trace
is realized with empty category, efficiency and superfluous ambiguity problems may
arise. In our design, we used a simple technique for raising slash feature, relying on two
properties: First is valid for most of the languages. Every trace should be subcategorized
by a head. Second is the free constituent order of Turkish. Since complement order is
relatively free in Turkish, we could assume that the missing item is the last constituent.
Because any constituent may be in the last (first in the surface but retrieved last)
position in the complement list. So we have introduced the following rule to introduce
trace instead of empty category:
62


(124) 
HEAD
head
1
SYNSEM | LOCAL | CAT SUBJ



SUBCAT


DTRS
COMP-DTR
−→
2
⊕

SYNSEM | LOCAL | CAT

1
DTRS | COMP-DTRS
select
1
HEAD
SUBCAT
head
3


2
4
LOCAL
NONLOCAL | INHERITED | SLASH
,
4
,
3
,
When the argument is the last item in the SUBCAT list, it is deleted, and the trace
is introduced. This solves the ambiguity in subcategorized constituents. However yet
another problem exists with the traces which may occur in adverbs, which are not
subcategorized for. In English, prepositions subcategorize for an NP so that the trace
could be generated from subcategorization. However two case suffixes -dA (locative) and
-(y)lA (instrumental) in Turkish produce nominal adjuncts that act as VP modifiers.
When they are missing, since they have no surface form, it is impossible to introduce
them by the rule above. We introduce the trace as an empty category for these two
cases:
(125)

PHON





SYNSEM




LOCAL



1

noun
CASE

CAT | HEAD 


inst ∨ locative
MOD | MODSYN | LOCAL | CAT
NONLOCAL | INHERITED | SLASH
1






HEAD verb 

SUBCAT


The second problem is the definition of the constraints for the use of wa and ga
strategies described in (118). When the slash value is introduced in the subject position
(subject daughter or one of its daughters is missing), wa strategy is used, otherwise ga
strategy is used. These are expressed as:
Relative Clause Principle
(126) a) "
SYNSEM
1
1
"
LOCAL | CAT
HEAD
SUBCAT
LOCAL | CAT | SUBJ
NONLOCAL | TO-BIND
##
subject-relative
=⇒
NONLOCAL | INHERITED | SLASH
2
63
2
b) "
"
SYNSEM
1
1
LOCAL | CAT
HEAD
SUBCAT
LOCAL | CAT | SUBJ
##
object-relative
=⇒
NONLOCAL | INHERITED | SLASH
NONLOCAL | TO-BIND
¬
2
2
In cases where the subject NP is incorporated, the wa strategy can be used even
though the gap is not the subconstituent of the subject. Such verbs are marked with
a boolean head feature called N-INCORP standing for the noun incorporation. The
main constraint on this type of relative clause is that the type of the noun in the subject
position should be indefinite (or nonreferential) because it is incorporated. The following
additional constraint solves the problem:
(127) c)

SYNSEM

1
1
h


LOCAL | CAT 
LOCAL | CAT | SUBJ
HEAD
SUBCAT
h
i


subject-relative
N-INCORP +
=⇒
LOCAL | CAT | ADJUNCTS | DEFINITE
i
−
When a relative clause satisfying these constraints is saturated, and its TO-BIND
feature is bound to the INHERITED feature, it acts as a noun modifier. The content
index of the modified noun is coindexed with the index of the gap to bind the semantic
features of the relative clause and the filler. The rest is handled by the adjunct schema.
Nested relative clauses can modify the same noun so that multiple gaps may be bound
to the same filler (128).
(128) annemin
sevdiğim kurabiyeler1
1 yaptığı
1 çok
mother-POSS
cook-Rel
much like-Rel cookie-Plu
‘the cookies that my mother cooked, that I like’
4.7
Substantive Predicates
As mentioned in Chapter 3, Turkish sentences may have verbal, existential or substantive
heads. Substantive predicates are formed by substantive heads with auxiliary (-DH-Agr,
-mHş-Agr) or copula suffixes. Syntactically, substantive heads subcategorize for an NP
which have the same semantic index. In other words, substantive head and the subcategorized NP describe the same nominal object (129a–b). Copula and agreement suffixes
marks the agreement of the categorized NP.
64
(129) a. Ben çok hastayım.
I
much ill-Cop(1Sg)
‘I’m too sick.’
b. Bütün kadınlar
çiçektir.
every woman-Plu flower-Cop(3Sg)
‘Every woman is a flower.’
Another design consideration is to distinguish predicative NP’s from the others.
First, this is necessary to determine whether a saturated NP forms a sentence or not.
Second, the same problem with the possessive NP exists for substantive predicates. A
saturated predicative NP should not be further modified by another adjective. For
these two reasons, we have added a boolean type head feature for substantial types
called PREDICATIVE. The following is a sample entry for a predicative noun “insanım”.
(130)

PHON
“insanım”

i

+ 


% Human-Cop(1Sg),‘I am a human’


h

noun

CAT HEAD PREDICATIVE

SYNSEM | LOCAL 

NP1sg 1

SUBCAT
CONT | INDEX
1
After this feature is defined, the following constraint is added to the constraint (117)
for the adjunct rule:
(131) SYNSEM | LOCAL | CAT | HEAD | MOD | LOCAL | CAT
1
h
SUBCAT
¬
1
i
HEAD
h
noun
PREDICATIVE
+
i
=⇒
In the implementation, substantive predicates are realized by a lexical rule which
maps lexical entry for a non-predicative substantive word to substantive predicate by
an auxiliary or copula suffix:
(132) PHON 1

"
h

subst
HEAD

PREDICATIVE
SYNSEM | LOCAL CAT


2
SUBCAT
CONT | INDEX

PHON
3
apply-cop( 1 , 5 )

SYNSEM | LOCAL
"
CAT
HEAD
SUBCAT
PREDICATIVE
NP[AGR
65
5

i#

− 

+
,INDEX
3
]⊕
7−→
2

#


Where apply-cop is a general predicate applying the copula suffix corresponding to
agreement feature marked with the second argument to the first argument and returning
the resulting string.
66
Chapter 5
ALE IMPLEMENTATION
In the implementation, we have used ALE (Attribute Logic Engine)[5]. ALE is an
integrated system of definite clause logic programming and phrase structure parsing.
All operations and declarations in ALE use typed feature structures as terms. ALE is
designed and suited for implementations of unification-based language formalisms.
ALE is a strongly typed language. Every structure must have a declared type.
Types are defined by an inheritance structure and subtype relation. Basic representation
scheme used is the typed feature structures. Types are assigned to appropriate featurevalue pairs. Type structure of ALE is very similar to the HPSG including properties
like inheritance, nesting, and well-typedness. However it is most restricted in favor of
efficiency and implementation considerations.
ALE allows definition of general constraints on types. One can put restrictions on
the feature structures of a particular type. Another feature of ALE is the definite
clauses in which all functionality of PROLOG definite clauses is provided with feature
structure unification instead of simple term unification. Also complex descriptions can
be simplified by the use of macros.
One of the most distinct features of ALE from other tools like TFS and CUF [17]
is the support for phrase structure grammars. ALE provides phrase structure rules to
be coded like Definite Clause Grammars of PROLOG. It has a built-in bottom-up chart
parser in addition to feature structure unification. DCG’s are top-down and depth-first.
However ALE parser works in a combined manner asserting edges to chart right to left
while applying rules left to right. ALE also allows lexical rules for dealing with lexical
redundancy. Lexical rules can be defined for inflectional or derivational morphology
as well as zero derivations like nominalization of adjectives. Morphological constraints
(suffixation, affixation etc.) can be controlled by some built in mechanisms or PROLOG
predicates. It also allows empty categories to be integrated into grammar.
67
5.1
Grammar Rules and Principles
We have four phrase structure rules, each corresponding to a schema that we have introduced in the preceding chapter. First two (108, 109) handle the complement retrieval
and subcategorization. Applying both rules cause superfluous parses due to the application order. The third is the adjunct schema (116) which handles the adjunct-head
relation. And the fourth is the rule introducing the slash (124).
Rules are coded by standard Immediate Dominance and Linear Precedence notation.
The application of rules are governed and constrained by a set of ALE definite clauses.
These include two simple clauses modifying the DAUGHTER and PHON features of the
mother phrase. The others are constraints and basic principles.
head-feature-principle applies the Head Feature Principle of HPSG; head
feature of the mother is structure shared with the head feature of the head daughter. selectlast and selectfirst implements the Subcategorization Principle of
HPSG. The surface form of the combined SUBCAT structure of list and sets with optional arguments is generated, one item is selected, and rest is returned as the SUBCAT
feature of the mother phrase. Figure 5.1 shows the source for the simplified versions of
these two principles.
%
% head-feature-principle(MothSign,HeadSign)
%
head-feature-principle(synsem:local:cat:head:X,synsem:local:cat:head:X)
if true.
%
% subject-retrieval-rule
%
subcat_retr1 rule
(Mother, synsem:local:cat:subcat:SubcatRest) ===>
cat> (Complement,synsem:CompSyn),
cat> (Head,synsem:local:cat:subcat:Subcat),
goal> (head-feature-principle(Mother,Head),
seleclast(CompSyn,Subcat,SubcatRest)).
Figure 5.1: Sample Source for Head Feature and Subcategorization Principles
Linear precedence and word order constraints of the system are realized by the definite clauses removeop and surface. The mixed subcat structure (cf. 104) consisting
of nested sets and lists with optional arguments is converted into permutations of the
surface form by these clauses. removeop produces omitted and existed permutations of
optional arguments, and surface produces the flattened lists from the resulting structure.
68
nonlocal-principle combines the NONLOCAL features of the daughters and modifies the mother. It inherits the gap information in one of the daughters from INHERITED
feature to the mother phrase. If TO-BIND feature of the head is not null it binds the gap
and applies the constraint for the relative clauses. adjunct-principle puts the constraints in schema (116) and two constraints introduced in (117) and (131). Grammar
rules are coded in file T.rule and principles are coded in T.clause in Appendix A.
5.2
Lexicon and Lexical Rules
Lexical redundancy becomes a crucially important problem in agglutinative languages
where a large number of derivations and inflections of a root word exists. It is almost impossible to store all derivations and inflections of Turkish words into a lexical database.
Therefore some sort of morphological analysis and application of lexical rules are essential. Also application of a lexical inheritance hierarchy could be used to deal with
redundancy.
ALE does not suggest a standart mechanism for implementing lexical type hierarchies. Two methods seem to be applicable: use of macros, and type hierarchies with
general constraints. Macro definitions in ALE allow for variable substitution and —not
recursive— nesting. Each node in the lexical hierarchy can be defined by a macro which
contains the calls for the parent macros. At the lexical level, items are defined by one
or more macros. Macros are expanded at run time, and ALE operates on the expanded
descriptions. Second solution is to construct a type hieararchy under the type word
which is the type of the lexical elements, and put general constraints on these types.
However, in this approach, it is impossible to assign a lexical item to several nodes in
the type hierarchy since each item can belong to only one type. Besides, ALE has a very
restricted type mechanism so multiple inheritence is very limited. On the other hand,
constraints are evaluted at compile time which is efficient compared to macros.
In implementation, we have used a lexical type hierarchy. We defined a lexical type
tree under the type word for the lexical entries. Each common class of lexical entries are
defined as a node in the tree (Figure 5.2). We have used “ l” as the last two characters
in the names of these lexical types to distinguish them from head types.
We also defined constraints on these types (see source files T.type and T.cons in
Appendix A). After all these declarations, defining a lexical entry as one of the types
above will apply all constraints associated with the supertypes along the path to word.
ALE supports lexical rules. At the feature structure level, unification and use of user
defined ALE definite clauses are provided. At the surface level, the user can define string
operations by concatenation and user defined PROLOG predicates. Use of PROLOG at
string level enables user to implement any kind of complex morphological phenomena
like affixation, vowel harmony, drops etc. Also complex structure changes in signs can
be coded by the help of the ALE’s definite clauses.
However, ALE lexical rules are inefficient for an agglutinative language like Turkish.
First, ALE applies all rules at compile time and asserts all generated combinations as
PROLOG predicates which consumes too much memory and increases compilation time
69
word
adverb_l
noun_l
common-l
proper-l
adjective_l
pronoun_l
relativized-l
quantitative_l
quantifier_l
article_l
demonstrative_l
verb_l
qualitative_l
infinitival_l
relative_l
subj_rel_l
adverbial_l finite_l
complement_l
obj_rel_l
Figure 5.2: Lexical hierarchy
considerably. Second, it implements all kinds of string-to-string mappings to handle a
wide range of languages. However this makes it very slow for a large number of lexical
rules and lexical entries. Both inefficiencies make the developement very difficult. So we
have tried to edit some portion of ALE source code and made some changes which will
apply lexical rules at runtime and apply lexical mappings only to closed class of words.
This increased the efficiency to a reasonable level.
We have coded lexical rules for nominal casess, possesive suffixes (which mark agreement in case of sentential complements and relative clauses), and noun relativizer suffix.
Zero derivations like adjective-to-noun promotion and production of non-referential object case of verbs are implemented by lexical rules. These lexical rules are defined in
the file T.lex rule (Appendix A). For reusability considerations, similar feature structure transformations are grouped into ALE definite clauses which are defined in the
file T.clause. For example, rules for application of case suffixes are implemented in
clause apply-case/3 and lexical rules for all cases call this clause with case passed as
argument.
In the appendix A, some part of grammar code is given. It requires Quintus or SICStus Prolog. Full system can be obtained via anonymous ftp from ftp.lcsl.metu.edu.tr
in path /pub/theses/sehitoglu-ms-96.tar.gz.
70
Chapter 6
CONCLUSION
In recent years, computational studies on Turkish have proliferated. These studies are
important in two respects: First, building foundations of linguistic description of Turkish
within the light of the contemporary linguistic theories. Second, providing basic tools
for natural language processing which has applications in computer science ranging from
simple text processing utilities to translation and learning tools.
HPSG is the synthesis of the some of the recent linguistics theories. It is a developing theory, and new principles and approaches are being introduced for expanding the
universal coverage. Being one of the most powerful among the other unification based
and phrase structure formalisms, it models the language in informational perspective
and describes the linguistic events by a set of universal principles and metarules. It is
a general theory trying to be as flexible as possible to cover principles of all natural
languages.
In this study, we have worked on a computational sign-based model of Turkish, following and adapting the HPSG framework. HPSG uses feature structures to describe
linguistic phenomena. This allows the grammar designers to concentrate on the constraints imposed by a particular language on a well-defined set of linguistic features.
This is in contrast to earlier context-free grammar rules, where language-specific rules
do not allow generalizations. Postulating principles and writing constraints on these
principles show how different languages model the same phenomenon in different ways.
To this end, we have analyzed and implemented the general principles such as subcategorization, adjunct-head selection, relative extraction. We have also studied the
principles such as word-order variation, pronoun and complement drops and unbounded
dependencies, which are particularly important for Turkish.
For the time being, the parser has not been combined with a lexical analyzer and
tested on a real corpus. Since most of the syntactic information is coded in the lexicon,
an intelligent mechanism for gathering all lexical entries for Turkish should be employed.
HPSG proposes solutions like lexical inheritance hierarchy, and lexical rules. Turkish is
an agglutinative language and has many syntactically effective and productive suffixes.
This means that there is more interactions between morphology and syntax, compared
71
to a language such as English.
The computational tool we have used for HPSG, ALE, supports lexical rules with
morphological analysis. However it is inefficient for running a grammar with large lexicon
and all lexical rules. As the main problem about lexicon, ALE does all lexical processing
at compile time and generates all possible results of lexical rules statically, which is
not suitable for agglutinative languages. Also, since morphological rules are defined in
PROLOG, they are very inefficient. We made some changes to apply lexical rules at run
time and make morphology a little bit faster. However, for an efficient lexical analysis,
use of an external lexicon and morphological analyser is necessary. Necessary interface
routines can be coded into PROLOG source code of ALE as the changes we have already
done.
Another approach could be integration of syntactic and morphological analysis. This
is achived by encoding morphological analysis combined with syntactic rules in the style
of HPSG principles. This is also desirable from the linguistic point of view; morphological
and syntactic phrasing can affect each other in a principled way.1
ALE has some drawbacks as well as powerful features. The strong typing cause
description domain to be restricted. Type hierarchies requiring latice-like multiple inheritances cannot be coded efficiently. Also it disallows the usage of atomic types without
type declaration. It has a unification based description language and type inferencing
mechanism provided with definite clauses with all functionality of PROLOG. However
ALE lacks some sort of overwrite operation especially in lexical rules which are procedural in nature. Overwrite operation changes some part of a feature rather than unifying
it. Such an operator may ease the formulations and descriptions of lexical rules.
We have limited semantic analysis of signs to minimum. Since HPSG is a complete
linguistic theory for both syntax and semantics, for a complete HPSG analysis of Turkish,
semantic principles and model should be analyzed.
1 for
an integrated analysis of morphology and syntax cf. [3]
72
REFERENCES
[1] Tahsin Banguoğlu. Türkçenin Grameri. Türk Dil Kurumu, 1986.
[2] Chris Barker, Jorge Hankamer, and John Moore. Wa and ga in turkish. In K. Dziwirek, P. Farrell, and E. Mejias-Bikandi, editors, Grammatical Relations. 1990.
[3] Cem Bozşahin and Elvan Göçmen. A categorial framework for composition in
multiple linguistic domains. In Proceedings of the Fourth International Conference
on Cognitive Science of NLP (CSNLP’95), Dublin, Ireland, July 1995.
[4] Joan Bresnan and R. Kaplan. The Mental Representation of Grammatical Relations,
chapter Lexical-Functional Grammar, pages 173–281. MIT Press, 1982.
[5] Bob Carpenter and Gerald Penn. The Attribute Logic Engine User’s Guide, Version
2.0. Carnegie Mellon University, Pittsburgh, August 1994.
[6] Jochen Dörre and Micheal Dorna. CUF: A Formalism for Linguistic Knowledge
Representation. Dyana, August 1993.
[7] E. E. Erguvanlı. The Function of Word Order in Turkish Grammar. PhD thesis,
UCLA, 1979.
[8] G. Gazdar, E. Klein, G. K. Pullum, and I. A. Sag. Generalized Phrase Structure
Grammar. Harvard University Press, 1985.
[9] Elvan Göçmen and Onur T. Şehitoğlu and Cem Bozşahin. An outline of turkish syntax. Technical Report 95–2, METU Department of Computer Engineering,
Ankara, Turkey, March 1995.
[10] Zelal Güngördü. A lexical-functional grammar for turkish. Master’s thesis, Dept.
of Comp. Eng., Bilkent University, July 1993.
[11] Zelal Güngördü and Kemal Oflazer. Parsing turkish using the lexical-functional
grammar. In Proceedings of COLING’94, The 15th Conference on Computational
Linguistics, Kyoto, Japan, August 1994.
73
[12] Jorge Hankamer and Laura Knecht. The role of the subject/non-subject distinction
in determining the choice of relative clause participle in turkish. In Harvard Syntax
and Semantics 2. Cambridge, 1976.
[13] Beril Hoffman. The Computational Analysis of the Syntax and Interpretation of
‘Free’ Word Order in Turkish. PhD thesis, University of Pennsylvania, 1995.
[14] Ray Jackendoff. X-syntax: A Study of Phrase Structure. MIT Press, Cambridge,
1977.
[15] Jonas Kuhn. Encoding HPSG grammars in TFS. Technical report, Institut für
Maschinelle Sprachverarbeitung, Universität Stuttgart, Germany, March 1993.
[16] Geoffrey L. Lewis. Turkish Grammar. Oxford University Press., Oxford, UK, 1967.
[17] Suresh Manandhar. CUF in context. Technical report, Human Comm. Research
Centre, University of Edinburgh.
[18] Robert H. Meskill. A Transformational Analysis of Turkish Syntax. Mouton, 1970.
[19] Teruko Mitamura, Hiroyuki Muska, and Marion Kee. The Generalized LR
Parser/Compiler Version 8.1: User’s Guide. Carnegie Mellon, April 1988.
[20] Carl Polard and Ivan A. Sag. Information Based Syntax and Semantics. CSLI,
1987.
[21] Carl Polard and Ivan A. Sag.
Chicago, 1994.
Head-driven Phrase Stucture Grammar.
CSLI
[22] W. Rounds and R. Kasper. A complete logical calculus for record structures representing linguistic information. In Proceedings of the IEEE Symposium on Logic
in Computer Science, June 1986.
[23] G. J. Schaaik. Studies in Turkish Grammar. PhD thesis, University of Amsterdam,
February 1995.
[24] Peter Sells. Lectures on contemporary Syntactic Theories. CSLI, 1985.
[25] Stuart M. Shieber. An Introduction to Unification Based Approaches to Grammar.
CSLI, 1986.
[26] Rasim Şimşek. Örneklerle Türkçe Sözdizimi. Kuzey Matbaacılık, 1987.
[27] Robert Underhill. Turkish Grammar. MIT Press, Cambridge, 1975.
74
Appendix A
PARSER SOURCE
A.1
Type Definitions
%=======================================================================
%
% @(#)T.type
Rev:1.6
1/7/96
%
%=======================================================================
% Type definitions
%=======================================================================
bot sub [bool,sign,null_synsem,cat,head,case,null_agr,per,num,posses,
list,char,set,tense,aux_tense,nonloc,null_adjstr,null_mod,
list_or_set_subcat,subcat_or_ne_set,subcat_or_ne_list,
psoa_arg,qfpsoa,sem_det,sem_obj,null_local,arg_type,
subcat,conx,null_local,nonlocal,daughters].
sign sub [lexical,phrase]
intro [phon:list_string,
synsem:synsem,
qstore:set_quant,
qretr:list_quant].
lexical sub [word].
word sub [noun_l,adj_l,adv_l,verb_l].
noun_l sub [common_l,proper_l,pronoun_l,relativized_l].
common_l sub [].
proper_l sub [].
pronoun_l sub [].
relativized_l sub [].
adj_l sub [quantif_l,article_l,demonstra_l,quantitive_l,
qualitative_l].
quantif_l sub [].
article_l sub [].
demonstra_l sub [].
75
quantitive_l sub [].
qualitative_l sub [].
adv_l sub [].
verb_l sub [finite_l,sadv_l,inf_l].
finite_l sub [].
sadv_l sub [].
inf_l sub [relcl_l,complement_l].
relcl_l sub [subj_rel_l,obj_rel_l].
subj_rel_l sub [].
obj_rel_l sub [].
complement_l sub [].
phrase sub []
intro [ dtrs: daughters ].
null_synsem sub [null,synsem].
null sub [].
synsem sub []
intro [local:local,
nonlocal:nonlocal].
null_local sub [null,local].
local sub []
intro [cat:cat,cont:sem_obj,conx:conx].
conx sub [].
nonlocal sub []
intro [inherited:nonloc,tobind:nonloc].
nonloc sub []
intro [slash:null_local].
daughters sub [hd_subj_st,hd_st]
intro [hd_dtr: sign,
comp_dtrs: list_sign,
spec_dtrs: list_sign].
hd_subj_st sub []
intro [subj_dtr:sign].
hd_st sub [].
bool sub [plus,minus].
plus sub [].
minus sub [].
cat sub []
intro [head:head,
subj:null_synsem,
adjuncts:null_adjstr,
subcat:list_or_set_subcat].
head sub [subst,prep,adverb,verb]
intro [mod:null_mod].
subst sub [adj,noun]
intro [pred:bool].
adj sub [determiner,quantitative_adj,qualitative_adj,questional_adj]
intro [countable:bool,gradable:bool].
determiner
sub [article,demonstrative_adj,quantifier].
article
sub [].
demonstrative_adj
sub [].
quantifier
sub [].
quantitative_adj
sub [number,distributive_adj,grouping_adj].
number
sub [cardinal,fractional].
cardinal
sub [].
76
fractional
distributive_adj
grouping_adj
qualitative_adj
questional_adj
sub [].
sub [].
sub [].
sub [].
sub [].
noun sub [common,proper_noun,pronoun]
intro [case:case,
agr:agr,
n_ind:agr,
rel:bool,
poss:posses].
common sub [].
proper_noun sub [].
pronoun sub [personal_pr,demonstrative_pr,reflexive_pr,indefinite_pr
,quantificational_pr,questional_pr].
personal_pr
sub [].
demonstrative_pr
sub [].
reflexive_pr
sub [].
indefinite_pr
sub [].
quantificational_pr
sub [].
questional_pr
sub [].
prep sub [].
adverb sub [dir_adv,dir_adv,temp_adv,manr_adv,quant_adv,sent_adv,quest_adv].
dir_adv sub []
intro [dir:direction].
temp_adv sub [t_unit_adv,pot_adv,t_per_adv].
t_unit_adv sub [].
pot_adv sub [].
t_per_adv sub [dayt,dayw,seas].
dayt sub [].
dayw sub [].
seas sub [].
manr_adv sub [qual_adv,rep_adv].
qual_adv sub [].
rep_adv sub [].
quant_adv sub [approx,comp,superl,excess].
approx sub [].
comp sub [].
superl sub [].
excess sub [].
sent_adv sub [].
quest_adv sub [].
verb sub [infinitival,adverbial,finite]
intro [tense:tense,neg:bool,vagr:null_agr,n_inc:bool].
infinitival sub [relative,complementary]
intro [vcase:case].
relative sub [subj_rel,obj_rel].
subj_rel sub [].
obj_rel sub [].
complementary sub [mak,ış,complemented].
mak sub [].
ış sub [].
complemented sub [].
adverbial sub [].
finite sub []
intro [aux_tense:aux_tense].
case sub [nom,obj,gen,loc,direction,ins].
nom sub [].
obj sub [].
gen sub [].
loc sub [].
77
direction sub [dat,abl].
dat sub [].
abl sub [].
ins sub [].
null_agr sub [null,agr].
agr sub []
intro [per:per,
num:num].
per sub [first,second,third].
first sub [].
second sub [].
third sub [].
num sub [sing,plur].
sing sub [].
plur sub [].
posses sub
none sub
poss sub
intro
[none,poss].
[].
[]
[by:agr].
null_adjstr sub [null,adjstr].
adjstr sub []
intro [qtfd:bool,dmstrtd:bool,rltvzd:bool,rltclsd:bool,qntfcd:bool,
qltfd:bool,non_ref:bool].
null_mod sub [null,mod].
mod sub []
intro [modsyn:synsem,modadj:null_adjstr].
tense sub [base,future,contin,pres,past,rep_past].
base sub [].
future sub [].
contin sub [].
pres sub [].
past sub [].
rep_past sub [].
aux_tense sub [null,hikaye,rivayet,condition].
hikaye sub [].
rivayet sub [].
condition sub [].
psoa_arg sub []
intro [argname: string,arg: arg_type].
arg_type sub [agr,psoa].
qfpsoa sub [property, relation]
intro [name:string].
property sub []
intro [inst:agr].
relation sub []
intro [args:list_psoa_arg].
sem_det sub [forall,exists,the].
forall sub [].
exists sub [].
the sub [].
sem_obj sub [nom_obj, psoa, quant].
nom_obj sub [npro, pron]
intro [index:agr,
78
restr:set_psoa].
npro sub [].
pron sub [ana, ppro].
ana sub [recp, refl].
recp sub [].
refl sub [].
ppro sub [].
quant sub []
intro [det:sem_det,
restind:npro].
psoa sub []
intro [quants:list_quant,nucleus:qfpsoa].
subcat sub [optionalcat,subcat_type].
optionalcat sub [opt,obl]
intro [s_arg:subcat_type].
opt sub [].
obl sub [].
subcat_type sub [char,synsem,sign].
list_or_set_subcat sub [set_subcat,list_subcat,list_xxx].
subcat_or_ne_set sub [subcat,ne_set_subcat].
subcat_or_ne_list sub [subcat,ne_list_subcat].
list sub [e_list,ne_list,list_cat,string,list_string,list_sign,
list_quant,list_xxx,list_psoa_arg].
e_list sub [].
ne_list sub [ne_list_cat,ne_string,ne_list_string,
ne_list_xxx,ne_list_sign,ne_list_quant,ne_list_psoa_arg]
intro [hd:bot,
tl:list].
ne_list_xxx sub [ne_list_subcat,ne_list_synsem].
list_cat sub [e_list,ne_list_cat].
ne_list_cat sub []
intro [hd:cat,
tl:list_cat].
string sub [e_list,ne_string].
ne_string sub []
intro [hd:char,
tl:string].
list_xxx sub [list_subcat,list_synsem,ne_list_xxx].
list_subcat sub [e_list,ne_list_subcat].
ne_list_subcat sub []
intro [hd: subcat_or_ne_set,
tl: list_subcat].
list_synsem sub [e_list,ne_list_synsem].
ne_list_synsem sub []
intro [hd:synsem,
tl:list_synsem].
list_string sub [e_list,ne_list_string].
ne_list_string sub []
intro [hd:string,
tl:list_string].
list_sign sub [e_list,ne_list_sign].
ne_list_sign sub []
intro [hd:sign,
tl:list_sign].
list_quant sub [e_list,ne_list_quant].
ne_list_quant sub []
intro [hd:quant,
tl:list_quant].
list_psoa_arg sub [e_list,ne_list_psoa_arg].
79
ne_list_psoa_arg sub []
intro [hd:psoa_arg,
tl:list_psoa_arg].
set sub [e_list,ne_set,set_char,set_subcat,set_psoa,set_quant].
ne_set sub [ne_set_char,ne_set_subcat,ne_set_psoa,ne_set_quant]
intro [el:bot,
els:set].
set_char sub [e_list,ne_set_char].
ne_set_char sub []
intro [el:char,
els:set_char].
set_subcat sub [e_list,ne_set_subcat].
ne_set_subcat sub []
intro [el: subcat_or_ne_list,
els: set_subcat].
set_psoa sub [e_list,ne_set_psoa].
ne_set_psoa sub []
intro [el: psoa,
els: set_psoa].
set_quant sub [e_list,ne_set_quant].
ne_set_quant sub []
intro [el:quant,
els: set_quant].
char sub [a,b,c,ç,d,e,f,g,ğ,h,ı,i,j,k,l,m,n,o,ö,p,q,r,s,ş,t,u,ü,v,w,x,y,z,].
a sub [].
........
.......
A.2
Phrase Structure Rules
%===============================================================
%
% @(#)T.rule
Rev:1.6
1/7/96
%
%===============================================================
% Grammar Rules
%===============================================================
subcat_retr1 rule
(Mother,phrase,phon:PhonMot,
synsem:local:cat:(subcat:SubMot,adjuncts:Adjs,subj:Subj),
dtrs:DtrsMot)
===>
cat> (Arg,phon:PhonArg,synsem:SynArg),
cat> (Head,phon:PhonHead,synsem:local:cat:(subcat:SubHead,adjuncts:Adjs,
subj:Subj)),
goal> (append(PhonArg,PhonHead,PhonMot),
head_feature_principle(Mother,Head),
sselectlast(SynArg,SubHead,SubMot),
combine_semantics(Head,Arg,Mother),
append_comp(DtrsMot,Head,Arg),
nonlocal_principle(Arg,Head,Mother)).
subcat_retr2 rule
(Mother,phrase,phon:PhonMot,
80
synsem:local:cat:(subcat:SubMot,adjuncts:Adjs,subj:Subj),
dtrs:DtrsMot)
===>
cat> (Head,phon:PhonHead,synsem:local:cat:(head:verb,
subcat:SubHead,adjuncts:Adjs,
subj:Subj)),
cat> (Arg,phon:PhonArg,synsem:SynArg),
goal> (append(PhonArg,PhonHead,PhonMot),
head_feature_principle(Mother,Head),
sselectlast(SynArg,SubHead,SubMot),
combine_semantics(Head,Arg,Mother),
append_comp(DtrsMot,Head,Arg),
nonlocal_principle(Arg,Head,Mother)).
adj_head rule
(Mother,phrase,phon:PhonMot,dtrs:DtrsMot)
===>
cat> (Adjunct,phon:PhonAdj),
cat> (Head,phon:PhonHead),
goal> (append(PhonAdj,PhonHead,PhonMot),
combine_semantics(Head,Adjunct,Mother),
head_feature_principle(Mother,Head),
adjunct_principle(Mother,Adjunct,Head),
append_spec(DtrsMot,Head,Adjunct),
nonlocal_principle(Adjunct,Head,Mother)).
slash rule
(Mother,phrase,phon:PhonMot,
synsem:(local:cat:(subcat:SubMot,subj:Subj),
nonlocal:(inherited:slash:Local,
tobind:slash:HT)),
dtrs:DtrsMot)
===>
cat>(Head,phon:PhonHead,synsem:(local:cat:(head:(noun;infinitival),
subcat:SubHead,
subj:Subj),
nonlocal:(inherited:slash:null,
tobind:slash:HT))),
goal> (append((PhonSl,[e_list]),PhonHead,PhonMot),
head_feature_principle(Mother,Head),
nonlocal_principle(synsem:Slsynsem,Head,Mother),
sselectlast((Slsynsem,local:(Local,
cat:head:(agr:per:third,n_ind:SlInd),
cont:index:SlInd),
nonlocal:(inherited:slash:Local,
tobind:slash:null)),SubHead,(SubMot,e_list)),
append_comp(DtrsMot,Head,(Slash,phon:PhonSl,qretr:e_list,qstore:e_list,
synsem:Slsynsem)),
combine_semantics(Head,synsem:local:cont:(index:SlInd,restr:e_list)
,Mother)).
A.3
Constraints and Macros
%===============================================================
%
% @(#)T.cons
Rev:1.8
1/7/96
%
%===============================================================
81
%
Constraints
%===============================================================
determiner cons (gradable: minus).
article
cons (countable: plus).
quantitative_adj cons (gradable: minus,countable: plus).
word cons (qretr:e_list,synsem:nonlocal:inherited:slash:null).
subj_rel cons (tense:base,vagr:null).
%obj_rel cons (tense:(future;past)).
mak cons (tense:base).
ış cons (tense:base,vagr:agr).
complemented cons (tense:(future;past),vagr:agr).
noun_l cons (synsem:local:cat:head:(n_ind:I,[agr] == [n_ind])).
common_l cons (synsem:local:(cat:(head:n_ind:I,
adjuncts:(qtfd:minus,
dmstrtd:minus,
rltvzd:minus,
rltclsd:minus,
qntfcd:minus,
qltfd:minus,
non_ref:plus)),
cont:index:I)).
pronoun_l cons (synsem:local:(cat:(head:n_ind:I,
adjuncts:(
qtfd:minus,
dmstrtd:minus,
rltvzd:minus,
rltclsd:minus,
qntfcd:minus,
qltfd:minus,
non_ref:minus)),
cont:index:I)).
quantif_l cons
(synsem:local:cat:head:(quantifier,
mod:(modsyn:(local:cat:(head:(common),
adjuncts:(qtfd:minus,
dmstrtd:minus,
rltvzd:minus,
rltclsd:A,
qntfcd:B,
qltfd:C))),
modadj:(qtfd:plus,
dmstrtd:minus,
rltvzd:minus,
rltclsd:A,
qntfcd:B,
qltfd:C,non_ref:minus)))).
demonstra_l cons
(synsem:local:cat:head:(demonstrative_adj,
mod:(modsyn:(local:cat:(head:(common),
adjuncts:(qtfd:minus,
dmstrtd:minus,
rltvzd:minus,
rltclsd:A,
qntfcd:B,
qltfd:C))),
modadj:(qtfd:minus,
dmstrtd:plus,
rltvzd:minus,
82
rltclsd:A,
qntfcd:B,
qltfd:C,
non_ref:minus)))).
qualitative_l cons
(synsem:local:cat:head:(qualitative_adj,
mod:(modsyn:(local:cat:(head:(common),
adjuncts:(qtfd:A,
dmstrtd:minus,
rltvzd:minus,
rltclsd:minus,
qntfcd:B,
non_ref:C))),
modadj:(qtfd:A,
dmstrtd:minus,
rltvzd:minus,
rltclsd:minus,
qntfcd:B,
qltfd:plus,
non_ref:C)))).
relativized_l cons
(synsem:local:cat:head:( mod:(modsyn:(local:cat:(head:(common),
adjuncts:(qtfd:A,
dmstrtd:B,
rltclsd:minus,
qntfcd:D,
qltfd:E)
)),
modadj:(qtfd:A,
dmstrtd:B,
rltvzd:plus,
rltclsd:minus,
qntfcd:D,
qltfd:E,
non_ref:minus)))).
subj_rel_l cons
(synsem:(local:(cat:(head:(subj_rel,
mod:(modsyn:(local:(cat:(head:(common,
n_ind:NInd),
adjuncts:(
qtfd:A,
dmstrtd:B,
rltvzd:minus,
qntfcd:D,
qltfd:E)
),
cont:(Cont,index:Ind))),
modadj:(qtfd:A,
dmstrtd:B,
rltvzd:minus,
rltclsd:plus,
qntfcd:D,
qltfd:E,
non_ref:minus)))),
cont:_),%index:Ind),
nonlocal:tobind:slash:(cat:head:(common,n_ind:NInd
)%,
% cont:Cont
))).
obj_rel_l cons
(synsem:(local:(cat:(head:(obj_rel,
mod:(modsyn:(local:(cat:(head:(common,
83
n_ind:NInd),
adjuncts:(
qtfd:A,
dmstrtd:B,
rltvzd:minus,
qntfcd:D,
qltfd:E)),
cont:Cont)),
modadj:(qtfd:A,
dmstrtd:B,
rltvzd:minus,
rltclsd:plus,
qntfcd:D,
qltfd:E,
non_ref:minus)))),
cont:index:NInd),
nonlocal:tobind:slash:(cat:head:(common,
n_ind:NInd),
cont:Cont))).
finite_l cons
(synsem:(local:(cat:head:finite),
nonlocal:tobind:slash:null)).
%===============================================================
%
% @(#)T.macro
Rev:1.5
1/7/96
%
%===============================================================
%
Macros
%===============================================================
common_noun macro
(common_l,
synsem:(local:(cat:(head:(common,
case:nom,
agr:(num:sing,
per:third),
mod:null,
n_ind:NInd,
pred:minus,
rel:minus,
poss:none),
subcat:e_list,
subj:null),
cont:(Cont,index:NInd)),
nonlocal:(inherited:slash:null,
tobind:slash:null)
),
qstore:e_list
).
opt(X) macro
(opt,s_arg:X).
obl(X) macro
(obl,s_arg:X).
np(Head,Ind) macro
(local:(cat:(head:(Head,noun,mod:null,rel:minus,pred:minus),
subcat:e_list),
cont:index:Ind),
nonlocal:( tobind:slash:null)
).
vp(Head,Cont) macro
84
(local:(cat:(head:(Head,mod:null),
subcat:e_list),
cont:Cont),
nonlocal:(tobind:slash:null)
).
slashinh(X) macro
(synsem:nonlocal:inherited:slash:X).
slashtob(X) macro
(synsem:nonlocal:tobind:slash:X).
f_phrase macro
(phrase,
synsem:local:cat:subcat:e_list,synsem:nonlocal:inherited:slash:null,
synsem:nonlocal:tobind:slash:null).
f_sent macro
(@f_phrase,synsem:local:cat:head:(finite;pred:plus)).
A.4
Definite Clauses
%===============================================================
%
% @(#)T.clause
Rev:1.9
1/7/96
%
%===============================================================
% Principles & Clauses
%===============================================================
%-------% Head Feature Principle:
%
Head Feature of the mother is token identical to head feature of
%
the Head.
head_feature_principle(synsem:local:cat:head:X,synsem:local:cat:head:X) if
true.
%-------% Combine Semantics: ( Head,Dtr,Mother)
%
combine_semantics( synsem:local:cont:(index:HInd,restr:Hrest),
synsem:local:cont:(index:DInd,restr:Drest),
synsem:local:cont:(index:HInd,restr:MRest)) if
appendset(Hrest,Drest,MRest).
combine_semantics( synsem:local:cont:(index:HInd,restr:Hrest),
synsem:local:cont:(Drest,psoa),
synsem:local:cont:(index:HInd,restr:MRest)) if
appendset(Hrest,(el:Drest,els:[]),MRest).
combine_semantics( synsem:local:cont:(nucleus:HNuc,quants:HQ),
synsem:local:cont:(DCont),
synsem:local:cont:(nucleus:HNuc,
quants:([(det:the,restind:DCont) |HQ]
true.
))) if
combine_semantics( synsem:local:cont:(nucleus:HNuc,quants:HQ),
synsem:local:cont:(psoa),
synsem:local:cont:(nucleus:HNuc,quants:HQ)) if true.
85
%--------% adjunct_principle(mother,adjunct,head)
%
adjunct_principle((synsem:local:cat:(subj:Subj,subcat:Subcat,adjuncts:MAdjs)),
(synsem:local:cat:(head:mod:(modsyn:(Mod),
modadj:MAdjs),
subcat:[])),
synsem:(Mod,local:cat:(CatH,subj:Subj,subcat:Subcat))) if
checkposs(Mod),
checksubst(CatH).
checkposs(local:cat:head: =\= noun) if true,!.
checkposs(local:cat:head:poss:none) if true.
checkposs(local:cat:(head:(poss:poss,pred:minus),subcat:ne_set)) if true.
checkposs(local:cat:(head:(poss:poss,pred:minus),subcat:ne_list)) if true.
checkposs(local:cat:(head:(poss:poss,pred:plus),subcat:tl:ne_set)) if true.
checkposs(local:cat:(head:(poss:poss,pred:plus),subcat:tl:ne_list)) if true.
checksubst(head: =\= noun) if true,!.
checksubst(head:(pred:minus)) if true.
checksubst((head:(pred:plus),subcat:ne_set)) if true.
checksubst((head:(pred:plus),subcat:ne_list)) if true.
%--------% nonlocal_principle
%
nonlocal_principle((@slashtob((local,LAdj)),@slashinh((LAdj))),
(@slashtob((HT,null)),@slashinh(null)),
(@slashtob(HT),@slashinh(null))) if true.
nonlocal_principle((@slashinh(null),@slashtob(null)),
(@slashinh(HI),@slashtob(HT)),
(@slashinh(HI),@slashtob(HT))) if true,!.
nonlocal_principle((@slashinh(null),@slashtob(null)),
(@slashtob((HT)),@slashinh(null)),
(@slashinh(null),@slashtob(HT))) if true.
nonlocal_principle((@slashinh(AI),@slashtob(null)),
(@slashtob((null)),@slashinh(null)),
(@slashinh(AI),@slashtob(HT))) if true.
nonlocal_principle((@slashinh(HT),Arg),
(@slashtob((HT,local)),Head),
(Mother,@slashtob(HT),@slashinh(HT))) if
check_rel(Arg,Mother).
check_rel((@slashinh(S)),
(synsem:(local:cat:(head:subj_rel,
subj:nonlocal:inherited:slash:S),
nonlocal:tobind:slash:S))) if true.
check_rel((@slashinh(S), synsem:local: =\= S),
(synsem:(local:(cat:(head:subj_rel,n_inc:plus,
subj:local:cat:adjuncts:non_ref:plus)),
nonlocal:tobind:slash:S))) if true.
check_rel((@slashinh(S)),
(synsem:(local:cat:(head:obj_rel,
subj:nonlocal:inherited:slash: =\= S),
nonlocal:tobind:slash:S))) if true.
86
%===============================================================
%
Clauses
%===============================================================
append([],Xs,Xs) if
true.
append([H|T1],L2,[H|T2]) if
append(T1,L2,T2).
appends(e_list,X,X) if true,!.
appends([A],(X,ne_set),[A|[X]]) if true,!.
appends(X,[],X) if true,!.
appends((X,set),(L,list),[X|L]) if true.
appends([X|Rx],Y,[X|Res]) if
appends(Rx,Y,Res).
appends((el:X,els:Rx),(Y,set),(el:X,els:Res)) if
appends(Rx,Y,Res).
listlast(X,(hd:X,tl:e_list),e_list) if true,!.
listlast(X,[H|T],[H|R]) if
listlast(X,T,R).
permut(e_list,e_list) if true , !.
permut((X,set),(X,set)) if true.
permut((el:X,els:R),(el:Y,els:(el:X,els:R2))) if
permut(R,(el:Y,els:R2)).
selectlast(Arg,(ne_list_synsem,Sursub),Reslist) if
listlast(Arg,Sursub,Reslist),!.
selectlast(Arg,Sub,Reslist) if
removeop(Sub,SubRem),
surface(SubRem,Sursub),
listlast(Arg,Sursub,Reslist).
selectfirst(Arg,(ne_list_synsem,[Arg|Rest]),Rest) if !,true.
selectfirst(Arg,Sub,Reslist) if
removeop(Sub,SubRem),
surface(SubRem,[Arg|Reslist]).
select(Arg,(s_arg:Arg),[]) if true.
select(Arg,(Arg,subcat_type),[]) if true.
select(T,(X,set),Z) if
permut(X,(el:Ct,els:Cr)),
select(T,Ct,Res),
appends(Res,Cr,Z).
select(T,[Xt|R],Z) if
select(T,Xt,Res),
appends(Res,R,Z).
appendset(e_list,(set,X),X) if true.
appendset((el:El,els:Rels),(set,S2),(el:El,els:Res)) if
appendset(Rels,S2,Res).
removeop([],[]) if true.
removeop(s_arg:X,@obl(X)) if true.
removeop([(opt)|R],R2) if removeop(R,R2).
removeop([H|R],[H2|R2]) if removeop(H,H2),removeop(R,R2).
removeop((el:(opt),els:R),R2) if removeop(R,R2).
removeop((el:H,els:R),(el:H2,els:R2)) if removeop(H,H2),removeop(R,R2).
87
surface(e_list,e_list) if true.
surface(X,[T|Rs]) if
select(T,X,R),surface(R,Rs).
append_comp((hd_dtr:Head,comp_dtrs:[],spec_dtrs:[],subj_dtr:Comp),
(Head,word,(synsem:local:cat:subj:Subj)),
(Comp,synsem:Subj)) if true,!.
append_comp((subj_dtr:Comp,comp_dtrs:Cdtrs,spec_dtrs:Sdtrs,hd_dtr:Hdtr),
(synsem:local:cat:subj:Subj,
dtrs:(comp_dtrs:Cdtrs,spec_dtrs:Sdtrs,hd_dtr:Hdtr)),
(Comp,synsem:Subj)) if true,!.
append_comp((hd_dtr:Head,comp_dtrs:[Comp],spec_dtrs:[]),
(Head,word), Comp) if true,!.
append_comp((hd_dtr:Hdtr,comp_dtrs:ResComp,spec_dtrs:Sdtrs),
(dtrs:(comp_dtrs:Comp1,hd_dtr:Hdtr,spec_dtrs:Sdtrs)),
Comp) if
append([Comp],Comp1,ResComp).
append_spec((hd_dtr:Head,comp_dtrs:[],spec_dtrs:[Adjunct]),
(Head,word), Adjunct) if true,!.
append_spec((hd_dtr:Hdtr,comp_dtrs:Cdtrs,subj_dtr:SUdtr,spec_dtrs:ResSpec),
dtrs:(hd_dtr:Hdtr,comp_dtrs:Cdtrs,subj_dtr:SUdtr,spec_dtrs:Spec1),
Spec) if
append([Spec],Spec1,ResSpec).
%=====================================================================
% Clauses applying the constraints of the Lexical Rules....
%=====================================================================
apply_case(
(word,
synsem:(local:(cat:(head:(common,
case:nom,
agr:(num:Num,per:third),
mod:Mod,
rel:(Rel,minus),
pred:(Pred,minus),
n_ind:NInd,
poss:Poss),
subcat:Subcat,
adjuncts:Adjuncts,
subj:Subj),
cont:Cont,
conx:Conx),
nonlocal:Nonlocal),
qstore:Qs) ,
(word,
synsem:(local:(cat:(head:(common,
case:Case,
agr:(num:Num,per:third),
mod:Mod,
rel:Rel,
pred:Pred,
n_ind:NInd,
poss:Poss),
subj:Subj,
adjuncts:Adjuncts,
subcat:Subcat),
cont:Cont,
conx:Conx),
88
nonlocal:Nonlocal),
qstore:Qs) , Case , CMod ) if check_case_mod(Poss,CMod).
check_case_mod(none,a) if true.
check_case_mod(by:per:first,a) if true.
check_case_mod(by:per:second,a) if true.
check_case_mod(by:per:third,b) if true.
apply_poss(
(word,
synsem:(local:(cat:(head:(common,
case:nom,
agr:(Agr,per:third),
mod:Mod,
rel:(Rel,minus),
pred:(Pred,minus),
poss:none),
subcat:Subcat),
cont:(index:Ind,restr:(el:EL)),
conx:Conx),
nonlocal:Nonlocal),
qstore:Qs) ,
(word,
synsem:(local:(cat:(head:(common,
case:nom,
agr:Agr,
mod:Mod,
rel:Rel,
pred:Pred,
n_ind:Ind,
poss:by:By),
subj:Subj,
subcat:Subcat2),
cont:(index:Ind,restr:(el:EL,els:(el:(nucleus:(relation,
name:[p,o,s,s,e,s,s],
args:
[(argname:[o,w,n,e,r],arg:Ind2),
(argname:[o,w,n,e,d],arg:Ind)]),
quants:e_list),els:e_list))),
conx:Conx),
nonlocal:Nonlocal),
qstore:Qs) , By ) if
appends([@obl((Subj,@np((case:gen,agr:By),Ind2)))],Subcat,Subcat2).
%--------------------------------------------------------------------apply_copula(
(word,
synsem:(local:(cat:(head:(common,
case:nom,
rel:(Rel,minus),
pred:minus,
poss:Poss),
subj:Subj,
subcat:Subcat),
cont:(C1,restr:Restr),
conx:Conx),
nonlocal:Nonlocal),
qstore:Qs) ,
(noun_l,
synsem:(local:(cat:(head:(common,
case:nom,
agr:(Agr),
mod:null,
89
rel:Rel,
pred:plus,
n_ind:Ind,
poss:Poss),
subj:Subj2,
subcat:Subcat2),
cont:C2,%(index:Ind,restr:Restr),
conx:Conx),
nonlocal:Nonlocal),
qstore:Qs), Agr) if
contentcop(C1,C2,Ind),
appends([@obl((Subj2,@np((case:nom,agr:Agr),_)))],Subcat,Subcat2).
%-----------------------------------------------------------------------apply_adj2noun(
(word,
synsem:(local:(cat:(head:((qualitative_adj;rel:plus),
mod:(modsyn:Mod,modadj:Modadj)),
subj:Subj,
subcat:Subcat),
cont:(index:Ind,Cont),
conx:Conx),
nonlocal:Nonlocal),
qstore:Qs),
(noun_l,
synsem:(Mod,local:(cat:(head:(mod:null,
case:nom,
agr:(per:third,num:sing),
rel:minus,
pred:minus,
n_ind:Ind,
poss:none),
subj:Subj,
subcat:Subcat),
cont:Cont,
conx:Conx),
nonlocal:Nonlocal),
qstore:Qs)) if true.
A.5
Lexicon
%===============================================================
%
% @(#)T.lex
Rev:1.10
1/7/96
%
%===============================================================
% Lexicon
%===============================================================
kırmızı --->
(qualitative_l,phon:[[k,ı,r,m,ı,z,ı]],
synsem:(local:(cat:(head:(countable:plus,gradable:plus,
mod:modsyn:(local:(cat:head:n_ind:NInd,
cont:(index:Ind)))),
subcat:[],
subj:null),
90
cont:(index:Ind,
restr:(el:(quants:e_list,
nucleus:(name:[r,e,d],inst:NInd)),
els:e_list)
)
),
nonlocal:tobind:slash:null)
).
ben --->
(word,phon:[[b,e,n]],
synsem:(local:(cat:(head:(personal_pr,
case:nom,
rel:minus,
agr:(num:sing,
per:first),
mod:null,
poss:none),
subcat:e_list,
subj:null),
cont:(npro,
index:(Ind,per:first,num:sing),
restr:e_list),
conx:conx
),
nonlocal:tobind:slash:null)
).
kapı --->
( @common_noun,
phon:[[k,a,p,ı]],
synsem:local:cont:(npro,
index:(agr,Ind,per:third,num:sing),
restr:(el:(nucleus:(name:[d,o,o,r],inst:Ind),
quants:[]),els:[]))
).
ev --->
(@common_noun,
phon:[[e,v]],
synsem:local:cont:(npro,
index:(agr,Ind,per:third,num:sing),
restr:(el:(nucleus:(name:[h,o,u,s,e],inst:Ind),
quants:[]),els:[])
)
).
gitti --->
(finite_l,
phon:[[g,i,t,t,i]],
synsem:local:(cat:(head:(finite,mod:null,
tense:past,
vagr:(Agr,(per:third,num:sing))),
subcat: @obl((Subj,@np((agr:Agr,case:nom),SInd))),
@opt(@np(case:abl,FInd)),
@opt(@np(case:dat,TInd)),
subj:Subj
),
cont:(quants:[],nucleus:(name:[g,o],args:[(argname:[g,o,e,r],arg:SInd),
(argname:[t,o],arg:TInd),
(argname:[f,r,o,m],arg:FInd)])
91
))).
giden --->
(subj_rel_l,
phon:[[g,i,d,e,n]],
synsem:(local:(cat:(head:(subj_rel,
vcase:nom),
subcat: @obl((Subj,@np(case:nom,SInd))),
@opt(@np(case:abl,FInd)),
@opt(@np(case:dat,TInd)),
subj:Subj
),
cont:restr:(el:(quants:[],nucleus:(name:[g,o],
args:[(argname:[g,o,e,r],arg:SInd),
(argname:[t,o],arg:TInd),
(argname:[f,r,o,m],arg:FInd)])
),els:[]))
)).
geldiği --->
(obj_rel_l,
phon:[[g,e,l,d,i,ğ,i]],
synsem:(local:(cat:(head:(obj_rel,
tense:past,
vagr:(Agr,(per:third,num:sing))),
subcat: @obl((Subj,@np((agr:Agr,case:gen),SInd))),
@opt(@np(case:abl,FInd)),
@opt(@np(case:dat,TInd)),
subj:Subj
),
cont:restr:(el:(quants:[],nucleus:(name:[g,o],
args:[(argname:[c,o,m,e,r],arg:SInd),
(argname:[t,o],arg:TInd),
(argname:[f,r,o,m],arg:FInd)])
),els:[]))
)).
söylüyor --->
(finite_l,
phon:[[s,ö,y,l,ü,y,o,r]],
synsem:(local:(cat:(head:(finite,
tense:past,
mod:null,
vagr:(Agr,(per:third,num:sing))),
subcat: @obl((Subj,@np((agr:Agr,case:nom),SInd))),
@obl(@vp((vcase:obj,complemented),Spsoa)),
@opt(@np(case:dat,TInd)),
subj:Subj
),
cont:(quants:[],nucleus:(name:[t,e,l,l],
args:[(argname:[t,e,l,l,e,r],arg:SInd),
(argname:[t,o],arg:TInd),
(argname:[w,h,a,t],arg:Spsoa)])
))
)).
empty
(word,phon:[[p,r,o]],
synsem:(local:(cat:(head:(pronoun,
case:(nom;gen),
agr:Agr,
mod:null,
n_ind:Agr,
pred:minus,
92
rel:minus,
poss:none),
subcat:e_list,
subj:null),
cont:(npro,
index:(Agr),
restr:[])),
nonlocal:(inherited:slash:null,
tobind:slash:null)
)).
A.6
Lexical Rules
%===============================================================
%
% @(#)T.lex_rule
Rev:1.9
1/7/96
%
%===============================================================
% Lexical Rules
%-% Morphological clauses used in lexical rules...
%--back(a).
back(ı).
back(o).
back(u).
kalin_hece([X]) :- back(X),!.
kalin_hece([X,_]) :- back(X).
front(e).
front(i).
front(ö).
front(ü).
ince_hece([X]) :- front(X),!.
ince_hece([X,_]) :- front(X).
wovel(X) :- front(X),!.
wovel(X) :- back(X).
backrounded(o).
backrounded(u).
b_r_hece([X]) :- backrounded(X),!.
b_r_hece([X,_]) :- backrounded(X).
frontrounded(ö).
frontrounded(ü).
f_r_hece([X]) :- frontrounded(X),!.
f_r_hece([X,_]) :- frontrounded(X).
backunrounded(a).
backunrounded(ı).
b_u_hece([X]) :- backunrounded(X),!.
b_u_hece([X,_]) :- backunrounded(X).
frontunrounded(e).
frontunrounded(i).
f_u_hece([X]) :- frontunrounded(X),!.
f_u_hece([X,_]) :- frontunrounded(X).
yumusa(p,b).
93
yumusa(ç,c).
yumusa(t,d).
yumusa(k,ğ).
kal_yum([X,Y],Yum) :- back(X),yumusa(Y,Yum).
ince_yum([X,Y],Yum) :- front(X),yumusa(Y,Yum).
f_u_yum([X,Y],Yum)
b_u_yum([X,Y],Yum)
f_r_yum([X,Y],Yum)
b_r_yum([X,Y],Yum)
::::-
frontunrounded(X),yumusa(Y,Yum).
backunrounded(X),yumusa(Y,Yum).
frontrounded(X),yumusa(Y,Yum).
backrounded(X),yumusa(Y,Yum).
%===============================================================
% Lexical rules...
%===============================================================
:-lex_rule_depth(4).
plural lex_rule
Cat1 **> Cat2
if apply_plural((Cat1,phon:[Phon]),(Cat2,phon:[Phon2])),
append(Phon,[,p,l,u],Phon2)
morphs
(X,L2) becomes (X,L2,lar) when kalin_hece(L2),
(X,L2) becomes (X,L2,ler) when ince_hece(L2).
%-accusative_a lex_rule
Cat1 **> Cat2
if apply_case((Cat1,phon:[Phon]),(Cat2,phon:[Phon2]),obj,a),
append(Phon,[,o,b,j],Phon2)
morphs
(X,[L]) becomes (X,[L],yı) when backunrounded(L),
(X,[L]) becomes (X,[L],yi) when frontunrounded(L),
(X,[L]) becomes (X,[L],yu) when backrounded(L),
(X,[L]) becomes (X,[L],yü) when frontrounded(L),
(X,[L1,L2]) becomes (X,L1,[Y],[ı]) when b_u_yum([L1,L2],Y),
(X,[L1,L2]) becomes (X,L1,[Y],[i]) when f_u_yum([L1,L2],Y),
(X,[L1,L2]) becomes (X,L1,[Y],[u]) when b_r_yum([L1,L2],Y),
(X,[L1,L2]) becomes (X,L1,[Y],[ü]) when f_r_yum([L1,L2],Y),
(X,L2) becomes (X,L2,[ı]) when b_u_hece(L2),
(X,L2) becomes (X,L2,i) when f_u_hece(L2),
(X,L2) becomes (X,L2,u) when b_r_hece(L2),
(X,L2) becomes (X,L2,[ü]) when f_r_hece(L2).
accusative_b lex_rule
Cat1 **> Cat2
if apply_case((Cat1,phon:[Phon]),(Cat2,phon:[Phon2]),obj,b),
append(Phon,[,o,b,j],Phon2)
morphs
(X,[L]) becomes (X,[L],nı) when backunrounded(L),
(X,[L]) becomes (X,[L],ni) when frontunrounded(L),
(X,[L]) becomes (X,[L],nu) when backrounded(L),
(X,[L]) becomes (X,[L],nü) when frontrounded(L).
%-possessive_3_s lex_rule
Cat1 **> Cat2
if
apply_poss((Cat1,phon:[Phon]),(Cat2,phon:[Phon2]),(num:sing,per:third)),
append(Phon,[,t,s,g],Phon2)
morphs
94
(X,[L]) becomes (X,[L],sı) when backunrounded(L),
(X,[L]) becomes (X,[L],si) when frontunrounded(L),
(X,[L]) becomes (X,[L],su) when backrounded(L),
(X,[L]) becomes (X,[L],sü) when frontrounded(L),
(X,[L1,L2]) becomes (X,L1,[Y],[ı]) when b_u_yum([L1,L2],Y),
(X,[L1,L2]) becomes (X,L1,[Y],[i]) when f_u_yum([L1,L2],Y),
(X,[L1,L2]) becomes (X,L1,[Y],[u]) when b_r_yum([L1,L2],Y),
(X,[L1,L2]) becomes (X,L1,[Y],[ü]) when f_r_yum([L1,L2],Y),
(X,L2) becomes (X,L2,[ı]) when b_u_hece(L2),
(X,L2) becomes (X,L2,i) when f_u_hece(L2),
(X,L2) becomes (X,L2,u) when b_r_hece(L2),
(X,L2) becomes (X,L2,[ü]) when f_r_hece(L2).
%-obj_rel_to_compl lex_rule
(obj_rel_l,
phon:[Phon],
synsem:(local:(cat:(head:(obj_rel,
vcase:nom,
neg:Neg,
vagr:Agr,
n_inc:N_Inc,
tense:Tense
),
subj:Subj,
adjuncts:Adj,
subcat:Subcat),
cont:restr:el:Rest,
conx:Conx))
) **>
(complement_l,
phon:[Phon],
synsem:(local:(cat:(head:(complemented,
vcase:nom,
neg:Neg,
mod:null,
vagr:Agr,
n_inc:N_Inc,
tense:Tense
),
subj:Subj,
adjuncts:Adj,
subcat:Subcat),
cont:Rest,
conx:Conx),
nonlocal:tobind:slash:null)
)
morphs
X becomes X.
%-relativizer lex_rule
Cat1 **> Cat2
if apply_reltvzr((Cat1,phon:[Phon]),(Cat2,phon:[Phon2])),
append(Phon,[,r,l,t],Phon2)
morphs
(X) becomes (X,ki).
%-copula1_s lex_rule
Cat1 **> Cat2
if apply_copula((Cat1,phon:[Phon]),(Cat2,phon:[Phon2]),(num:sing,per:first)),
95
append(Phon,[,c,o,p],Phon2)
morphs
(X,[L2]) becomes (X,[L2],yım) when backunrounded(L2),
(X,[L2]) becomes (X,[L2],yim) when frontunrounded(L2),
(X,[L2]) becomes (X,[L2],yum) when backrounded(L2),
(X,[L2]) becomes (X,[L2],yüm) when frontrounded(L2),
(X,[L1,L2]) becomes (X,L1,[Y],[ı,m]) when b_u_yum([L1,L2],Y),
(X,[L1,L2]) becomes (X,L1,[Y],[i,m]) when f_u_yum([L1,L2],Y),
(X,[L1,L2]) becomes (X,L1,[Y],um) when b_r_yum([L1,L2],Y),
(X,[L1,L2]) becomes (X,L1,[Y],[ü,m]) when f_r_yum([L1,L2],Y),
(X,L2) becomes (X,L2,[ı],m) when b_u_hece(L2),
(X,L2) becomes (X,L2,im) when f_u_hece(L2),
(X,L2) becomes (X,L2,um) when b_r_hece(L2),
(X,L2) becomes (X,L2,[ü],m) when f_r_hece(L2).
%-copula3_s lex_rule
Cat1 **> Cat2
if apply_copula((Cat1,phon:[Phon]),(Cat2,phon:[Phon2]),(num:sing,per:third)),
append(Phon,[,c,o,p],Phon2)
morphs
(X,[L1,L2]) becomes (X,L1,[L2],tır) when b_u_yum([L1,L2],Y),
(X,[L1,L2]) becomes (X,L1,[L2],tir) when f_u_yum([L1,L2],Y),
(X,[L1,L2]) becomes (X,L1,[L2],tur) when b_r_yum([L1,L2],Y),
(X,[L1,L2]) becomes (X,L1,[L2],tür) when f_r_yum([L1,L2],Y),
(X,L2) becomes (X,L2,dır) when b_u_hece(L2),
(X,L2) becomes (X,L2,dir) when f_u_hece(L2),
(X,L2) becomes (X,L2,dur) when b_r_hece(L2),
(X,L2) becomes (X,L2,dür) when f_r_hece(L2).
adj_promotion lex_rule
Cat1 **> Cat2
if apply_adj2noun((Cat1,phon:[Phon]),(Cat2,phon:[Phon2])),
append(Phon,[,a],Phon2)
morphs
X becomes X.
%--non_ref_object lex_rule
(verb_l,
phon:[PhonV],
synsem:(local:(cat:(head:HeadV,
subj:SubjV,
adjuncts:AdjV,
subcat:Subcat1V),
cont:ContV,
conx:ConxV),
nonlocal:NonlocalV)
) **>
(verb_l,
phon:[PhonV],
synsem:(local:(cat:(head:HeadV,
subj:SubjV,
adjuncts:AdjV,
subcat:[ SubcatRV,
@obl((local:(cat:(head:(common,
case:nom,
96
agr:AgrN,
mod:ModN,
rel:(RelN),
pred:(PredN),
n_ind:NIndN,
poss:PossN),
subcat:SubcatN,
adjuncts:(AdjunctsN,
non_ref:plus),
subj:SubjN),
cont:ContN,
conx:ConxN),
nonlocal:NonlocalN))]
),
cont:ContV,
conx:ConxV),
nonlocal:NonlocalV)
)
if selectlast(
(local:(cat:(head:(common,
case:obj,
agr:AgrN,
mod:ModN,
rel:(RelN),
pred:(PredN),
n_ind:NIndN,
poss:PossN),
subcat:SubcatN,
adjuncts:AdjunctsN,
subj:SubjN),
cont:ContN,
conx:ConxN),
nonlocal:NonlocalN), Subcat1V, SubcatRV)
morphs
X becomes X.
97
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