Construction-specific properties of syntactic subjects in Icelandic and German JÓHANNA BARÐDAL* Abstract This paper discusses the syntactic similarities and di¤erences in the behavior of subject-like obliques in the Obl–V–(XP) construction in Icelandic and German. Research on this construction so far has suggested that the subject-like oblique behaves as a syntactic subject in Icelandic, but as an object in German. Data from German are presented which show that the subject-like oblique in fact passes almost all the subject tests, with some restrictions. The di¤erences between Icelandic and German are therefore much smaller, and the similarities much greater, than predicted by analyzing them as subjects in Icelandic and objects in German. A comparison between Icelandic and German further reveals that the subject criteria cannot be applied across two as closely related languages as Icelandic and German, and they cannot be consistently applied even within the same language. Therefore, grammatical relations like ‘‘subject’’ and ‘‘object’’ should be regarded, not as universal, not as language-speciﬁc, but as CONSTRUCTIONSPECIFIC relations. It is shown that the di¤erence between Icelandic and German resides in Obl–V–(XP) predicates being reluctant to occur in ‘‘elliptic’’ constructions in German, whereas their occurrences in such constructions in Icelandic are less restricted. This correlates with di¤erences in the frequency of Obl–V–(XP) predicates in the two languages, suggesting that the construction exists at di¤erent levels of schematicity in Icelandic and German. This is expected on a usage-based account in which frequency is taken to be an important determinant of the language system. Keywords: Icelandic; German; oblique subjects; argument structure; subject tests; syntactic relations vs. syntactic roles; Radical Construction Grammar; frequency; schematicity. Cognitive Linguistics 17–1 (2006), 39–106 DOI 10.1515/COG.2006.002 0936–5907/06/0017–0039 6 Walter de Gruyter 40 J. Barðdal 1. Introduction This article focuses on the Obl–V–(XP) construction in Icelandic (1a) and German (1b), highlighting some of the problems caused by the assumption that a uniform/universal category ‘‘subject’’ exists. In contrast, this article will argue for a Radical Construction Grammar approach (cf. Croft 2001) in which the concept of global grammatical relations is systematically abandoned. Instead, it is assumed that categories like subjects and objects are not only language-speciﬁc, but even construction-speciﬁc categories. (1) a. b. Icelandic Mér er kalt. me.dat is cold ‘I’m cold.’ German Mir ist kalt. me.dat is cold ‘I’m cold.’ For Icelandic (Sigur¶sson 1989; Zaenen et al. 1985) and German (Reis 1982; Zaenen et al. 1985) together, the following thirteen subject criteria have been used: ﬁrst position in declarative clauses, subject-verb inversion, ﬁrst position in subordinate clauses, ‘‘subject-to-object raising,’’ ‘‘subject-to-subject raising,’’ long distance reﬂexivization, clause-bound reﬂexivization, control inﬁnitives, conjunction reduction, nominative case, verb agreement, deletion in imperatives and deletion in telegraphic style (see Table 2 in Section 3.1 below). The ﬁrst nine tests have been assumed for Icelandic, while the last seven have been used for German. Only three tests are common for both languages. Because of this di¤erence in which constructions have been assigned the status of being subject criteria in Icelandic and German, the subject-like obliques in (1) above are analyzed as objects in German, while their Icelandic equivalents are analyzed as subjects. This has led to a view of Icelandic and German as typologically widely distinct languages, with a subsequent downgrading of the similarities between them. These empirical facts give rise to three problems: – – Subjects do not exhibit the same syntactic behavior in closely related languages, a problem for theories that posit a universal category ‘‘subject.’’ A closer survey of the German Obl–V–(XP) construction reveals that subject-like obliques actually pass most of the syntactic tests that have been regarded as subject criteria in German. Therefore, the subject criteria cannot be consistently applied even within the same language. Construction-speciﬁc properties of syntactic subjects – 41 Using di¤erent constructions in di¤erent languages to deﬁne subjects is theoretically inconsistent and is an example of methodological opportunism (Croft 2001: 30–32, 41–44). The best way to adequately account for the empirical facts presented here is to abandon the concept of universal grammatical relations. This is the approach taken by Radical Construction Grammar, where constructions are assumed to be the basic units of language and the behavior of each argument is speciﬁed for the construction it is a part of. On such an account categories like subjects and objects are derived from the construction as a whole, and thus represent part–whole relations and not part– part relations (cf. Section 5 and Kay 1997 on the di¤erence between the two). Following Croft (2001), traditional syntactic relations will be referred to with lower-case letters (subject) in the remainder of this article, while syntactic roles will be referred to with upper-case letters (Subject). I begin by giving my deﬁnition of the Obl–V–(XP) construction in Icelandic and German in Section 2, where I also show that the construction has the same structure and semantics in both languages. In Section 3 I give a systematic overview of the subject criteria in Icelandic and German in order to explain why the same criteria have not been applied in both languages. I also provide examples of the construction in Icelandic and German in order to illustrate how the subject-like oblique behaves with regard to these criteria. I show that the di¤erence between Icelandic and German is smaller than that assumed in the prevalent literature, and certainly smaller than predicted by the analysis that subject-like obliques are subjects in Icelandic but objects in German. The data relevant for this argument involve reﬂexivization, conjunction reduction and control constructions, including a systematic questionnaire survey of native speakers’ judgments on documented control inﬁnitives in German (Section 4). After discussing the theoretical problems brought about by the empirical ﬁndings presented here I turn to an outline of a Radical Construction Grammar solution to the problem (Section 5), and how it instigates di¤erent kinds of research questions, hitherto unexplored (Section 6). The ﬁndings of the present article show that the main di¤erence between Icelandic and German is found in ‘‘elliptic’’ constructions, in that subject-like obliques show a reluctance to be left unexpressed in German, whereas their Icelandic counterparts do not show this reluctance to the same degree. This is expected on a usage-based account, a model in which frequency is assumed to be one of the main determinants of the language system (see Section 6). Thus, I suggest that the status of the Obl–V–(XP) construction in German is such that it takes an intermediate position between Icelandic, where the construction is still frequent and psychologi- 42 J. Barðdal cally real in the minds of speakers, and English, where only a couple of lexicalized relics of the construction still exists. 2. The Obl–V–(XP) construction I assume that the basic unit of language is the construction, i.e., a formmeaning (or a form-function) pairing (cf. Bar¶dal 1999, 2001a, 2001c; Bar¶dal and Molnár 2003; Fillmore et al. 1988, Goldberg 1995; Kay and Fillmore 1999; Michaelis and Ruppenhofer 2001, and many more). More speciﬁcally, I assume that all form-meaning pairings are constructions of their own, and that a grammar of language consists of constructions and a network deﬁning the relations between them. On such an account, all linguistic units classify as constructions, from the smallest morpheme to complex argument structure and sentence type constructions. This is laid out in Table 1 below (Croft and Cruse 2004: 255): Table 1. The syntax-lexicon continuum Construction type Traditional name Examples Complex and (mostly) schematic Complex, substantive verb Complex and (mostly) substantive Complex but bound Atomic and schematic Atomic and substantive syntax subcategorization frame idiom morphology syntactic category word/lexicon [sbj be-tns V-en by obl] [sbj consume obj] [kick-TNS the bucket] [noun-s], [verb-tns] [dem], [adj] [this], [ green] In other words, all linguistic form can be paired up with a meaning/ function, sometimes a meaning/function of its own and sometimes a meaning/function that is shared across forms. Constructions can, moreover, be divided into two subtypes (cf. Bar¶dal 2001a: 62–63; Bar¶dal and Molnár 2003: 234–235; Croft and Cruse 2004: 253–254; Tomasello 1998: 481–482): (i) more general constructions, of which the meaning of the whole is derivable from the meaning of the parts, for example the Ditransitive construction, and (ii) more specific constructions, in which case the meaning of the whole is not a sum of the meaning of the parts, but idiosyncratic, like the What is X doing Y-construction, as discussed by Kay and Fillmore (1999). By Obl–V–(XP) construction I refer to the constructions in Icelandic and German in which the ﬁrst or leftmost argument of the argument structure [Obl–(XP)] is not in the nominative but is found in accusative, dative or, more rarely, in the genitive case. I subsume the three under the label oblique, meaning ‘‘non-nominative.’’ This means that the Construction-speciﬁc properties of syntactic subjects 43 construction has three subconstructions, namely the Obl dat –V–(XP) construction, the Oblacc –V–(XP) and the Obl gen V–(XP) construction. The discussion in this article is conﬁned to the two ﬁrst, as the Obl gen – V–(XP) construction is very rare in Icelandic and obsolete in German. Thus, the construction is always lexically ﬁlled with a subject-like dative or accusative in both languages, and a predicate which can be a simple verb (see example  below), an adjectival predicate ( above), a compositional predicate (2a), and others. There is also a variation as to the number of arguments which can occur in the construction. In addition to the subject-like oblique, it can be ﬁlled with another direct argument (example  below), a PP (3), a subordinate clause (4), or the subject-like oblique is the only argument as in the examples in (1) above: (2) a. b. (3) a. b. (4) a. b. Icelandic Mér fellur hann me.dat falls he.nom ge¶. liking German Mir gefällt der me.dat ge-falls the.nom ‘I like that guy Christian.’ to my liking.’ Kristján mjög Christian.nom very vel well ı́ in Christian sehr gut.1 Christian very well or: ‘That guy Christian is very much Icelandic Mér bý¶ ur vi¶ essum óhreinu neglum. me.dat disgusts with these dirty nails German Mir ekelt vor den schmutzigen Fingernägeln. me.dat disgusts for the dirty ﬁnger-nails ‘I feel disgusted by the(se) dirty ﬁnger nails.’ Icelandic Mig grunar a¶ etta bo¶i ekki gott. me.acc suspects that this bodes not good ‘I suspect that this does not bode anything good.’ German Mir schwant dass es nichts Gutes bringt. me.dat suspects that it nothing good brings ‘I suspect that nothing good will come out of this.’ Verbs selecting for dative objects also instantiate the Obl–V–(XP) construction when they are passivized, as shown with the verbs mótmæla and widersprechen ‘contradict’ in (5) below: 44 J. Barðdal (5) a. b. Icelandic Honum var mótmælt. him.dat was contradicted German Ihm wurde widersprochen. him.dat be(came) contradicted ‘He was contradicted.’ These two verbs occur in the Nom–V–Dat argument structure construction in the active form in both Icelandic and German but in the Obl dat – V–(XP) construction in the passive form (see Bar¶dal and Molnár 2003 for a construction-based analysis of the passive in the North Germanic languages and more generally for an aspectual account of diathesis). Instances of mótmæla and widersprechen ‘contradict’ in the active form with the dative object topicalized to ﬁrst position are naturally not included in my deﬁnition of the Obl–V–(XP) construction as they are topicalization of the Nom–V–Dat construction, yielding Dat–V–Nom surface order. Thus, topicalizations of the Nom–V–Dat argument structure construction are distinct from the Dat–V–Nom argument structure construction in both Icelandic and German.2 The Obl–V–(XP) construction is documented in all the Germanic daughter languages, including Gothic, Old English, Old High German, Old Dutch, Old Swedish, Old Danish and Old Norse-Icelandic, although it has gone lost in all the modern languages except for Modern Icelandic, Faroese and German (cf. Bar¶dal 2001b: 196–206, 2004; Eythórsson and Bar¶dal 2005 and the references therein). In both Icelandic and German, the Obl dat –V–(XP) subconstruction is much more common than the Oblacc –V–(XP) subconstruction (cf. Bar¶dal 2004: 109–110, 120–124). When counting both adjectival and verbal predicates, including di¤erent lexical entries of the same verbal stems, Obl dat –V–(XP) predicates amount to 700 in Icelandic, while Oblacc –V–(XP) predicates are approximately 200. In German, however, the distinction between the Obl dat –V– (XP) and the Oblacc –V–(XP) construction has been neutralized because of a massive exchange of predicates between the two constructions (Seefranz-Montag 1983: 162). Therefore, by subsuming the two into one in German, it seems that existing Obl–V–(XP) predicates are approximately 80–100 (Bar¶dal 2004: 137). Moreover, Icelandic has around 750 passive Obl dat –V–(XP) predicates (excluding ditransitives), while German exhibits around 100 (Bar¶dal 2002: 90; Maling 2002: 2). The Obl–V–(XP) construction is lexically ﬁlled in Icelandic and German with either cognate predicates as in (1–2) above, or with synonymous verbs as in (3–5). A semantic analysis of the active Obl–V–(XP) Construction-speciﬁc properties of syntactic subjects 45 predicates in Icelandic and German reveals the following thirteen classes of verbs, of which the ﬁrst eleven classes are common to both languages (Bar¶dal 2004): 1. Verbs of emotion: e.g., ‘feel good/bad’ Ic. lı́¶a vel/illa, G. gut/ schlecht sein; ‘be sorry’ Ic. ykja leitt, G. verdrießen; ‘regret’ Ic. vera eftirsjá, G. reuen, . . . 2. Verbs of cognition: e.g., ‘suspect’ Ic. bjó¶a ı́ grun, G. schwanen; ‘have in mind’ Ic. vera efst ı́ huga, G. vorschweben; ‘get an idea’ Ic. detta ı́ hug, G. einfallen, eine Idee kommen; ‘remember’ Ic. vera ı́ fersku minni; ‘understand’ Ic. skiljast, . . . 3. Verbs of perception: e.g., ‘taste’ Ic. smakkast, G. schmecken, munden; ‘appear in vision’ Ic. birtast; ‘appear’ Ic. vir¶ast, G. vorkommen, . . . 4. Verbs expressing idiosyncratic attitudes, e.g., ‘be indi¤erent’ Ic. vera sama, G. egal/gleichgültig sein; ‘be OK for sb.’ G. recht sein; ‘be (im)possible for sb.’ Ic. vera (ekki) au¶i¶, . . . 5. Verbs denoting bodily states: e.g., ‘feel pain’ Ic. blæ¶a, G. weh tun; ‘feel queasy’ Ic. vera óglatt, G. übel sein; ‘freeze’ Ic. vera kalt, G. frieren, frösteln, . . . 6. Verbs denoting changes in bodily states: e.g., ‘start to freeze’ Ic. ver¶a kalt, G. kalt werden; ‘get better/worse (from an illness)’ Ic. batna/versna, . . . 7. Verbs denoting personal properties and innate tendencies: e.g., ‘be natural for sb.’ Ic. vera e¶lislægt, G. angeboren sein; ‘be typical for sb.’ G. eigen sein; ‘have a loud voice’ Ic. liggja hátt rómur, . . . 8. Verbs of gain: e.g., ‘beneﬁt’ Ic. græ¶ast, G. frommen; ‘be o¤ered sth’ Ic. bjó¶ast; ‘receive’ Ic. berast, . . . 9. Verbs of success and/or performance: e.g., ‘succeed’ Ic. heppnast/ lánast, G. gelingen/glücken; ‘do well/badly’ Ic. ganga vel/illa, G. gut/schlecht gehen; ‘make progress’ Ic. fara fram, . . . 10. Verbs of failure or mistake: e.g., ‘fail’ Ic. misheppast, G. missglücken; ‘overlook’ Ic. yﬁrsjást; ‘get hindrance’ Ic. seinka, . . . 11. Verbs of decline: e.g., ‘deteriorate’ Ic. hnigna; ‘subside’ Ic. aﬂétta; ‘fall thick on the ground’ Ic. snjóa, rigna, G. schneien, regnen, . . . 12. Verbs of ontological existence: e.g., ‘be in a particular manner’ Ic. vera hátta¶, . . . 13. Verbs of social interaction: e.g., ‘be friends’ Ic. vera vel til vina; ‘not get along’ Ic. lenda saman, . . . The last two classes of verbs of existence and social interaction are the smallest classes in Icelandic, with only a few predicates in each, whereas there are innumerous examples of, for instance, verbs of emotion and 46 J. Barðdal attitudes (Bar¶dal 2004). The category of Obl–V–(XP) predicates in Icelandic and German is a radial category (cf. Lako¤ 1987) with experiencebased predicates being the largest class, and the other classes either occupying adjacent regions in semantic space, or being related to the prototypical class via shared (partial) semantics, metaphorical extension or pragmatic inference (for the details of this analysis and a more elaborate discussion, the reader is referred to Bar¶dal 2004). There is thus a clear family resemblance between the subclasses of Obl–V–(XP) predicates in both Icelandic and German. This can be represented in a simpliﬁed graphic version as in Figures 1 and 2, for Icelandic and German, respectively. Observe that the semantic map for German is a miniature of the map for Icelandic. In other words, the Obl–V–(XP) construction in German is Figure 1. A semantic map of the Obl dat –V–(XP) construction in Icelandic Figure 2. A semantic map of the Obl–V–(XP) construction in German Construction-speciﬁc properties of syntactic subjects 47 a proper subset of the construction in Icelandic. This is of course expected given the common historical origin of the construction in both languages, and its considerably lower type frequency in German. To summarize, the Obl–V–(XP) construction in Icelandic and German di¤ers from the ordinary default argument structure construction in the two languages in that the subject-like argument is not case marked in the nominative but in dative or accusative. The type frequency of the construction is considerably higher in Icelandic than in German, both in the active and the passive diathesis. The category of active predicates instantiating the construction shows a radial structure, with experience-based predicates being central while the other subclasses are either extensions of the prototype or of other less central subclasses. Hence, the Obl–V– (XP) construction has, more or less, the same structure and semantics in both languages. This raises the question whether the syntactic behavior of subject-like obliques in the Obl–V–(XP) argument structure construction is also the same in Icelandic and German, or whether there are profound di¤erences between them, supporting the assumption that the two languages are as typologically di¤erent as the literature claims. Thus, I now proceed to a comparison of the assumed subject tests in these languages and the syntactic behavior of the subject-like oblique with regard to these tests. 3. The subject criteria As stated in Section 1 above, of the thirteen most widely discussed subject criteria in Icelandic and German only three have been unanimously assumed as valid in both languages, which in turn has given rise to a major typological distinction between Icelandic and German in the syntactic literature. In this section I discuss the remaining ten criteria and why they have been refuted in each language. Table 2 below gives an overview of all the subject criteria and their alleged (non-)applicability in Icelandic and German. I ﬁrst review the criteria that have been accepted for Icelandic but not for German (Subsection 3.1) and then the criteria accepted for German but not Icelandic (Subsection 3.2). The three criteria that have been taken to be common for both languages are discussed in Section 4 below, and new data against the traditional analysis of German are presented. 3.1. The Icelandic subject criteria The ﬁrst four tests, ﬁrst position in declarative clauses, subject-verb inversion, ﬁrst position in subordinate clauses, and ‘‘subject-to-object raising,’’ 48 J. Barðdal Table 2. Subject criteria in Icelandic and German 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. Icelandic German H H H H H H H H H * * * * * * H H H H H H H First position in declarative clauses Subject-verb inversion First position in subordinate clauses Subject-to-object raising Subject-to-subject raising Long distance reﬂexivization Clause-bound reﬂexivization Conjunction reduction Control inﬁnitives Nominative case Verb agreement Deletion in telegraphic style Deletion in imperatives * * * * have not been used as subject tests in German. The reason is that German word order is sensitive to information structure, allowing for various scrambling alternations (cf. Reis 1982: 191) which do not exist in Icelandic. In addition, OV word order in clauses containing an auxiliary or a ‘‘raising-to-object’’ verb in German obscures the picture even further: (6) a. b. (7) a. b. Icelandic Ég hef keypt bókina. I.nom have bought book-the.acc German Ich habe das Buch gekauft. I.nom have the.acc book bought ‘I(’ve) bought the book.’ Icelandic Bókina hef book-the.acc have German Das Buch habe the.acc book have ég keypt. I.nom bought ich gekauft. I.nom bought The example in (6a) shows that the nominative subject precedes the ﬁnite verb whereas the accusative object follows the non-ﬁnite verb in an Icelandic active declarative clause containing an auxiliary. Example (6b) illustrates that the equivalent German active declarative clause has the accusative object between the two verbs (OV word order). When the accusative object is topicalized to ﬁrst position, the nominative subject inverts with the verb and occurs immediately following the ﬁnite verb. As a Construction-speciﬁc properties of syntactic subjects 49 consequence, it is impossible to distinguish between the neutral Active declarative construction (6b) and the Topicalization construction (7b) in German since the distribution of the two arguments is the same relative to the verb. This is not a problem in Icelandic since the Topicalization construction has the accusative object preceding the verb cluster and the nominative subject between the two verbs (7a), while in the neutral Active declarative construction the two arguments are placed on either side of the verb cluster (7a). This is laid out in the following schema: (8) (9) Icelandic a. X–Aux–V–X b. X–Aux–X–V Active declarative construction Topicalization construction German a. X–Aux–X–V b. X–Aux–X–V Active declarative construction Topicalization construction Therefore, on a positional approach where subjects and objects are assumed to occupy a certain position in the structure, a comparison between the various word order constructions, like the Active declarative and the Topicalization construction, has not been considered fruitful for a language like German. However, it is possible to distinguish between the two constructions in German with other means than word order. First, the function of the two constructions is not the same; they di¤er with regard to which argument is the most discourse-prominent argument. In the ordinary Active declarative clause, it is the subject which is the most discourse-prominent argument, whereas in the Topicalization construction, it is the object. Second, the intonation structure of the two constructions is di¤erent in that the topicalized clause-initial accusative object typically bears nuclear stress, while clause-initial nominative subjects in ordinary active declarative clauses do not. In contrast, inverted pronominal subjects, as in (7b) can only be stressed if they are contrastive, while nuclear stress on the object in (6b) constitutes neutral sentence intonation. Therefore, given an analysis where constructions are recognized as independent linguistic objects with their own distinctive meaning/function and intonation pattern, it also becomes possible to identify the arguments as either being in their canonical subject vs. object positions, or being topicalized vs. inverted. With regard to the Obl–V–(XP) construction in German, it has been acknowledged, at least since Lenerz (1977: 113–116), that they represent neutral word order in German. In other words, they display the intonation pattern of ordinary Active declarative clauses with neutral information structure, but do not show up with the intonation typical of 50 J. Barðdal Topicalization constructions.3 In this regard, subject-like obliques in German behave as ordinary nominative subjects. Turning now to the fourth criterion in Table 2 above, ‘‘subject-toobject raising,’’ this criterion is not considered a subject test by Reis (1982: 192–193), presumably because of German OV word order. The term ‘‘raising-to-object’’ is here used about inﬁnitive clauses where the subject of the lower non-ﬁnite verb behaves syntactically as the object of the matrix verb. Although German does not have the prototypical believe-class of ‘‘raising-to-object’’ verbs, it still has raising verbs of the causative type, such as lassen ‘let’ (cf. Hoberg 1981: 79–81) which is cognate to the Icelandic ‘‘raising-to-object’’ verb láta ‘let.’ Both of these select for an inﬁnitive clause, with the nominative subject selected by the lower verb showing up in the accusative case, assigned by the matrix verb (see Haspelmath 2001: 70 and Wunderlich to appear for more ‘‘raisingto-object’’ verbs with Obl–V–(XP) constructions in German and a discussion thereof ): (10) (11) German Ich lasse ihn eine I.nom let him.acc a.acc ‘I make him eat a newspaper.’ Tageszeitung essen. newspaper eat.inf Icelandic Ég læt hana bor¶a dagbla¶. I.nom let her.acc eat.inf newspaper.acc ‘I’ll have her eat a newspaper.’ In (10–11) the nominative subject of the lower verb ‘eat’ shows up in the accusative case when embedded under the matrix verb ‘let’ in both Icelandic and German. With regard to the Obl–V–(XP) construction, however, it is a well-known fact of Icelandic that, as in the passive, the subject-like oblique of the Obl–V–(XP) argument structure construction maintains its case marking in this sentence-type construction. This is shown in (12) for the Obl–V–(XP) predicates detta ı́ hug ‘get an idea’ and lı́¶a vel ‘feel good’: (12) Icelandic a. Þú lætur ér alltaf detta eitthva¶ you.nom let yourself.dat always fall.inf something hug. mind ‘You always get new ideas.’ b. Láttu ér lı́¶a vel! let-you you.dat feel.inf well ‘Make sure to feel good!’ or: ‘Take care of yourself.’ ı́ in Construction-speciﬁc properties of syntactic subjects 51 As the examples in (13) show, the subject-like oblique of the Obl–V–(XP) construction in German also maintains its dative case with ‘let’, exactly as in Icelandic. The German Obl–V–(XP) predicates, einfallen ‘get an idea’ and gut gehen ‘do well,’ serve as examples: (13) German a. Du lässt dir immer etwas Neues you.nom let yourself.dat always something new einfallen. come-in-mind.inf ‘You’re always getting new ideas.’ b. Lass dir gut gehen! let yourself.dat well go.inf ‘Make sure to do well!’ or: ‘Take care of yourself.’ (www.mtb-news.de/forum/p64625.html, 2003) However, not only can subject-like obliques of the Obl–V–(XP) construction occur in between the two verbs in ‘‘subject-to-object raising’’ constructions in German, but ordinary dative objects of transitive verbs also occur there with ‘let’, as illustrated in (14) below. A comparison of the German examples in (13) and (14) thus shows that distributional properties with ‘let’ do not distinguish between subjects and objects of the embedded verb, as either argument can occur in between the matrix and the non-ﬁnite verb in German. In contrast, in the Icelandic example in (15) the object of the embedded verb follows the non-ﬁnite verb. (14) German Lass dir nicht raten! let you.dat not advise.inf ‘Don’t let anybody advise you.’ (15) Icelandic Láttu rá¶leggja ér eitthva¶! let-you advise.inf you.dat something ‘Get advice from somebody!’ The ﬁfth test, ‘‘subject-to-subject raising,’’ is not considered a valid subject test either in German because not only can the nominative subject occupy the ﬁrst slot with such ‘‘raising verbs,’’ but also other material, like a dummy es ‘it, there,’ adverbials like heute ‘today,’ and the subject-like oblique of the Obl–V–(XP) construction (cf. Reis 1982: 192): (16) German a. Heute today scheint seems mal really gearbeitet worked zu to werden. be(come).inf 52 J. Barðdal b. c. ‘People really seem to be busy today.’ (Reis 1982: 192) Es scheint gearbeitet zu werden. it seems worked to be(come).inf ‘Some work seems to be done here.’ (Reis 1982: 192) Den Kindern scheint kalt zu sein. the.dat children.dat seems cold to be.inf ‘The children seem to be freezing.’ (www.katharinawinkler.de/weihnachten.pdf, 2001) Likewise, as evident from (17), adverbials, a dummy a¶ and the subjectlike oblique of the Obl–V–(XP) construction can also occur in this position with ‘‘raising’’ verbs like seem in Icelandic: (17) Icelandic a. Í dag vir¶ist vera miki¶ unni¶. today seems be.inf much worked ‘Much work seems to be done today.’ b. Þa¶ vir¶ist vera miki¶ unni¶. it seems be.inf much worked ‘Much work seems to be done here.’ c. Börnunum vir¶ist vera kalt. children-the.dat seems be.inf cold ‘The children seem to be freezing.’ The Icelandic examples in (17) are exact parallels of the German ones in (16). This has nevertheless not discredited ‘‘subject-to-subject raising’’ as a subject criterion in Icelandic. On the contrary, the fact that the subjectlike oblique of the Obl–V–(XP) construction, as well as ordinary nominative subjects, can occur in the position preceding ‘seem’ has been taken as evidence for the subject status of subject-like obliques in Icelandic (Bar¶dal 2001a, b; Rögnvaldsson 1996; Sigur¶sson 1989). Consider also examples in which some other material than the subject occurs in ﬁrst position with ‘‘raising-to-subject’’ verbs.4 In such cases, the subject-like oblique occurs between the two verbs in both Icelandic and German, exactly as nominative subjects: (18) German a. Heute scheint den Kindern kalt zu today seems the.dat children.dat cold to b. Í dag vir¶ist börnunum vera kalt. today seems children-the.dat be.inf cold ‘The children seem to be freezing today.’ sein. be.inf Construction-speciﬁc properties of syntactic subjects 53 This is regarded as subject-verb inversion in Icelandic, but as OV word order in German, since subject-like obliques are traditionally analyzed as objects in that language. The sixth criterion, long distance reﬂexivization, is applicable in Icelandic as shown in (19), but inapplicable in German since long distance reﬂexivization does not exist in that language. This is illustrated with the ditransitive verb geben ‘give’ in (20): (19) (20) Icelandic Hansi vildi a¶ Anna gæﬁ séri Hans wanted that Anna would-give self.dat ‘Hans wanted Anna to give him some cake.’ smá köku. little cake German a. Hansi wollte, dass Anna ihmi ein bischen Torte Hans wanted that Anna him.dat a tiny cake gäbe. would-give ‘Hans wanted Anna to give him some cake.’ b. *Hansi wollte, dass Anna sich i ein bischen Torte Hans wanted that Anna self.dat a tiny cake gäbe. would-give The Icelandic reﬂexive pronoun sér in the subordinate clause in (19) is coreferential with the nominative subject of the main clause, thus the label long distance reﬂexivization. In German, in contrast, the nominative subject of a matrix clause can only bind an anaphoric pronoun in a subordinate clause (20a) and not a reﬂexive pronoun (20b). Hence, since German does not use a reﬂexive here, there is no subject test either. 3.2. The German subject criteria I now turn to the criteria that have been assigned the status of being subject properties in German but not Icelandic, listed as 10–13 in Table 2 above. These criteria are nominative case, agreement, deletion in imperatives and deletion in telegraphic style. When investigating the syntactic status of subject-like obliques in the Obl–V–(XP) construction in Icelandic, the property of being case marked as nominative has been excluded for obvious reasons (cf. Andrews 1976; Sigur¶sson 1989 and subsequent work; Zaenen et al. 1985). Furthermore, properties that can be shown to correlate with nominative, such as verb agreement (Sigur¶sson 1990–1991 and later work), have not been regarded as subject properties either since they, a priori, exclude everything 54 J. Barðdal but nominative. This is the reason that neither case nor agreement has been used as subject criteria in Icelandic.5 The third criterion, deletion in imperatives, has not been considered a subject property either in Icelandic since many Obl–V–(XP) predicates do not have the right semantics to occur as imperatives (cf. Rögnvaldsson 1996: 48, also pointed out by Barnes 1986: 25 for Faroese). Consider (21) below: (21) a. b. #Have a divine vision! #Feel good! Exclamations of this type involving Obl–V–(XP) predicates are expressed either with the conjunctive or as embedded under ‘let’ in Icelandic. Moreover, the Imperative construction in Icelandic ‘‘univerbates’’ the imperative form of the verb and a nominative deictic pronoun. In (22) the underlined -¶u and -(i)¶i are cliticized forms of the nominative pronouns ú (2p.sg.) and i¶ (2p.pl.), respectively: (22) a. b. Far¶u! go-you.2p.sg.nom ‘Leave!’ Fari¶i! go-you.2p.pl.nom ‘Leave!’ This means that the Imperative construction in Icelandic is formally only compatible with nominative subject predicates. Thus, Obl–V–(XP) predicates selecting for subject-like obliques are not only excluded from occurring in the Imperative construction in Icelandic on semantic grounds but also on formal grounds. Regarding the last criterion, deletion in telegraphic style (cf. Reis 1982: 190), I know of no discussion of it in the literature on syntactic subjecthood in Icelandic. However, the subject-like oblique of the Obl–V–(XP) construction in Icelandic passes this test, whereas it’s German counterpart does not: (23) Icelandic Fór ı́ bı́ó ı́ gær, leiddist Ø.nom went in theatre in yesterday, Ø.dat was-bored alveg hræ¶ilega. quite horribly ‘Went to the movie theatre yesterday . . . was horribly bored.’ (24) German * Muß morgen Ø.nom must tomorrow die the Prüfung exam machen, make, Ø.dat Construction-speciﬁc properties of syntactic subjects 55 graut schrecklich. fear horribly Intended meaning: ‘Have to take the exam tomorrow . . . am horribly nervous.’ These data show that the subject-like oblique of the Obl–V–(XP) construction can be left unexpressed in telegraphic style in Icelandic without that resulting in ungrammaticality, whereas the same is not true for German. 3.3. Conclusion To summarize so far, with regard to word order distribution, ‘‘raising-tosubject’’ and ‘‘raising-to-object,’’ German subject-like obliques behave as nominative subjects in German and as nominative and oblique subjects in Icelandic. These distributional properties, however, have not been considered subject properties in German. Also, long distance reﬂexivization is inapplicable since there are no long distance reﬂexives in German. Subject-like obliques in both languages are not in the nominative case and agreement is found with the nominative in both languages. Finally, subject-like obliques in neither language can be left unexpressed in imperatives. A clear di¤erence, however, between Icelandic and German is found with deletion in telegraphic style, as subject-like obliques in Icelandic are easily left unexpressed in such constructions while their German equivalents are not. In next section I turn to the three tests which have been regarded as subject tests in both German and Icelandic. These are clause-bound reﬂexivization, conjunction reduction and control inﬁnitives. Observe that in such closely related languages as Icelandic and German less than one third of the assumed universal subject properties, as discussed by Keenan (1976), are regarded as valid tests for both languages. These facts severely undermine the whole concept of a universal subject; obviously, if the subject tests are not applicable across languages, then the category is not the same category across languages. The argument here is of course not that it is impossible to distinguish between subjects and objects within one and the same languages. Surely, this is possible for all ‘‘conﬁgurational’’ languages. The argument is rather that if subjects do not behave in the same way across languages, then the category of subject is clearly languagespeciﬁc and not universal. Instead, what is universal is the fact that languages distinguish between these categories of arguments and not the exact behavior of the categories themselves. In other words, if scholars agree that behavioral properties of subjects really are behavioral properties of subjects, they must also be consistent with themselves and accept 56 J. Barðdal the consequences of the fact that the behavioral properties leak across constructions and across languages. The approach suggested later in this article takes this problem seriously and proceeds from the fact that subjects vary in their syntactic behavior not only across languages but also across argument structure constructions within one and the same language. A second problem is that di¤erent scholars have assumed di¤erent constructions to be criterial of subjecthood, partly because of languagespeciﬁc di¤erences but partly, it seems, to suit their own theoretical purposes: nominative case has been excluded as a subject criterion in Icelandic since the goal has been to investigate the syntactic behavior of subject-like non-nominatives, whereas nominative case and verb agreement have been deﬁned as subject criteria in German, thus a priori excluding the subject-like oblique in the Obl–V–(XP) construction. Deletion in imperatives and telegraphic style have been considered subject properties in German on the basis of the fact that subject-like obliques do not pass these tests, only nominatives. ‘‘Raising-to-subject’’ has not been considered a subject test either in German since subject-like obliques share this syntactic behavior with nominative subjects. These are two examples of both cross-linguistic methodological opportunism, and languageinternal methodological opportunism (cf. Croft 2001: 30–32, 41–44). Cross-linguistic methodological opportunism is manifested as di¤erent criteria being used for subjects in di¤erent languages, and languageinternal methodological opportunism manifests itself as some properties being assigned the status of being criterial without a principled way of making the choices. The choices are based on the theoretical preferences of the researcher in question (see also Bar¶dal 2000a for a survey of the methodological opportunism found in research on subjecthood in Old Scandinavian). 4. The three subject tests common for both Icelandic and German The subject tests that have been discussed as valid in both Icelandic and German are only clause-bound reﬂexivization, conjunction reduction and control inﬁnitives. Let us now review the data, relevant to these tests that have been discussed in the literature on the Obl–V–(XP) construction in Icelandic and German. Haider (2005: 26) argues with regard to reﬂexivization that only subjects can bind reﬂexive direct objects in German, whereas both subjects and objects can bind reﬂexive PPs. In (25) below, the object ihm ‘him’ can either bind a reﬂexive or an anaphoric pronoun within the PP: (25) Ich I habe have ihm i him.dat Geschichten stories.acc über about sichi /ihni self.acc/him.acc Construction-speciﬁc properties of syntactic subjects 57 erzählt. told ‘I told him stories about himself.’ Hence, the fact that subject-like obliques can bind reﬂexive PPs in German, as in (26) below, does therefore not unanimously single them out as subjects. (26) a. b. . . . und was man für Angst hat und wie and what one for fear has and how einem i graust vor sich selberi . . . one.dat fears for self self ‘. . . and the fears one has, and how terriﬁed one is by oneself . . .’ (www.andrip.de/kind/gutacht/2423gean.rtf, 1992) Ihm i gefallen Geschichten über sichi . him.dat ge-fall stories about self ‘He likes stories about himself.’ (Stepanov 2003: 6) es it With regard to Conjunction reduction, Zaenen et al. (1985: 477) present the German examples in (27) and (29–30), following the analysis in Cole et al. (1980).6 The examples below show that the subject-like oblique in Icelandic can be left unexpressed in conjunction reduction (27a), while the corresponding subject-like oblique in German cannot (27b): (27) a. b. Icelandic Hann i kom og hjálpa¶. i var he.nom came and Ø.dat was helped ‘He came and was helped.’ German *Er i kam und i wurde geholfen. he.nom came and Ø.dat was helped Intended meaning: ‘He came and was helped.’ These examples must be compared with equivalent examples where a conjoined nominative subject is left unexpressed in conjunction reduction: (28) a. Icelandic Hann i kom ı́ ﬂýti he.nom came in hurry strax aftur. immediately again en but i Ø.nom urfti had a¶ fara to leave 58 J. Barðdal b. German Er i kam schnell vorbei, aber i mußte he.nom came quickly to-here but Ø.nom had-to gleich zurück. immediately back ‘He came by in a hurry but had to leave again immediately.’ The examples in (28) show that a nominative subject in second conjuncts in Icelandic and German can be left unexpressed on identity with a nominative subject of a ﬁrst conjunct. Subject-like obliques can also be left unexpressed on identity with a nominative subject in Icelandic (27a) but not in German (27b). Zaenen et al. (1985) also point out that the Icelandic examples in (29a) and (30a) illustrate that the unexpressed argument of a control inﬁnitive corresponds to a subject-like oblique of the passive Obl–V–(XP) vera hjálpa¶ ‘be helped’ in Icelandic, whereas such correspondence with the subject-like oblique of geholfen werden in German is impossible (29b, 30b): (29) a. b. (30) a. b. Icelandic A¶ vera hjálpa¶ er gott.7 to pro.dat be.inf helped is good ‘It is good to be helped.’ German * Geholfen zu werden ist pro.dat helped to be(come).inf is Intended meaning: ‘It is nice to be helped.’ angenehm. agreeable Icelandic Ég vonast til a¶ vera hjálpa¶. I.nom hope for to pro.dat be.inf helped ‘I hope to be helped.’ German *Ihm/*Er ho¤t geholfen zu him.dat/he.nom hopes pro.dat helped to werden. be(come).inf Intended meaning: ‘He hopes to be helped.’ Similar examples of conjunction reduction and control constructions in German are also discussed by Haspelmath (2001), Fischer and Blaszczak (2001), Fanselow (2002), Sigur¶sson (2002b), Stepanov (2003), Bayer (2004), Haider (2005) and Wunderlich (to appear). Construction-speciﬁc properties of syntactic subjects 59 On the basis of the examples above, Zaenen et al. conclude that the subject-like oblique in the Obl–V–(XP) construction is a syntactic subject in Icelandic but a syntactic object in German (1985: 479). This position is generally held by the contemporary linguistic community (see for instance Askedal 2001; Cole et al. 1980; Faarlund 1990, 2001; Sigur¶sson 1989; Smith 1994, 1996; and many more). The problem with such an analysis is that the subject-like oblique of the Obl–V–(XP) construction in German does not behave particularly like an object. This has been pointed out by Seefranz-Montag (1983: 166–167), Bar¶dal (1997: 47), Haspelmath (2001: 67–75), Fanselow (2002), Stepanov (2003), Bayer (2004) and Wunderlich (to appear), and is ﬂeshed out in more detail below. I now turn to a discussion and presentation of examples involving reﬂexivization, conjunction reduction and control, i.e., the three criteria that have been regarded as subject tests in both Icelandic and German. I present German data relevant to all three subject tests, data that have not ﬁgured at all in the previous literature on the syntactic properties of subject-like obliques in German (except in earlier [working] papers of mine and Eythórsson’s). 4.1. Reﬂexivization In contrast to Haider’s claims, the subject property of clause bound reﬂexivization in Icelandic and German is not in and of itself that subjects bind reﬂexives but rather that subjects must bind reﬂexives while objects do so only optionally. This optionality of objects controlling reﬂexives holds for both Icelandic and German, shown in (25) above for German, and does not exist for subjects in neither language: (31) a. Icelandic Hún i hræ¶ist sigi /*hanai . she.nom fears self.acc/her.acc b. German Sie i fürchtet sichi /*siei she.nom fears self.acc/her.acc ‘She fears herself.’ vs.: ‘She fears her.’ Therefore, as evident from the examples in (31), syntactic subjects in Icelandic and German must bind reﬂexives and cannot bind anaphoric pronouns. The same is also true for subject-like obliques in both languages: (32) a. Icelandic Honum i falla ı́ him.dat fall in sigi /*hanni . self.acc/him.acc ge¶ liking sögur stories um about 60 J. Barðdal b. German Ihm i gefallen Geschichten über sichi /*ihni . him.dat ge-fall stories about self.acc/ihm.acc ‘He likes stories about himself.’ vs.: ‘He likes stories about him.’ These examples show that there is in fact a di¤erence between the binding abilities of subjects and objects and that the subject-like oblique of the Obl–V–(XP) construction patterns with ordinary nominative subjects and not objects in both Icelandic and German. 4.2. Conjunction reduction With regard to conjunction reduction, another criterion of subjecthood common to both German and Icelandic, Seefranz-Montag (1983: 167) shows that the subject-like oblique of ekeln ‘feel disgusted’ can be the unexpressed argument in conjunction reduction, but only if it is omitted on identity with another subject-like oblique (33a). The same is true for the subject-like oblique of dürsten ‘thirst’ (34b) and grauen ‘shudder’ (33c). Conjunction reduction is explicitly discussed as a subject property in Reis (1982: 190–191), Sigur¶sson (2002b: 694–695), Bayer (2004) and Wunderlich (to appear). (33) German a. Mich i schauderte und i ekelte. me.acc felt-horriﬁed and Ø.acc felt-disgusted ‘I felt horriﬁed and disgusted.’ b. Mich i hungert nach Brot und i dürstet me.acc longs for bread and Ø.acc thirsts Wasser. water ‘I long for bread and water.’ c. Mir i wird(’s) schlecht und i graut(’s) me.dat is-it bad and Ø.dat fear-it der Zukunft. the future ‘I feel sick and worry about the future.’ nach for vor for Compare these examples with a corresponding one where an object is treated as an antecedent of the unexpressed argument in conjunction reduction: (34) *Ich sah ihni und mich nicht. i sah I saw him.acc and Ø.nom saw me not Intended meaning: ‘I saw him and he did not see me.’ Construction-speciﬁc properties of syntactic subjects 61 This example clearly shows that a nominative subject cannot be left unexpressed in conjunction reduction on identity with an accusative object. The question arises whether conjunction reduction is perhaps a case test in German rather than a subject test. Consider the examples below: (35) *Er verzieh mir i und i graute davor, sein He forgave me.dat and Ø.dat fear for his Vertrauen zu mißbrauchen. trust to misuse Intended meaning: ‘He forgave me and I am scared of misusing his trust.’ (Seefranz-Montag 1983: 167) (36) *Ich sah ihn i und i hungert. I.nom saw him.acc and Ø.acc hungers Intended meaning: ‘I saw him and he was hungry.’ If conjunction reduction were only sensitive to morphological case irrespective of syntactic relations, obviously the subject-like obliques of grauen and hungern, respectively (35–36) should have the ability to be left unexpressed on identity with an object of a preceding clause, provided that the case marking is the same. Such omission, however, is ungrammatical in German. Haider (2005: 26–27) points out that topicalized objects can be left unexpressed in a second conjunct as long as both objects are preverbal. This is shown in (37) below: (37) Ihm i hat kein Rat geholfen und i wird him.dat has no advice.nom helped and Ø.dat will keiner schaden können. no-one.nom hurt can ‘No advice has helped him and no-one will be able to harm (him).’ However, if the subject-like oblique of the Obl–V–(XP) construction really were an object, it should be omissible if the relevant accusative and dative objects in (35–36) are preverbal. As shown in (38–39) this is not borne out in German (38) *Mir i verzieh er und mißbrauchen. (39) *Ihn i sah ich und i i graute davor, sein Vertrauen zu hungert. Haider further argues that omission in second conjuncts can only be taken as positive evidence for subject behavior if the subject of the ﬁrst conjunct is inverted, i.e., not located in ﬁrst position. Only then is it 62 J. Barðdal certain that the examples do not represent fronted object drop (as in  above). However, on the traditional analysis that the nominative in Dat– V–Nom constructions is a subject, one would certainly expect a nominative subject in a second conjunct to be omissible on identity with such an (allegedly inverted) nominative in ﬁrst conjunct. This expectation, however, is not borne out either: (40) *Mir gefällt der Peter i wirklich und i ist I.dat ge-falls the.nom Peter really and Ø.nom is damit sehr zufrieden. with-that very happy Intended meaning: ‘I really like Peter and (he) is very happy with that.’ Finally, there are fully acceptable examples of subject-like obliques in German being left unexpressed on identity with an inverted subject-like oblique in ﬁrst conjuncts, despite Haider’s claim to the contrary: (41) Deswegen hungert mich i nach Brot because-of-that longs me.acc for bread dürstet nach Wasser. thirsts for water ‘Because of that I long for bread and water.’ und and i Ø.acc The examples discussed in this section show that conjunction reduction is sensitive to both morphological case and syntactic relations in German, with subject-like obliques of the Obl–V–(XP) construction patterning with unambiguous subjects but not with unambiguous objects. 4.3. Control constructions The last subject test traditionally considered as being valid for both Icelandic and German (and other languages) is the control-inﬁnitive test (cf. Bayer 2004; Cole et al. 1980; Fanselow 2002; Fischer and Blaszczak 2001; Haspelmath 2001; Reis 1982: 188–190; Sigur¶sson 1989, 1991, 2002a; Stepanov 2003; Wunderlich to appear; Zaenen et al. 1985; and others). As discussed, and shown in (29–30) above, examples of Obl–V–(XP) constructions embedded under control verbs are always presented as being ungrammatical in German but fully grammatical in Icelandic. However, in spite of their (alleged) ungrammaticality, attested examples of both active and passive Obl–V–(XP) predicates can be found in control inﬁnitives in German: (42) a. Hier sind wir noch halb sinnlich, und es ist äusserst naturwidrig, hier alles verleugnen wollen, was Gott dem physischen Menschen Construction-speciﬁc properties of syntactic subjects b. 63 zum Labsal und zur Erfrischung hie und da am Pfade unserer Wallfarth aufgetischt hat: aber den Lebensweg darum pilgern, um an diesen Erquickungsorten zu schmausen, das ist so verächtlich, daß man das Auge davon this is so disgusting, that one the eye from-it abwenden muß, um nicht übel zu turn must, in-order pro.dat not sick to werden. be(come).inf ‘Here we are still half sensuous, and it is very much against nature to abstain from everything here that the Lord has served the physical person for comfort and refreshment here and there on the path of our pilgrimage: but to take a pilgrimage on the path of life in order to feast at these rest places, that is so disgusting that one has to turn (the eye) away in order not to feel sick.’ (home.t-online.de/home/dr.erich. mertens/STILLIN2.htm, 1789) Häuﬁg ist die gesamte Alltagsbewältigung behinderter Menschen auf Assistenz angewiesen, vom Aufstehen, Waschen, Anziehen über Essen und Bewegen. Die Betro¤enen bauen fast immer ein Vertrauensverhältnis zu ihren Betreuern auf. Potenzielle Täter nutzen das freundschaftliche Verhältnis häuﬁg aus, um gezielt die Bedürfnisse des behinderten Menschen auszuforschen. Je größer die Abhängigkeit, umso größer ist die Gefährdung. Wie soll man Berührungen auch vermeiden, wenn auch die intimsten Handlungen nicht alleine bewerkstellig werden können? Ein Recht für geistig wie körperlich behinderte a right for mentally as physically disabled Frauen, nur von Frauen bei intimen women pro.dat only by women at private Handlungen assistiert zu werden, gibt es in activities assisted to be(come).inf exists there in der Bundesrepublik . . . nicht. the Federal-Republic . . . not. ‘In coping with their everyday life, disabled people are often forced to seek assistance, from the moment they get up, wash, get dressed and with eating and moving around. These people almost always build up a relationship of trust with their carers. Potential o¤enders often take advantage of this friendly relationship with the speciﬁc aim to gather information about the needs of the disabled person. The greater the dependency, the greater the threat. How is one supposed to avoid contact, if 64 J. Barðdal c. d. even the most personal activities cannot be performed in privacy? The right for mentally and physically disabled women to only be assisted by women when engaged in private activities does not exist . . . in Germany.’ (www.freitag.de/2002/45/02450402.php, 2002) Denn ein Teil dieser Erkenntnisse, die mathematischen, ist im alten Besitze der Zuverlässigkeit, und gibt dadurch eine günstige Erwartung auch für andere, ob diese gleich von ganz verschiedener Natur sein mögen. Überdem, wenn man über den Kreis der besides if one about the sphere the Erfahrung hinaus ist, so ist man sicher, experience over is, so is one sure pro.dat durch Erfahrung nicht widersprochen zu through experience not contradicted to werden. be(come).inf ‘Because a part of this knowledge, the mathematical one, has always possessed reliability, and by means of this it provides a favorable expectation for others, even though these may be of a quite di¤erent nature. Besides, if one has left the sphere of experience, one can be certain not to be contradicted by experience.’ (www.gutenberg2000.de/kant/krva/krva003.htm, 1781) Der ewig läuﬁge Richard Kimble, der dicke Captain Kirk, der ﬂedermäusige Flatterheini Batman—sie alle kamen zurück. Sogar den ollen Zossen Black Beauty ließ man letztens noch einmal ein paar Pferdeäpfel auf die Leinwand abseilen, bevor er zu seiner letzten Autogrammstunde in die Freibank trabte. Aber warum klappt das TV-Recycling ei[g]entlich nur im Ausland? ‘‘Mission: Impossible’’ wird ein Mega-Hit auf der ganzen Welt—aber wo bleibt zum Beispiel ‘‘MS-Franziska—Der Film’’? Oder ‘‘Manni, der Libero, returns’’? Haben wir Deutschen etwa keine weggeworfenen have we Germans part non away-thrown Serien, die es wert wären, series, which it worthy were pro.dat wiederverwertet und nostalgisch gehuldigt zu re-used and nostalgically embraced to werden? be(come).inf ‘Richard Kimble, constantly on heat, the fat Captain Kirk, the bat-like ﬂibbertigibbet Batman—they all returned. They even Construction-speciﬁc properties of syntactic subjects 65 had the old hackney Black Beauty drop some dung one last time on the movie screen recently, before it trotted to its last autograph session at the shambles. But why does TV recycling only work abroad? ‘‘Mission: Impossible’’ has become a worldwide mega-hit—but where, for example, is ‘‘MS-Franziska— the Movie’’? Or ‘‘Manni, the Fullback, returns’’? Don’t we Germans have any comic series down the drain which are worthy of being put to good use again and embraced nostalgically?’ (www.bei-gertrud.de/ok/klw_9618.html, 1996) The examples in (42) contain the inﬁnitive form of the passive Obl–V– (XP) predicates assistiert werden ‘be assisted,’ widersprechen werden ‘be contradicted,’ gehuldigt werden ‘be embraced’ and the inﬁnitive of the active Obl–V–(XP) predicate, übel sein ‘feel sick.’ These predicates standardly occur in the Obl–V–(XP) construction and not in the nominative subject construction in German, so the unexpressed argument corresponds to a subject-like oblique and not a nominative subject. The examples in (42) therefore show that the subject-like oblique of Obl–V–(XP) predicates in German shares this particular property—the ability to be left unexpressed in control inﬁnitives—with nominative subjects, despite claims to the contrary in the literature. Moreover, I have searched for control inﬁnitives of approximately 100 types of active and passive Obl– V–(XP) predicates on the World Wide Web and found examples of at least ﬁfteen types, of which the oldest one is from 1781 ad (42c). A similar search on Icelandic web sites has also revealed that only a fraction of all Obl–V–(XP) predicates in Icelandic are found as being embedded under control verbs in texts on the World Wide Web. There are, however, several questions that arise at this point: – – – – Are the passive examples in (42) above of Obl–V–(XP) predicates embedded under control verbs really passive Obl–V–(XP) predicates in the language of these particular speakers and not, say, nominative passives? Are the examples at all representative of German, given that they were found on the World Wide Web and need therefore not manifest the competence of native German speakers? How do we know that these examples aren’t just ‘‘on-line performance errors?’’ Is the ability to be left unexpressed in control inﬁnitives really a subject property and not a property of objects? I will now go systematically through the facts related to these questions. To begin, consider the following examples: 66 (43) (44) J. Barðdal a. *Er i verzieh mirj he.nom forgave me.dat helfen. help.inf b. *Er i verzieh mirj he.nom forgave me.dat helfen. help.inf Intended meaning: ‘He forgave a. *Ihm i gefällt him.dat ge-falls b. *Ihm i gefällt him.dat ge-falls Intended meaning: ‘He statt instead-of pro.nom statt instead-of eri he.nom i j zu Ø.dat to j zu Ø.dat to me instead of helping me.’ es i i zu it pro.nom Ø.dat to es eri i zu it he.nom Ø.dat to likes to help himself.’ helfen. help.inf helfen. help.inf Examples (43–44) illustrate that an unambiguous dative object cannot be left unexpressed in a control inﬁnitive on identity with a dative object of a preceding clause (43), or on identity with a subject-like oblique of the Obl–V–(XP) construction (44), neither instead of the subject (a) nor in parallel with it (b). This is not only true for languages in general, as far as I am aware, but also for those of my German informants who accepted the control inﬁnitives in (42) above (see below). Therefore, since the ability to be left unexpressed in control inﬁnitives does not apply to objects but only to subjects, it is clear that the subject-like obliques of the Obl– V–(XP) construction pattern with unambiguous subjects and not with unambiguous objects. Several of my German discussants, when shown examples like the ones in (42) out of context, have not accepted them as good German. This is of course in line with the fact that sentences like those in (29–30) have been presented as ungrammatical in the literature. I have therefore only included here examples which I feel reasonably sure are formulated by native German speakers and left out a number of examples which are clearly phrased by non-native speakers. That does, however, not exclude the possibility that these sentences may be ‘‘performance errors.’’ This of course raises the fundamental question of how to distinguish between performance errors and marginally acceptable data. This question is particularly meaningful given the fact that one person’s marginalia is probably other people’s performance errors (cf. Bar¶dal and Eythórsson 2005). On a closer inspection, it turns out that several of the relevant examples are phrased by creative writers, academics and journalists, and are found in literary texts, academic prose, biographies and newspapers. Example (42a) is from a lecture given by Prof. Johann Heinrich Jung-Stilling in Construction-speciﬁc properties of syntactic subjects 67 1789, entitled Rede über den Werth der Leiden (‘Lecture on the signiﬁcance of su¤ering’). Example (42c) is from the famous philosopher, Immanuel Kant, and his earlier edition of Kritik der reinen Vernunft (‘Critique of pure reason’) from 1781. Example (42b) is formulated by a political sociologist and occurs in a text published in the weekly newspaper Freitag ‘Friday.’ And there are more examples of Obl–V–(XP) predicates embedded under control verbs occurring in texts composed by the literate section of the German population. Since many of my examples go completely unnoticed when presented in context to native speakers, they hardly qualify as ‘‘performance errors’’ or ‘‘stylistic jokes,’’ but deserve to be taken seriously as genuine examples of a highly infrequent usage pattern.8 It has been repeatedly pointed out for Modern Icelandic that Obl–V– (XP) constructions embedded under control predicates are extremely rare in language use and that a huge corpus is needed to ﬁnd them (Bar¶dal 2000a: 102; Bar¶dal and Eythórsson 2003b: 461, 2005: 5; Eythórsson and Bar¶dal 2005: 837–838; Rögnvaldsson 1991: 372, 1996: 50). I myself have only been able to ﬁnd such examples on the World Wide Web. Recent research on corpora has shown that the World Wide Web can be regarded as a representative language corpus, despite the fact that it is both unbalanced and uncontrolled for. Keller et al. (2002) and Keller and Lapata (2003) have shown that there exists a positive correlation between frequency and acceptability judgments in that the more frequent a lexical-syntactic combination is in the British National Corpus (BNC), the higher it is rated in acceptability by native speakers. And vice versa, that infrequent lexical-syntactic combinations are judged less acceptable by native speakers. Keller and Lapata (2003) compared two di¤erent corpora with the World Wide Web and found that the correlation is strongest for the Web. They interpret their ﬁndings in such a way that the Web is in fact the best corpus to use when searching for lowfrequency marginally-acceptable data, and that the gigantic size of the Web compensates for the fact that it is both unbalanced and uncontrolled for. It thus seems clear that if Obl–V–(XP) constructions can only marginally embed under control predicates in German, the best chances of ﬁnding such examples are on the World Wide Web, since it is the largest language corpus available to us.9 With a corpus of the size of the Web, together with a qualitative assessment of the texts from which the examples are extracted, one would think that the reliability of my control-inﬁnitive examples has been ﬁrmly established. However, in order to eliminate any remaining uncertainty, I have systematically run a questionnaire survey, as a part of a larger investigation (cf. Bar¶dal and Eythórsson 2005 and Eythórsson and Bar¶dal 2005), 68 J. Barðdal containing examples of the kind in (42), with native German speakers. All the examples used in the questionnaire were presented in context, and in order to avoid priming e¤ects, either negative or positive, each participant was only confronted with three examples. The participants were asked to read the text and state: (i) whether the text represents idiomatic German, (ii) how they felt about the particular inﬁnitive in question, and (iii) whether the non-ﬁnite verb is a dative-assigning verb in their language or not (see Appendix).10 The survey took place simultaneously in four German-speaking cities, Bochum, Jena, Saarbrücken and Vienna, with 130 native-speaker participants who were either students of German or English at their respective universities. All answers that were incomplete with regard to the second and/or the third question were omitted from the statistics (as such an incompleteness makes it uncertain whether these particular speakers use the predicates in the Obl–V–(XP) construction or not), leaving behind 336 observations. The examples in the questionnaire were twelve in total, representing eight di¤erent Obl–V–(XP) predicates, including three of the examples in (42). The remaining nine examples are listed, and numbered consecutively, in the Appendix below. From the answers, a three-point judgment scale emerged, i.e., good/OK, strange/awkward and bad/wrong. The results are presented in Table 3 (with numbers referring to the examples in this article and in the Appendix). The ﬁgures in Table 3 show clearly that Obl–V–(XP) constructions, when embedded under control predicates, are judged acceptable by a Table 3. Native-speaker judgments of attested examples of control inﬁnitives in German good/OK 42b 42c 42d A1 A2 A3 A4 A5 A6 A7 A8 A9 strange bad/wrong Total N % N % N % N % assistiert zu werden widersprochen zu werden gehuldigt zu werden übel zu werden widersprochen zu werden vergeben zu werden widersprochen zu werden geholfen zu werden gefolgt zu werden widersprochen zu werden gekündigt zu werden geholfen zu werden 11 6 16 2 2 5 6 4 10 4 19 9 34.4 21.4 64.0 6.2 6.6 20.0 19.4 12.5 38.5 13.8 86.4 36.0 5 5 1 3 3 2 8 2 2 4 2 5 15.6 17.9 4.0 9.4 10.0 8.0 25.8 6.3 7.7 13.8 9.1 20.0 16 17 8 27 25 18 17 26 14 21 1 11 50.0 60.7 32.0 84.4 83.4 72.0 54.8 81.2 53.8 72.4 4.5 44.0 32 28 25 32 30 25 31 32 26 29 22 25 100 100 100 100 100 100 100 100 100 100 100 100 Total 94 28.0 42 12.5 200 59.5 336 100 Construction-speciﬁc properties of syntactic subjects 69 part of the German population in spite of their alleged ungrammaticality. The acceptability rates range from 6.2–86.4 percent, depending on the verb. In fact, the statistical di¤erences between the examples are highly signiﬁcant ( p < 0:000), suggesting that there may be lexical restrictions on which Obl–V–(XP) predicates are accepted as embedded under control verbs in German.11 The verb kündigen ‘dismiss’ (example  in Appendix) is most widely accepted, then huldigen ‘embrace’, then folgen ‘follow’ ( in Appendix), helfen ‘help’ ( in Appendix) and assistieren ‘assist.’ The token frequency of these lexemes from the Mannheim corpus was measured against the acceptability rates and no correlation was found. It therefore seems that token frequency can be ruled out as a possible factor a¤ecting Obl–V–(XP) predicates’ ability to be embedded in control constructions in German. The ﬁgures in Table 3 give the judgments of those German speakers who claim that they use these verbs with a subject-like oblique and not a nominative subject. There may of course be some German speakers who use the passive verbs in the nominative subject construction. However, this investigation only takes into consideration speakers who explicitly claim that they use these passive verbs in the Obl–V–(XP) construction. Hence, the unexpressed syntactic subject in the examples in (42) above and the Appendix must correspond to a subject-like oblique and not a nominative subject, at least for these speakers (see also Bar¶dal and Eythórsson 2005: 8–10 and Eythórsson and Bar¶dal 2005: 854 for examples showing that both Immanuel Kant and H. J. Jung Stilling used widersprechen werden and übel werden consistently with a subject-like oblique and not a nominative subject).12 Further statistical analysis of the responses reveals that there is a correlation between acceptability rates and the participants’ ﬁeld of study at university, in that the German majors were much less accepting of the examples above than the other group of students. This is shown in Table 4. The correlation between acceptability and ﬁeld of study at university is highly signiﬁcant ( p < 0:000). The chances are only one against 1,000 that Table 4. Acceptability rates across ﬁeld of study at university good/OK strange bad/wrong Total N % N % N % N % English German 53 41 38.1 20.8 23 19 16.5 9.6 63 137 45.3 69.5 139 197 100 100 Total 94 28.0 42 12.5 200 59.5 336 100 70 J. Barðdal this is due to coincidence. There are several possible explanations for this. One relates to the fact that the survey was administrated by the students’ university professors, which for the German majors means that they were not only being ‘‘tested’’ in their own subject but also by the professors in that subject. The German majors may therefore have felt more pressure to answer ‘‘correct’’ than the other group. Another explanation is that the German majors (who were advanced students) may perhaps have been exposed to more language prescriptivism during their studies than the English majors (who were beginners), and may thus be inclined to be as strict (or perhaps even stricter) in their judgments than normative prescriptive standards require. The more observant reader may have noticed that there is a clear difference in acceptability between the two examples of geholfen werden in Table 3 above, a di¤erence which is statistically signiﬁcant ( p < :014). This means that lexical restrictions do not explain all the variation found within the data set. However, the variation can be shown to be related to the text registers’ degree of formality. A three-point scale turned out to be su‰ciently di¤erentiated to capture this variance: (i) an informal register (message boards, chat rooms, etc.), (ii) a formal register (philosophical texts, sociopolitical texts, etc.) and (iii) a third category of texts which showed mixed properties between formal and informal registers (popular texts of elevated style, for instance). The results are given in Table 5. The table shows that while 41.4 percent of the examples from the formal texts were judged acceptable, only 6.5 percent of the examples belonging to the informal registers were judged equally acceptable. Also, while only 44 percent of the examples from the formal register were rejected a whole 84 percent of the examples from the informal register were rejected. This di¤erence is statistically signiﬁcant ( p < 0:000), showing that the more formal the text is the higher the chances are that native speakers will accept the control inﬁnitives, and the less formal the text is the higher the chances are that native speakers will reject them. Table 5. Acceptability rates across registers good/OK strange bad/wrong Total N % N % N % N % Informal register Mixed register Formal register 4 35 55 6.5 24.8 41.4 6 17 19 9.7 12.1 14.3 52 89 59 83.9 63.1 44.4 62 141 133 100 100 100 Total 94 28.0 42 12.5 200 59.5 336 100 Construction-speciﬁc properties of syntactic subjects 71 This correlation could mean that in reality the participants were only rating the acceptability/formality of the register, as formal texts are in more conformity with prescriptive norms of standard High German. Several of the participants who accepted the examples above claimed that the occurrence of Obl–V–(XP) constructions in control inﬁnitives is highly colloquial and not accepted in standard High German. However, there is an alternative explanation to this correlation, namely that there may be some very subtle semantic, pragmatic or grammatical restrictions on whether and how Obl–V–(XP) predicates embed under control verbs in German, and if so these restrictions are perhaps violated most in the informal colloquial registers (cf. the discussion on the Icelandic examples in  below), resulting in lower acceptability rates for the informal registers. Marginal constructions in general are presumably subject to extensive restrictions, irrespective of the construction type and the language in question. Yet another factor that may inﬂuence the acceptability of German control inﬁnitives is the category of the controller, i.e., the argument that the unexpressed subject-like oblique is omitted on identity with. In two out of twelve examples, the subject-like oblique is left unexpressed on identity with a nominative ﬁrst person pronoun in the preceding clause, in two cases on identity with an argument retrievable on the basis of the context, in two cases on identity with an indeﬁnite, generic man ‘one,’ in two cases on identity with a nominative noun, in one example on identity with a PP, and in three instances on identity with a relative pronoun. The distinction made here between nominative personal pronouns and nominative nouns is motivated by the fact that only pronouns are case marked in German, not nouns (cf. the discussion around the examples in  below). The relation between acceptability judgments and the category of the controller is given in Table 6. Table 6. Acceptability rates across categories of controllers good/OK strange bad/wrong Total N % N % N % N % Nominative pronoun Contextual reference Generic man Nominative noun Prepositional phrase Relative pronoun 4 8 11 16 11 44 6.4 13.3 20.8 28.1 34.4 61.1 6 6 7 10 5 8 9.7 10.0 13.2 17.5 15.6 11.1 52 46 35 31 16 20 83.9 76.7 66.0 54.4 50.0 27.8 62 60 53 57 32 72 100 100 100 100 100 100 Total 94 28.0 42 12.5 200 59.5 336 100 72 J. Barðdal The ﬁgures in Table 6 show that there is a clear correlation between how acceptable the control inﬁnitives are and the category of the controller. Subject-like obliques left unexpressed on identity with a nominative personal pronoun in the preceding clause are absolutely judged worst, while those left unexpressed on identity with a relative pronoun are judged best, with the other four categories falling in between, in an ascending order ( p < 0:000). Observe also that all the examples discussed above, with the possible exception of (42a), contain fairly stative or non-agentive control predicates, like ‘be worthy of,’ ‘have the feeling,’ ‘feel certain,’ etc. Verbs like ‘promise,’ or ‘hope to’ for that matter, do not seem to be among the control verbs which embed the Obl–V–(XP) construction in German language use. Needless to say, I am not questioning the grammaticality judgments of the German control-inﬁnitive examples in (29b) and (30b) above that have been extensively discussed and refuted in the literature. The claim made by Cole et al. (1980), Reis (1982), Zaenen et al. (1985) and others that the subject-like oblique of the Obl–V–(XP) construction cannot be left unexpressed in control inﬁnitives in German is certainly not taken out of the blue. Note, though, that in (30b) the controller is a nominative personal pronoun and in (29b) it is retrievable on the basis of the context, i.e., the two most dispreferred identity relations. Nevertheless, the examples presented here show that Obl–V–(XP) predicates may well occur in control inﬁnitives with the unexpressed argument corresponding to the subject-like oblique, and facts of German are therefore more nuanced than hitherto believed. At the moment, the restrictions and nonrestrictions on the occurrence of Obl–V–(XP) predicates in control constructions in German are far from clear, and I believe that this topic is worthy of a paper of its own. We do know, however, that this ability may be verb dependent in German, register dependent, controller dependent, or all three. Moreover, those of the participants who stated their reason for not accepting the examples in the questionnaire survey claimed that it was because the subject-like oblique must be overt in their language. I return to this topic in Section 6 below. 4.4. Conclusion My examination in this section has shown that the behavior of the subject-like oblique in the Obl–V–(XP) construction in Icelandic and German is quite similar. In fact it is more similar than expected on the analysis that it is a subject in Icelandic and an object in German. The comparison above also shows that the di¤erences between Icelandic and German have been grossly overestimated, calling into question the Construction-speciﬁc properties of syntactic subjects 73 established view that Icelandic and German belong to two widely distinct typological classes of languages with Obl–V–(XP) constructions. I have also shown that the three tests which have been regarded as subject tests for both Icelandic and German, the ability to control reﬂexives, deletion in conjunction reduction and control inﬁnitives, are not conﬁned to nominative subject arguments in German as hitherto assumed and argued in the literature. Instead, Obl–V–(XP) predicates must control reﬂexives (as opposed to anaphors) in both Icelandic and German, they can be left unexpressed in second conjuncts in German provided that it is on identity with another subject-like oblique, and they can marginally occur in control constructions in German. There are thus only three eminent di¤erences between Icelandic and German: First, the Obl–V–(XP) construction cannot occur in telegraphic style in German, while such examples are grammatical in Icelandic. Second, Conjunction reduction is sensitive to both syntactic relations and morphological case in German, while in Icelandic it is only sensitive to syntactic relations. Third, there are some restrictions on the occurrence of Obl–V–(XP) predicates in control constructions in German, restrictions which do not seem to exist in Icelandic, at least not according to the previous literature on this topic. However, as I discuss in Section 6 below, it is not possible to embed all Obl–V–(XP) constructions under all control predicates in Icelandic either. It therefore seems that the object analysis of the subject-like oblique in the Obl–V–(XP) construction in German is problematic in, at least, four respects: – – – – The object analysis is not based on enough research on the syntactic behavior of the subject-like oblique in the Obl–V–(XP) construction in German. The object analysis is based on a selective choice of subject criteria and not founded on principle. The object analysis is not based on a comparison of the actual behavior of syntactic objects; thus syntactic objecthood is treated as a dustbin category. The object analysis makes wrong predictions about the syntactic behavior of the subject-like oblique in the Obl–V–(XP) construction since it does not behave as a canonical object. The problem which arises is that irrespective of whether the subject-like oblique of the Obl–V–(XP) construction in German is analyzed as a subject or an object, as the terms are deﬁned within most current syntactic theories, it will make false predictions about its behavior. Arguing that the subject-like oblique is a subject predicts that it can freely occur in control inﬁnitives and conjunction reduction, as is the case with ordinary 74 J. Barðdal nominative subjects. However, analyzing it as an object predicts that it should not occur at all in these constructions. Thus, the data presented here show that the subject tests cannot even be consistently applied within the same language, as di¤erent tests will yield di¤erent results for di¤erent argument structure constructions. This general fact is of course well known in the literature on grammatical relations, although it has not been taken seriously so far. Radical Construction Grammar, on the other hand, not only acknowledges this but takes it as its point of departure. At this juncture, it should be mentioned that it has recently been suggested that the subject-like oblique of the Obl–V–(XP) construction in Russian and other languages that display similar constructions is neither a syntactic object nor an oblique subject, but rather a so-called I-nominal, which is taken to equate an indirect object (Moore and Perlmutter 2000: 375). Sigur¶sson in his work (2002b: 695) also states that ‘‘German subject-like non-nominatives seem to be best analyzed as I-nominals.’’ The problem I have with this analysis is that the subject-like oblique of the Obl–V–(XP) construction is assigned a status similar to that of indirect objects without it being established at all that it behaves as an indirect object. In fact, there are examples in Moore and Perlmutter’s article which show that their I-nominals do not behave as indirect objects at all (2000: 379–380), thus their analysis is ad-hoc, not even warranted by the data they present themselves. Also, on this account, the category of Inominals is no more appropriate than that of objects since it is not independently deﬁned either, but is a dustbin category. The Icelandic and German facts discussed in this section can easily be accounted for in a non-reductionist construction grammar, i.e., in a theory of grammar which does not reduce syntactic structure to primitive atomic units, but takes the clausal construction to be the basic unit in a grammatical description. Such an approach aims at giving a description of language which captures generalizations, not only at the most abstract schematic level of description, but also lower-level generalizations. In that way, we are not forced to either make broad generalizations, which abstract away from irregularity, or to only generalize about a small subset of the data. Radical Construction Grammar, together with the usagebased and the network models of grammar, provides us with the tool to capture generalizations at all levels. I now turn to a RCG solution to the problem addressed in this article. 5. Radical construction grammar Radical Construction Grammar is a variant theory of Construction Grammar designed to deal with the typological diversity of constructions Construction-speciﬁc properties of syntactic subjects 75 across languages (cf. Croft 2001). It rejects the idea that grammatical relations and parts of speech are universals as such but instead it takes semantic and symbolic structure to represent the universals underlying seemingly compatible constructions in di¤erent languages. On this approach, constructions are language speciﬁc and thus, grammatical relations and parts of speech are not only language speciﬁc but also construction specific. Consequently, notions of grammatical relations and parts of speech cannot be transferred from one language description to another, since each language consists of its own inventory of constructions and relations between constructions. This of course does not prevent constructions, in genetically or areally related languages, from being identiﬁable as formal and semantic analogs of each other, as is the case with the Obl–V–(XP) construction in Germanic. It only precludes us from transferring facts about a construction in one language to the description of an equivalent construction in another language. Radical Construction Grammar aims at representing language and grammar in a typologically and psycholinguistically realistic way. It is a usage-based full entry model in which redundant information about constructions is stored at all levels (cf. Goldberg 1995: 74; Langacker 1988, 2000). Radical Construction Grammar makes use of taxonomic links in the constructional hierarchy (Croft 2001: 25–29, Croft and Cruse 2004: 283–289), which in essence means two things. First, utterances are instantiations of di¤erent (parental) constructions, so mir ist kalt in (1b) above is both an instance of the Obl–V–(XP) argument structure construction and the ordinary Active declarative construction. Second, constructions are stored at di¤erent levels of schematicity, ranging from highly schematic skeletal constructions to concrete lexically- and semantically-ﬁlled instantiations, with various intermediate levels of schematicity in between the two (cf. Bar¶dal 2001a, 2006/2007; Croft 2003). This is a consequence of speakers’ ability to organize constructions into categories, exactly as all other linguistic entities. Radical Construction Grammar is a non-reductionist theory which takes the clausal construction as a whole to be the primitive of language and thus takes all other categories, such as syntactic roles and parts of speech, to be derived from the whole. Radical Construction Grammar does away with syntactic relations, and posits instead syntactic roles, semantic relations and symbolic relations. More generally, semantic relations correspond grossly to argument linking rules in other current theories, symbolic relations correspond grossly to general semantic interpretation rules, whereas syntactic roles correspond to syntactic relations. Constructions are, moreover, formally accounted for by specifying a semantic/functional level (SEM-part) and 76 J. Barðdal Figure 3. The semantic and symbolic relations between the parts of a construction Figure 4. Part–whole relations vs. part–part relations a syntactic/formal level (SYN-part). The semantic relations hold between the components of the SEM-part. The symbolic relations (cf. Langacker 1987, 1991) hold between the SYN-part and the SEM-part of a construction, and between the individual elements of the SYN-part and the individual components of the SEM-part. This can be illustrated as in Figure 3, where the dashed lines, labeled s, stand for the symbolic relations between the individual elements and components of the SYN- and the SEM-parts and between the SYN- and the SEM-parts as a whole. The solid line, labeled r, stands for the semantic relation holding between the components of the SEM-part. There is, however, a fundamental di¤erence between syntactic roles and syntactic relations. The notion of syntactic relations entails that there is a direct relation between two syntactic elements of an utterance (part–part relation), whereas the notion of syntactic roles entail that there is a relation between one part of an utterance to the utterance as a whole (part– whole relation) (cf. Kay 1997). This is illustrated in Figure 4. The larger boxes represent ordinary argument structure constructions, labeled C, whereas the smaller boxes represent the parts. The leftmost Construction-speciﬁc properties of syntactic subjects 77 ﬁgure illustrates that a part a has a certain role or function in the construction C as a whole, whereas the rightmost ﬁgure illustrates that there is a relation between parts a and b, irrespective of the construction C which the parts instantiate. The leftmost ﬁgure gives the Radical Construction Grammar view of grammatical relations (as syntactic roles), while the rightmost ﬁgure illustrates the traditional view of grammatical relations (as syntactic relations). More speciﬁcally, the Transitive construction has two arguments and a verb, thus it has three parts: A, B and C. How do we know what the grammatical relations between these parts are? In most current theories it is assumed that the relations are as follows: [A [B–C]], where B is the predicate, C is the object and A is the subject. The only problem with this analysis is that it is based on semantic grounds and not syntactic. Consider the following possibility: [[A–B] C]. How do we know which two parts belong closer together syntactically than the others? We cannot derive the answer to this question with the aid of syntax, but only with the aid of semantics (cf. Croft 2001: ch. 4). Therefore, since this information can be derived from the semantic and the symbolic structure, it does not need to be represented in the syntax. The word order of an ordinary argument linking construction is captured by assuming part–whole relations as the following: [[A]–[B]–[C]]. The utmost square brackets represent the construction as a whole, and the inner square brackets around each part represent that it is a part within this whole. Thus, part–whole relations is all that is needed to represent the internal order of the elements, together with an inventory of the constructions of the language, and the categorization mechanism required to recognize the parts across related constructions. Radical Construction Grammar assumes that the behavior of the parts of a construction is speciﬁc to, and thus speciﬁed for, each construction. This is a consequence of the fact that in real language use parts of a speciﬁc construction only occur in constructions and not in isolation. Thus, grammatical relations like the subject and the object relation cannot be deﬁned irrespective of constructions. They are construction speciﬁc and are thus best accounted for as syntactic roles. This is supported by the fact that the so-called ‘‘subject tests’’ only measure a certain syntactic argument’s occurrence and behavior in language-speciﬁc constructions (as the subject tests are themselves constructions of their own). Abstracting away from di¤erent technical deﬁnitions across frameworks, the underlying assumption within most current linguistic theories is that a label like ‘‘subject’’ entails a relation between the so-called subject and the predicate, and a label like ‘‘object’’ entails a relation between the verb and the object, i.e., part–part relations. However, since Radical 78 J. Barðdal Figure 5. The behavior of the parts across di¤erent constructions Construction Grammar only employs part–whole relations and not part– part relations the syntactic representation within RCG di¤ers drastically from these. Subjects and objects, within Radical Construction Grammar, are syntactic roles and thus di¤erent kinds of relations, which emphasize the language-speciﬁc and construction-speciﬁc properties of utterances (and will thus be represented with upper-level letters in the remainder of this article). This entails that an ordinary Intransitive active declarative clause construction has a Verb and a Subject, and in languages like German and Icelandic the Subject precedes the Verb in the linear order. For the Topicalization construction and the Question construction in Icelandic and German, however, the Subject and the Verb invert with each other. This word order is independently speciﬁed for these constructions. Therefore, if a speaker knows his/her language, s/he also knows the inventory of constructions in that language, and how the constructions are related to each other, including how the various parts can be identiﬁed across constructions. This can be illustrated as in Figure 5. Therefore, in a grammar consisting of an inventory of constructions, together with speciﬁcations on how the constructions are related to each other and how the various parts behave in these constructions, the behavior of the subject-like oblique in the Obl–V–(XP) construction in German can easily be accounted for. Such an account would contain a link from the relevant predicates, like grauen ‘fear’ and gefallen ‘like, be to sb’s liking, to please,’ to the Obl–V–(XP) argument structure construction and then a link from that argument structure construction to the various word order and sentence type constructions they can instantiate. The same holds for the Nominative subject construction. The verbs occurring in that construction are linked to it, like for instance the simple verb kommen ‘come,’ with a subsequent link to the various word order and sentence type constructions. The di¤erence between the Obl–V–(XP) and the Nominative subject construction is that their distribution in certain Coordinating, Control and Telegraphic style constructions is not the same. In this way, Radical Construction-speciﬁc properties of syntactic subjects 79 Figure 6. The distribution pattern of a network representation Construction Grammar can accurately model the linguistic knowledge of German speakers without postulating either universal grammatical relations, or even language-speciﬁc grammatical relations, which wrongly predict that the behavior of the subject-like oblique in the Obl–V–(XP) construction should either be identical to the nominative of the Nominative subject construction, or be like that of Objects. Such a model is illustrated in Figure 6 for German. In a network model, the general property of the Nominative subject construction, to occur in all kinds of sentence type constructions, is captured by a link between the Nominative construction and these constructions. Thus, this broad generalization is captured by links in the model. The lower-level generalization, that a certain subclass of predicates in German occurs with an Oblique Subject, is captured by a link between these predicates and the Obl–V–(XP) construction. Moreover, the subject-like behaviors of the subject-like oblique, such as the property to invert with the verb, to control reﬂexivization, etc. are captured with relevant links. Conversely, the lack of prototypical subject behavior is captured by not positing a link between the Obl–V–(XP) construction and the ordinary Control and Conjunction reduction constructions. The fact that conjunction reduction seems to be sensitive to both syntactic roles and morphological case is captured by a weak link (dashed lines) from the Obl–V–(XP) construction to, not the Conjunction reduction construction itself, but to a subconstruction of it involving a subject-like oblique. Figure 7 is a detail of Figure 6, showing the two subconstructions of the Conjunction reduction construction, i.e., the Nominative subject 80 J. Barðdal Figure 7. The Conjunction reduction construction and its subconstructions in German construction and the Obl–V–(XP) construction, which are partially instantiated by di¤erent verbs in German. Verbs occurring in the Nominative subject argument structure construction occur in the leftmost subconstruction whereas verbs selecting for the Obl–V–(XP) argument structure construction occur in the rightmost subconstruction in Figure 7. Returning to Figure 6, the bold-faced box around the Nominative subject construction shows that this construction is more entrenched (in the sense of Bybee 1985, 1995, 2001; Bybee and Thompson 1997; Langacker 1988, 2000) in the minds of speakers than the Obl–V–(XP) construction, as more verbs occur in it than in the Obl–V–(XP) construction. Thus, generalizations across constructions, both general and lower level, are captured by the network model and links between constructions. A model of the grammar of Icelandic speakers would be similar to the one illustrated in the ﬁgures above for German speakers except that the links that are weak or missing in German are present in Icelandic. To exemplify, consider Figure 8 below. Since omission in second conjuncts in Icelandic is not restricted to predicates having the same morphological case on the Subject, no subconstructions need to be modeled. Instead, the highest schematic construction, with no restrictions on the morphological case of the Subject, captures the facts of conjunction reduction in Icelandic. Figure 8. The conjunction reduction construction in Icelandic On a Radical Construction Grammar account nominative arguments of the Nominative subject construction and subject-like obliques of the Obl–V–(XP) construction are both analyzed as Subjects, as the behavior of the subject-like oblique is in fact much more similar to the nominative of the Nominative subject construction than the accusative or the dative Construction-speciﬁc properties of syntactic subjects 81 of Nom–V–Acc or Nom–V–Dat constructions in both Icelandic and German, although the category of subject in RCG is based on construction-speciﬁc behavioral properties, and is neither a universal nor a language-speciﬁc category. The fact that the distribution of the Obl–V–(XP) argument structure construction across sentence-type constructions in German di¤ers from the distribution of the Nominative subject construction in German and Icelandic and from the Obl–V–(XP) construction in Icelandic conﬁrms that an approach like that of Radical Construction Grammar is prompt. It is needed to adequately account for the fact that there are subject-like non-nominative arguments of speciﬁc argument structure constructions in certain languages which neither behave fully like ordinary subjects of these languages nor like objects. This is achieved by abandoning, not only the postulation that grammatical relations are universal, but also the postulation that there is a uniform subject category across argument structure constructions within one and the same language. Moreover, the dichotomy between subjects and objects is departed from without that leading to a need to stipulate additional categories of the type found for instance in Relational Grammar, labeled I-nominals by Moore and Perlmutter (2000). In this sense, the present account is not only more parsimonious than the RG account but it is also descriptively more accurate than that. It is also descriptively more accurate than the LFG account suggested by Zaenen et al. (1985) and the GB account argued for by Sigur¶sson (1989), since both of these assign object status to the subjectlike oblique. Moreover, the various theoretical explications recently suggested by Fanselow (2002), Stepanov (2003), Bayer (2004), Haider (2005) and Wunderlich (to appear) have not led either to the empirical insights gained by the present study, namely that Obl–V–(XP) predicates can indeed restrictively occur in both Conjunction reduction and Control constructions in German. 6. The more extensive question One of the main empirical results of the research presented in this article is that the di¤erence in the distribution/behavior of the subject-like oblique in the Obl–V–(XP) construction in Icelandic and German is much smaller than predicted by theories that posit universal categories of subjects and objects. In fact, my comparison has revealed that the similarities are greater than the di¤erences: Subject-like obliques in both languages occur in ﬁrst position in ordinary Active declarative clause constructions and in Subordinate clause constructions, and both invert with the verb in the Topicalization construction, assuming that function and 82 J. Barðdal intonation can be used to di¤erentiate between the two in German (see Section 3.1). Subject-like obliques in both languages are ‘‘raised’’ to subject in ‘‘subject-to-subject raising,’’ both can be embedded under certain ‘‘subject-to-object raising’’ verbs and both must control clause-bound reﬂexivization. In neither language does the subject-like oblique carry nominative case or control verb agreement, nor can Obl–V–(XP) predicates occur in the Imperative construction. The di¤erences found between the Obl–V–(XP) construction in Icelandic and German thus only involve conjunction reduction, certain control inﬁnitives, and occurrences in telegraphic style. I have, furthermore, illustrated that conjunction reduction is sensitive to both morphological case and syntactic roles, and thus that the subject-like oblique of an Obl–V– (XP) predicate can correspond to the argument left unexpressed in second conjuncts on identity with the ﬁrst Subject given that they have the same morphological case. I have also presented examples of Obl–V–(XP) predicates occurring in control constructions, examples which demonstrate beyond doubt that, at least for some German speakers, subject-like obliques can be left unexpressed on various types of identity relations. Thus, there is no motivation for analyzing the relevant Icelandic subject-like obliques as subjects and the German ones as objects. It is more natural to analyze them as Oblique Subjects as they ﬁll the syntactic role of Subject in Obl– V–(XP) constructions in both Icelandic and German. Having established that, a more extensive question that arises is why the distribution of Obl– V–(XP) predicates across sentence types should be more restricted in German than in Icelandic, given the structural and lexical similarities of the two languages. To the best of my knowledge, the most prominent answer to this question found in the literature so far has been that one is a subject and the other is an object.13 This analysis has been put forth only at the cost of the behavioral similarities between the Nominative Subject and the Oblique Subject in German. Therefore, a Radical Construction Grammar solution, in which the similarities, as well as the di¤erences, in the behavior of the subject-like oblique in the Obl–V–(XP) construction in Icelandic and German are acknowledged, generates di¤erent kind of research questions and di¤erent answers. I now turn to these questions. The ﬁrst question is what the constructions, in which the behavior of the Oblique Subjects in Icelandic and German di¤er, have in common. Consider the overview in Table 7. The di¤erence between the constructions numbered 1–6 and those numbered 7–10 is that the last four are all ‘‘elliptic’’ in the sense that the Subject is left unexpressed. The question is thus whether it isn’t a descriptively more adequate generalization to claim that the di¤erence between Icelandic and German is not one Construction-speciﬁc properties of syntactic subjects Table 7. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 83 Syntactic similarities and di¤erences in the behavior of Oblique Subjects in Icelandic and German First position in declarative clauses Subject-verb inversion First position in subordinate clauses Subject-to-object raising Subject-to-subject raising Reﬂexivization Conjunction reduction Control inﬁnitives Deletion in telegraphic style Deletion in imperatives Icelandic German H H H H H H H H H H H H H H H * * * * * of grammatical relations but rather a di¤erence in the ability to be ‘‘elliptic?’’ That is, Obl–V–(XP) predicates in German di¤er from their Icelandic counterparts in being reluctant to occur in ‘‘elliptic’’ constructions, such as conjunction reduction, control inﬁnitives, telegraphic style and imperatives. Obl–V–(XP) predicates in Icelandic do not share this reluctance, at least not to the same degree. In Icelandic, Obl–V–(XP) predicates cannot occur in all control constructions (see examples  below), and they cannot occur in the Imperative construction, for di¤erent reasons, however (see Section 3.2). This is the main di¤erence in the syntactic behavior of Oblique Subjects in Icelandic and German. A follow-up question is why Oblique Subjects should be more reluctant to be ‘‘elliptic’’ in German than in Icelandic. One possible answer is that when the Oblique Subject is left unexpressed important morphological/ semantic information is lost. This was acknowledged by Thráinsson (1979: 470) in a discussion on Icelandic and the fact that many Icelandic speakers prefer Obl–V–(XP) predicates embedded in subordinate clauses than in control constructions. Thus, many speakers prefer (45a) over (45b): (45) Icelandic a. Ég vonast til a¶ mér ver¶i hjálpa¶. I hope for that me.dat will-be helped ‘I hope that I’ll be helped.’ b. Ég vonast til a¶ ver¶a hjálpa¶. I hope for to pro.dat will-be.inf helped ‘I hope to be helped.’ This preference of Icelandic speakers is certainly unexpected given the fact that control constructions like (45b) are always discussed as being 84 J. Barðdal perfectly grammatical in Icelandic. This raises the question whether all possible examples of Obl–V–(XP) predicates embedded under control verbs are equally acceptable in Icelandic. This is not the case. I have, in my research, come across attested examples which are not grammatical in my language. Three such examples are given in (46). (46) a. b. c. Hlutfall nemenda ı́ 5.–10. bekk sem eru frekar proportion students in 5–10 grade who are rather e¶a mjög sammála vı́ a¶ ykja vænt or very agreeing it to pro.dat feel.inf a¤ection um skólann sinn, about school their a¶ samskipti nemenda og fullor¶inna séu gó¶ ı́ skólanum og a¶ krakkarnir ı́ bekknum séu gó¶ir vinir. ‘The proportion of students in 5–10 grade who agree [with the statement] that they care about their school, that the interaction between the teachers and the students is good in the school, and that the children are on friendly terms with each other.’ (www.rvk.is/upload/ﬁles/Bornin_i_borginni_lokaskyrsla.pdf ) Ég átti nú egar heimili me¶ mömmu sem ótti vænt um mig og tvo bræ¶ ur sem ég gat leiki¶ mér vi¶, og a¶ra hvolpa sem stoppu¶ u stundum vi¶, stöldru¶u vi¶ um stund, og fóru sı́¶an sı́na lei¶. Mig langar ekki a¶ fara neitt anna¶. Loksins kom ég a¶ húsi essara indæla eldra ﬁnally came I to house these lovely older fólks og au gáfu mér a¶ bor¶a og reyndu people and they gave me to eat and tried a¶ ykja vænt um mig . . . to pro.dat feel.inf care about me ‘I already had a home with my mother who loved me and with my two brothers whom I could play with, and the other puppies who stopped by occasionally for a while, before they went their way. I don’t want to go anywhere else. Finally, though, I came to the house of this lovely older couple and they fed me and tried to care about me . . .’ (www.shihtzu-in-iceland.com/soguhornid.html, 2003) En svona ı́ alvöru tala¶ á er ekkert sni¶ ugt a¶ ér skuli lı́¶a svona illa . . . ú ert me¶ svo margt spennandi framundan og sı́¶an ertu lı́ka svo sæt og skemmtileg!!! Ég veit! hættu bara a¶ lı́¶a illa . . . I know! stop-you just to pro.dat feel.inf bad Construction-speciﬁc properties of syntactic subjects 85 ‘But seriously, it is not a good thing that you feel so bad . . . There are so many exciting things ahead of you, and you’re also so sweet and fun to be with!!! I know! Just stop feeling bad . . .’ (kaninka.net/halla/005637.html, 2003) The fact that I ﬁnd these examples unacceptable provides motivation for carrying out a questionnaire survey like the German one (described in Section 4.3), among native speakers of Icelandic. Such a survey, with the three examples above, reveals that not all speakers agree on their acceptability (cf. Bar¶dal and Eythórsson 2005: 15–20). Table 8 gives the acceptability judgments, collected from 36 university students, majoring in English and Icelandic at the University of Iceland, all native speakers, yielding a total of 87 observations: Table 8. Native-speaker judgments of attested examples of control inﬁnitives in Icelandic good/OK 46a a¶ ykja 46b a¶ ykja 46c a¶ lı́¶a Total N % 16 5 2 57.1 17.8 6.5 23 26.4 strange N bad/wrong Total % N % N % 7 8 9 25.0 28.6 29.0 5 15 20 17.9 53.6 64.5 28 28 31 100 100 100 24 27.6 40 46.0 87 100 The acceptability rates range from 6.5 percent to 57 percent, with the ﬁrst example in (46) judged best and the last one judged worst ( p < 0:000). In fact, 64.5 percent of the participants judged (46c) as unacceptable. This shows that documented control constructions involving Obl–V–(XP) predicates are not unanimously judged well-formed in Icelandic. In fact, exactly as with German, there is a statistically signiﬁcant correlation between acceptability rates and ﬁeld of study at university, with students majoring in Icelandic being more restricted in their judgments than those majoring in English ( p < :002). This is shown in Table 9. Moreover, exactly as in German, there is a correlation between acceptability rates and register, as example (46a) stems from a formal register, (46b) from a less formal register and (46c) from an informal register. Observe also that the controller in the matrix clause is a relativized argument in (46a), but a nominative personal pronoun in the two other examples. These parallels in the behavior of Oblique Subjects in German and Icelandic suggest an implicational hierarchy for types of identity relations 86 J. Barðdal Table 9. Acceptability rates across ﬁeld of study at university good/OK strange bad/wrong Total N % N % N % N % English Icelandic 11 12 45.8 19.0 9 15 37.5 23.8 4 36 16.7 57.2 24 63 100 100 Total 23 26.4 24 27.6 40 46.0 87 100 on the basis of which an Oblique Subject can be left unexpressed in Control constructions: (47) The Control Construction’s Identity Relation Hierarchy: relative pron. < PP < noun < generic pron. < context-based ref. < nom. pers. pron. The Control construction’s identity relation hierarchy predicts that if an Oblique Subject can be left unexpressed on identity with a nominative personal pronoun of a matrix clause, then it can also be left unexpressed on all the other identity relations in that language, in an ascending order. And, as its corollary, the ability to be left unexpressed on identity with a relativized argument in a language does not imply an ability to be left unexpressed on the other identity relations lower on the scale. This implicational hierarchy is based on comparative data from Germanic only, which of course means that further research is needed to establish its wider validity. Returning to the examples in (46) above, observe that the control predicates in (46b–c) are both purposive, while the one in (46a) is stative. This may further contribute to the unacceptability of these examples as Obl– V–(XP) predicates are known for being low on the agentivity scale and for lacking an ability to be construed as conveying intentionality of the referent denoted by the Oblique Subject (cf. Bar¶dal 2001a: 53–58, 2001b: 35–39, 100–101, 2004; Bar¶dal and Eythórsson 2005: 29–31). They are thus incompatible in meaning with purposive control predicates (and some of the most prototypical control verbs discussed, for instance, by Jackendo¤ and Culicover 2003 for English). Furthermore, Barnes (1986: 26–27) reports that Faroese speakers also di¤er in their acceptability judgments of Obl–V–(XP) predicates in both the Conjunction reduction and the Control construction in Faroese. The following examples of control inﬁnitives, for instance, were judged only marginally possible by Barnes’ informants: Construction-speciﬁc properties of syntactic subjects (48) 87 Faroese a. ??Eingin bey¶ sær til at dáma no-one.nom o¤ered self for to pro.dat like.inf hana. her ‘No one o¤ered to like her.’ b. ?At skorta mat er ræ¶ uligt. to pro.dat lack.inf food is terrible ‘To lack food is terrible.’ Observe that the matrix clause in (48a) is both purposive and contains a nominative pronoun as a controller, while the example in (48b) is generic non-purposive, with the Oblique Subject being left unexpressed on identity with a referent retrievable from the context. Recall that nominative reference and context-based reference are dispreferred identity relations, according to the Control construction’s identity relation hierarchy in (47) above. Therefore, these facts show that not only German speakers vary in their acceptability judgments of Obl–V–(XP) predicates in Control constructions, such lack of agreement on acceptability is also found in both Icelandic and Faroese. In conclusion, the di¤erences between Icelandic and German regarding the behavior of the Oblique Subject in Control constructions are not as gross as suggested in the literature: In German the Oblique Subject can be left unexpressed in certain Control constructions, though not in all. In Icelandic the Oblique Subject cannot be embedded under purposive control verbs. It is therefore clear that research on the distribution of Obl–V– (XP) constructions in a language cannot be conﬁned to their occurrence with purposive control verbs, as for instance in Moore and Perlmutter in their investigation of certain Obl–V–(XP) constructions in Russian (2000: 398), but has to take other less agentive Control constructions into consideration. In fact, the ﬁndings of the present article call for a much more thorough research on Control constructions in both Icelandic and German, and other languages, than hitherto undertaken in the literature. What is more, these ﬁndings call for a revision of the Subject construction hierarchy suggested by Croft (2001: 155–157), both the simpliﬁed (49) and the full version (50): (49) The Simpliﬁed Subject Construction Hierarchy: behavioral constructions < coding constructions (50) The Subject Construction Hierarchy: coordination < purposive < relativization < verb agreement < case marking 88 J. Barðdal The Germanic data do not support the internal order between the Conjunction reduction construction and the Control construction in (50), (assuming that Croft is really referring to the general category of Control constructions and not only to the subcategory of Purposive control constructions). In contrast, the evidence from Germanic suggests that Control constructions (perhaps with the exception of purposive constructions) and conjunction reduction seem to have the same status in Germanic. They seem to be equally well- or ill-formed in Icelandic, German and Faroese. Thus, from the behavioral and coding properties of Oblique Subjects in Germanic, the following Subject construction hierarchy emerges: (51) The Germanic Subject Construction Hierarchy: other behavioral constructions < ‘‘elliptic’’ constructions < coding constructions In other words, behavioral constructions must be divided into two categories, ‘‘elliptic’’ constructions and others, as Oblique and Nominative Subjects in Germanic clearly show distinct behavior with regard to these two types of argument tracking constructions. Let us now return to the hypothesis that Oblique Subjects resist being left unexpressed in ‘‘elliptic’’ constructions in German because of the morphological/semantic information encoded by the Oblique Subject. This hypothesis is sustained by the fact that the Oblique Subject can only be left unexpressed on identity with another Oblique Subject in German Conjunction reduction constructions, as shown in (33c), repeated here for convenience: (33) c. German Mir wird(’s) schlecht und graut(’s) me.dat is-it bad and Ø.dat fear-it der Zukunft. the future ‘I feel sick and fear for the future.’ vor for Thus, the morphological/semantic information is maintained in the Oblique Subject in the ﬁrst conjunct. With regard to Control constructions in German, all the participants of the questionnaire survey reported on in Section 4.3 who gave a reason for rejecting the relevant control constructions did so on the basis of morphological case marking, arguing that the examples were bad because the dative was lacking. This is a clear indication that not only Conjunction reduction but also Control constructions in German are sensitive to morphological case. There is in fact independent evidence for the assumption that the reluctance of Obl–V–(XP) predicates to occur in ‘‘elliptic’’ constructions in Construction-speciﬁc properties of syntactic subjects 89 German may be due to their morphological case marking. Bayer et al. (2001), for instance, argue that there is a requirement in German that dative case must be morphologically overt. They present examples of non-inﬂecting nominals and topic drop, amongst others, in favor of this argument. The following examples show that sentences in which the dative is represented by a non-inﬂected nominal are ungrammatical (52a), whereas corresponding sentences with an overt and visible dative are grammatical (52b) (Bayer et al. 2001: 473): (52) German a. *Otto hat allerlei Unsinn widersprochen. Otto has a-lot nonsense contradicted b. Otto hat allerlei solchem Unsinn widersprochen. Otto has a-lot such.dat nonsense contradicted ‘Otto has contradicted a lot of (such) nonsense.’ German also allows topic drop of topical arguments as long as they are not in the dative case. The examples in (53) below show that a preposed Accusative Object may undergo topic drop, while a preposed Dative Object may not: (53) a. b. Das the Geburtstagsgeschenk hat Peter gut gefallen, birthday-present has Peter good ge-fallen hat er gemocht. Ø.acc has he liked ‘The birthday present was really to Peter’s liking, he loved (it).’ Dem Peter hab’ ich an seinem Geburtstag eine the Peter have I at his birthday a Freude gemacht, * hab’ ich ein Buch gegeben. pleasure made Ø.dat have I a book given Intended meaning: ‘Peter, I made happy on his birthday, I gave (him) a book.’ Furthermore, in a recent experiment on relative clauses in German, Vogel and Frisch (2003) have shown that in free relative clauses where the morphological case assigned by the matrix verb does not coincide with the morphological case assigned by the subordinate verb, German speakers show statistically signiﬁcant variation in which case form they prefer as overtly marked. Generally, they are more accepting of examples in which the dative overrides the accusative than of examples where the accusative overrides the dative. Thus, German speakers favor examples where the dative is morphologically spelled out. These facts of non-inﬂecting nominals, topic-drop and free relative clauses in German all suggest that there 90 J. Barðdal is a requirement in German that dative case must be visible. They, thus, provide further support for my hypothesis that Obl–V–(XP) predicates in German may be reluctant to occur in ‘‘elliptic’’ constructions because the Oblique Subject must be visible. Why, then, is there no such requirement in Icelandic that datives must be morphologically overt? If facts of German are due to dative being more marked than nominative and accusative, or lower in the universal case hierarchy, as Vogel and Frisch suggest, datives should be equally ‘‘low’’ in the case hierarchy in Icelandic. This is not the case, however, as Oblique Subjects in Icelandic are much more at ease with being ‘‘elliptic’’ than their German counterparts. I believe that this di¤erence reﬂects a di¤erence in the entrenchment of the construction between the two languages, in that the Obl–V–(XP) construction in Icelandic is much more entrenched in the minds of speakers than the German Obl–V– (XP) construction, as both its type and its text frequency is higher in Icelandic. As already mentioned in Section 2 above, the types occurring in the Obl–V–(XP) construction in Icelandic are approximately 900 (700 Dative Subject types and 200 Accusative Subject types), whereas Obl–V– (XP) predicates in contemporary German are relatively few in number, perhaps between 80–100 in total (cf. Bar¶dal 2001b: 136, 2004: 109– 110). This count includes both simple Dat–V–Nom verbs and compositional predicates, like those with sein ‘be’ together with adjectives or nouns (like in  above). Verbs selecting for Dative Objects in the active form, like mótmæla and widersprechen ‘contradict,’ are also higher in type frequency in Icelandic than in German: According to Maling (2002: 2), transitive verbs selecting for Dative Objects are at least 750 in Icelandic, whereas they are only approximately 140 in German. This count of German, however, includes at least some Obl–V–(XP) predicates since subject-like obliques have traditionally been considered to be objects. Therefore, the amount of true Dative object verbs is probably somewhat lower. Moreover, certain Obl–V–(XP) predicates are high in text frequency in Icelandic: Oblique Subjects have been measured as 5.5 percent of Subjects in some written genres, and their text frequency is even higher in spoken genres, i.e., 7 percent (Bar¶dal 2001b: 89). About their type and text frequency in German, Seefranz-Montag states (1984: 541): Only a few relics of ‘‘subjectless’’ constructions (with es-variants) are still unrestrictedly acceptable in contemporary standard German: mich friert, mir graut, mich ekelt, mir schwindelt. However, even these relics are in the process of being contextually restricted. They are increasingly conﬁned to written elevated style, they are used mainly by elder speakers and only in some regional varieties of German. In colloquial spoken German, on the other hand, and especially in Construction-speciﬁc properties of syntactic subjects 91 substandard varieties of the language, ‘‘subjectless’’ expressions are almost completely eliminated: Younger speakers prefer ich friere, ich hab kalt to mich friert and even to mich friert es, they would hardly use mir graut, but ich hab’n Horror vor, not mich gelüstet (arch.) but ich hab Lust auf, not mir schwindelt but ich bin schwindlig/ich hab’n Drehwurm, not mir gefällt das but ich ﬁnde das gut/da fahr ich drauf ab and similar (slang) expressions . . . (Seefranz-Montag 1984: 541) The fact that it is easier for Icelandic speakers to leave the Oblique Subject unexpressed in ‘‘elliptic’’ constructions than for German speakers falls directly from the assumption that the Obl–V–(XP) construction is more deeply entrenched in Icelandic than in German. This is a natural conclusion within grammatical models based on language use, in which frequency is assumed to be one of the main determinants of the structure of the language system (Barlow and Kemmer 2000; Bar¶dal 2001a–b, 2003; Bar¶dal and Eythórsson 2005; Bybee 1985, 1995, 2001; Bybee and Hopper 2001; Bybee and Thompson 1997; Haspelmath 2004; Langacker 1987, 1988, 1991, 1999, 2000). On a usage-based account the language system is an emerging and dynamic system, based on non-linguistic experience, sensitive to and shaped by the frequency of the input. This language system can change and evolve during the life span of a speaker. On a usagebased account the Obl–V–(XP) construction is much more integrated into the Icelandic system, due to its higher type and text frequency, than in the German language system, where it is low in type frequency. Thus it can also occur in a wider range of ‘‘elliptic’’ constructions in Icelandic. The di¤erences in the behavior of the Obl–V–(XP) construction in Icelandic and German are therefore expected on a usage-based account. On a bottom-up approach, the Obl–V–(XP) construction in Icelandic can be assumed to exist at three di¤erent levels of schematicity. First, it exists at the level of verb-specific constructions (in Croft’s 2003 terminology) which contains all the individual predicates that select for the Obl–V–(XP) argument frame. This is the lowest level of schematicity, given in Figure 9 below, and thus the level ﬁlled with the most lexical material and the most speciﬁed semantic information. Second, the Obl–V– (XP) construction in Icelandic also exists at the level of verb-subclass specific constructions, as emotive verbs can be divided into the following six subclasses of verbs of emotion, cognition, perception, idiosyncratic attitudes, bodily states and verbs denoting changes in bodily states (listed in Section 2 above). This is the second lowest level given in Figure 9 below. Finally, the construction can also be assumed to exist one level higher, i.e., on the level of verb-class specific constructions, where the larger subclass of verbs of emotion is on par with the other seven verb-class speciﬁc constructions of verbs denoting personal properties, gain, success/ 92 J. Barðdal performance, failure/mistake, decline, existence and verbs of social interaction, respectively. This is the third lowest level in Figure 9 below (which, by the way, is only meant to give an overview of the structure of the Obl–V–(XP) construction in Icelandic, thus several details are omitted). Figure 9. Levels of schematicity of the Obl–V–(XP) construction in Icelandic Observe that the vertical axis of Figure 9 represents the lexicality vs. the schematicity levels of constructions. That is, lower level constructions exist in the network as lexically-ﬁlled instantiations, whereas higher-level constructions are more abstract and schematic, and thus less ﬁlled with lexical and semantic material. The levels of schematicity at which each construction is assumed to exist, is a function of the construction’s type frequency and semantic coherence (cf. Bar¶dal 2006/2007). Hence the existence of the Obl–V–(XP) construction at the three lowest levels of schematicity reveals a relatively high degree of entrenchment of the construction in Icelandic, at least compared to German. This is further supported by psycholinguistic evidence showing that the construction displays a low degree of productivity in Icelandic, as it is extendable to nonce verbs with similar meanings (Bar¶dal 2000b). In German, in contrast, Obl–V–(XP) predicates are only 80–100 and although Figure 2 in Section 2 shows that there are predicates in German which fall into the subclasses of verbs denoting changes in bodily states, personal properties, gain, success/performance, failure/mistake and verbs of decline, the instances in each class are extremely few, only one or two at most. It is thus doubtful whether these instances make up a subclass of their own in the minds of German speakers. It is clear, however, that the Construction-speciﬁc properties of syntactic subjects 93 Figure 10. Levels of schematicity of the Obl–V–(XP) construction in German construction in German exists as a verb-subclass speciﬁc construction as the majority of German Obl–V–(XP) predicates fall into the larger subclass of verbs of emotion. This is illustrated in Figure 10. As the construction is much lower in type frequency in German, and the instances make up much fewer semantic classes, the construction exists at a lower level of schematicity in German. Hence, the construction is also less entrenched in German than in Icelandic. Consider now, for the sake of the argument, the English forms methinks and me-seems. These are two frozen idiomatic instances of an obsolete Obl–V–(XP) construction in English, not being in a synchronic relation to anything else in the grammar. They, thus, only exist at the lowest level of schematicity in the minds of those English speakers who still have passive knowledge about them, and the Subject slot is also lexically ﬁlled. This can be represented as in Figure 11. Figure 11. Levels of schematicity of the Obl–V–(XP) construction in English Given that me-thinks is a lexicalized relic in English, no-one would expect it to occur in ‘‘elliptic’’ constructions. In fact, if me-thinks only exists as a verb-speciﬁc frozen instance of an obsolete Obl–V–(XP) construction its occurrences in ‘‘elliptic’’ constructions where the Subject is left unexpressed is ruled out, simply because the me-part is a lexicalized, and thus an obligatory, part of the expression. However, the Obl–V–(XP) construction in German di¤ers from me-thinks in English as the Oblique Subject in German can occur in all person/number forms, and not only in ﬁrst person singular. The Obl–V–(XP) construction is also much higher in type frequency in German than in English (80–100 vs. 2). Thus, the construction exists as a verb-speciﬁc and a verb-subclass speciﬁc construction. As the Obl–V–(XP) construction in Icelandic exists at an even higher and more schematic level in speakers’ minds than in German, the construction is also less ‘‘lexical’’ in Icelandic than in German. In other words, the Oblique Subject in German is more dependent on its Obl–V– (XP) predicate than in Icelandic, where it exists as a more schematic entity less dependent on its concrete lexically-ﬁlled instantiations. In that 94 J. Barðdal sense, the Obl–V–(XP) template or schema is built into the Icelandic system at a higher level, with less need for overt case marking because of the overall stronger entrenchment of the construction in the system. That is the reason, I believe, for why Oblique Subjects are more easily ‘‘elliptic’’ in Icelandic than in German. The present analysis predicts that Oblique Subjects’ ability to be left unexpressed correlates with the type frequency of the construction in a language. Further research is of course needed to establish a wider typological validity of this prediction than only for the languages discussed here. It is a fact, however, that the type frequency of the construction is also low in Modern Faroese (Bar¶dal 2004: 110). A count based on the examples and the lists in Petersen (2002) and Eythórsson and Jónsson (2003) reveals only 50–60 Obl–V–(XP) predicates as being in use in contemporary Faroese. The prediction of the present analysis is thus borne out for Modern Faroese. To summarize, the status of the Obl–V–(XP) construction in German is at an intermediate level between Icelandic and English. In Icelandic, the construction is relatively high in type frequency compared to German, thus it exists as a fairly schematic construction, maintaining the ability to occur in various ‘‘elliptic’’ constructions in Icelandic. In German, however, the Obl–V–(XP) construction is closer to falling into disuse. It exists as verb-speciﬁc lexically-ﬁlled low-level instantiations of the construction, and as a verb-subclass speciﬁc construction of emotion verbs, which in turn restricts its occurrence in ‘‘elliptic’’ constructions. Finally, in English, the construction is only found in a couple of frozen expressions, thus it is expected to be incompatible with occurring in ‘‘elliptic’’ constructions. These data show that there is a correlation between the type frequency of predicates instantiating the Obl–V–(XP) construction in a language and the occurrences of these predicates in ‘‘elliptic’’ constructions. Such a correlation is predicted on a usage-based approach to language and grammar. 7. Summary In this article I have compared the Obl–V–(XP) construction in Icelandic and German to the Nominative subject construction in these languages, and I have compared the Obl–V–(XP) construction across Icelandic and German. I started by giving a survey of the conventional subject tests used in these languages, a survey which revealed great inconsistencies, exemplifying both cross-linguistic methodological opportunism and language-internal methodological opportunism. Cross-linguistic methodological opportunism is manifested as di¤erent criteria being used for Construction-speciﬁc properties of syntactic subjects 95 deﬁning subjects in di¤erent languages, whereas language-internal methodological opportunism manifests itself as some properties being assigned the status of being criterial in a particular language without a principled way of making the choices. This comparison supports the position that grammatical relations do not exist as universal categories. I have also presented examples of Reﬂexivization, Conjunction reduction and Control constructions in German which have not ﬁgured in the earlier literature, examples which show that the di¤erence between the Obl–V–(XP) construction in Icelandic and German is much smaller than assumed in the literature, and lesser than predicted by the analysis that the subject-like oblique is a subject in Icelandic but an object in German. Thus, the typological distinction between Icelandic and German assumed in the literature is unfounded. However, analyzing the subject-like oblique in German as a subject will also yield wrong predictions about its behavior, since it is not identical to the behavior of the nominative in the Nominative subject construction in German. I have suggested a Radical Construction Grammar solution to this problem, in which universal grammatical relations of subjects and objects are systematically abandoned. The behavior of the subject-like oblique in the Obl–V– (XP) construction can be captured by assuming that constructions are language-speciﬁc and that syntactic roles are construction-speciﬁc. Radical Construction Grammar accounts for the grammatical knowledge of speakers by assuming that constructions form a network in which both broad generalizations and lower-level generalizations are captured with links in the network model. Finally, liberating oneself from the yoke of the subject–object dichotomy allows for di¤erent kind of research questions, such as what the German sentence-type constructions have in common in which the Obl–V– (XP) construction is reluctant to occur. This shift in research focus has revealed that the main di¤erence in the syntactic behavior of the Oblique Subject in Icelandic and German is that it is more reluctant to be ‘‘elliptic’’ in German than in Icelandic. There are certainly some restrictions on their occurrence in Icelandic but to a much lesser degree than in German. Particularly with regard to Control constructions, there are several factors that seem to inﬂuence their well-formedness in Germanic, such as lexical factors, register, di¤erences in identity relations and speakers’ ﬁeld of study at university. Finally, Oblique Subjects’ ability to be ‘‘elliptic’’ correlates with the frequency of the Obl–V–(XP) construction in Icelandic and German, a fact predicted by theoretical models based on language use, i.e., models which take frequency to be one of the most important determinants of the language system. The status of the Obl–V–(XP) construction in German takes an intermediate position between the status 96 J. Barðdal of the construction in Icelandic, where it still has high enough type frequency to be recognized as an integrated part of the system, and English, where the construction is a relic only found in a couple of frozen expressions. Received 6 December 2004 Revision received 10 October 2005 University of Bergen, Norway Appendix Please read the following text and answer the questions below: (1) Mir geht es in Bezug auf die erotische Liebe ähnlich wie ‘‘Liddle Sister’’. Mit meinen drei Ex-Freundinnen hab ich nur Probleme. Die erste hat mich wahrscheinlich verarscht. Die zweite auch. Ich war bei ihr zu ‘‘besuch’’ :naughty: :D Ihr hab ich das von der ersten vorher erzählt. Wir haben n bisschen rumgemacht. Am nächsten Tag hat sie in der Schule vor Freunden geleugnet, dass wir zusammen wären, mir hat sie aber weiter die Freundin vorgegaukelt. Ich Idiot hab auch nicht auf meine Freunde gehört, die haben mir das vom leugnen gesagt, und ich wollt es nicht glauben. Jedenfalls hab ich jetzt Probleme ne Frau anzufassen, ohne an sie zu denken und übel zu werden (also im Magen). Die dritte, das war meine Schuld, ich hab mich nicht gemeldet. Letztens hab ich gehört, sie will mich schlagen, weil ich sie ‘‘Dorfmatratze’’ genannt haben soll. Ich weiß aber ganz genau, dass ich Ex-Freundinnen im Nachhinein nicht beleidige. Das hat keinen Stil. Zum Glück hab ich sie bisher noch nicht wiedergetro¤en . . . und danach war gar nichts mehr . . . Does the text in (1) represent idiomatic German? How do you feel about the inﬁnitive with übel werden? Does übel werden take a dative in your language? Does that a¤ect your answer, if so how? (Please use the back of this sheet to elaborate on your answer if you need to!) (2) Hi Noti, diesen PB-Code kann VB nie packen. Sonst hätte ich eine VBCode dazu gemacht. Comþ ist mit solchen multiblen Objects, die rein vom OS herrühren ziemlich überfordert. DlgListDir wird in VB fehlschlagen, auch dann wenn Form und Listbox deﬁnitiv deklariert sind und vorhandene Objekte darstellen. Ich brenne ja darauf widersprochen zu werden ------ Cu Construction-speciﬁc properties of syntactic subjects 97 (3) Glück im Horror Obwohl er zunächst ganz anders aussieht, ist dies ein Film über das Glück. Er beschreibt, wie Glück daraus entstehen kann, dass man sich dem reinen Horror aussetzt. Folgt man dieser Geschichte von Vergebung und Erlösung, dann gilt die paradoxe Weisheit: Man muss seine Familie hinter sich lassen, um seine Familie zu gründen. Man muss vergeben können, um vergeben zu werden. Denn selbst der Henker ist ein Opfer. (4) Die Übergänge sind ﬂießend und die Betriebstreue—wie er es nannte— ist für beide gleichermaßen gegeben. ‘‘Mit diesen Fristen sollen sie uns nur kommen . . .’’, sagte er. O¤ensichtlich macht ihm die Flut der dann anstehenden Kündigungsschutzklagen wenig Angst. Peter Werner von der IG Metall kündigte Sammelklagen an. Kündigungen sind nicht da, um angenommen zu werden. Kündigungen sind da, um widersprochen zu werden. Und die Chancen bei einem Widerspruch stehen keinesfalls schlecht. (5) Mitglied werden Was bedeutet es für mich, Mitglied zu sein? Mitglied bei uns zu sein bedeutet, sich in unserer Gemeinschaft wohlzufühlen. Mitglied bei uns zu sein bedeutet aber auch, das Gefühl zu haben, geholfen zu werden. Mitglied bei uns zu sein bedeutet, jemandem zu helfen, egal ob das ein paar aufmunternde Worte sind oder einen Käfermotor wieder zum knattern zu bringen. Interessengemeinschaft bedeutet, jedes Mitglied setzt sich für den Erhalt luftgekühlter Volkswagen ein. (6) Verzeiht mir, ich habe mich ein wenig in meinen Worten verloren . . . Zurück zur Geschichte. Samhir, die Zeit des Gleichgewichtes von Morua und Silior, ging jedoch allmählich dem Ende zu. Niemand weiß genau, warum, doch irgendwann begannen die beiden Urmächte miteinander zu konkurrieren . . . Auf eine Zeit des Überﬂusses folgte eine Zeit wüster Zerstörung, und viele Geschöpfe wurden dahingera¤t. Dann kam es erneut zu einer Zeit, in der das Leben selbst aufzubegehren schien, nur um wieder von einem Massensterben an düsteren Tagen und stockdunklen Nächten gefolgt zu werden. Samhir war vorüber, nun tobte ein Krieg zwischen Mächten, gegen die niemand etwas ausrichten konnte–außer die Große Mutter selbst . . . (7) Der folgende Ausschnitt aus dem Interview mit einem freien Drehbuchautor verweist auf diese ‘‘Einsamkeit des Respondenten’’: Wie war das für dich, diese Fragen? (lange Pause) Ja, ich meine, es ist interessant. Ich denke, ich werde selten so mal gefragt und hab die Möglichkeit, mich dazu zu äußern, unwidersprochen.’’ Nicht unterbrochen und nicht widersprochen zu werden bedeutet in diesem Falle auch, kaum eine Reaktion zu erhalten. Man muss schon Spaß 98 J. Barðdal am Monolog haben, um dies—wie im obigen Ausschnitt—positiv zu wenden und als im Vergleich zum normalen Gespräch größere Entfaltungsmöglichkeit zu interpretieren. (8) Krankheit—Rechte und Pﬂichten des Arbeitnehmers Wer krank ist, ist eigentlich schon unglücklich genug dran. Für viele ist es außerordentlich lästig, daß sie, wenn sie krank sind, noch zusätzliche Pﬂichten gegenüber ihrem Arbeitgeber haben. Trotzdem sollte man diese Pﬂichten unbedingt erfüllen, damit man keine Rechtsnachteile erleidet. Wer sich nicht an die gesetzlich vorgeschriebenen Regeln hält, riskiert, keine Lohnfortzahlung zu bekommen oder evtl. gar gekündigt zu werden. Bitte beachten Sie daher im Falle Ihrer Erkrankung: (9) ‘‘Ich bin nichts, der Herrgott ist alles. Ich will weder Geld noch Gold, was ich will und kann, ist, allen Menschen helfen und heilen. Wer den Herrgott verleumdet, ist es nicht wert, geholfen zu werden.’’ Das schrieb Bruno Gröning (1906–1959) im Jahr 1949 über sich selbst. Er sah sich als Vermittler göttlicher Botschaften, durch die er heilen könne. Solange der Mensch die göttlichen Gesetze befolge, erfahre er nur Gutes wie Gesundheit, Zufriedenheit, Freude und Glück. Notes * This article is based on an earlier working paper (Bar¶dal 2002); it appears here in a substantially altered version. First, I thank Werner Abraham, Katrin Axel, Balthazar Bickel, Hans Boas, Ute Bohnacker, Miriam Butt, Bill Croft, Ulrike Demske, Tonya Kim Dewey, Edith Ekberg, Sam Featherston and his assistant at Tübingen University, Stefan Huber, Ulrike Jaeckel, Carsten Jopp, Jóhannes G. Jónsson, Haralampos Kalpakidis, Thomas Klein, Knud Lambrecht, Valéria Molnár, Reimar Müller, Jens Neumann, Falco Pfalzgraf, Christer Platzack, Günter Radden, Háj Ross, Halldór Á. Sigur¶sson, Kendra Willson, the editor, Adele Goldberg, and two reviewers, Jan-Ola Östman and an anonymous reviewer, for comments, discussions and/or judgments. Second, I thank Werner Abraham in Vienna, Doris Schönefeld in Bochum, Ulrike Demske in Saarbrücken and Beate Hampe in Jena for running the German version of the control-inﬁnitive questionnaire in their classes in April–May 2004, Jóhannes G. Jónsson, Sigrı́¶ur Sigurjónsdóttir and Matthew Whelpton for running the Icelandic version of it in their classes in April–May 2005, and Christer Johansson for help with the relevant statistics. Third, I am indebted to the audiences at the RCG workshop at the 7th ICCL in Santa Barbara, 27 July 2001, the Linguistics Circle Meeting, Dept. of Germanic Studies, UTAustin, 25 January 2002, the Linguistics Circle of Denton, UNT, 22 February 2002, the Research Seminar at the Dept. of German, Lund University, 29 April 2003, the Noon Colloquium, Dept. of German, UCBerkeley, 7 October 2005 and the audience at the Linguistics Colloquium, Dept. of Linguistics, University of British Columbia, 21 October 2005. Finally, I thank my friend and colleague, Thórhallur Eythórsson, for traveling with me along this thorny, yet rewarding, ‘‘oblique subject’’-path, and for inspiring research collaboration and endless support during these last years. The research presented in this article is funded in part by STINT Construction-speciﬁc properties of syntactic subjects 1. 2. 3. 4. 99 (The Swedish Foundation for International Cooperation in Research and Higher Education) during my a‰liation at the Department of Scandinavian Languages, Lund University. Author’s email address: [email protected] In the linguistic literature, the German Dat–V–Nom predicate gefallen is usually translated into English as either ‘like’ or ‘please.’ Since the meaning of gefallen seems to cover the semantic range from ‘like’ via ‘be to sb’s liking’ to ‘please,’ I will consistently gloss it here as ge-fall in the glossing (second) line but give the appropriate English translation in the translation (third) line. One exception to this is found with alternating predicates which can occur either in the Dat–V–Nom or the Nom–V–Dat argument structure construction in both Icelandic and German (cf. Bar¶dal 2001a; Eythórsson and Bar¶dal 2005). There is, however, one exception to this, namely with alternating predicates when the nominative is a ﬁrst or second person pronoun (cf. Eythórsson and Bar¶dal 2005: 860–865). My deﬁnition of ‘‘raising to subject’’ with scheinen ‘seem’ in German does of course not include constructions in which the subject-like dative is selected by scheinen itself, as in the examples below, but only constructions where scheinen occurs with no arguments of its own, as in example (16). The same pattern holds for vir¶ast, the Icelandic equivalent of scheinen: (i) Mir scheint als ob . . . me.dat seems as if (ii) Mér vir¶ist sem . . . me.dat seems as-if ‘To me it seems as if . . .’ 5. There are nevertheless some alleged di¤erences in agreement between Icelandic and German in that German always displays agreement with the nominative in Dat–V– Nom constructions while Icelandic deviates from that when the nominative object is a ﬁrst or second person pronoun (cf. Sigur¶sson 2002a). However, as argued by Bar¶dal and Eythórsson (2003a) and Eythórsson and Bar¶dal (2005: 860–865), Dat–V–Nom constructions are ungrammatical anyway in Icelandic when the nominative is a pronominal ﬁrst or second person argument because of a person restriction blocking nominative ﬁrst or second person pronominal arguments as objects, not because of the agreement. This analysis is supported by two facts: (i) Examples of ﬁrst and second person nominative objects are also ill-formed with default agreement, which is otherwise a structural possibility in Icelandic. That is unexpected if their ungrammaticality is only due to restrictions on ﬁrst or second person agreement, as Sigur¶sson argues. (ii) There is another class of Dat–V–Nom predicates in Icelandic, namely alternating predicates, not discussed by Sigur¶sson (2002a), which shows the same agreement pattern as in German. In such examples, indeed, the nominative behaves as a syntactic subject in Icelandic and not as an object. 6. For a critical discussion of the ideas suggested and defended in Cole et al. (1980), see Eythórsson and Bar¶dal (2005). 7. In order to distinguish control inﬁnitives from raising inﬁnitives I gloss the missing argument of control inﬁnitives as pro in this article. This has no theoretical implications from my side but only signals which argument of a ﬁnite clause the missing argument corresponds to. 8. One of my discussants has even gone so far as to argue that the German speakers who formulated these sentences must be ‘‘partially illiterate.’’ To this I can only respond 100 9. 10. 11. 12. 13. J. Barðdal that if Immanuel Kant and Prof. H. J. Jung-Stilling were ‘‘partially illiterate,’’ there is not much German data left for linguistic research. In fact Keller et al. estimated in 2002 that the English part of the Web might be approximately 330–980 times larger than BNC which is 100 million words, and my guess is that it has grown explosively since then. I realize that these instructions probably made the participants particularly attentive to the fact that the predicates in question are Obl–V–(XP) predicates. Therefore, if this particular language use is not in conformity with prescriptive standards, I will unfortunately have made the participants aware of that (and risked lower rates for the acceptability judgments as a consequence). However, in order to ﬁnd out whether the reluctance of these predicates to occur embedded under control verbs is related to case marking, it was necessary to formulate the instructions in this way. The statistics reported on in this article were calculated using Pearson Chi-Square. As all the categories here are nominal (except for ‘‘judgment’’ which is ordinal, however with three values only), non-parametric tests are the only relevant tests. Other nonparametric tests yield similar p-values for all the variables discussed here and the same level of signiﬁcance. It could now be argued that the German speakers, formulating and accepting the control inﬁnitives discussed here, may use these verbs with a subject-like oblique in all utterances except those which require the subject to be covert, as in control inﬁnitives. In those cases, then, the same German speakers use these verbs with a(n invisible) nominative subject. Such an assumption, however, would be highly controversial, and if accepted it would simply (and erroneously) lead to the exclusion of control inﬁnitives as a subject test not only in Icelandic but in all other languages as well. The whole linguistic argument of control inﬁnitives has indeed evolved around showing that the unexpressed argument in control constructions corresponds to a subject-like oblique in ﬁnite clauses. This has been shown with case agreement in secondary predicates in, at least, Icelandic (Sigur¶sson 1991), Russian (Franks and Hornstein 1992) and Hungarian (Tóth 2000, here cited by Landau 2004). Therefore, if the scenario outlined above would be regarded as plausible, it means that uncontroversial evidence from secondary predicates in various language families must be disregarded, leaving us with dubious analyses of Icelandic and all other languages which are considered by the contemporary linguistic community as having oblique subjects. Also, one would expect native speakers to realize themselves if they are using these verbs with two di¤erent case constructions. However, from all the comments I have gathered no sign of such ‘‘complementary case assignment’’ is evident, which again severely undermines the creditability of the analysis. Finally, such an analysis is only a logical possibility within formal theoretical models which assume that the unexpressed argument has an invisible casemarked representation in the sentence and that it is this invisible representation which bears (an equally invisible) case. On less formal accounts, like the present one, where the unexpressed argument is understood on the basis of the predicate’s argument structure, analyzing the unexpressed argument of Obl–V–(XP) predicates as a nominative in control constructions only is not a possibility (as these verbs do not occur in the nominative subject construction in ﬁnite clauses in the language of these speakers). For di¤erent kind of generative analyses, however, recently suggested in the literature, see Fanselow (2002), Stepanov (2003), Bayer (2004), Haider (2005) and Wunderlich (to appear). Fanselow argues against categories like subjects and objects but suggests instead that the di¤erence between Icelandic and German lies in di¤erent structural positions of the subject-like oblique. It is not clear how the variation found in Control constructions in German, documented in 4.3 above, would be satisfactorily accounted Construction-speciﬁc properties of syntactic subjects 101 for in his model. Presumably, di¤erent speakers have subject-like obliques of di¤erent verbs occurring in di¤erent positions, and even the same speaker would have the subject-like oblique of the same verb occurring in di¤erent positions, depending on registers or the nature of the controller in the matrix clause. On such an account, the original generalization supposedly captured by the two di¤erent positions Fanselow assumes is lost anyway. Stepanov (2003) argues that the di¤erence between Icelandic and German lies in the fact that the unexpressed argument of Control constructions (PRO in his terminology) must be ‘‘lexical’’ in Icelandic but ‘‘non-lexical’’ in German. Observe that this is intuitively the opposite of my proposal below that the subject-like oblique in Icelandic is less ‘‘lexical’’ than it is in German since it exists at a higher level of schematicity, due to the higher type frequency of the Obl–V–(XP) construction in Icelandic. Because of that, they are also more integrated in the grammar of Icelandic speakers and can thus more easily be left unexpressed in Icelandic than in German. Bayer (2004) argues that the subject-like oblique is not a syntactic object in German, but a ‘‘non-subject external argument,’’ as opposed to Icelandic where it is both a syntactic subject and an external argument. This proposal is similar to the RG account in Moore and Perlmutter (2000) in that it departs from the subject–object dichotomy only by stipulating additional argument categories, a category which in this case is of a rather unclear syntactic status. Both Bayer (2004) and Haider (2005) suggest that the existence of oblique subjects in a language is related to the OV–VO word order parameter, and that German does not have oblique subjects as it is an OV language, as opposed to Icelandic which is a VO language. This analysis raises the question of how to analyze subject-like obliques in languages which have acquired either a uniform OV structure, like German, or a uniform VO structure, like Icelandic, as both these languages have not been uniformly one or the other throughout their recorded histories. It seems on this analysis that the subject-like oblique would be a subject in VO-clauses in Old Icelandic and Old High German but an object in OV-clauses. Wunderlich (to appear), in his optimality-inspired account, reduces the notion of subject to ‘‘argument ranking’’ and ‘‘nominative case’’ in order to account for the paradox in the behavior of the dative and the nominative of Dat–V–Nom predicates in German, as either argument posits some subject-like behavior. On my approach, the notion of subject can be maintained as a language-speciﬁc and construction-speciﬁc notion and the ‘‘alternating’’ behavior of Dat–V–Nom predicates is captured by acknowledging the fact that we are indeed dealing with two argument structure construction, namely the Dat–V–Nom and the Nom–V–Dat construction, and not only one argument structure construction as Wunderlich assumes (cf. Eythórsson and Bar¶dal 2005). References Andrews, Avery D. 1976 The VP complement analysis in Modern Icelandic. Proceedings of the North East Linguistic Society 6, 1–21. Askedal, John Ole 2001 Oblique subjects, structural and lexical case marking: Some thoughts on case assignment in North Germanic and German. In Faarlund, Jan T. (ed.), Grammatical Relations in Change. Amsterdam/Philadelphia: John Benjamins, 65–97. 102 J. Barðdal Bar¶dal, Jóhanna 1997 Oblique subjects in Old Scandinavian. Working Papers in Scandinavian Syntax 60, 25–50. 1999 Case in Icelandic—A construction grammar approach. TijdSchrift voor Skandinavistiek 20(2), 65–100. 2000a The subject is nominative! On obsolete axioms and their deep-rootedness. In Lindberg, Carl-Erik and Ste¤en Nordahl Lund (eds.), 17th Scandinavian Conference of Linguistics. Odense: Institute of Language and Communication, 93–117. 2000b Case assignment of nonce verbs in Icelandic. SKY Journal of Linguistics 13, 7–28. 2001a The perplexity of Dat–Nom verbs in Icelandic. Nordic Journal of Linguistics 24, 47–70. 2001b Case in Icelandic—A Synchronic, Diachronic and Comparative Approach. Lund: Department of Scandinavian Languages. 2001c The role of thematic roles in constructions? Evidence from the Icelandic inchoative. In Holmer, Arthur, Jan-Olof Svantesson and Åke Viberg (eds.), Proceedings of the 18th Scandinavian Conference of Linguistics 2000. Lund: Department of Linguistics, 127–137. 2002 ‘‘Oblique subjects’’ in Icelandic and German. Working Papers in Scandinavian Syntax 70, 61–99. 2003 Case and argument structure of novel verbs of communication in Icelandic. In Delsing, Lars-Olof, Cecilia Falk, Gunlög Josefsson and Halldór Á. Sigur¶sson (eds.), Grammar in Focus: Festschrift for Christer Platzack 18 November 2003, II. Lund: Department of Scandinavian Languages, 25–35. 2004 The semantics of the impersonal construction in Icelandic, German and Faroese: Beyond thematic roles. In Abraham, Werner (ed.), Focus on Germanic Typology. Berlin: Akademie Verlag, 101–30. 2006/2007 Syntactic Productivity: Evidence from Case and Argument Structure in Icelandic. (Constructional Approaches to Language.) Amsterdam: John Benjamins. Bar¶dal, Jóhanna and Thórhallur Eythórsson 2003a Icelandic vs. German: Oblique subjects, agreement and expletives. Chicago Linguistic Society 39(1), 1–14. 2003b The change that never happened: The story of oblique subjects. Journal of Linguistics 39(3), 439–472. 2005 Case and control constructions in German, Faroese and Icelandic: Or how to evaluate marginally-acceptable data? Working Papers in Scandinavian Syntax 75, 1–36. Bar¶dal, Jóhanna and Valéria Molnár 2003 The passive in Icelandic—Compared to Mainland Scandinavian. In Hetland, Jorunn and Valéria Molnár (eds.), Structures of Focus and Grammatical Relations. Tübingen: Niemeyer, 231–260. Barlow, Michael and Suzanne Kemmer (eds.) 2000 Usage-Based Models. Houston: Athelstan. Barnes, Michael 1986 Subject, nominative and oblique case in Faroese. Scripta Islandica 38, 3–35. Bayer, Josef 2004 Non-nominative subjects in comparison. In Bhaskararao, P. and K. V. Subbarao (eds.), Non-Nominative Subjects. Amsterdam: John Benjamins, 31–58. Construction-speciﬁc properties of syntactic subjects 103 Bayer, Josef, Markus Bader and Micheal Meng 2001 Morphological underspeciﬁcation meets oblique case: Syntactic and processing in German. Lingua 111, 465–514. Bybee, Joan L. 1985 Morphology: A Study of the Relation between Meaning and Form. Amsterdam/Philadelphia: John Benjamins. 1995 Regular morphology and the lexicon. Language and Cognitive Processes 10(5), 425–455. 2001 Phonology and Language Use. Cambridge: Cambridge University Press. Bybee, Joan and Paul Hopper (eds.) 2001 Frequency and the Emergence of Linguistic Structure. Amsterdam: John Benjamins. Bybee, Joan and Sandra Thompson 1997 Three frequency e¤ects in syntax. Berkeley Linguistic Society 23, 65– 85. Cole, Peter, Wayne Harbert, Gabriella Hermon and S. N. Sridhar 1980 The acquisition of subjecthood. Language 56, 719–743. Croft, William 2001 Radical Construction Grammar: Syntactic Theory in Typological Perspective. Oxford: Oxford University Press. 2003 Lexical rules vs. constructions: A false dichotomy. In Cuyckens, Hubert, Thomas Berg, René Dirven and Klaus-Uwe Panther (eds.), Motivation in Language: Studies in Honour of Günter Radden. Amsterdam/Philadelphia: John Benjamins, 49–68. Croft, William and D. Alan Cruse 2004 Cognitive Linguistics. Cambridge: Cambridge University Press. Eythórsson, Thórhallur and Jóhanna Bar¶dal 2005 Oblique subjects: A common Germanic inheritance. Language 81(4), 824– 881. Eythórsson, Thórhallur and Jóhannes G. Jónsson 2003 The case of subjects in Faroese. Working Papers in Scandinavian Syntax 72, 207–231. Faarlund, Jan Terje 1990 Syntactic Change: Toward a Theory of Historical Syntax. Berlin/New York: Mouton de Gruyter. 2001 The notion of oblique subject and its status in the history of Icelandic. In Faarlund, Jan T. (ed.), Grammatical Relations in Change. Amsterdam/ Philadelphia: John Benjamins, 99–135. Fanselow, Gisbert 2002 Quirky subjects and other speciﬁers. In Kaufmann, Ingrid and Barbara Stiebels (eds.), More than Words: A Festschrift for Dieter Wunderlich. Berlin: Akademie Verlag, 227–250. Fillmore, Charles J., Paul Kay and Mary Kay O’Connor 1988 Regularity and idiomaticity in grammatical constructions: The case of let alone. Language 64, 501–538. Fischer, Susann and Joanna Blaszczak 2001 Diachronic perspective of quirky subjects. In Bhaskararao, P. (ed.), Working Papers of the International Symposium on ‘‘Non-Nominative Subjects,’’ Tokyo, December 18–21, 2001. Tokyo: Institute for the Studies of Languages and Cultures of Asia and Africa, 42–56. 104 J. Barðdal Franks, Steven and Norbert Hornstein 1992 Secondary predication in Russian and proper government of PRO. In Larson, Richard, Sabine Iatridou, Utpal Lahiri and James Higginbotham (eds.), Control and Grammar. Dordrecht: Kluwer, 1–50. Goldberg, Adele E. 1995 Constructions: A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press. Haider, Hubert 2005 How to turn German into Icelandic—and derive the OV–VO contrast. Journal of Comparative Germanic Syntax 8, 1–53. Haspelmath, Martin 2001 Non-canonical marking of core arguments in European languages. In Aikhenvald, Alexandra Y., R. M. W. Dixon and Masayoki Onishi (eds.), Non-Canonical Marking of Subjects and Objects. Amsterdam: John Benjamins, 53–83. 2004 Explaining the ditransitive person-role constraint: A usage-based approach. Constructions 2/2004 (http://www.constructions-online.de). Hoberg, Ursula 1981 Die Wortstellung in der geschriebenen deutschen Gegenwartssprache. München: Max Hueber Verlag. Jackendo¤, Ray and Peter W. Culicover 2003 The semantic basis of control in English. Language 79(3), 517–556. Kay, Paul 1997 Construction grammar feature structures (revised). http://www.icsi.berkeley. edu/~kay/bcg/FSrev.html. Kay, Paul and Charles Fillmore 1999 Grammatical constructions and linguistic generalizations: The what’s X doing Y? construction. Language 75, 1–34. Keenan, Edward 1976 Towards a universal deﬁnition of ‘‘subject.’’ In Lee, Charles N. (ed.), Subject and Topic. New York: Academic Press, 303–334. Keller, Frank and Maria Lapata 2003 Using the Web to obtain frequencies for unseen bigrams. Computational Linguistics 29(3), 459–484. Keller, Frank, Maria Lapata and Olga Ourioupina 2002 Using the Web to overcome data sparseness. In Hajic, J. and Y. Matsumoto (eds.), Empirical Methods in Natural Language Processing. Philadelphia: University of Pennsylvania and the Association for Computational Linguistics, 230–237. Landau, Idan 2004 The scale of ﬁniteness and the calculus of control. Natural Language and Linguistic Theory 22, 811–877. Lako¤, George 1987 Women, Fire and Dangerous Things: What Categories Reveal about the Mind. Chicago: Chicago University Press. Langacker, Ronald W. 1987 Foundations of Cognitive Grammar, vol. I: Theoretical Prerequisites. Stanford: Stanford University Press. 1988 A usage-based model. In Rudzka-Ostyn, Brygida (ed.), Topics in Cognitive Linguistics. Amsterdam/Philadelphia: John Benjamins, 127–161. Construction-speciﬁc properties of syntactic subjects 1991 1999 2000 105 Foundations of Cognitive Grammar, vol. II: Descriptive Application. Stanford: Stanford University Press. Grammar and Conceptualization. Berlin/New York: Mouton de Gruyter. A dynamic usage-based model. In Barlow, Michael and Suzanne Kemmer (eds.), Usage-Based Models of Language. Stanford: CSLI Publications, 1–63. Lenerz, Jürgen 1977 Zur Abfolge nominaler Satzglieder im Deutschen. Tübingen: Günter Narr. Maling, Joan 2002 Icelandic verbs with dative objects. Working Papers in Scandinavian Syntax 70, 1–60. Michaelis, Laura A. and Josef Ruppenhofer 2001 Beyond Alternations: A Constructional Model of the German Applicative Pattern. Stanford: CSLI Publications. Moore, John and David M. Perlmutter 2000 What does it take to be a dative subject? Natural Language and Linguistic Theory 18, 373–416. Petersen, Hjalmar P. 2002 Quirky case in Faroese. Fró¶skaparrit 50, 63–76. Reis, Marga 1982 Zum Subjektbegri¤ im Deutschen. In Abraham, Werner (ed.), Satzglieder im Deutschen. Vorschläge zur syntaktischen, semantischen und pragmatischen Fundierung. Tübingen: Gunter Narr, 171–211. Rögnvaldsson, Eirı́kur 1991 Quirky subjects in Old Icelandic. In Sigur¶sson, Halldór Á. (ed.), Papers from the Twelfth Scandinavian Conference of Linguistics. Reykjavı́k: Institute of Linguistics, University of Iceland, 369–378. 1996 Frumlag og fall a¶ fornu [Subject and case in Old Icelandic]. Íslenskt mál 18, 37–69. Seefranz-Montag, Ariane von 1983 Syntaktische Funktionen und Wortstellungsveränderung. Die Entwicklung ‘‘subjektloser’’ Konstruktionen in einigen Sprachen. München: Wilhelm Fink Verlag. 1984 ‘‘Subjectless’’ constructions and syntactic change. In Fisiak, Jacek (ed.), Historical Syntax. Berlin/New York: Mouton Publishers, 521–553. Sigur¶sson, Halldór Ármann 1989 Verbal syntax and case in Icelandic. Doctoral Dissertation. Lund University. [Reprinted 1992 by Institute of Linguistics, University of Iceland, Reykjavı́k]. 1990–1991 Beygingarsamræmi [Agreement]. Íslenskt mál 12–13, 31–77. 1991 Icelandic case-marked pro and the licencing of lexical arguments. Natural Language and Linguistic Theory 9, 327–363. 2002a Agree and agreement: Evidence from Germanic. Working Papers in Scandinavian Syntax 70, 101–156. 2002b To be an oblique subject: Russian vs. Icelandic. Natural Language and Linguistic Theory 20, 691–724. Smith, Henry 1994 ‘‘Dative Sickness’’ in Germanic. Natural Language and Linguistic Theory 12, 675–736. 1996 Restrictiveness in Case Theory. Cambridge: Cambridge University Press. 106 J. Barðdal Stepanov, Arthur 2003 On the ‘‘quirky’’ di¤erence Icelandic vs. German: A note of doubt. Working Papers in Scandinavian Syntax 71, 1–32. Thráinsson, Höskuldur 1979 On Complementation in Icelandic. New York: Garland Publishing. Tomasello, Michael 1998 Cognitive linguistics. In Bechtel, William and George Graham (eds.), A Companion to Cognitive Science. Massachusetts and Oxford: Blackwell Publisher, 477–487. Tóth, Ildiko 2000 Inﬂected Inﬁnitives in Hungarian. Doctoral Dissertation, Tilburg University, Tilburg. Vogel, Ralf and Stefan Frisch 2003 The resolution of case conﬂicts: A pilot study. Linguistics in Potsdam 21, 91–103. Wunderlich, Dieter to appear The force of lexical case: German and Icelandic compared. In Hanson, Kristin and Sharon Inkelas (eds.), The Nature of the Word: Essays in Honor of Paul Kiparsky. Massachusetts: MIT Press. Zaenen, Annie, Joan Maling and Höskuldur Thráinsson 1985 Case and grammatical functions: The Icelandic passive. Natural Language and Linguistic Theory 3, 441–483.
* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project