Subject Case in Turkish nominalized clauses

Jaklin Kornfilt

This chapter discusses the notions of morphological (surface) case and abstract Case, showing the empirical and theoretical motivation for each. The discussion of morphological case presents the dimensions of cross-linguistic variation found in this domain, and outlines the main tendencies in the expression of case. The notion of abstract Case is used to predict the distribution of overt and non-overt nominal forms, and is considered one of the fundamental abstract syntactic relations in linguistic theory. The chapter presents a brief survey of theoretical approaches to Case in formal grammar and then discusses the main ways of modeling Case in nominative-accusative and ergative languages. 1. Morphological case, abstract Case, and the need for Case Theory Certain constituents in clause structure are known to determine the form and/or position of other clausal constituents. In particular, verbs and adpositions determine the morphological form of their associated nouns. For example, i...

The main idea suggested in this paper is that subordinate clauses need to be Case licensed (namely embedded under KPs) and that various patterns of "fusion" within a layered functional skeleton obliterate that process. I will provide two types of empirical support for this proposal. First, I will show that this proposal correctly allows for nominal elements to stand in for a whole embedded clause (using the mechanism of Phrasal Spell-Out, Starke 2009; Neeleman & Szendroi 2007). Second, I will show that a wide range of typologically unrelated languages overtly exhibit Case marking on complementizers. Also, direct evidence against the identity of relative pronouns and complementizers / subordinators, possibly suggested -prima facie- by the layered model proposed here, will be provided here with the aid of data from Akkadian, Germanic languages and West Iranian languages.

This paper deals with the Udmurt, Komi-Zyryan and Meadow Mari non-finite clauses headed by the suffix-m. These non-finite clauses can be used in relative, adjunct and argument position in the matrix clause. According to previous studies, the subject of these non-finite clauses can be either in the nominative or in the genitive, and this case alternation has to do with semantic and, partly, with syntactic factors. It has been observed that in Udmurt, the subject must bear the genitive if the non-finite clause is in argument position. In this study, I present some counterexamples to this generalisation. However, I argue that these are deverbal compounds. I try to extend the compound analysis to Komi-Zyryan and Meadow Mari as well. Additionally, I suggest that the position of the non-finite clause determines the case-marking of the subject. Finally, I briefly discuss the notion of the so-called " juxtaposed possessive construction ". Most of the presented (Udmurt) data come from own fieldwork.

Subject Case in Turkish nominalized clauses* Jaklin Kornfilt 1. Introduction and summary It is well-known that the distinction between adjuncts and arguments plays an important role in syntax. For example, arguments can be extracted more easily than adjuncts out of syntactic islands. Furthermore, adjunct domains tend to be syntactic islands, while argument clauses tend not to have island properties (abstracting away from syntactic subjects). In this paper, I claim that the argument-adjunct distinction can also play a role in determining the Case on the subject of a particular syntactic domain. It is the status as an adjunct versus as an argument of that domain which can determine, I claim, the type of subject Case. This paper is also a case study in the interactions of morphology and syntax, as it claims that overt Agr(eement) 1 determines subject Case (but only where Agr is licensed itself in this capacity). Another aspect of the morphology-syntax interaction shown here is absence of a one-to-one relationship between syntactic and morphological Case: while morphological Genitive indeed reflects licensed nominal subject Case, morphological Nominative (possibly by virtue of being phonologically null) reflects both licensed verbal subject Case and default Case. The specific proposals made in this paper are listed below: 1.I claim that Turkish has three types of overt subjects: Those that bear genuine subject Case, those that bear default Case, and those which are Case-less. “Genuine subject Case” is licensed by a designated Case licenser; for Turkish, this is the overt Agr(eement) marker. Such subject Case can be Nominative or Genitive in Turkish, depending on the categorial features of Agr. 130 Jaklin Kornfilt Default Case is possible as a last resort strategy, when subject Case is not licensed for an overt subject, and when no other licenser can license another appropriate Case (e.g. an ECM verb licensing Accusative). Case-less subjects will be discussed briefly, as well; these are non-specific, and they are less mobile than the other two types of subjects. 2.The proposed interaction between the argument-adjunct asymmetry and the designated subject Case licenser, i.e. overt Agr, is implemented in the following way: Agr needs to be licensed itself in order to function as a subject Case licenser. This can happen in three ways: A.Categorially, i.e. via matching category features: A verbal Agr is licensed in a fully verbal extended projection, and a nominal Agr is licensed in a fully nominal extended projection.2 B.However, where there is a categorial mismatch, Agr must be licensed differently. This is when the argument-adjunct asymmetry comes into play: An argument domain bears a thematic index (cf. the proposal in Rizzi 1994 that arguments bear a “referential” index, while adjuncts don’t); this index is inherited by the Agr (if there is one) that heads the argument domain in question. 3 I assume that it is such indexation which licenses a categorially unlicensed Agr as a subject Case licenser. Thus, if Agr does not match its clause categorially, it is only where that clause is an argument that Agr will be able to license subject Case; where the domain is an adjunct, a categorially mismatched Agr cannot license subject Case. We thus correctly predict the existence of argument-adjunct asymmetries with respect to subject Case in categorially hybrid clauses, as well as the absence of such asymmetries in categorially homogeneous clauses. C.There is another way for a categorially mismatched Agr to receive an index and thus to get licensed as a subject Case licenser: via predication with an external head, i.e. when the domain headed by that Agr receives an index via predication (in headed operatorvariable constructions like relative clauses and comparatives), and Subject case in Turkish nominalized clauses 131 when, once again, the Agr head inherits the index of the clause in question. 3.In all other instances (i.e. where there is no Agr, or where an existing, but categorially unlicensed Agr cannot receive an index by either “referential” θ-marking or under predication), no genuine subject Case is possible. The clause will have either a PRO subject or, if it has an overt subject, that subject will be in a default Case rather than in a genuine subject Case. The paper discusses the issue of default Case and proposes criteria determining when default Case is possible and when it is not. It further proposes that the morphological realization of default Case may differ across languages; e.g. it is Accusative in English, while it is Nominative in Turkish. 4.Coming back to subject Case, it is licensed locally within the extended functional projection of the clause; no clause-external nominal element is involved in this licensing—at least not directly, as the licenser of subject Case. 5.The account proposed is compatible with approaches where AgrP is an independent projection (Pollock 1989, Kornfilt 1984), but also with approaches where Agr is positioned within the head of another functional projection, e.g. of the head of a Fin(iteness)P (cf. Rizzi 1997), as long as Agr is housed in a projection separate from TAM (i.e. Tense, Aspect, Mood). 6.This paper is, at the same time, a case study concerning the two most widely used nominalization types in Turkish, with respect to genuine subject Case. The argument-adjunct asymmetry mentioned in 2. is observed in one type of nominalization only (i.e. the indicative type) and not the other (i.e. the subjunctive type). The account proposed claims that, while both types of subordinate domains are DPs, only indicatives are also CPs. This explains the sensitivity of indicatives to “CP-level” phenomena and to θ-marking, and the lack of such sensitivity in non-indicative subordination. The paper is organized as follows: Section 2 presents the two main asymmetries and establishes the relevance of Agr for subject Case. Section 3 offers a basic account of subject Case. Section 4 extends that account to predication. Section 5 draws preliminary 132 Jaklin Kornfilt conclusions. Section 6 discusses the nature of default Case. Section 7 proposes an explanation for when default Case may or may not be allowed. Section 8 discusses two rival approaches to the first asymmetry (i.e. the asymmetry between arguments and adjuncts in nominalized factive clauses) and presents counterarguments. Section 9 summarizes this study’s conclusions and mentions some speculations. The paper is written in a general Principles and Parameters framework without focussing on formalistic issues, to enhance readability by an audience of the kind that attended the workshop where this work was presented. For the same reason, I have not formulated my account in strictly Minimalistic terms. 2. Basic facts: Different types of clauses I now turn to an exposition of the basic facts, starting with different types of subordinate clauses. Embedded clauses in Turkish typically are not tensed, and are traditionally said to be nominalized to varying degrees. However, some subordination, even of the head-final type, is fully verbal; I start my discussion of subject Case with that type. The main point of this section will be to show that genuine subject Case is licensed by Agr, and that TAM morphology does not play a role in this regard. More specifically, I shall claim that there is only one kind of genuine subject Case, irrespective of its morphological realization as Nominative or Genitive: subject Case licensed by Agr. Depending on the categorial features of this Agr as [+N] or [+V], this subject Case will be realized as Genitive (=nominal subject Case) or Nominative (=verbal subject Case), respectively. Subject case in Turkish nominalized clauses 133 2.1. “Verbal clauses”—“verbal” to the fullest 2.1.1. Indicatives “Regular” indicative root clauses exhibit a rich array of TAM markers, as well as (predicate-subject) agreement markers. The latter come from a particular agreement paradigm which is the most widely used one for verbal predicates4 in the language.5 (1) exemplifies this type, with the TAM marker as the future tense: (1) Sen yar›n akﬂam ev -de yemek you(SG) (NOM) tomorrow evening home -LOC food piﬂir-ecek -sin cook-FUT -2.SG ‘You will cook food at home tomorrow evening.’ Notice that the subject is in the Nominative. (In Turkish, there is no phonologically realized Nominative morpheme; I assume here that the syntactic Nominative corresponds to a null morpheme.) Identical clauses can be found as subordinate clauses when used as quotations, but also as “regular” subordinate clauses with a number of matrix verbs; the following two examples illustrate these two situations, in the order mentioned: (2) a.[Sen yar›n akﬂam ev -de yemek you(SG) (NOM) tomorrow evening home -LOC food piﬂir -ecek -sin ] diye duy -du -m cook -FUT -2.SG ‘saying’ hear -PAST -1.SG ‘I heard “you will cook food at home tomorrow evening”.’ b.[Sen yar›n akﬂam ev -de yemek you(SG) (NOM) tomorrow evening home -LOC food piﬂir -ecek -sin ] san -›yor -um cook -FUT -2.SG believe-PRSPROG -1.SG ‘I believe you will cook food at home tomorrow evening.’ 134 Jaklin Kornfilt Exceptional Case Marking [ECM]-constructions provide evidence that it is not TAM—morphology which is responsible for licensing of the subject and of its Case.6 This can be seen clearly in the contrast between fully verbal subordinate clauses like (2)b., where we just saw a subordinate clause exhibiting fully verbal TAM as well as fully verbal Agr morphology, and a corresponding subordinate clause also exhibiting fully verbal TAM morphology, but no Agr morphology. (3)a. is similar to (2)b. in showing that a fully verbal subordinate clause with verbal TAM morphology and with verbal Agr morphology has a Nominative subject, this time with a subordinate clause that exhibits progressive aspect and simple past tense: (3) a.[Sen dün sabah ev -de yemek you(SG)(NOM) yesterday morning home -LOC food piﬂir -iyor -du -n ] san -d› -m cook -PROG -PAST -2.SG believe-PAST -1.SG ‘I believed (that) you were cooking food at home yesterday morning.’ The next example exhibits an interesting pattern: the subordinate clause has the identical (verbal) aspect and tense combination, but it lacks Agr marking: (3) b.[Sen -i dün sabah ev -de yemek you(SG) -ACC yesterday morning home -LOC food piﬂir -iyor -du ] san -d› -m cook -PROG -PAST (no Agr) believe-PAST -1.SG ‘I believed you to have been cooking food at home yesterday morning.’ This last example shows that when Agr is absent, the subject cannot show up in the appropriate subject Case, which would be the verbal subject Case in this instance, i.e. in the Nominative. Instead, where the matrix verb is one of a small number of ECM verbs, Accusative is licensed by that verb. Subject case in Turkish nominalized clauses 135 It should be noted that many speakers accept ECM-like constructions with overt Agr, as well: (3) c.[Sen -i dün sabah ev -de yemek you(SG) -ACC yesterday morning home -LOC food piﬂir -iyor -du -n ] san -d› -m cook -PROG -PAST -2.SG believe-PAST -1.SG ‘I believed you to have been cooking food at home yesterday morning.’ I shall not, in the context of this paper, address the issue of the nature of ECM in Turkish in detail, nor in the status of (3)c. It is possible, for example, that while (3)b. is a genuine instance of ECM, (3)c. exemplifies a phonologically empty subject (i.e. pro) copy in the subordinate clause, with the Accusative DP actually raised into the matrix (cf. Moore 1998). For the purposes of this paper, the important point is the following: all speakers accept (3)b., with an Accusative subject under absence of Agr in the verbal subordinate clause, and no speaker would accept (3)d., where the Agr element is missing, yet where the embedded subject is in the Nominative: (3) d.*[[Sen dün sabah ev -de yemek you(SG) (NOM) yesterday morning home -LOC food piﬂir -iyor -du ] san -d› -m cook -PROG -PAST (no Agr) believe-PAST -1.SG Intended reading: ‘I believed (that) you [Nom.] were yesterday morning.’cooking food at home yesterday morning.’ Note that both in the fully grammatical (3)b. and in the completely ungrammatical (3)d., the embedded predicate bears its regular TAM morphology, i.e. in this instance, markers for progressive aspect and for past tense. Therefore, the licenser of Nominative subjects, i.e. of the verbal subject Case, cannot be the verbal TAM morphology. 136 Jaklin Kornfilt The second part of our conclusion must therefore be as follows: the licenser of the Nominative in root as well as embedded verbal clauses is the (verbal) Agr marker. Our general observation so far, then, is as follows: 1. There is a strict correlation between (verbal) Agr and (verbal) subject Case, i.e. Nominative; 2. There is no correlation at all between (verbal) TAM morphology and the (verbal) subject Case. From a cross-linguistic point of view, it is not a novel observation that ECM is not limited to infinitival subordinate clauses (while it is so limited in some languages, e.g. English). There are some languages where ECM can apply to subjunctive clauses, and some where ECM is possible even into clauses with tense and agreement; the latter is the case, for example, in Modern Greek. The Turkish facts are of special interest nevertheless. First of all, the subordinate clauses into which ECM may apply are indicative, not subjunctive. It is well-known that subjunctive clauses are more “transparent” than indicative ones with respect to a number of syntactic phenomena, ECM being only one of them. The same is true with respect to anaphoric binding, for example. Secondly, in Modern Greek, the morphological infinitival has been lost. Tensed forms of verbs are therefore used instead of the infinitival; in such instances, they can be said to be “fake” tenses. For example, the citation form of verbs is tensed, with subject agreement. Therefore, it is not too surprising that ECM should be able to apply into a tensed subordinate clause that also exhibits predicate-subject agreement.7 In contrast, the tenses in Turkish verbal subordinate clauses with ECM are genuine. Turkish does have a morphological infinitive; therefore, tensed forms of the verb are not used in place of the infinitive, and are genuinely tensed. Thus, the fact that Nominative subjects are only possible in fully verbal subordinate clauses in the presence of verbal Agr, and that verbal genuine Tense forms do not license overt Nominative subjects when Agr is absent, is significant cross-linguistically, as well as for Turkish individually. Subject case in Turkish nominalized clauses 137 I shall make the further assumption that (a licensed) Agr must be in C so as to act as a Case licenser. That the level of CP is involved can be seen by the contrast between extractions out of ECM-clauses versus fully finite (in the sense of George and Kornfilt 1981) verbal clauses: (4) a.??/*[Ali-nin [sen -i ei yaz -dı ] Ali-GEN you -ACC write-PAST san -dı¤-ı ] mektupi believe -FN-3.SG letter Intended reading: ‘*the letter which Ali believes you to have written’8 b.[Ali -nin [sen ei yaz -dı -n ] Ali -GEN you write -PAST -2.SG san -dı¤-ı ] mektupi believe -FN-3.SG letter ‘the letter which Ali believes you wrote’ I now turn to another type of fully verbal subordinate clause, namely to subjunctives. 2.1.2. Subjunctives There is a predicate form in root clauses which is called the Optative or Subjunctive; I shall use the second form. This form takes different predicate-subject Agr forms than Indicatives; however, these forms are verbal, as well, and thus differ from nominal Agr forms. The subject is in the Nominative. These clauses are illustrated by the next example: (5) a.Ben bugün yemek piﬂir -e -yim I (NOM) today food cook -SUBJNCT -1.SG ‘I should/ought to cook food today; Let me cook food today.’ 138 Jaklin Kornfilt Just as we saw for Indicatives, Subjunctive clauses can also be embedded and show up in a form completely identical to a root clause: (5) b. [Ben bugün yemek piﬂir -e -yim ] I (NOM) today food cook -SUBJNCT -1.SG isti -yor -um want -PRSPROG -1.SG ‘I want to cook [that I should cook] food today; I want for myself to cook food today.’ The facts here are just as expected; whether in root or embedded clauses, the Agr form is verbal, and it licenses verbal subject Case, i.e. the Nominative, on a subject. In the next subsection, I turn to nominalized embedded clauses. In that subsection, I shall aim at establishing the same correlation between Agr and subject Case for such clauses that we saw in fully verbal clauses. Another aim of the discussion will be to establish a categorial difference between the two main types of embedded nominal clauses—a difference which I claim plays a central role in the licensing of nominal Agr as a subject Case licenser. More specifically, I will claim that while nominal subjunctive clauses are homogeneously nominal, nominal indicative clauses are categorially hybrid, with a nominal Agr sandwiched between a verbal TAM layer and a verbal CP (or Force Phrase) layer. Therefore, while nominal Agr is fully licensed within the fully nominal subjunctive clause, it is not so licensed within the hybrid indicative clause and therefore needs another licensing mechanism. Subject case in Turkish nominalized clauses 139 2.2. Non-tensed argument clause types 2.2.1. Similarities and differences Turkish has a few different “nominalization” types. For the purposes of this paper, the term “nominalization” is being used to refer to an extended clausal projection with some nominal functional layers that represent “nominalization” (for which diagrams will be shown later in the paper), and not to lexically derived deverbal nouns.9 I illustrate the two main types of syntactic nominalization here, namely the “factive” (i.e. indicative) and “non-factive” (i.e. subjunctive) types. “Factive” (indicative) nominalized embedded clause: (6) a.[Sen -in dün sabah ev -de yemek you -GEN yesterday morning home -LOC food piﬂir -di¤ -in ]-i duy -du -m cook -FN-2.SG-ACC hear -PAST -1.SG /san -dı -m /believe -PAST -1.SG ‘I heard/believed that you had been/were cooking/cooked /had cooked food at home yesterday morning.’ “Non-factive” (subjunctive) nominalized clause: (6) b.[Sen -in yar›n ev -de yemek you -GEN tomorrow home -LOC food piﬂir -me -n ]-i isti -yor -um cook -NFN10 -2.SG -ACC want -PRSPROG -1.SG ‘I want for you to cook food at home tomorrow; I want that you should cook food at home tomorrow.’ These two types of nominalized embedded clauses exhibit some similarities as well as differences. 140 Jaklin Kornfilt I start with the similarities: the subject in both types is in the Genitive, i.e. in what I have been calling the nominal subject Case. This is just as expected under the correlation I have posited here, because Agr is also nominal in both. Both types are ultimately, i.e. in their highest functional layer(s), DPs which need Case just like any DP. Such Case is licensed by a structurally higher Case licenser and is realized, in both clausal types, as the last morpheme in the morphological sequence of the nominalized predicate. A further point of similarity concerns the morphological sequence within the predicate. In both types, the “factive” and the “nonfactive” nominal morphemes appear in the morphological slot in which TAM morphemes show up in fully verbal clauses; this can be seen by comparing the examples presented so far for fully verbal versus nominal clauses. These nominal morphemes share not just the morphological slot, but also certain semantic properties with the corresponding TAM morphemes: mood properties like indicativity versus subjunctivity are similar (hence my terms of nominal indicative and subjunctive). Furthermore, as we shall see presently, the indicative nominal marker can also express a vestige of tense. Yet another property in which the two types of clauses are similar is in exhibiting the full argument structure of their respective predicates. From this point of view, these two types of nominalized clauses are not different from fully verbal clauses; similar adjuncts can show up, as well. I now turn to the differences between these two nominal clauses. The non-factive clauses are more nominal than the factive ones in a number of ways. I shall therefore be claiming in this paper that nonfactive, i.e. subjunctive nominal clauses are homogeneously DPs; factive, i.e. indicative nominal clauses, on the other hand, are, at the same time, CPs (or, in the terminology of Rizzi 1997, Force Phrases), i.e. they have at least one “high” functional layer with verbal features within a nominal functional projection, i.e. within a DP (in addition to a “low” verbal functional layer, i.e. the TAM layer). Subject case in Turkish nominalized clauses 141 1. I take Tense to be part of a verbal property. As we shall see presently, nominal indicatives can be overtly marked for future versus non-future tense (-DIK: non-future, -(y)AcAK: future 11). This verbal property is congruent with positing a higher CP-layer, which has verbal features, as well. In contrast, Subjunctive nominal clauses have only one marker, -mA, and are thus neutral for tense. The non-future nominal indicative was exemplified above; the next example illustrates the future tense nominal indicative: (6) c.[Sen -in ev -de yemek you -GEN home -LOC food piﬂir -ece¤ -in ]-i duy -du -m cook -FUTN -2.SG -ACC hear -PAST -1.SG ‘I heard that you will cook food at home.’ As the translation makes clear, the embedded nominal indicative clause is independent from the root clause with respect to tense. In other words, the embedded clause has its own tense features. This contrasts with embedded nominal subjunctive clauses; with respect to tense, these depend on the clause they are embedded under: (6) d.[Sen -in ev -de yemek you -GEN home -LOC food piﬂir -me -n ]-i isti -yor -um cook -NFN -2.SG -ACC want -PRSPROG -1.SG /iste -di -m /isti -yece¤ -im /want -PAST -1.SG /want -FUT -1.SG ‘I want/wanted/will want for you to cook food at home.’ These examples have shown us that nominalized indicatives have tense (albeit by far not as richly so as in fully verbal clauses) and have thus verbal properties, in contrast to nominal subjunctive clauses that lack tense completely and thus lack corresponding verbal properties. 142 Jaklin Kornfilt 2. In addition to lacking (functional) verbal properties, subjunctive nominals have certain nominal properties which are absent in nominal indicatives. Subjunctive nominalizations can, with varying degrees of success, be pluralized and can also co-occur with certain determiners, e.g. with demonstratives, while neither is possible with indicatives: (6) e.**[Hasan -›n bu durmadan kumarhane -ye Hasan -GEN this constantly casino -DAT kaç -t›k -lar -›n ] -ı escape -FN -PL -3.SG -ACC duy -ma -mıﬂ -tı -m hear -NEG -PERF -PAST -1.SG Intended reading: ‘I hadn’t heard (about) these constant runnings (away) of Hasan to the gambling casino.’ (6) f. ?(?)[Hasan-›n bu durmadan kumarhane -ye Hasan -GEN this constantly casino -DAT kaç -ma -lar -›n ] -dan escape -NFN -PL -3.SG -ABL hoﬂlan -m› -yor -um like -NEG -PRSPROG -1.SG ‘I don’t like these constant runnings (away) of Hasan to the gambling casino (i.e. that Hasan should run to the casino constantly).’ 3. There is very suggestive evidence showing that nominal indicatives are CPs, while nominal subjunctives are not: non-factive nominalized clauses cannot host WH-operators, i.e. they can neither act as embedded WH- or Yes/No-questions, nor can they function as modifying clauses in relative clause constructions. Factive nominalized clauses can be used in all of those functions, arguing that they are CPs (and thus have a Spec, CP position that can host an operator), albeit dominated by DP, while non-factive nominalized clauses are homogeneously DPs and consequently don’t have a qualifying Specifier position for the operators in question. 12 Subject case in Turkish nominalized clauses 143 The following examples illustrate the contrasting properties of nominal indicative versus subjunctive embedded clauses with respect to embedded (i.e. narrow-scope) WH-questions and with respect to relative clauses, in this order. (7) [yeme¤ -i kim -in piﬂir -di¤-in ]-i food -ACC who -GEN cook -FN-3.SG -ACC sor -du -m /duy -du -m ask -PAST -1.SG /hear -PAST -1.SG /söyle-di -m /tell -PAST -1.SG ‘I asked/heard/told who had cooked the food.’ (8) *[yeme¤ -i kim -in piﬂir -me -sin ]-i food -ACC who -GEN cook -NFN -3.SG -ACC söyle-di -m tell -PAST -1.SG Intended reading: ‘I said who should cook the food.’ The indicative nominal clause in (7), by virtue of being a CP, has a position in which a WH-operator is licensed: the Spec, CP position. Please note that in recent approaches (such as the one proposed in Rizzi 1997) in which the CP is layered further into a number of distinct functional projections, this position could plausibly be the Specifier position of a Force Phrase. What’s important here is that nominal indicative clauses would include such a projection, and thus its Specifier position, as a position which is qualified to host a WHoperator, while subjunctive nominal clauses, by virtue of not having such a functional projection, cannot host a WH-operator. (Please note that this statement will be generalized soon, so as to include any operator.) Note also that there is nothing wrong with the nominal subjunctive clause in combination with the matrix predicate in (8) per se; this is illustrated in the next example, where the corresponding declarative nominalized subjunctive is fine in the same syntactic context: 144 Jaklin Kornfilt (9) [yeme¤ -i Ali-nin piﬂir -me -sin ]-i food -ACC Ali -GEN cook -NFN -3.SG-ACC söyle -di -m tell -PAST -1.SG ‘I said that Ali should cook the food.’ The ungrammaticality of (8) is thus clearly due to the fact that the subjunctive clause is an embedded interrogative and that there is a question operator there for which the clause does not offer an appropriate position. It is instructive to observe that the desired reading in (8) can be expressed, but with some additional means—namely involving the indicative: it is necessary to embed the subjunctive under an appropriate nominal indicative clause; e.g.: (10) [[yeme¤ -i kim -in piﬂir -me -si ] food -ACC who -GEN cook -NFN -3.SG gerek -ti¤ -in ]-i söyle-di -m (be) necessary -FN-3.SG -ACC tell -PAST -1.SG ‘I said for whom it was necessary to cook the food.’ Now, the embedded interrogative has become larger, i.e. it is the nominal indicative clause that dominates the subjunctive clause. As we said earlier, the indicative clause does, by virtue of being a CP (or a Force Phrase) have the necessary specifier position for the interrogative operator, and the result is fine. Similar facts hold for Y/N questions, as well; I shall not illustrate those, due to constraints on space. Finally, it is interesting to note that the issue is not just one of licensing a [+WH] operator via [+WH] features of a qualifying functional head. This is because a similar dichotomy between nominal indicative versus subjunctive clauses can be observed with respect to relative clauses, as well. The operator in relative clauses is not [+WH]. Therefore, the issue is not merely one of [+WH] features being licensed by a particular functional head (or not being so licensed), but rather of having a functional projection whose Subject case in Turkish nominalized clauses 145 Specifier position is able to host any operator that can enter into an operator-variable relationship. The following examples illustrate the contrast between the two types of nominal embedded clauses with respect to relative clause constructions. I start with indicative RCs: (11) [Ali-nin ei piﬂir -di¤ -i ] yemeki Ali-GEN cook -FN-3.SG food ‘The food Ali cooked’ Subjunctive RCs don’t exist (with the exception of irrealis RCs, to be discussed later): (12) *[Ali -nin ei piﬂir -me -si ] yemeki Ali -GEN cook -NFN -3.SG food Intended reading: ‘The food Ali should cook’ Embedding the subjunctive RC under an appropriate indicative “saves” the utterance: (13) [[Ali-nin ei piﬂir -me -sin ] -i söyle -di¤-im ] Ali -GEN cook -NFN -3.SG -ACC tell -FN-1.SG yemeki food ‘The food which I said Ali should cook’ The explanation for the ungrammaticality of (12), as well as the reason for why embedding the ungrammatical subjunctive modifier clause under an indicative saves the utterance in (13) carry over from the discussion of embedded interrogatives—under the proposed extension from [+WH] operators to any operator, with the corresponding extension from a functional head with [+WH] features to one whose categorial features enable it to AGREE with any operator, i.e. an extension to a C-head or a Force-head. 146 Jaklin Kornfilt The following diagrams integrate the proposals made in the previous discussion, starting with a subjunctive nominalized clause. (14) a. KP g K' AgrP DPi ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ 5 5 @ Agr' 5 DPi ¡ ¡ ¡ ¡ ¡ MP 5 M' 4 VP 3 DPi V' ¡ 2 DP V ! ! @ g M ¡ ¡ ¡ ¡ K ¡ ¡ ¡ Agr ¡ (=Fin) ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ Ali nini ti ti kitab -› oku -ma -sın -ı Ali -GEN book -ACC read -NFN -3.SG -ACC ‘for Ali to read the book’ [=‘for Ali’s reading the book’] (as a direct object) (Adapted from Borsley and Kornfilt 2000:108) Subject case in Turkish nominalized clauses 147 The following is a rough representation for an indicative nominalized clause: (14) b. CP g C' KP g K' 4 K ¡ ¡ ¡ 5 ¡ AgrP C (=Force) ¡ ¡ DPi Agr' ¡ 5 ¡ ¡ MP Agr ¡ ¡ 5 (=Fin) ¡ ¡ DPi M' ¡ ¡ ¡ ¡ 4 ¡ ¡ ¡ ¡ VP M ¡ ¡ ¡ ¡ 3 ¡ ¡ ¡ ¡ ¡ DPi V' ¡ ¡ ¡ ¡ ¡ ¡ 2 ¡ ¡ ¡ DP V ¡ ¡ ¡ @ ! ! @ g Ali nini ti ti kitab -› oku du¤ -un -u Ali -GEN book -ACC read -FN -3.SG -ACC ‘(that) Ali read the book’ (as a direct object) 5 These rough representations are very similar, with the exception of a CP- (or Force-Phrase)-layer between the AgrP (or Finiteness Phrase) and the KP, the highest layer of the nominal clause (i.e. a Case Phrase) in the representation for the nominal indicative clause 148 Jaklin Kornfilt —a layer which is missing in the representation of the fully nominal subjunctive clause, as discussed. As a consequence, the nominal Agr finds itself between two verbal phrasal layers in the nominal indicative clause, while it is surrounded by fully nominal layers in the subjunctive clause. The significance of this difference for the proposed account of subject Case will be central for my account; a discussion of this significance will be initiated in section 3. Is it a coincidence that there is a CP (or a Force Phrase) in a clause where there are also Tense features? The answer is no. As mentioned in the earlier discussion, Tense features are verbal, and so are the features of the CP (or of the Force Phrase); thus, there is a categorial agreement between these layers. This analysis predicts that we would not get lack of Tense if a CP-layer is present which exhibits CP-related syntactic properties. This prediction is indeed fulfilled in Turkish, as we saw. I expect it also to hold cross-linguistically, and, as far as I know, it does. 2.2.2. The importance of nominal Agr for genuine subject Case and infinitival clauses I now turn to the importance of nominal Agr in licensing subject Case. A subset of matrix predicates that subcategorize for subjunctive argument clauses also co-occur with infinitival argument clauses. Such clauses share with the previously illustrated nominalized clauses the property of being Case-marked: (15) Beni [PROi karanl›k -ta sokak -lar -da I darkness -LOC street -PL -LOC yürü -mek ]-ten kork -ar -›m walk-INF -ABL fear -AOR -1.SG ‘I am afraid to walk in the streets in the dark.’ Subject case in Turkish nominalized clauses 149 Such infinitival clauses cannot bear overt Agr markers. Note that there is no such marker between the infinitival marker and the Case marker on the clause. Overt subjects are not possible in infinitivals, no matter what their Case is: (16) *Ben [k›z -›m /k›z -›m -›n I daughter -1.SG[NOM] /daughter-1.SG -GEN karanl›k-ta sokak -lar -da yürü -mek] -ten darkness -LOC street -PL -LOC walk -INF -ABL kork -ar -›m fear -AOR -1.SG Intended reading: ‘I am afraid for my daughter to walk in the streets in the dark.’ I claim here that these two observations are linked to each other; in other words, infinitival clauses have no Agr, and it is therefore that the only possible subject in such clauses is PRO. (This statement will be refined later.) For an utterance like (16) to be grammatical with an overt subject, the embedded predicate must be marked with the non-factive nominalization marker, instead of with the infinitive (thus preserving subjunctive Mood), and it must also bear overt Agr morphology: (17) Ben [k›z -›m -›n karanl›k -ta I daughter-1.SG -GEN darkness -LOC sokak -lar -da yürü -me -sin ]-den street -PL -LOC walk -NFN -3.SG -ABL kork -ar -›m fear -AOR -1.SG ‘I am afraid for my daughter to walk in the streets in the dark.’ Again, the observations are linked: Where there is Agr, there is also an overt subject, and that subject appears in the appropriate genuine subject Case. This is the Genitive, i.e. the nominal subject Case, because the Agr itself has nominal features. PRO cannot show 150 Jaklin Kornfilt up in the presence of Agr. For the purposes of this paper, I shall not be concerned with the specifics of why this is not possible; the two main types of answers would be either that this is due to some implementation of the original PRO-Theorem (cf. Chomsky 1981), or else to an inappropriate Case being licensed for the subject position if that position is occupied by PRO (cf. Chomsky and Lasnik 1991). Either way, the presence of Agr would preclude the presence of a PRO-subject. We thus explain the two correlations we have observed in this discussion of nominal clauses: Correlation A: when there is infinitival morphology, there is no Agr; no overt subject possible (because no Case of any type—or a Case of an inappropriate type—is licensed); the only possible subject is PRO. Correlation B: when there is instead nominal subjunctive morphology (which has the same Mood as the infinitival), there also is overt (nominal) Agr; now, an overt subject with nominal subject Case (i.e. Genitive) shows up; no PRO subject is possible. We are now able to collapse the correlations we have set up for verbal and for nominal clauses into one overall correlation: For both nominal and fully verbal clauses: where overt Agr shows up, the overt subject is licensed via the corresponding (i.e. nominal or verbal) subject Case (i.e. Genitive or Nominative, respectively), depending on the nominal versus verbal features of the Agr. Without Agr, no genuine subject Case of any sort is possible. Having thus concluded a preliminary discussion of verbal as well as nominal argument clauses, I turn to adjunct clauses. 2.3. Adjunct clauses 2.3.1. Indicative adjunct clauses with nominal Agr Both the nominal indicative and the nominal subjunctive clause types which we discussed as argument clauses can also appear as adjuncts. Those are usually complements of postpositions, but they can also Subject case in Turkish nominalized clauses 151 occur without a postposition. The two asymmetries mentioned in the introduction surface when we compare these with the corresponding argument clauses just discussed. I start the discussion of adjunct clauses with examples of nominal indicative clauses as objects of postpositions, followed by examples of the same type of clause without any postposition, but still as adjuncts. Comparison of these adjunct nominal indicatives (whether with or without postpositions) with their argumental counterparts will establish the first asymmetry. The main property to notice about these adjunct nominal indicatives is that their subjects are not in the Genitive, as expected, but rather in the Nominative—or, if we don’t want to prejudge the issue at this point, we can say that the subjects are bare. The first three examples illustrate this for nominal indicatives that are postpositional objects, and the latter two exhibit the same fact for the same clause type, used as adjuncts, but without a postposition: (18) [[Sen yemek piﬂir -di¤-in ] için ] ben you(SG) (NOM) food cook -FN-2.SG because I konser -e gid -ebil -di -m concert-DAT go -ABIL -PAST -1.SG ‘Because you cooked, I was able to go to the concert.’ (19) [[Sen yemek piﬂir -di¤-in ] -e you(SG) (NOM) food cook -FN-2.SG -DAT göre ] hepiniz ev -de kal -acak -s›n›z according to all+you home -LOC stay -FUT -2.PL ‘Given that you cooked, all of you will stay at home.’ (20) [[Ben yemek piﬂir -di¤-im ] -den dolay›] I (NOM) food cook -FN-1.SG -ABL because konser -e gid -e -me -di -m concert -DAT go -NegABIL -NEG -PAST -1.SG ‘Because I cooked, I was unable to go to the concert.’ 152 Jaklin Kornfilt (21) [[Ben yemek piﬂir -di¤-im ] -den konser -e I (NOM) food cook -FN-1.SG -ABL concert-DAT gid -e -me -di -m go -NegABIL -NEG -PAST -1.SG ‘Because I cooked, I was unable to go to the concert.’ (22) [Sen konser -e git -ti¤ -in ] -de ben you(SG) (NOM) concert-DAT go -FN-2.SG -LOC I ev -e dön -üyor -du -m home -DAT return -PROG -PAST -1.SG ‘When you were going to the concert (at your going to the concert), I returned home.’ This appears to be a problem for the account I proposed so far. Note that in all of these examples, the nominal indicative clause does include a nominal Agr marker, and that all of these clauses do carry some kind of θ-role, albeit an adjunct θ-role. Thus, one would assume that some sort of thematic index would be assigned to the clause and be inherited by the nominal Agr, thus enabling it to license subject Case. However, this would incorrectly predict Genitive subjects here, as this would be the licensed subject Case. I would like to claim that this problem is only apparent. I shall propose in this paper that the subjects of adjunct indicative clauses are in a default Case (and not in a genuine, licensed subject Case), because the Agr element is not licensed by a primary θ-index. I follow Grimshaw (1990) in distinguishing primary from secondary θ-roles—a distinction which directly corresponds to the one in Rizzi (1994) between referential and non-referential indexation, which is a distinction drawn between arguments and adjuncts. In line with this distinction, I refine the proposal I made earlier about licensing a nominal Agr which is not categorially licensed. I had proposed that such an Agr needs to carry a θ-index to be licensed as a subject Case licenser. I now constrain that proposal: this θ-index must be that of a primary θ-role in Grimshaw’s sense, i.e. a “referential” index in Rizzi’s sense. While any θ-index might be sufficient to license Agr as a subject Case marker in some other languages, it is clear that for Subject case in Turkish nominalized clauses 153 Turkish, there must be such a constraint imposed on the type of indexation.13 I shall return to the issue of default Case in subsection 2.3.3., by showing that it applies to adjunct clauses without any Agr element, as well, thus providing independent motivation for this mechanism in the grammar. An account of my overall approach to subject Case will be offered in section 3. The present subsection has illustrated the first asymmetry mentioned in the introduction; the next subsection illustrates the second asymmetry, i.e. the fact that the first asymmetry is found only in the categorially hybrid nominal indicatives but not in the fully nominal subjunctives. 2.3.2. Subjunctive adjunct clauses with nominal Agr Subjunctive nominal adjunct clauses contrast with indicative nominalized adjunct clauses with respect to their subjects: those subjects are in the Genitive, just as they are in corresponding argument clauses: (23) [[Sen -in yemek piﬂir -me -n ] için ] ben you(SG) -GEN food cook -NFN -2.SG for I ev -de kal-dı -m house -LOC go -PAST -1.SG ‘I stayed at home so that you should cook (for you to cook).’ The Genitive subject here contrasts with the corresponding Nominative subjects in the examples (18) through (22), where nominal indicatives were exemplified as adjuncts. The contrast between (23) and (18) is particularly instructive in this regard, as the same postposition, i.e. için, shows up in both (albeit with different semantics, due to the different factivity and Mood differences between the embedded clauses); the nominal Agr morphology is the same in both, as well. Yet, the subject of the embedded indicative clause in (18) is in the Nominative (i.e. default) case, while the 154 Jaklin Kornfilt subject of the embedded subjunctive clause is in the Genitive (i.e. genuine nominal subject) case in (23). I propose that the reason for this contrast is the one mentioned earlier, e.g. in the introduction: the nominal Agr element is licensed via the categorially matching nominal features within its own clause in nominal subjunctive clauses such as in (23). I have argued in subsection 2.1.2. that these clauses are indeed homogeneously nominal, and that especially their TAM morphology is [+N, -V], thus “harmonizing” with the corresponding feature values of the nominal Agr morphology. As a consequence, the nominal Agr in nominal subjunctives always licenses genuine (nominal) subject Case, i.e. Genitive, irrespective of the argument or adjunct status of the clause. In other words, an Agr licensed categorially within its clause does not need further licensing via any sort of indexation, thematic or otherwise; this is why nominal subjunctive clauses don’t exhibit sensitivity to the adjunct versus argument distinction with respect to subject Case. In contrast, Agr is not categorially licensed clause-internally in nominalized indicative clauses. This is why it needs licensing via indexation, as discussed in the previous subsection, and why nominal indicative clauses show sensitivity to their adjunct versus argument status. There is some independent evidence for my proposal that the lack of sensitivity observed for nominal subjunctive clauses with respect to the argument/adjunct asymmetry (due to the ability of its Agr to license genuine nominal subject case irrespective to that asymmetry) is made possible by the homogeneously nominal categorial features of the domain headed by such nominal Agr. This evidence comes from regular possessive phrases. Possessive phrases are similar to nominalized clauses with respect to the nominal Agr morphology on the head, i.e. on the nominal with the semantics of “possessed”. Interestingly, the specifier of such phrases (i.e. the nominal with the semantics of “possessor”), which is the nominal that corresponds to the subject of nominal phrases, bears Genitive marking, irrespective of the argument or adjunct status of the entire possessive phrase. In the Subject case in Turkish nominalized clauses 155 following pair of examples, the possessive phrase is an argument in (24), and an adjunct in (25). Please note that the possessor, Ali, is marked with the Genitive in both instances: (24) Hasan [Ali -nin kitab-ın ] -ı Hasan Ali -GEN book-3.SG -ACC ‘Hasan read Ali’s book.’ oku -du read -PAST (25) Hasan kitab -ı [[Ali -nin kız -ı ] için] Hasan book -ACC Ali -GEN daughter -3.SG for al -dı buy -PAST ‘Hasan bought the book for Ali’s daughter.’ This fact is exactly as predicted by the approach outlined above. The nominal Agr morphology is licensed by the nominal features within its own phrase; obviously, the nominal head of a possessive phrase is unambiguously and fully nominal. Therefore, such Agr morphology does not need any other licensing and is thus insensitive to any indexing that would originate from a thematic or predicational indexer. In the following section, I discuss adjunct clauses without an overt Agr element and show that such clauses independently motivate the assumption of a default Case mechanism for subjects. 2.3.3. Adjunct clauses without any Agr Typically, adjunct clauses that lack overt Agr morphology on their predicates also don’t have an overt subject; these embedded subjects have the properties of PRO. 14 Note that in the following two examples with PRO-subjects, the subject of the adjunct clause takes on the reference of the overt subject in the main clause obligatorily: 156 Jaklin Kornfilt (26) a.Oyai dün bütün gün çal›ﬂ -t›. Oya yesterday all day work -PAST [PROj/*i makale-yi yaz -ar -ken ] article -ACC write -PRES.PART -‘while’ Ahmetj ›sl›k çal -›yor -du Ahmet whistle play -PROG -PAST ‘While writing the article, Ahmet was whistling.’ (The only person writing the article can be Ahmet, even though Oya was mentioned in the discourse, and would be pragmatically the likelier antecedent for PRO in this discourse.) b. [PROi makale-yi yaz -ar -ken ] article -ACC write -PRES.PART -‘while’ beni ›sl›k çal -aca¤ -›m I whistle play -FUT -1.SG ‘While (I will be) writing the article, I will be whistling.’ In contrast, the pro-subject in corresponding clauses with overt Agr morphology may also take on the reference of other antecedents: (27) Oyai çok özverili bir insan.[[proi/j dün yemek Oya very selfless a person yesterday food piﬂir -di¤-i ] için ] Ahmetj konser -e cook-FN-3.SG because Ahmet concert-DAT gid -ebil -di go -ABIL -PAST ‘Oya is a very selfless person. Because she/he cooked yesterday, Ahmet was able to go to the concert.’ Although syntactically, Ahmet is a closer antecedent to the prosubject, the discourse-antecedent Oya is pragmatically the likelier antecedent and is thus the preferred indexer. (Without the first sentence about Oya, the indexer of pro is, of course, Ahmet.) For our purposes, these sentences establish the difference between PRO- and pro-subjects. Note, at the same time, that PRO-subjects co-occur with predicates that lack overt Agr, while pro-subjects need Subject case in Turkish nominalized clauses 157 overt Agr to be licensed and identified. (Kornfilt 1996b discusses additional criteria to distinguish these two empty categories in Turkish.) An interesting observation about Agr-less adjunct clauses is that an overt subject can show up in the position of PRO in such examples: (28) [Meral makale -yi yaz -ar -ken ] ben ›sl›k Meral article -ACC write-AOR -‘while’ I whistle çal -›yor -du -m play -PROG -PAST -1.SG ‘While Meral (was) writing the article, I was whistling.’ Similar facts are found with other morphologies found in such tenseless adjunct clauses, as well: (29) [sen konser -e gid -ince ] ben ev -e you(SG) concert-DAT go -‘when’ I home -DAT dön -dü -m return -PAST -1.SG ‘When you went to the concert, I returned home.’ (30) [sen konser -e gid -eli ] beﬂ saat ol -du you(SG) concert-DAT go -‘since’ five hour be -PAST ‘It’s been five hours since you went to the concert.’ The factive nominalization morpheme -DIK can, albeit infrequently, also serve as part of the predicate of such an adjunct clause with either a PRO-subject or an overt subject, when followed (in its Agr-less form) by the locative morpheme -DA: (31) a.[PRO her gel -dik-te ] Ali ben -im -le each come -FN-LOCAli I -GEN -with kavga ed -er quarrel do -AOR ‘Ali quarrels with me every time [he] come[s].’ 158 Jaklin Kornfilt b. [sen her gel -dik -te ] Ali ben -im -le you(SG) each come -FN-LOC Ali I -GEN -with kavga ed -er quarrel do -AOR ‘Ali quarrels with me every time you come.’ (Adapted from Lewis 1967: 183) The interesting question that arises here is: how can it be that both PRO-subjects and overt subjects are possible here? We saw earlier that Agr-less infinitival clauses that are arguments (rather than adjuncts) allow only PRO-subjects, while nominalized clauses with overt Agr allow only overt (or pro-) subjects. In other words, for argument clauses, PRO- and overt subjects are in complementary distribution. However, this complementary distribution obviously breaks down for adjunct clauses. In other words, PRO- and overt subjects are in free variation for Agr-less adjunct clauses. In order to explain this observation, we have to address the basic issue of how overt subjects receive Case in these adjunct clauses that lack overt Agr. My proposal is that such Case is due to a mechanism of default Case that applies as a last resort. In other words, when no Case licenser is available for an overt subject, this last resort mechanism applies. Note that in these Agr-less adjunct clauses, there indeed is no Case licenser for an overt subject: there is neither overt Agr as such a licenser, nor Tense (even if we had not ruled out Tense previously in such capacity), as these clauses have no independent tense and take on the tense interpretation of the root clause. I shall come back to the issue of default Case; at present, it is sufficient to say that such Agr-less clauses with overt subjects that do not bear genuine, licensed subject Case establish the necessity of default Case in Turkish. Given that the grammar of the language needs this mechanism, default Case can also be appealed to when accounting for the overt subjects in categorially hybrid clauses that are adjuncts and which do have overt Agr. The common denominator of both types of clauses, i.e. hybrid clauses with and without overt Agr, is that they are adjuncts, i.e. that they lack primary θ−roles. Subject case in Turkish nominalized clauses 159 Thus, I propose that there is a correlation between that lack and the necessity of default subject Case in both types of clauses. I now turn to an overall account of licensed subject Case in Turkish. 3. An account of subject Case 3.1. A sketch of a proposal, and one previous proposal In the previous discussion, I have proposed that: 1. there is one single licensed subject Case (with the possibility of default Case when there is no licenser available; necessary constraints on default Case shall be discussed later); I further claimed that this single subject Case may have different morphological realizations (in Turkish, those would be Nominative and Genitive); 2. that the licenser of the genuine, licensed subject Case is the overt Agr morphology, and that 3. the Agr morphology has to be licensed itself in order to function as a licit subject Case licenser. The correlations between overt Agr of both categorial types and the corresponding subject Case pointed out so far show us that the first two claims above are convincing. What about the third claim? In the Minimalist Program, elements that don’t have semantic features (“interpretable features”), or else which have them redundantly, are imperfections (cf. Chomsky 2002). Agr is one such element, since whatever semantic features it has are already contained in the co-indexed subject DP. The importance of such an element—and of the syntactic position AGR in which the Agr morpheme is situated—has therefore to be motivated. This is what I have tried to establish in this paper so far— i.e. I have ruled out alternative sources for licensed subject Case, thus motivating the existence of Agr as a subject Case licenser. In the same spirit as in the Minimalist Program, I would argue that because an entity like AGR (and its corresponding morphology, i.e. Agr) with its uninterpretable features is undesirable in general, its existence must be licensed all the more, when its categorial features 160 Jaklin Kornfilt are in conflict with those of its syntactic environment. Thus, I have proposed a primary source of licensing via matching categorial features; where there is categorial mismatch, I have proposed licensing via referential indexing (in Rizzi’s sense). This, in turn, is made possible either via thematic indexing, as we have seen so far, or via predicational indexing, as we shall see later in the paper. The idea that an Agr element, categorially mismatched locally within its own clause, needs to be licensed from the outside, has a predecessor, albeit not an identical one. Raposo (1987) proposes for inflected infinitives in European Portuguese (EP) the following generalization: nominal Agr needs Case itself. I offer here one citation to this effect: “...Agreement (Agr) in [the inflected infinitive’s] Infl node must be Case-marked, if it is to assign nominative Case to the subject of its clause.” (Raposo 1987: 85.) Raposo mentions the nominal nature of Agr in these instances as a motivating factor for his claim, just presented, that such Agr has to be “Case-marked” from outside in order to have its own “Caseassignment” potential be activated. But why should a nominal Agr be any less able to license subject Case than a verbal Agr? Raposo analyzes inflected infinitives in EP as CPs; in such a syntactic domain, a nominal Agr would indeed cause conflict of categorial features and thus would need licensing itself, if we look at the EP facts from the perspective of the approach I have suggested. The licensing of the nominal Agr, Raposo proposes, takes place through the Case on the whole infinitival clause, and via subsequent percolation of that Case down to the nominal Agr that heads the inflected infinitival, and which, in Raposo’s account, has risen to the C-head of the CP. It is important to note that the Case “assigned” to the overt subject in these EP inflected infinitives is not the same as the Case on the “switched on” Agr in most instances: while the overt subject bears subject Case (in EP, this is the Nominative, despite the syntactically nominal nature of Agr15), the external Case on the CP (and thus on its Agr-head) is, in most instances, the Accusative (abstracting away Subject case in Turkish nominalized clauses 161 from inflected infinitives as sentential subjects); this is illustrated by the following example: (32) Eu lamento [os deputados ter -em trabalhado pouco] ‘I regret the deputies to-have-Agr worked little.’ (Raposo 1987: 87, his [7]a.) In other words, the subject Case licensing mechanism proposed for EP by Raposo is not Case transmission, as suggested for some English gerunds in Reuland (1983). Instead, we have here an Agr, licensed by any Case, in turn licensing subject Case. Given that no Case transmission takes place here, the question arises as to why the licensing of Agr as a Case licenser should be due to an external Case, and whether other ways of such external licensing of Agr might be conceivable. In the following sections, I shall show that at least for Turkish, external Case as the factor that activates an Agr element (i.e. an Agr not activated internally within a clause or a phrase) does not work. Therefore, a different external factor is needed, and I have proposed indexation by a primary θ−role (and shall add indexation by predication) as such a factor. Note, however, that Raposo’s and my proposals are motivated by a similar way of thinking: where Agr is not “legitimate” within its own local domain, it must get legitimized by virtue of heading a syntactic domain which, in turn, is a necessary, even obligatory, constituent in its own domain. In Turkish, such a constituent would be an argument of a verb or of a noun, as opposed to an adjunct of a verb. We shall see that adjunct clauses of nouns, i.e. modifier clauses in relative clause (RC) constructions (as opposed to adjuncts of verbs) are also treated as “necessary” constituents in Turkish. This might be a parametric dimension along which languages might differ and whose investigation I leave for future research. Returning to a formalization of the general ideas just discussed, the “necessary” constituents are, I propose, indexed: either by a (primary) θ−role for argument clauses, or by a predicational index, 162 Jaklin Kornfilt for the modifying clauses in RCs. This index percolates down to the Agr-head, thus activating it as a subject Case licenser. 16 I now turn to showing that in Turkish, it isn’t the Case on the Agr that activates it as a subject Case licenser. 3.2 Problems with licensing of Turkish Agr via Case Three types of problems have to be acknowledged that a Case-based account of the kind proposed by Raposo for EP would have to face with respect to subject Case in Turkish: 3.2.1. Instances of licensed nominal Agr without structural Case Turkish categorially hybrid indicative clauses appear in nouncomplement constructions: (33) [Ali-nin i [proi aile -sin ] -i Ali-GEN family -3.SG -ACC terket -ti¤ -i ] söylenti -si abandon-FN-3.SG rumor -CMPM ‘the rumor that Ali abandoned his family’ A noun does not check the Case of its complement (or at least not structural Case); in this respect, it is different from either verbs or adpositions. This is also shown here by the fact that there is no overt Case on the nominalized complement clause of the noun. But, just like a verb, a noun assigns primary θ−roles. This explains the Genitive on the subject of the categorially hybrid clausal Ncomplements. At the same time, such examples motivate the approach to licensed subject Case proposed here, i.e. one as based on indexation via primary θ−roles, and against an approach based on licensing of Agr via Case on that element. Subject case in Turkish nominalized clauses 3.2.2. 163 Instances of licensed verbal Agr without structural Case Further motivation for Agr as the subject Case licenser, as well as for its own licensing via θ−role-based indexation, is offered by the existence of tensed complement clauses of nouns: (34) [[beni [proi aile -m ] -i I [NOM] family -1.SG -ACC terket -ti -m ] söylenti -si ] abandon -PAST-1.SG rumor -CMPM ‘the rumor that I abandoned my family’ Given the fully verbal, tensed nature of the noun-complement clause, it is clear that neither the clause nor its (verbal) Agr-head receive any Case from the external noun. Nevertheless, the overt subject of that clause has subject Case, namely Nominative. This shows that the local, verbal Agr element is licensed as a subject Case licenser; given the verbal nature of that local Agr, the appropriate subject Case is the Nominative, and this is what we find here. These facts about noun-complement clauses are problematic for any account like Raposo’s, where Agr as a licenser is activated by external Case on that Agr. 3.2.3. Instances of unlicensed nominal Agr with adpositional Case Another type of problem is posed by the existence of categorially hybrid clauses as complements of postpositions. Such examples were discussed previously, in section 2.3.1., where we saw that in such constructions, the subject does not receive the expected subject Case, despite the existence of an Agr element. (See examples [18] through [22] in that section.) Note that in such examples, postpositions do assign Case to their complement clauses, and thus to the Agr element heading such clauses. This is especially obvious in examples (19) through (22), where the clause bears overt Case, irrespective of the presence or absence of an overt postposition. Thus, (clause-)external 164 Jaklin Kornfilt Case on Agr and the clause that it heads does not license that Agr as a subject Case marker, making default Case necessary. On the other hand, the common denominator between all the examples in 2.3.1. (i.e. [18] through [22]) is the fact that the categorially hybrid clauses do not bear any primary θ−role. I therefore submit that the account I have proposed here is corroborated by these examples, while a Case-based account is refuted by them. 4. An additional subject Case licensing mechanism: Indexation by predication in headed operator-variable constructions 4.1. Overtly headed relative clauses Agr has an additional option of receiving a “referential” index (cf. Rizzi 1994), if θ-role assignment is not an option. This is through predication. Relevant examples are relative clauses (overtly headed as well as Free Relatives) and comparatives (which, formally, are similar to Free Relatives in Turkish). In order to show that a (somewhat) separate treatment of RCs is necessary, I would first like to demonstrate that the structure of RCs is different from that of noun-complement constructions just discussed. First, I would like to show that the modifying clause in RCs is an adjunct of the head, not a complement of the head: (35) [Ali-nin geçen gün dükkân -dan al -d›¤-› ] bu Ali -GEN past day shop -ABL buy -FN-3.SG this ﬂahane vazo magnificent vase ‘this magnificent vase which Ali bought at the store the other day’ Subject case in Turkish nominalized clauses 165 Note that the modifying clause precedes the demonstrative, and compare this property to a corresponding clause as a complement of a head noun: (36) ﬂu [[Ali -nini [proi aile -sin ] -i that Ali -GEN family -3.SG -ACC terket -ti¤ -i ] söylenti -si ] abandon -FN -3.SG rumor -CMPM ‘that rumor that Ali abandoned his family’ Here, the clause follows the demonstrative. Note that the order found in RCs is not possible in noun-complement constructions: (37) *[Ali -nini [proi aile -sin ] -i Ali -GEN family -3.SG -ACC terket -ti¤ -i ] ﬂu söylenti -si abandon -FN-3.SG that rumor -CMPM Intended reading: ‘that rumor that Ali abandoned his family’ This shows that the modifying clause in RCs is merged higher than the corresponding clause in noun-complement constructions. Thus, these examples motivate the analysis of the noun-complement clauses in a way appropriate to their label, i.e. the clause is the complement of the noun and is therefore merged closer to that noun as compared to the modifying clause in RCs which is an adjunct rather than a complement of the head noun and is therefore merged higher in the structure. But if the modifying clause in RCs is not a complement of the head noun, then it also does not receive a primary θ−role from that noun, and thus the Agr heading the clause does not receive an appropriate index. Yet, the subject of the clause does show up in the appropriate subject Case, i.e. it is in the Genitive. This means that the Agr is indexed, after all, but not via a marking based on a primary θ−role. Instead, I suggest that the indexation needed is achieved via a predication relation between the modifying clause and the head noun. Early mention of such a predication rule in RCs and in Left 166 Jaklin Kornfilt Dislocation constructions can be found in Chomsky (1977), where such a rule is taken to express a general notion of “aboutness” (cf. Chomsky 1977: 81). The same predication relation, I suggest, holds in the following two examples: (38) a. the [ sad ] man b.the man [who[ is sad]] In turn, this predication relation is, in a sense, similar to that we find within a clause between a predicate, i.e. a VP (or an I’) and the subject: (39) a. The man [coughed] b.The man [is sad ] There is some recent work in which at least some instantiations of this kind of predication is related to θ−role assignment, as in Williams (1994). There, predication between a predicate and a subject as well as predication between a nominal head and its adjectival modifier is taken to involve θ−role assignment. It is interesting to see an approach where the relationship between a modifier (i.e. an adjunct) of a nominal head and that head (as in [38]a.) is viewed as one involving θ−role assignment. If this view is on the right track, then we have exactly the type of natural class of constructions that we have been aiming for in this paper: complements of verbs and of nouns, along with (for Williams, only adjectival) adjuncts of nouns, to the exclusion of adjuncts of verbs. The task that remains is to also introduce relative clause modifiers into that natural class. Williams (1994) excludes them, for reasons that it would take us too far afield to discuss here. I would like to suggest that modifier clauses of nominal heads should have the same relationship to those heads as modifier adjectives; in other words, if there is a predication relationship between the modifier and the head in (38)a. i.e. a relationship based on θ−role assignment, in accordance with the suggestions in Williams 1994, then the same Subject case in Turkish nominalized clauses 167 relationship should also hold between the modifying clause and the head in (38)b. As a matter of fact, the traditional labelling of such modifier clauses as adjective clauses goes along with this idea. The similarity between these two kinds of noun modification is even more obvious in other languages, Turkish being one of them: (40) a. [üzgün] adam sad man ‘the sad man’ b. [[ei üzgün ol -an ] Opi] adami sad be -REL.PART man ‘the man who is sad’ If my suggestion is correct, there is predication between modifier and nominal head in both of (40) a. and b., and in both of them (following Williams 1994 at least in spirit), this predication is based on θ −role assignment. I further suggest that this sort of θ−role assignment is “primary” in a sense similar to Rizzi’s primary θ−roles, because it restricts the reference of the DP-head; therefore, the indexation that encodes the predication relationship is a “referential” one. This is where we find the ultimate similarity between those instances where an argument clause receives a “referential” index, namely via a primary θ−role, and those instances where an adjunct clause likewise receives a referential index via predication, the latter also based on (primary) θ−role assignment between head and modifier. The examples of RCs just discussed have subject targets and thus don’t possess subjects in need of Case. But note that once we insure such indexation on the modifier clause in an RC, the account developed here for subject Case applies to those RCs that do have a subject, i.e. in RCs whose target is a non-subject, as the one in (35), repeated here for convenience as (41): 168 Jaklin Kornfilt (41) [Ali-nin geçen gün dükkân -dan al -d›¤ -› ]j Ali-GEN past day shop -ABL buy -FN17 -3.SG [bu ﬂahane vazo]j this magnificent vase ‘this magnificent vase which Ali bought at the store the other day’ I have encoded the predication relation at issue via indexation. The index on the modifying clause would, as outlined earlier for argument clauses, percolate down to the (nominal) Agr element which is not licensed internal to the clause, as it conflicts with the “verbal” features of the predicate in this categorially hybrid clause. Agr, now licensed via its predication index, licenses the appropriate subject Case on the subject; due to the nominal nature of this Agr, subject Case is realized as the Genitive. A similar account would hold for comparative constructions whose head would be an overt or covert quantificational phrase. Before turning to those and to Free Relatives, I would like to point out that indexation between modifying clause and nominal head is a necessary but not sufficient condition for genuine, licensed subject Case to be realized; the presence of overt Agr is crucial. To see this, I show examples of nominalized irrealis relative clauses without an Agr marker; relative clauses with nominal future tense morphology are such an example. It is important to note that in such instances, no overt subject is possible any longer; the only possible subject is PRO: (42) a.[[[PRO san -a ei ver -ecek] Opi] bir you(sg.)-DAT give -FUTN a vazo] bul -a -m› -yor -um vase find -NegABIL -NEG -PROG -1.SG ‘I am unable to find a vase to give you.’ Subject case in Turkish nominalized clauses 169 b. *[[[ben /ben -im san -a ei I [NOM] / I -GEN you(sg.) -DAT ver -ecek] Opi] bir vazo] give -FUTN a vase bul -a -m› -yor -um find-NegABIL -NEG -PROG -1.SG Intended reading: ‘I am unable to find a vase for me to give you.’ Whether the overt subject is in the Nominative or in the Genitive, the last example remains ungrammatical in the absence of Agr. The ungrammatical example can be rescued by having the future tense in the categorially hybrid modifying clause be followed by nominal Agr: (43) [[[Ben-im san -a ei ver -ece¤ -im ] Opi] I -GEN you(sg.) -DAT give -FUTN-1.SG bir vazo] -yu dün bul -du -m a vase -ACC yesterday find-PAST -1.SG ‘I found yesterday a vase I’ll give you.’ These sets of examples offer a clear illustration of the significant role of Agr in the licensing of overt subjects, and we find the same correlation in relative clauses between Agr and overt subjects with licensed subject Case as we found in argument hybrid clauses; when Agr is present, we find overt subjects with licensed subject Case; when Agr is absent, only PRO is possible as a subject. 4.2. Indicative adjunct clauses with Genitive subjects: Free relatives and comparatives Note that nominalized indicative clausal postpositional complements do exist whose subjects are carriers of Genitive, i.e. of the genuine nominal subject Case: 170 Jaklin Kornfilt (44) [[Ayﬂe -nin duy -du¤ -un ] -a göre ] Sare Ayﬂe-GEN hear -FN -3.SG -DAT according to Sare deprem -de vefat et -miﬂ earthquake -LOC death do -REP.PAST ‘According to what Ayﬂe heard, Sare died in the earthquake.’ (45) Piyanist bu parça -y› [[Pollini -nin pianist this piece -ACC Pollini -GEN göster -di¤ -i ] gibi] çal -d› show -FN -3.SG like play -PAST ‘The pianist played this piece like Pollini showed (i.e. in the way in which P. showed it to be played).’ (46) Ali [[baba -s›n -›n iste -di¤ -i ] kadar ] Ali father -3.SG -GEN want -FN -3.SG as-much-as baﬂar› -l› ol -a -ma -m›ﬂ success -with become -NegABIL -NEG -REP.PAST ‘(It is said that) Ali wasn’t able to become as successful as his father wanted.’ Note that the subjects of the nominalized factive clausal complements of the postpositions is in the Genitive rather than in the default Nominative Case, in contrast with what we saw in (apparently) similar adjunct clauses in section 2.3.1. What is the difference? All of the postpositions in the last three examples have either comparative semantics, or else the construction can be interpreted as a (free) RC. More specifically, I suggest that (44) and (45) are Free Relatives (FRs), while (46) is a comparative construction. Among a number of competing analyses for comparatives, one widely accepted analysis has been to view comparative constructions as involving an operator, in a sense similar to relative clauses (cf. Bresnan 1973 and 1975; for an account of Turkish comparatives along these lines, cf. Knecht 1976). The translations of these last three examples are suggestive: (44), a Free Relative: ‘According to what (i.e. on the basis of the things that) Ayﬂe heard, ...’; (45), another Free Relative: ‘The pianist played this piece like the way Subject case in Turkish nominalized clauses 171 which Pollini showed’; (46), a comparative construction: ‘Ali wasn’t successful as much as, i.e. to the extent that his father wished’. Note also that similar facts hold for other comparatives: -DAn fazla ‘more than’, -DAn az ‘less than’ etc.; due to space limitations I shall not illustrate those. It is particularly interesting to compare (44) with (19), since the same postposition (göre) is used in both, yet the subject of the postposition’s clausal object is bare in (19), but has Genitive Case in (44). The reason is, I claim, that we have a Free Relative in (44)—and therefore crucially—an operator and a (phonologically empty) head18, leading to the presence of the Genitive. In (19), there is no reason to assume the presence of an operator, nor that of a nominal head. The most appropriate translation of the postposition göre in (19) is ‘given that’, rather than ‘according to X’, as in (44). All we have in (19) is the clausal complement within a Postpositional Phrase, with the whole PP being an adjunct of the matrix verb—hence the lack of Genitive, despite nominal, rich Agr; instead, the default, bare Case is found on the subject. Two common denominators of relative clauses and comparatives are the presence of an operator and the presence of a head of the whole construction. I suggest that it is a predication relationship between the head and the clause which activates the Agr heading the clause (by co-indexing the external nominal or quantifier phrase head of the construction with the clause, and with inheritance of the index on the clause by the nominal Agr- head of the clause). In other words, what we see here is exactly the same kind of “activation” of nominal Agr in categorially hybrid indicative clauses which we observed in overtly headed relative clauses and which we accounted for by proposing referential indexation on the nominal Agr, based on predication between the head of the RC and the modifying clause. This predication is made possible by the moved operator (relative or comparative), which turns the clause into an “open” clause. The only difference is that here, the respective heads of the constructions are phonologically empty. The Genitive marking on the subjects is therefore just as expected. 19 172 Jaklin Kornfilt From a typological perspective, it is interesting to note that a somewhat similar proposal has been made in the literature for Japanese. Watanabe (1996) proposes to analyze some of the Japanese “Ga-No conversion” contexts as relative clauses and comparative constructions, and he claims that the operator movement in such constructions makes Genitive marking on the subject possible. While this is a different account from mine, as it does not appeal to predication and to referential indexing, it is nonetheless very suggestive that, when observed from a particular perspective, Japanese and Turkish should have rather similar phenomena. It is possible that indexation via predication, with concomitant subject Case licensing, is a particular parameter, while indexation via θmarking is another one. Japanese might have a positive value for the first, European Portuguese for the second, and Turkish for both. Clearly, this is a fascinating area for further research. 5. Preliminary conclusions My account has been based on the following proposals: 1. Genuine subject Case is licensed by Agr which is itself licensed. 2. The primary type of Agr-licensing is via matching categorial feature in the local domain. In absence of this licensing, we may have: 3. Licensing by (referential) indexation, instantiated by either primary θ-role marking or by predication (the latter also based on θ-role assignment, if Williams 1994 is correct). 4. The type of subject Case is determined by the category of the licensing Agr (Nominative for verbal Agr and Genitive for nominal Agr). 5. Where Agr is absent, or where it is not licensed as a subject Case licenser, no subject Case is licensed, even where conditions for indexation are met otherwise. Overt subjects receive default Case instead. Subject case in Turkish nominalized clauses 6. 173 What is the Case of the non-Genitive subjects in adjunct clauses? The question now arises about the nature of the default Case I have proposed. Is this simply a morpho-phonologically unrealized general Case, or is it the Nominative? Given that the Nominative in Turkish has no overt realization, this is a legitimate query. I shall conclude that the default Case is indeed the Nominative. There is some independent evidence for my conclusion. One type of such evidence is provided by Left-Dislocation constructions, and especially in non-Case matched contexts. In Left-Dislocation constructions, the dislocated element can either exhibit the same Case as the corresponding constituent in the clause, or the default Case, i.e. it can be bare (in other words, I am claiming that the “bare” dislocated constituent is in the Nominative); but it cannot be in the Accusative, if the corresponding constituent in the clause is not Accusative: (47) Ali (-yi) mi? Ben kendisin-i Ali (-ACC) Y/N I himself -ACC üç ay -d›r gör -me -di -m three month -since see -NEG -PAST -1.SG ‘(About) Ali, I haven’t seem him for (the last) three months.’ (48) Ali (*-yi) mi? Ben kendisin-den çok Ali (-ACC) Y/N I himself -ABL very kork -ar -›m fear -AOR -1.SG ‘(About) Ali, I am very much afraid of him.’ This is in contrast to English, where the default Case appears to be Accusative: (49) a. Who’s there?—It’s me. b.Who’s there?—*It’s I.20 174 Jaklin Kornfilt Chomsky (2001) offers a typology of Case which is, in part, similar to his older proposals (cf. Chomsky 1981) in including structural and inherent Case. But an important addition is the notion of default Case, i.e. Case licensed not by any particular licenser, but rather assigned independently of such licensing relationships. Examples like those in (49) are offered as illustrations of this notion. 21 My proposal to analyze the dislocated subjects in (47) and (48) (where they are phonologically “bare”) as well as the “bare” subjects of adjuncts without operators and nominal heads (i.e. overt subjects which I have claimed bear default Case) as being in the Nominative Case accords well with this recent approach. The basic default Case assignment/checking mechanism would be the same in English and Turkish; the only difference would be in the actual morphological realization of the default Case: Accusative in English, Nominative in Turkish. The fact that Nominative is morphologically realized as a zero morpheme makes it, I suggest, even more plausible as a default Case. Another source of missing overt Case marking on nominals is lack of specificity: a non-specific nominal does not bear the expected structural Case morpheme (i.e. Accusative or Genitive). This phenomenon has been discussed in the literature; discussion and further sources can be found in Enç (1991), Dede (1986), Tura (1986), Erguvanl›-Taylan (1984), as well as Kornfilt (1984) and (1995a). Such non-specific, morphologically “bare” nominals must usually be immediately pre-verbal; they cannot scramble away from that position, although Turkish is otherwise rather word-order free. I will illustrate with Genitive subjects and their “bare”, non-specific counterparts: (50) [Araba -n›n yol -dan geç -ti¤ -in ] -i car -GEN road -ABL pass -FN-3.SG -ACC gör -dü -m. see -PAST -1.SG ‘I saw that the car went by on the road.’ In this example, we find the Genitive subject in its canonical, sentence-initial position. A corresponding non-specific subject, “bare” Subject case in Turkish nominalized clauses 175 morphologically, cannot show up in this canonical subject position; instead, it must be in immediate pre-verbal position: (51) a.[yol -dan bir araba geç -ti¤ -in ] -i road -ABL a car pass -FN-3.SG -ACC gör -dü -m. see -PAST -1.SG ‘I saw that a car (non-specific, non-referential) went by on the road.’ (The subject may be focussed, but it does not have to be.) b.*[bir araba yol -dan geç -ti¤ -in ] -i a car road -ABL pass -FN-3.SG -ACC gör -dü -m. see -PAST -1.SG Intended reading: ‘I saw that a car (non-specific, nonreferential) went by on the road.’ Similar facts hold in existentials—this is expected, as the “semantic” subjects of existentials are obviously non-specific: (51) c.[Garaj -da beﬂ araba ol -du¤ -un ] -u garage-LOC five car be -FN -3.SG -ACC bil -iyor -um know -PRSPROG -1.SG ‘I know that there are five cars in the garden.’ d. *[Beﬂ araba garaj -da ol -du¤ -un ] -u five car garage -LOC be -FN -3.SG -ACC bil -iyor -um know -PRSPROG -1.SG Intended reading: ‘I know that there are five cars in the garden.’ In all of these examples, we would expect the subjects to show up in the Genitive, but they are “bare”, i.e. Case-less, instead. 176 Jaklin Kornfilt Specific subjects which have, in my account, undergone default Case marking (and are, morpho-phonologically speaking, “bare” as well), behave differently with respect to word order, i.e. they can show up in canonical subject position: (52) [bu çocuk ev -de kal -d›¤-› ] için Ali this child house -LOC stay -FN-3.SG because Ali iﬂ -e gid -ebil -di. work-DAT go -ABIL -PAST ‘Ali could go to work because this child stayed at home.’ In this respect, they pattern with Genitive subjects (cf. example [50]) as well as with Nominative subjects: (53) bu çocuk ev -de kal -d› this child house -LOC stay -PAST ‘This child stayed at home.’ Conclusion: there are different types of morphologically bare subjects. While bare non-specific subjects lack (structural) Case (in all of the instances just observed, this would be the Genitive) and are, probably due to that reason, fixed in their pre-verbal position, subjects that are bare but carry default Nominative Case behave like regular, genuinely Nominative subjects in verbal clauses as well as their Genitive counterparts in nominal clauses. The issue of nonspecific, bare DPs is orthogonal to the issue of licensed versus default Case. Any nominal phrase with non-lexical Case, irrespective of the licensed or default nature of such Case, is (usually) realized in a Case-less, bare form, as in (51) a. or c. 7. When default Case may or may not be licensed It is important to set up constraints on the application of default Case; otherwise, default Case would apply in all instances when licensed Case does not, and we would lose our explanations of, for Subject case in Turkish nominalized clauses 177 example, the complementary distribution between overt subjects and PRO-subjects in infinitival clauses. I therefore start with those. 7.1. Infinitivals I propose that the infinitival marker -mAK is actually -mA + K, with -m A in M(ood) position, and -K in AGR (=Fin) position. This proposal captures the morpho-phonological similarity between the inflected nominal subjunctives (i.e. -mA + Agr) and the non-inflected infinitivals (-mA+K). It further captures the fact that, semantically, the Mood of infinitives and that of the inflected non-factives is similar, i.e. they are both subjunctive. The analysis further accounts for the fact that the verbs selecting for infinitival clauses are similar to those that select subjunctive clauses. Note that - K never expresses ϕ -features; I analyze it, consequently, as [-Agr]. This licenses “null Case” (cf. Lasnik and Chomsky 1991), i.e. the Case for licensing PRO, but not for licensing overt DPs. Now, default Case is blocked: default Case can’t apply when any Case (of the type that needs to be licensed by a designated licenser) is actually licensed; this holds for “genuine subject Case” (e.g. for the Genitive in nominal clauses) as well as for the “null Case” of Lasnik and Chomsky, i.e. the Case special for PRO-subjects. In this way, the complementary distribution between overt subjects and PRO in infinitivals and the correlation of this distribution with the presence versus absence of an appropriate Agr element can be maintained.22, 23 I now turn to other syntactic domains that lack Agr—domains where it is desirable to block application of default Case. 7.2. Other Agr-less domains 7.2.1. In ECM-contexts When discussing ECM-constructions, I suggested that there is no “Csystem” dominating the embedded clause. Hence there also is no 178 Jaklin Kornfilt “Fin”-head (corresponding to AGR), not even negatively specified in the way I just proposed for infinitives (cf. George and Kornfilt 1981, where, for Turkish, Agr determines Finiteness). Consequently, Accusative Case is licensed by the higher (ECM-) verb. We now have a licensed Case; therefore, default Case is not needed and hence not allowed, given that it is a strictly last resort mechanism. 7.2.2. In adjunct domains headed by other forms than the infinitive, but no Agr We discussed adjunct clauses that lack Agr. The CP-status of such clauses is unclear: extraction judgements are murky. I would like to make the following assumption about such clauses: “Fin” exists in their clausal architecture, but it is underspecified, as it lacks Agr features. No other Case licensing is possible (the way it is in infinitivals as well as in ECM-constructions). Consequently, default Case applies, as a last resort. 7.3. Adjunct domains with Agr (and no predicational indexation) No primary θ−role is assigned to this syntactic domain. If such a domain is not involved in predication (i.e. if such an adjunct clause is not a relative clause or if it is not a comparative), and if there also is no clause-internal categorial congruence between Agr and the rest of the predicate, Agr will not be licensed as a subject Case licenser. No other licensing is possible. As a consequence, default Case applies as a last resort. Subject case in Turkish nominalized clauses 8. 179 Other treatments of the argument/adjunct asymmetry based on subject Case There have been very few discussions of the phenomena presented in this paper. I am aware of two: one earlier and one later than my work. I start with the earlier one. 8.1. Kennelly (1990) She proposes that it is Tense that assigns Nominative to the subject of a clause, rather than Agr. For her, indicative nominal markers (i.e. the markers which I have glossed as F[active] N[ominal] and which I analyzed as giving rise to categorially hybrid predicates) equal Tense. She claims that external θ − role assignment blocks Nominative assignment; motivated by the need to receive Case, the subject raises to Spec, DP (i.e. the DP which Kennelly assumes is the proper category of nominal clauses); in that higher position, the subject receives Genitive, simply as a consequence of being the specifier of a DP. My main objection to this account is conceptual: why should thematic marking block regular subject Case? To my knowledge, this phenomenon, i.e. θ-marking as blocking Case checking, has never been observed or referred to for other languages and/or for other phenomena. If anything, θ-marking either presupposes or makes possible Case assignment/checking; this has been usually assumed for inherent Case, for example.24 As for structural Case, I am not aware of an instance where thematic marking blocks such Case. We saw earlier that for Raposo, it is in “Case-marked”, i.e. largely θmarked, domains that a nominal clausal head is able to assign a subject Case to its subject. Thus, the main proposal is not motivated and goes against general assumptions and cross-linguistic facts. In addition, the data Kennelly uses to bolster her analysis are problematic and, to my knowledge, are not shared by many native speakers. Due to space considerations, I shall not discuss her proposals further, given the gravity of the conceptual problems with 180 Jaklin Kornfilt it. It is probably due to these reasons, and I suspect especially due to the problems with the data, that this proposal has seldom, if ever, been referred to or used as the basis of later work. In any case, given the problems with it that I just sketched, I think we can safely reject this approach. 8.2. Aygen (2002) This work postdates the presentation of my paper at the Leipzig workshop related to this volume, as well as Kornfilt (2001) and (2002). Given certain similarities between it and my work, as well as between it and Lees (1965), I shall devote some space to a discussion of its main aspects. Aygen (2002) devotes attention to factive “nominalizations” only, i.e. the type which I have claimed to be categorially hybrid. She notes the main contrast between argument and adjunct factive clauses in Turkish—the contrast that has been the focus of this paper, too: the subjects of the former are Genitive, but those of the latter are Nominative. Note that in terms of adjunct clauses, Aygen considers only those that are marked with the factive marker, and which do have Agr. (In other words, she does not discuss the Agr-less predicates in adjunct clauses; in terms of argument clauses as well as adjunct clauses, she considers only the factive nominal clauses, i.e. there is no account of the non-factive, fully nominal clauses.) She reasons that, given that Agr is present in both argument and adjunct clauses, yet the Case on the subjects is different, Agr couldn’t possibly be involved in Case assignment to subjects. This first premise is faulty; the conclusion does not necessarily follow. As we have seen in the present paper, an approach is possible where Agr (or any representation of ϕ-features) checks for subject Case only in particular configurations. This is the reasoning I have followed here. Raposo’s approach to European Portuguese takes this same direction of reasoning, as well, as did Reuland’s (1983) earlier one to English. Subject case in Turkish nominalized clauses 181 However, although Aygen’s premise is faulty, it is nevertheless true that the initial observation of the asymmetry just mentioned might indeed be open to an approach that does not involve Agr at all in subject Case assignment—especially if all the arguments in favor of Agr in this respect (and which were discussed earlier in the present paper) are not considered, due to a very narrow focus on just the asymmetry at hand. I shall therefore briefly sketch Aygen’s account. Aygen (2002) proposes that in Turkish, Case on the subject is licensed by neither the [+Tense] features of T nor the ϕ-features of AGR, but by a Case feature on C. This feature is responsible for Genitive subjects: in relative clauses and in noun-complement clauses, some agreement relationship between either the external nominal head of such DPs or perhaps the D-head of the (high) DP and the CP licenses a Genitive feature on the CP, and, indirectly, on the subject of the CP. Indicative argument clauses are claimed by Aygen to be nouncomplement clauses with an abstract nominal external head; adjunct indicative clauses are claimed not to be externally headed. To illustrate the latter claim, I use the following two examples; (54) would be claimed to have a structure similar to (55): (54) Ben[Hasan -›n gel -di¤-in ] -i I Hasan -GEN come -FN-3.SG -ACC bil -iyor -um know -PRSPROG -1.SG ‘I know that Hasan came.’ (55) Ben [[Hasan-›n gel -di¤ -i ] I Hasan-GEN come -FN -3.SG gerçe¤ -in ] -i bil -iyor -um fact -CMPM-ACC know -PRSPROG -1.SG ‘I know the fact that Hasan came.’ The internal argument of the verb in (55) is a complex DP, i.e. a noun-complement clause with its head noun. According to Aygen, the Genitive is licensed by indirect agreement (C-N agreement) with 182 Jaklin Kornfilt a nominal head. This indirect agreement is mediated via the CP, i.e. the complement clause, which “agrees” with the head. She claims that in (54), the object clause is actually an instance of a complex DP, as well, with a phonologically unrealized nominal head. Hence, the Genitive subject of that clause is similarly accounted for. The idea that nominalized argument clauses are actually complements of phonologically unrealized nominal heads was, to my knowledge, first proposed in Lees (1965) for Turkish. There, both factive and non-factive nominalizations were analyzed in this way, although slightly different phrase structures were attributed to each construction. This interesting proposal has some drawbacks, however, and some of the criticism I shall raise against Aygen’s approach to Genitive subjects in factives will concern Lees’s original proposal, as well. First, concerning Aygen’s proposals, i.e. the licensing of a Genitive subject via an “agreement” relation between the C-head of a complement (or, more generally, argument) CP and the nominal head (or perhaps even the D-head) of a dominating DP: this proposal would make sense for languages where “concord”, i.e. agreement between the head noun (or D) and the complement in terms of certain features (e.g. ϕ-features, or Case) obtains. Turkish, however, has no concord. The features of a nominal head do not spread within the DP—neither to complements, nor to modifiers. The latter point is relevant with respect to relative clauses, where the modifier clause is not a complement, but an adjunct of the nominal head. For those, too, it is not plausible to assume an “agreement” relationship between the nominal head and the modifier clause, given that no agreement between modifier and nominal head is ever found elsewhere, either. Also, as the examples in the current study show, there is no overt agreement between the nominal head and either a complement clause (as in noun-complement constructions) or between a nominal head and a modifier clause in a relative clause construction. Furthermore, while the proposal appears to unify relative clauses and noun-complement clauses by positing this “agreement” relation between a nominal head (overt or covert) and a (complement or Subject case in Turkish nominalized clauses 183 adjunct) clause, this unification is only apparent. Complements of nouns are structurally in a different position from adjuncts of nouns, as we saw in some detail earlier in the present paper; we saw that the modifier clause in a relative clause is attached higher with respect to the head than is a complement clause in a noun-complement construction. Thus, even if some sort of (abstract) “agreement” did obtain, we would be looking at different “agreement” relations for these two constructions. One way out of this would be to posit raising of the subject to Spec, DP of the higher DP, i.e. of the DP associated with the external head N. While there is no evidence in Turkish for such raising in the syntax, LF-raising is a possibility. For Japanese, this has been proposed by Miyagawa (1993), with considerable explanatory success. I shall offer some evidence to counter any analysis imputing Genitive subject Case licensing directly to an external nominal for Turkish, thus suggesting a property of grammar which is open to parametric variation. In addition to these conceptual and empirical problems, there are problems concerning the evidence offered for parts of the analysis, namely that nominalized complement clauses are externally headed DPs. I shall consider only a few of those that are particularly clearcut and shall start by considering one type of such evidence, namely scrambling to post-verbal positions. 8.2.1. Problems for post-verbal scrambling It is well-known that Turkish allows backgrounded constituents to scramble to post-verbal positions. Such scrambling from out of an embedded clause to the very end of the root clause is not too bad for most speakers: 184 Jaklin Kornfilt (56) ?[Hasan -›n ti nihayet kaç -t›¤ -›n ] -› Hasan -GEN finally escape -FN-3.SG -ACC duy -du -m kar› -s›n -dani hear -PAST -1.SG wife -3.SG -ABL ‘I heard that Hasan finally ran away from his wife.’ Given that the scrambled constituent has not attached to the argument clause, but rather elsewhere (i.e. to the root clause), there is no reason (other than subjacency, with mild effects) to predict ungrammaticality, especially given that I am not assuming that the argument clause has a nominal head. This contrasts with overtly headed factive clauses. In such constructions, scrambling to root-final position deteriorates: This contrasts with overtly headed factive clauses: (57) ??/*[[Hasan -›n ti nihayet kaç -t›¤ -› ] Hasan -GEN finally escape -FN -3.SG söylenti -sin ] -i duy -du -m rumor -CMPM-ACC hear -PAST -1.SG kar› -s›n -dani wife -3.SG -ABL ‘I heard the rumor that Hasan finally ran away from his wife.’ Aygen would wrongly predict the same status of acceptability for both, given that the host of scrambling in (56) is externally headed in her approach, just as its counterpart is headed in (57). The clear contrast between these two examples sheds further doubt on her approach. (Note also that this last contrast is problematic for Lees 1965, as well.) Similarly revealing are examples where the whole argument clause has been scrambled to verb-final position in the root clause: Subject case in Turkish nominalized clauses 185 (58) tj Duy -du -m [[Hasan-›n nihayet hear -PAST -1.SG Hasan -GEN finally kar› -s›n -dan kaç -t›¤ -›n] -› ]j wife -3.SG -ABL escape -FN -3.SG -ACC ‘I heard that Hasan finally ran away from his wife.’ In such examples, post-verbal scrambling of a constituent of the subordinate clause is fine: (59) tj Duy -du -m [[Hasan-›n ti nihayet hear -PAST -1.SG Hasan-GEN finally kaç -t›¤ -›n ] -› ]j kar› -s›n -dani escape -FN -3.SG -ACC wife -3.SG -ABL ‘I heard that Hasan finally ran away from his wife.’ This is just as expected in any approach in which this type of subordinate clause is not headed. The fact that it is an argument clause is not a problem, either, given that the clause is not in argument position, similarly to extraction facts in English, where a syntactic island like a sentential subject does not exhibit island effects when it is extraposed, as shown in Ross (1967). (For discussion of Turkish facts of postverbal scrambling out of subordinate clauses in different positions, cf. Kornfilt 1998.) However, the full grammaticality of (59) for many speakers is a serious problem for Aygen’s approach, which would predict it to be ungrammatical, for the reasons already discussed: the scrambled subordinate clause is, in that approach, headed. This problem is compounded by the ill-formedness of corresponding examples where there is an overt head: 186 Jaklin Kornfilt (60) ??/*tj Duy -du -m [[Hasan-›n ti nihayet hear -PAST -1.SG Hasan-GEN finally kaç -t›¤ -› ] söylenti -sin -i ]j escape -FN -3.SG rumor -CMPM-ACC kar -s›n -dani wife -3.SG -ABL Intended reading: ‘I heard the rumor that Hasan finally ran away from his wife.’ For Aygen, there should be no difference between the perfectly fine (59) and the ill-formed (60); again, this is a problem for Lees (1965), as well. 8.2.2. Problems for distribution Nominalized clauses can differ in their distribution according to whether they have an external nominal head or not. Only two systematic differences (among a number of similar subcategorizational differences) are considered here: factive versus non-factive nominalized clauses as objects versus subjects of psychological predicates. 1. Psychological predicates allow both the factive and the nonfactive nominalization types as complements, without any difference in semantics. (61) a.[Ali -nin ev -den kaç -ma -sın ] -a Ali-GEN home -ABL flee -NFN -3.SG -DAT üzül -dü -m sadden -PAST -1.SG ‘I was saddened at Ali’s running away from home.’ Subject case in Turkish nominalized clauses 187 b.[Ali -nin ev -den kaç -tı¤ -ın ] -a Ali-GEN home -ABL flee -FN-3.SG -DAT üzül -dü -m sadden -PAST -1.SG Same translation as in the previous example. However, when an external noun shows up, only the factive gerund is well-formed for factive semantics: (62) a.??/*[[Ali -nin ev -den kaç -ma (-sı)] Ali-GEN home -ABL flee -NFN -3.SG söylenti -sin ] -e üzül -dü -m rumor -CMPM -DAT sadden -PAST -1.SG Intended reading: ‘I was saddened at the rumor of Ali’s running away from home.’ b.[Ali -nin ev -den kaç -tı¤ -ı ] Ali-GEN home -ABL flee -FN-3.SG söylenti -sin -e üzül -dü -m rumor -CMPM-DAT sadden -PAST -1.SG ‘I was saddened at the rumor of Ali’s running away from home.’ 2. With the same type of predicates, only the non-factive gerundive is well-formed as subject, despite indicative semantics; however, when such a sentential subject is externally headed, only the factive gerund is well-formed for indicative semantics: (63) a.[Ali -nin ev -den kaç -ma -sı ] ben -i Ali-GEN home -ABL flee -NFN -3.SG I -ACC üz -dü sadden -PAST ‘Ali’s running away from home saddened me.’ 188 Jaklin Kornfilt b.*[Ali-nin ev -den kaç -tı¤ -ı ] ben -i Ali-GEN home -ABL flee -FN-3.SG I -ACC üz -dü sadden -PAST Intended reading: ‘Ali’s running away from home saddened me.’ (64) a.??/*[[Ali -nin ev -den kaç -ma (-sı)] Ali-GEN home -ABL flee -NFN -3.SG söylenti -si ] ben -i üz -dü rumor -CMPM I -ACC sadden -PAST Intended reading: ‘The rumor of Ali’s running away from home saddened me.’ b.[[Ali -nin ev -den kaç -tı¤ -ı ] Ali -GEN home -ABL flee -FN-3.SG söylenti -si ] ben -i üz -dü rumor -CMPM I -ACC sadden -PAST ‘The rumor of Ali’s running away from home saddened me.’ Once again, these examples are problematic for both Aygen (2002) and Lees (1965), as they clearly show that the distribution of nominalized clauses with external nominal heads is different from the distribution of their counterparts without external nominal heads. 8.2.3. Existing correlations not captured There are clear correlations that hold between subject Case types and local Agr types. These hold at an observational level and are independent from any analytical bias: 1. Nominative subjects in argument clauses (as well as root clauses) are possible only when verbal Agr is present locally; 2. Genitive subjects in both argument and adjunct clauses are possible only when nominal Agr is present locally. The present study has offered illustrations of both generalizations. In both instances, the presence or absence of external nouns is completely irrelevant. Therefore, any approach to the first asymmetry (i.e. the Subject case in Turkish nominalized clauses 189 argument-adjunct asymmetry in categorially hybrid nominalized factive clauses) that rejects Agr as an important factor in determining subject Case in Turkish is problematic, as is any approach that attributes primary importance to an external nominal head as determining Genitive Case. 8.2.4. Correlations posited that do not exist Aygen (2002) claims that Genitive subjects are possible only when the clause has an external nominal head, or else, where there is no such overt head, where a nominal head is potentially possible, because the position is there structurally. In consequence, she claims that when a Genitive subject is not possible, an external nominal head is not possible, either. Likewise, when there is an external nominal head, the subject of the clause should always be Genitive. Both correlations are counterexemplified by a variety of constructions: 8.2.4.1. Indicative nominalized existentials with non-Genitive subjects I shall start with one type of construction which is discussed in Aygen (2002) as furnishing support for her analysis (and as supposedly being problematic for one aspect of my approach—an aspect of the analysis which I had presented in Leipzig as well as in Kornfilt 2001). This is the existential construction in nominalizations. In Turkish existentials, the subject is to the immediate left of the verb; in this respect, it is similar to other non-specific, non-referential subjects which we saw earlier in this paper. Another similarity is that such a subject cannot be marked with the Genitive; instead, it has to be morphologically bare with respect to Case. While there is a special existential verb in fully verbal clauses, the predicate in 190 Jaklin Kornfilt nominalized existentials is the “light verb” ol ‘be’, which takes the regular nominalization inflections we have discussed in this study. An indicative nominalized existential follows as an illustration: (65) Ali [bahçe -de bir ejderha ol -du¤ -un ] -u Ali garden-LOC a dragon be -FN -3.SG -ACC duy -du hear -PAST ‘Ali heard that there is a dragon in the garden.’ Aygen (2002) claims that, because the subject of existential subjects in indicative nominalizations cannot be in the Genitive, such clauses cannot show up with an external nominal head. She claims that examples of the following sort are ungrammatical: (66) Ali [bahçe -de bir ejderha ol -du¤ -u ] Ali garden -LOC a dragon be -FN -3.SG söylenti -sin -i duy -du rumor -CMPM-ACC hear -PAST ‘Ali heard the rumor that there is a dragon in the garden.’ Aygen (2002) further claims that the supposed ungrammaticality of (66) is a problem for my approach. This is because she imputes to that approach the prediction that the subject in (66) should be Genitive: due to the θ-marking which the gerund clause would receive from the external noun, the agreement would receive a referential index and thus, so she claims, would license Genitive subject Case, contrary to fact. First of all, my approach does not make this prediction. The nominal Agr in (66) does not AGREE with the subject, but rather with an expletive pro. In other words, this is a “fake”, default Agr. Secondly, both in the current study and in its precursor presentations, I made clear that non-specific subjects (which include existential subjects) cannot be morphologically marked for structural Case (and thus for Genitive), even if such a structural Case should be licensed syntactically. Subject case in Turkish nominalized clauses 191 A third, and even more important, point is that Aygen’s idiolect appears to be exceptional in rejecting examples like (66). Such examples are perfect for me, and they are perfect for all the native speakers whom I have consulted. (All the individuals in the list of speakers in the footnote of acknowledgements were consulted about these examples and similar ones, and they all found them to be flawless.) 8.2.4.2. In the absence of Agr, only PRO is licensed as a subject; an overt DP is not, irrespective of its Case Turning to problematic examples not mentioned in Aygen (2002), we saw earlier that irrealis relative clauses cannot have any overt subject, Genitive or otherwise; the only possible subject is PRO. The next two examples illustrate this point once again, for the readers’ convenience: (67) a.[[PRO san -a ver -ecek] bir vazo] you -DAT give -FUTN a vase bul -du -m find-PAST -1.SG ‘I found a vase to give you.’ b.*[[Ben /Ben -im san -a ver -ecek] I (NOM) / I -GEN you -DAT give -FUTN bir vazo] bul -du -m a vase find -PAST -1.SG Intended reading: ‘I found a vase for me to give you.’ In the presence of Agr, the same FUTN marker expresses Future/ Indicative; a Genitive subject is licensed—due to the presence of the Agr element: 192 Jaklin Kornfilt (68) [[Ben-im san -a ver -ece¤ -im] vazo] -yu I -GEN you -DAT give -FUTN-1.SG vase -ACC bul -du -m find-PAST -1.SG ‘I found the vase I am going to give you.’ Remember that Aygen claims that the nominal head of the relative clause licenses Genitive subjects, and that the Agr element of a clause is irrelevant for the subject being licensed via Case. She would therefore predict that the version of (67)b. with a Genitive subject should be grammatical. Thus, both the fact that irrealis relative clauses cannot have overt subjects (as shown by the examples in [67]) and the contrast with future tense relative clauses that do have Genitive subjects (as shown by [68]) are problematic for Aygen (2002) but just as expected under the approach developed in the present study, as comparison of these constructions shows the importance of Agr for subject Case licensing as well as the irrelevance of an external nominal head. 8.2.4.3. Determining factor not Mood by itself (in general); e.g. noun-complement clauses with the same Mood, but different categorial features Aygen (2002) mentions in passing that Mood plays a role in determining the subject Case, too, but does not make explicit in what way this would interact with the subject Case determination via the external nominal head. But suppose that we do pursue this idea. We would have to say that irrealis mood somehow blocks the Genitive Case licensed by the external nominal head, while indicative mood does not do so. Vague and unlikely as this proposal is, it would draw the correct distinction between indicative/future and irrealis relative clauses. However, it is clear that Mood does not determine subject Case marking in Turkish in general. This can be seen clearly by contrasting (69) with (70): Subject case in Turkish nominalized clauses 193 (69) [Ben-im aile -m -i terket -ti¤ -im ] I -GEN family -1.SG -ACC abandon-FN-1.SG söylenti -si rumor -CMPM ‘the rumor that I abandoned my family’ (70) [Ben aile -m -i terket -ti -m ] I (NOM) family -1.SG -ACC abandon-PAST-1.SG söylenti -si rumor -CMPM ‘the rumor that I abandoned my family’ (i.e. same as in the previous example) (69) illustrates a nominalized indicative noun-complement clause, while (70) exemplifies a fully verbal, but also indicative noun-complement clause. Both have obviously an external nominal head. But only (69) has a Genitive subject, while (70) has a Nominative subject. There is no difference in Mood. However, there is a difference in the local marker for Agr: it is nominal in (69), thus licensing Genitive subject Case, and verbal in (70), thus licensing Nominative subject Case. Thus, these facts and their contrast are just what my approach predicts. However, this contrast, and especially the Nominative (rather than Genitive) subject in (70) are problematic for Aygen (2002). Her analysis predicts Genitive subjects for both constructions, due to the external nominal head in both. Since there is no Mood difference between the two, her analysis cannot take recourse to a Mood-based determination of subject Case, either. We thus see that what matters for subject Case and its overt realization are the categorial features of Agr.25 8.2.4.4. Another relevant construction: The nominalized indicative clause as a postpositional complement Yet another type of construction that is discussed in Aygen (2002) in the context of the supposed correlation Genitive subject—external 194 Jaklin Kornfilt nominal head is the nominalized indicative clause as a postpositional complement. We saw that such clauses have Nominative rather than Genitive subjects (unless they are in a predicational relationship). Aygen gives examples showing that no nominal head is possible when the subject of such a clause is in the Nominative; to be wellformed, the subject of such a clause must be in the Genitive: (71) *[[Hasan anla -dı¤-› ] ﬂey -e göre ] Hasan understand -FN-3.SG thing-DAT according to herkes anla -yacak everybody understand -FUT Intended reading: ‘According to the thing that Hasan understood, everybody will understand.’ (Aygen 2002: example [15 a.]; glosses and translation slightly changed) (72) [[Hasan-›n anla -dı¤ -› ] ﬂey -e Hasan-GEN understand -FN -3.SG thing -DAT göre ] herkes anla -yacak according to everybody understand -FUT ‘According to the thing that Hasan understood, everybody will understand.’ (Aygen 2002: example [16]; glosses and translation slightly changed) (73) [[Hasan haber -i anla -dı¤-›n ] -a Hasan news -ACC understand -FN-3.SG -DAT göre ] herkes anla -yacak according to everybody understand -FUT ‘Given that Hasan understood the news, everybody will.’ (Aygen 2002: example [17]; glosses and translation slightly changed) This triplet does not establish that the Genitive subject is due to the external noun in (72). The account I have proposed in this study explains these facts too, and without all the problems that go along Subject case in Turkish nominalized clauses 195 with Aygen’s analysis. In (71) and (72), we have a relative clause. Predication between ﬂey ‘thing’ and the modifier clause would referentially index the clause and thus turn the nominal Agr into a licenser of Genitive. This is why (72) is well-formed, but (71), without the Genitive, is ill-formed. In (73), the clause does not receive a referential index from anywhere. Since this is an indicative, and thus categorially hybrid, clause, its nominal Agr is not licensed to be a Genitive licenser clause-internally, either. Therefore, no genuine subject Case is licensed, and default Nominative applies instead. A further problem with Aygen’s claim that her analysis explains the existence of such triplets is the following: why should verbs subcategorize for noun-complement constructions (whether overtly or covertly headed by an external noun), while postpositions subcategorize for nominalized clauses without external nominal heads? Why should any kind of (factive) nominalized clause reject an external nominal head when the clause is an adjunct—whether as an adjunct itself, or as part of an adjunct, when it is subcategorized by a postposition? Aygen (2002) does not attempt to explain or motivate this difference, while basing her (as shown here, problematic) analysis on the assumption of this difference. 8.2.5. Problems with scope facts If the Genitive subject in indicative nominalized clauses somehow AGREEs with an external head noun, one would expect for suitable subjects to be able to take scope over that head noun, at least optionally. Miyagawa (1993) shows that in Japanese, this is indeed an option for Genitive subjects, but not for Nominative subjects of nominal complement clauses. Turkish examples differ in this respect: 196 Jaklin Kornfilt (74) Ali veya Veli -nin parti -ye gel -ece¤ -i Ali or Veli -GEN party-DAT come -FUTFN -3.SG ihtimal -i yüz -de elli -den yüksek. probability -CMPM hundred-LOC fifty -ABL high ‘The probability that Ali or Veli will come to the party is greater than fifty per cent.’ (75) Ali veya Veli -nin parti -ye gel -di¤-i Ali or Veli -GEN party-DAT come -FN-3.SG ihtimal -i yüz -de elli -den yüksek. probability -CMPM hundred-LOC fifty -ABL high ‘The probability that Ali or Veli came to the party is greater than fifty per cent.’ For those speakers who accept these constructions as well-formed, the head noun has scope over the Genitive subject; the reverse scope is not possible. In other words, for those speakers who accept them, these examples can mean: ‘The probability that either Ali or Veli came/will come to the party is greater than fifty per cent.’ However, they cannot mean: ‘Either the probability that Ali came to the party is greater than fifty per cent, or the probability that Veli came to the party is greater than fifty per cent.’ This strongly suggests that the Genitive subject is not in a nonlocal (or indirect) AGREE relation with the external head ihtimal ‘probability’, nor has it risen to the specifier position of that external head or of its associated D. In Japanese, such raising might be possible or necessary, as the complement clauses of nouns (or the modifier clauses of relative clauses) are not nominalized themselves. They neither have a nominal Agr, nor do they have TAM morphology with nominal features. Therefore, an analysis imputing Genitive licensing capabilities to the external head is plausible for Japanese, with the concomitant scope effects. In Turkish, however, the clause has local morphology with nominal categorial features, and thus Genitive subject Case can be licensed by closer nominal elements than an external noun. As a consequence, attempts to impute Genitive licensing to such an external Subject case in Turkish nominalized clauses 197 nominal head fail when confronted with syntactic challenges, as we have seen. 9. Conclusions and some speculations I have claimed in this paper that the subject-predicate agreement morphology of subordinate clauses in Turkish may license overt subjects in the following ways: 1. If the Agr element is itself licensed categorially within its clause, it also licenses genuine subject Case (and thus an overt subject). An Agr element is licensed if it occurs in a morphological sequence with categorially fully specified TAM markers and is of the same categorial type as those and as the higher functional projections. 2. If the Agr is not licensed in this way (which typically is the case when a nominal Agr shows up after a categorially underspecified TAM marker, i.e. the factive future or non-future markers and under a CP=ForceP, i.e. a verbal functional projection), an overt subject can be licensed in one of two ways: A. If the Agr bears a referential index (in the sense of Rizzi 1994), it is licensed itself and can license genuine subject Case. Such referential index can be inherited by Agr in one of two ways: 1. Via a primary θ-role that the clause, headed by Agr, receives; 2. Via predication between such a clause and a co-indexed head. B. Where neither categorial homogeneity nor indexation enables an Agr to assign subject Case (and where there also is no negative Agr element that licenses null Case), default Case is assigned instead. In Turkish, default Case is Nominative. In contrast to two studies on the same issue (but limited to factive subordination only), one preceding the current study temporally and the other following its earlier incarnations, this larger perspective on subordination and subjects in subordinate domains shows that it is unnecessary to make otherwise unmotivated assumptions, e.g. the assumption that subject Case is blocked under θ-marking, or that nominalized argument clauses have all an abstract nominal head and are noun-complement clauses. Furthermore, subject Case is not 198 Jaklin Kornfilt licensed by Tense/Aspect or Mood per se—the latter only together with Agr, which itself determines, in Turkish, finiteness. This study, then, shows us that in a morphologically rich language like Turkish, not only does Agr express ϕ-features, but also categorial distinctions which are reflected in the subject Case it licenses. Another result is that θ-marking, whose importance for extraction has long been established theoretically and cross-linguistically, has been claimed here to also play a central role in determining subject Case. It would be interesting to find additional cross-linguistic evidence; the facts of European Portuguese are very suggestive in this regard. It was important, when referring to the Case of subjects, to distinguish between genuine and default subject Case. Genuine subject Case (whether Nominative or Genitive) depends on the category of the Agr that licenses it. Subjects with genuine subject Case are in complementary distribution with PRO or Accusative subjects (depending on the presence versus absence of Agr), while subjects with default Case are in free variation with PRO in the absence of Agr (other than in infinitives and irrealis relative clauses, where there is negative Agr), and such subjects are independent from the presence of Agr with respect to Case. It is important, when categorizing Case, not to draw the lines simply according to the overt appearance of Case (i.e. Nominative versus Genitive), but according to the way in which Case is licensed—something which may, but not always does, coincide with the overt realization of Case. There is one potential problem with the account I proposed: For the factive nominalized, categorially hybrid clauses, I proposed an analysis based on the claim that they are CPs—in contrast with the fully nominal, non-hybrid, non-factive clauses which I claimed lack a CP-level. I further claimed that in factive nominalized clauses, the nominal Agr raises to the C-head of the clause and thus inherits the index that the entire CP receives; it is this index that enables Agr to license genuine subject Case. The potential problem mentioned is typological: Sabel (1996) points out that in languages with raising of a verb or of a verbal Subject case in Turkish nominalized clauses 199 element to C in non-finite clauses, no WH-movement is possible, as the raised element obliterates the [+WH] features in C, making it impossible for the Spec, CP to host WH-elements (cf. Sabel 1996: 297). Yet, I showed that in Turkish, it is the factive nominalized clause where raising Agr takes place and where RCs and embedded WH-questions are possible, i.e. in whose Spec,CP the WH-operator would be hosted. I suggest that the problem is only apparent. In all of the languages listed by Sabel in this context—all of them Indo-European languages—there is obviously complementary distribution between a C-position occupied by a complementizer and one into which a verb or other predicational morphology has been raised; thus, it is plausible to suggest, as Sabel does, that raising erases [+WH] features otherwise expressed by a complementizer. However, in a language like Turkish, there is never an overt complementizer introducing nominalized clauses. The nominalized predicate, raised into C with the Agr, includes the Mood marker—i.e. the factive marker, and it is this marker that encodes the hybrid category and the associated CP-status of the clause, thus acting as a clause-typing marker of the sort exhibited by Indo-European languages in the shape of a complementizer. Therefore, Agr (+Mood) raising to C in Turkish does not obliterate [+WH] features in C; such raising might in fact be motivated by the need to actually activate those features.26 In the body of the paper, I have suggested that the approach to licensed subject Case, based on referential (in the sense of Rizzi 1994) indexing (or lack thereof) on a categorially hybrid clause may be extended to other languages, under appropriate parametrization. For example, while in Turkish, the indexing must indeed be referential in the appropriate sense, i.e. must express a primary θ-role or predication (following Williams 1994, for whom predication is linked to θ-marking), in European Portuguese, the indexing can express a secondary θ-role, as well, but (probably) cannot express predication. In both languages, the subject Case licenser is Agr, appropriately indexed, and raised to C. In other languages, too, it has been proposed that an inflectional element like Agr (or, depending on 200 Jaklin Kornfilt the language, Tense) may license subject Case if it is raised to C. This was mentioned, in passing, for Modern Greek. Bayer (1983–84) has proposed such an account for Bavarian; similar proposals exist for other languages, too. This means that the raised Agr can reach into the lower functional projection, i.e. into the AgrP (or TenseP, depending on the language). Thus, we have a configuration and mechanism somewhat similar to ECM-constructions, where it is a designated verb that reaches into a “deficient” clause—in Turkish, the deficiency being expressed by the lacking Agr. Here, in the indexed clauses with their Agr (or Tense) raised to C, the clause is similarly deficient, due to the raising of its head to C.27 In Japanese, there is no inflectional element such as Agr or Tense. I won’t speculate here on the nature of the regular subject Case, i.e. the Nominative, in Japanese. However, perhaps due to the lack of an appropriate inflectional subject Case licenser, Japanese has the possibility of an external nominal head of clauses licensing a nominal subject Case, i.e. the Genitive. Turkish, on the other hand, by virtue of having nominal as well as verbal inflectional heads, can have subject Case licensed locally, i.e. within the clause. This is particularly instructive when the Agr is nominal, as (under appropriate indexation) a nominal subject Case can be licensed locally, without needing recourse to an external nominal head, in contrast with Japanese, where such an external head is needed for Genitive licensing. We may say, then, that the nominal Agr in Turkish acts, in a sense, as the Japanese external nominal head. Thus, in turn, we may reach a new understanding of why Agr must, or even is able to, bear referential indexing in Turkish: it is a (small) noun.28 Subject case in Turkish nominalized clauses 201 Notes * This paper corresponds to the presentation at the Workshop on morphologically rich languages, held within the DGfS conference that took place in Leipzig, in February/March 2001. It is also related to Kornfilt (2002), but is very different from that paper: the coverage of the present paper is larger, as it looks at non-factive as well as factive nominalizations. Also, the approach taken here is different: while here, indexation of Agr is limited to referential indexation, the just mentioned work does not do so and thus runs into problems. Furthermore, (referential) indexation of Agr in operator-variable constructions is performed here via predication between the clause and a nominal (phrasal) head, while indexation of Agr was done via Spec-Head agreement within CP in the aforementioned work. The approach based on categorial features is new here, as is discussion and criticism of some other work that addresses one of the asymmetries studied here. I would like to thank the workshop organizers, Uwe Junghanns and Luka Szucsich, for inviting me to the workshop and for their patience with the drawn-out progress of this paper. I owe a particular debt of gratitude to both organizers (and editors of this volume) for their close reading of a preliminary draft of this paper, and for their comments; two anonymous referees also provided insightful and useful comments, and I thank them. I also wish to thank Noam Chomsky for discussion of this material, and especially of the issue of default Case in adjunct domains. I am grateful to the DGfS for providing travel funds that made my participation at the workshop possible. I owe a special debt of gratitude to a number of Turkish native speakers for adding to the pool of native speakers’ judgements: Çi¤dem Bal›m, Akgül Baylav, Cemal Beﬂkardeﬂ, Demir Dinç, Cem Mansur, Alp Otman, Bengisu Rona, Mehmet Yan›lmaz, Ayﬂe Yazgan. I further thank Mark Brown for his help with the formatting of this paper. In addition, I would like to thank the audiences of a number of related presentations: in 1999 at the MPI EVA, at the University of Jena, the University of Venice, the University of Paris at Jussieu, and at Bo¤aziçi University; in 2001, in addition to the DGfS workshop in Leipzig, at the Altaic workshop at MIT, and at the CUNY Graduate Center; in 2002, at the In the Mood conference at Frankfurt University, at ZAS Berlin, at Cornell University, and at the MPI EVA in Leipzig. Among those audiences, I would like to thank, in particular, the following individuals: Artemis Alexiadou, Josef Bayer, John Bowers, Guglielmo Cinque, Peter Cole, Chris Collins, Marcel den Dikken, Günther Grewendorf, Jacqueline Guéron, Gabriella Hermon, Sumru Özsoy, Shigeru Miyagawa, Jean-Yves Pollock, Luigi Rizzi, Joachim Sabel, Eser Erguvanlı Taylan, and John Whitman. All shortcomings of the resulting study are to be blamed on the author. 202 Jaklin Kornfilt 1. In this paper, I shall use Agr for the overt morphological agreement marker, and AGR for the related syntactic position. I am using the term "extended projection" in the sense of Grimshaw (1991). However, contrary to that work, I do allow categorially mixed extended projections, especially for nominal predicates; for a discussion, see Borsley and Kornfilt (2000). See also Stowell (1981), Sabel (2002) for indexation of arguments via θmarking. In the concluding section, I speculate that in Turkish, the Agr element is a true nominal, similar to an external noun, and as such it is expected that it can and will inherit the θ-index of the domain that it heads. “Verbal predicate” refers to predicates whose functional projections are fully verbal (rather than mixed, i.e. including nominal layers, as is the case in the nominalized subordinate clauses which we will be discussing shortly). Thus, predicate adjectives and predicate nouns fall under the term “verbal predicate”, as they include either a copula, or else some sort of auxiliary, e.g. ol ‘be, become’, et ‘make”, etc., whereby these “light” verbs have their own verbal functional projections—unless they are nominalized, in which case such adjectival or nominal predicates would, of course, fall under the term “nominal predicate”. For agreement paradigms in Turkish, the reader is referred to reference grammars of Turkish, e.g. Lewis (1967), Kornfilt (1997). For discussion, see Kornfilt (1977) and Kornfilt (1996a). For an account of Turkish ECM, proposing distinct derivations for clauses with versus without overt Agr, see Moore (1998). “Genuine” tense does seem to be the Nominative Case licenser in Modern Greek. In ECM-constructions, only the present tense, i.e. the citation form, can show up; alternation with other tenses is not possible, and neither is a Nominative subject. Thus, non-alternating present tense is “fake”—and so is Agr in these forms that mimic the infinitive. Iatridou (1993) suggests that in Classical Greek, which did have an infinitive form, Agr was the subject Case assigner, while Tense is the subject Case assigner in Modern Greek. The correlation is suggestive for a possible parametrization, as Greek seems to have undergone a change from a Turkish-type language (i.e. with an Agrless, special infinitive form, and with Agr as the subject Case licenser) to a language without a dedicated infinitive form, where Agr, in those constructions where it does show up, is “fake” for purposes of subject Case licensing. One of the anonymous reviewers raises the objection that if this analysis of ECM constructions is correct for Turkish, it would incorrectly predict corresponding sentences to be ungrammatical in English. But, as a matter of fact, most native speakers I have consulted indeed judged examples like the translation of (4)a. to be ill-formed. Also, crucially, those speakers, while 2. 3. 4. 5. 6. 7. 8. Subject case in Turkish nominalized clauses 9. 10. 11. 12. 203 not allowing for WH-movement out of ECM-infinitivals, do allow for WHextraction out of Control infinitivals, i.e. from infinitivals that do have CPstatus. For such speakers, then, the analysis carries over to English straightforwardly. There were, to be sure, also some speakers who accepted such examples in English. I shall not attempt offering an account of the idiolects of those latter speakers. Note also that Iatridou (1993) suggests that Tense, the subject Case licenser in Modern Greek, must be in C so as to fulfill its Case-licensing function. Kornfilt and Greenberg (2000) give examples for both lexically and syntactically derived nominalizations as well as criteria to distinguish these two types. One criterion is argument structure. I assume that lexical nominalization can change the argument structure of a predicate, while syntactic nominalization cannot. Consequently, externalizing an internal argument can happen only via passive morphology in syntactic nominalizations, while lexical nominalization can effect such externalization directly, i.e. without passive morphology. Furthermore, sentence-level adverbs are possible with such predicates, in contrast with lexical nominalizations. Further discussion and examples can be found in Kornfilt (2000a) and in Borsley and Kornfilt (2000). An early work where criteria distinguishing lexical and syntactic nominalization for English is Chomsky (1970). For a concise, but very insightful and influential early generative treatment of nominalizations in Turkish, see Appendix C of Lees (1968). The gloss NFN (nominal non-factive) here refers not to the semantics, but to the morphology of this marker. For example, as will be shown later (in section 8), psychological predicates can take both “factive” and “nonfactive” nominalized forms with factive/indicative semantics. Following general Turkological practice, I use capital letters to represent phonologically determined alternations. Vowel alternations are determined by Vowel Harmony, and consonant alternations by a number of phenomena, e.g. devoicing of obstruents, and conversion of /K/ into /¤/. There is another way to accommodate the extraction-based contrasts sketched here: both types of nominal embedded clauses would be analyzed as CPs (embedded within DPs), but only the indicatives would have a [+WH] or, more generally, [+Operator] feature, while the nominal subjunctives would be [-Operator]. I have proposed such analyses in the past; see, e.g., Kornfilt (1993) and (1995b). Sabel (1996) makes a similar proposal for a number of other languages. While this analysis covers the data, I think that the proposal made in the present paper is less ad-hoc and more convincing, as it is in line with a number of other facts, i.e. those presented earlier in this section, showing categorial differences between the two nominal clause types. However, the question arises as to how an operator can be 204 13. 14. 15. 16. Jaklin Kornfilt extracted out of a non-factive nominal clause, given lack of a CP-layer. Examples like (10) and (13) illustrate the possibility of such extraction out of NFN-clauses when embedded under FN-clauses (which do have a CPlayer). I suggest that the AgrP-layer (i.e. the Finite ness layer) provides an escape hatch for the extracted operator; however, it is only the CP-layer (i.e. the Force layer) that provides the target position, i.e. the “resting place”, for such an operator (as opposed to a mere escape hatch position). Another apparent problem concerning adjunct clauses are posed by examples like the two last ones, where the adjunct clause is marked with a Case whose provenance is not immediately clear: if the superordinate predicate does not assign a θ-role to the clause, it also does not check for its Case. I propose that the Locative and Ablative here are licensed not via the regular mechanisms, but semantically. Following Larson (1985), I have claimed in Kornfilt (2000b) that not all Case is assigned by a θ-role assigner or via specifier–head agreement, but rather that some configurations require, for semantic reasons, certain Cases. The Locative and Ablative are required here to convey the semantics of something like "at a specific point in time", "for a particular reason". These Cases have to be overt, because otherwise the appropriate semantic interpretation could not be assigned to the respective clauses. For detailed discussion of the properties of PRO as contrasted with other empty categories in Turkish, see Kornfilt (1996b). For present purposes, the most important property of PRO is that of Control; pro, in contrast, is not controlled. European Portuguese evidently doesn’t make an overt, morphological distinction between nominal versus verbal Agr. Consequently, the subject Case licensed by Agr is of only one single shape, i.e. that of Nominative. The question arises here whether European Portuguese might also benefit from the index-based approach proposed here, thus enabling us to abandon the Case-based approach proposed by Raposo, and bringing it closer to the approach being adopted here for Turkish. One anonymous reviewer discusses a paradigm in European Portuguese which suggests that for this language, at least, it is Case rather than indexing due to θ−marking which licenses nominal Agr as a subject Case licenser. S/he shows that the preposition antes ‘before’ cannot assign Case and therefore needs the presence of de ‘of’, as a Case assigner for nominal prepositional complements. S/he further shows that antes may take fully tensed clausal complements, introduced by the complementizer que, without de. In contrast, when antes takes inflected infinitival clauses as its complement, then de becomes obligatory. The reviewer interprets these facts as meaning that in the latter instance, de is needed to license the Subject case in Turkish nominalized clauses 17. 18. 19. 205 nominal Agr of the inflected infinitive as a subject Case licenser via Case, because indexation of Agr via θ−marking is ensured even without de. Without knowing how uninflected infinitival clauses without overt subjects behave in such contexts, I am not sure how to evaluate such facts. It is clear that EP differs from Turkish in—among other properties—allowing nonprimary θ−roles (like those assigned by prepositions) to activate Agr in general, in contrast to Turkish, where such activation by non-primary θ−roles is not possible. Such a non-primary θ−role would be assigned to the inflected infinitive that is the complement of antes. It is possible that de is needed here not for the activation of Agr, but as a Case licenser for the entire infinitival clause, made nominal by its nominal Agr-head, and that activation of Agr is achieved by indexation as I have proposed—with the parametric difference to Turkish that a non-primary θ−role can achieve this indexation in EP. Further investigations along this line of inquiry must be left to future research. Relative clauses with non-subject targets bear the regular factive nominalization morphology, as expected (cf. Kornfilt 1984, among others, versus the established usage of terming this morphology object relativization). RCs whose targets are subjects have a special nominalization marker, as have RCs whose targets are contained within larger subjects, and RCs with targets in impersonal constructions. These facts have been discussed, with different proposals, by Underhill (1972), Hankamer and Knecht (1976), and Kornfilt (2000a), among others. Coindexation between the moved operator and the associated C is exploited in Kornfilt (2000a) to explain the occurrence of this special nominalization marker in these particular configurations. Kornfilt (1995b) proposes an analysis of Free Relative Clauses in Turkish such that there is a nominal head position in these constructions to which the relativization operator moves and to which, if the construction does have an Agr element, this element adjoins. Thus, Free RCs in Turkish are actually not headless. To be more exact than the formulation in the text, the head isn’t phonologically empty, either, when there is overt Agr, as the head is occupied with some phonological material (i.e. that of the Agr element), if the analysis in this older work is on the right track. Aygen (2002) analyzes examples like (44) and (45) as Free Relatives, as well. (She does not discuss comparatives.) She refers to Kornfilt (2001) as having made such a proposal, but also refers to additional sources as having offered the same analysis. Those references among her list which I could locate (leaving out an MA-thesis by B. Öztürk, which I have been unable to locate so far) do not offer such an analysis. The items in question are: Hankamer (1972), Sezer (1991) and an earlier version of Sezer (2002), and Kennelly (1996). Öztürk (2002), which appears to include relevant parts of 206 20. 21. 22. 23. 24. 25. 26. Jaklin Kornfilt her thesis, does mention the relevant Genitive/Nominative contrasts in nominalized indicative clauses as postpositional complements, but does not offer any account of that contrast, other than claiming that it shows the irrelevance of nominal Agr in determining subject Case—a claim which my current study argues against. It should be mentioned that Kornfilt (2001) was a presentation similar to the one at the earlier Leipzig workshop which forms the basis of the current study. This example is ill-formed in most stylistic registers but very formal ones. For a detailed study of default Case (under a similar, but not identical, view of this notion), see Schütze (2001). He offers a characterization which is similar to the one I have proposed in the text: “The default case forms of a language are those that are used to spell out nominal expressions (e.g., DP) that are not associated with any case feature assigned or otherwise determined by syntactic mech anisms.” (Schütze 2001: 206.) Another recent application of this notion is to be found in Szucsich (2002), who proposes the Instrumental as a default Case for adjuncts in certain constructions in Russian and other Slavic languages. This analysis of the infinitival morphology predicts that not only argument clauses, but also adjunct clauses that are headed by infinitival morphology and lack relevant indexation should display only PRO subjects and never overt subjects. This prediction is correct. This analysis also extends to irrealis relative clauses, illustrated in (42). The marker -(y)AcAK, glossed as FUTN, would now be analyzed as two morphemes: -(y)AcA under Mood, and -K under Agr, with the latter bearing negative value for agreement, just as with infinitives. This analysis would also draw a distinction between -(y)AcAK as the genuine Future Tense Nominal with indicative mood, and -(y)AcA(-K) as a (nominal) irrealis mood. In addition to more recent sources like Chomsky (2001), early sources of this assumption are Chomsky (1986) and Pesetsky (1982). The fact that nominal possessive phrases (which clearly lack any kind of Mood) are not sensitive to the argument-adjunct distinction and always exhibit Genitive specifiers, as we saw earlier, further shows that it is categorial features and not Mood features that determine the choice of licensed subject Case. Hiraiwa (2001) assumes raising of a verbal complex to C, as well. He does so for Japanese, in so called 'Ga-No conversion' instances, i.e. for instances where a Nominative subject can optionally show up in the Genitive. He also briefly discusses Genitive subjects in Turkish, for which he proposes similar raising. Our proposals were obviously made independently of each Subject case in Turkish nominalized clauses 27. 28. 207 other, as I have been made aware of that study only recently. I would like to thank John Whitman for having drawn my attention to it. Note, however, that our proposals, similar as they are, do differ from each other. I am assuming the raising of a verbal complex to C/D, but not only in those instances where Genitive is licensed (as Hiraiwa does), but everywhere. Thus, Nominative as a genuine subject Case is licensed by a raised verb, too. It is referential indexation on the raised verbal complex that includes an overt Agr element which licenses the appropriate subject Case, not raising per se. Furthermore, Hiraiwa is wrong in claiming that in Turkish, an overt complementizer blocks verb raising and thus Genitive subject Case. While Turkish does have (right-branching) subordination introduced by complementizers and with Nominative subjects, we have seen in this paper that leftbranching subordination without complementizers and with fully verbal predicates is possible, too. Lack of complementizer should make raising possible, according to Hiraiwa, and thus license Genitive subjects. Instead, the subject is Nominative. This shows that it is not raising of the verb to C per se that licenses a particular subject Case, but rather the category of the inflected verb complex, and, in particular, the category of the Agr. I would also like to point out that raising of a predicate to C in Turkish was proposed , as far as I know, for the first time in Kural (1993), with different motivation than my proposal in this paper. I am grateful to Marcel den Dikken for pointing out the similarity between the subject Case licensing mechanism proposed here and that found in ECM constructions. den Dikken raised this similarity as a problem. However, given the widely assumed nature of a Case-licensing predicate as having raised to C (and thus licensing subject Case in ECM-like configurations) in widely differing languages such as in Bavarian, European Portuguese, and Modern Greek, I view this aspect of my approach as unproblematic. I am grateful to Chris Collins for a suggestion along similar lines, after a presentation of this material at Cornell University. If Agr is a nominal head, what does it mean to say that Agr can be verbal, in those instances where I posited a verbal Agr? "Verbal" Agr would then simply mean a nominal agreement element which AGREEs with the verbal predicate in category features. Likewise, what I have called a "nominal" Agr is a nominal head which AGREEs with the categorial features of its phonological host, i.e. a nominal predicate or a nominal head of a domain. In categorially hybrid clauses, the Agr bears [+N] categorial agreement features which are in congruence with the higher K, but these features conflict with the verbal features of the Tense and C-layers of the clausal architecture. 208 Jaklin Kornfilt Abbreviations 1. 2. 3. ABIL ABL ACC ADV AGR Agr AgrP AOR CAUS CMPM DAT DP DVN FN FUT FUTN GEN K KP LOC M MP MST N NP NegABIL NEGN NF NFN NP Op PASS PL PRES.PART PROF First person Second person Third person Abilitative Ablative Accusative Adverbial Agreement as a syntactic node Agreement (as a morpheme); agreement in general Agreement Phrase Aorist Causative Compound marker Dative Determiner phrase Deverbal noun Factive nominalization Future Future nominalization Genitive Case as a syntactic node Case Phrase (as a functional syntactic projection) Locative Mood Mood Phrase Modern Standard Turkish Noun; nominal as a distinctive feature Noun Phrase Negative abilitative Negative nominalizer Nominal functional category Non-factive nominalization Noun phrase Operator Passive Plural Present participle Professional suffix Subject case in Turkish nominalized clauses PROG PRSPROG REL.PART REP.PAST RES SUBJNCT SG TAM V VBL.CONJ VF VP Progressive Present progressive Relativization participle Reported past Resultative Subjunctive Singular Tense/Aspect/Mood Verb; verbal as a distinctive feature Verbal conjunction Verbal functional category Verb phrase 209 210 Jaklin Kornfilt References Aygen, Gülﬂat 2002 Subject case in Turkic subordinate clauses: Kazakh, Turkish and Tuvan. In: Masako Hirotani (ed.), Proceedings of NELS 32, 563–580. Amherst: University of Massachusetts, GLSA. Bayer, Josef 1983–84 COMP in Bavarian syntax.The Linguistic Review 3: 209–274. Borsley, Robert D. and Jaklin Kornfilt 2000 Mixed extended projections. In: Robert D. Borsley (ed.), The Nature and Function of Syntactic Categories, 101–131. New York/ San Diego: Academic Press. Bresnan, Joan 1973 The syntax of comparative clause constructions in English. Linguistic Inquiry 4: 275–343. Bresnan, Joan 1975 Comparative deletion and constraints on transformations. Linguistic Analysis 1: 25–74. Chomsky, Noam 1970 Remarks on nominalization. In: Roderick A. Jacobs and Peter S. Rosenbaum (eds.), Readings in English Transformational Grammar, 184–221. Boston: Ginn. Chomsky, Noam 1977 On Wh-Movement. In: Peter W. Culicover, Thomas Wasow and Adrian Akmajian (eds.), Formal Syntax, 71–132. New York: Academic Press. Chomsky, Noam 1981 Lectures on Government and Binding. Dordrecht: Foris. Chomsky, Noam 1986 Knowledge of Language: Its Nature, Origin, and Use. New York: Praeger. Chomsky, Noam 2001 Derivation by phase. In: Michael Kenstowicz (ed.), Ken Hale: A Life in Language, 1-52. Cambridge, Mass.: MIT Press. Chomsky, Noam 2002 On Nature and Language. (Adriana Belletti and Luigi Rizzi, eds.) Cambridge: Cambridge University Press. Chomsky, Noam and Howard Lasnik 1991 Principles and parameters theory. In: Joachim Jacobs, Armin von Stechow, Wolfgang Sternefeld and Theo Vennemann (eds.), Syntax: an International Hand-book of Contemporary Research. Berlin: de Gruyter. Subject case in Turkish nominalized clauses 211 Dede, Müﬂerref 1986 Definiteness and referentiality in Turkish verbal sentences. In: Dan I. Slobin and Karl Zimmer (eds.), Studies in Turkish Linguistics, 147–164. Amsterdam /Philadelphia: John Benjamins. Enç, Mürvet 1991 The semantics of specificity. Linguistic Inquiry 22: 1–25. Erguvanlı-Taylan, Eser 1984 The Function of Word Order in Turkish grammar. [University of California publications in Linguistics, vol. 106.] Berkeley and Los Angeles: University of California Press. George, Leland and Jaklin Kornfilt 1981 Finiteness and boundedness in Turkish. In: Frank Heny (ed.), Binding and Filtering, 104–127. London: Croom Helm, and Cambridge, Mass.: MIT Press. Grimshaw, Jane 1990 Argument Structure. Cambridge, Mass.: MIT Press. Grimshaw, Jane 1991 Extended Projections. Ms., Brandeis University. Hankamer, Jorge 1972 Analogical Rules in Syntax. Proceedings of CLS 8: 111–123. Hankamer, Jorge and Laura Knecht 1976 The role of the subject/non-subject distinction in determining the choice of relative clause participle in Turkish. Proceedings of NELS VI, Montréal WPL 6: 123–135. Hiraiwa, Ken 2002 On Nominative-Genitive conversion. In: A Few from Building E39: MIT Working Papers in Linguistics 39: 67–126. Iatridou, Sabine 1993 On Nominative Case assignment and a few related things. In: Papers on Case & Agreement II: MIT Working Papers in Linguistics 19: 175–196. Kennelly, Sarah 1990 Theta control in Turkish. In: GLOW Newsletter 24: 72–73; abstract. Kennelly, Sarah 1996 Turkish subordination: [–Tense, –CP, +Case]. In: Modern Studies in Turkish: Proceedings of the 6th International Conference on Turkish Linguistics. Eskiﬂehir University, Turkey; 55–75. (Copyright 1996; conference held in 1992.) Knecht, Laura 1976 Turkish comparatives. In: Judith Aissen and Jorge Hankamer (eds.), Harvard Studies in Syntax and Semantics 2, 279–358. 212 Jaklin Kornfilt Kornfilt, Jaklin 1977 A note on subject raising in Turkish. Linguistic Inquiry 8:4. 736–742. Kornfilt, Jaklin 1984 Case marking, agreement, and empty categories in Turkish. Ph.D. dissertation, Harvard University. Kornfilt, Jaklin 1993 Infinitival WH-constructions and complementation in Turkish. In: Eurotyp Working Papers, Group 3: Subordination and Complementation, vol. 4, 66–83. Manchester: University of Manchester, as a European Science Foundation Working Papers volume. Kornfilt, Jaklin 1995 a Scrambling and incorporation in Turkish. In: Artemis Alexiadou, Nanna Fuhrhop, Paul Law and Sylvia Löhken (eds.), FAS Papers in Linguistics 1: 56–65. Berlin: FAS. Kornfilt, Jaklin 1995 b Constraints on free relative clauses in Turkish. In: Artemis Alexiadou, Nana Fuhrhop, Paul Law and Sylvia Löhken (eds.), FAS Papers in Linguistics 4, 36–57. Berlin: FAS. Kornfilt, Jaklin 1996 a NP-Movement and 'Restructuring'. In: Robert Freidin (ed.), Current Issues in Comparative Grammar, 121-147. Dordrecht: Kluwer. Kornfilt, Jaklin 1996 b Turkish and configurationality. In: Bengisu Rona (ed.), Current issues in Turkish linguistics 1: 111–125. Ankara: Hitit. (Copyright: London: SOAS, 1996; related conference held in 1990.) Kornfilt, Jaklin 1997 Turkish. London: Routledge. Kornfilt, Jaklin 1998 On rightward movement in Turkish. In: Lars Johanson (ed.), The Mainz Meeting: Proceedings of the Seventh International Conference on Turkish Linguistics, 107–123.Wiesbaden: Harrassowitz Verlag. Kornfilt, Jaklin 2000 a Some syntactic and morphological properties of relative clauses in Turkish. In: André Alexiadou, Paul Law, André Meinunger and Chris Wilder (eds.), The Syntax of Relative Clauses, 121–159. Amsterdam/Philadelphia: John Benjamins. Subject case in Turkish nominalized clauses 213 Kornfilt, Jaklin 2000 b Postpositions and adverbs: a case study in syntactic categories. In: Çi¤dem Balım and Colin Imber (eds.), The Balance of Truth: Essays in Honour of Professor Geoffrey Lewis, 217–237. Istanbul: The Isis Press. Kornfilt, Jaklin 2001 Subjects and their Case in Turkish/Turkic embeddings. Paper presented at the First Altaic Workshop, MIT. Kornfilt, Jaklin 2002 Functional projections and their subjects in Turkish clauses. In: Eser Erguvanl›-Taylan (ed.), The Verb in Turkish, 183–212. Amsterdam: Benjamins. Kornfilt, Jaklin and Gerald Greenberg 2000 Changing argument structure without voice morphology: A concrete view. In: Asl› Göksel and Celia Kerslake (eds.), Studies on Turkish and Turkic Languages, 51–56. Wiesbaden: Harrassowitz Verlag. Kural, Murat 1993 V-to(-I-to)-C in Turkish. In: Filippo Beghelli and Murat Kural (eds.), UCLA Occasional Papers in Linguistics 11: 1–37. Larson, Richard 1985 Bare-NP adverbs. Linguistic Inquiry 16: 595–621. Lees, Robert B. 1965 Turkish nominalizations and a problem of ellipsis. Foundations of Language 1–2; 112–121. Lees, Robert B. 1968 The Grammar of English Nominalizations. Bloomington/The Hague: Indiana University/Mouton. [Fifth printing; original publication in 1963.] Lewis, Geoffrey L. 1967 Turkish Grammar. Oxford: Oxford University Press. Miyagawa, Shigeru 1993 Case-checking and minimal link condition. In: Case and Agreement II, MIT Working Papers in Linguistics 19: 213–254. Moore, John 1998 Turkish copy-raising and A-chain locality. Natural Language and Linguistic Theory 16: 149–189. Öztürk, Balk›z 2002 Turkish as a non-pro-drop language. In: Eser Erguvanl›-Taylan (ed.), The Verb in Turkish, 239–259. Amsterdam: Benjamins. Pesetsky, David 1982 Paths and Categories. Doctoral dissertation, MIT. 214 Jaklin Kornfilt Pollock, Jean-Yves 1989 Verb movement, universal grammar, and the structure of IP. Linguistic Inquiry 20: 365–424. Raposo, Eduardo 1987 Case theory and Infl-to-Comp: The inflected infinitive in European Portuguese. Linguistic Inquiry 18: 85–109. Reuland, Eric 1983 Governing -ing. Linguistic Inquiry 14: 101–136. Rizzi, Luigi 1994 Argument/Adjunct (a)symmetries. In: Guglielmo Cinque, Jan Koster, Jean-Yves Pollock, Luigi Rizzi, and Raffaella Zanuttini (eds.), Paths Towards Universal Grammar: Studies in Honor of Richard S. Kayne, 361–376. Washington, D.C.: Georgetown University Press. Rizzi, Luigi 1997 The fine structure of the left periphery. In: Liliane Haegeman (ed.), Elements of Grammar, 281–337. Dordrecht: Kluwer. Sabel, Joachim 1996 Restrukturierung und Lokalität. Universelle Beschränkungen für Wortstellungsvarianten. Berlin: Akademie Verlag, Studia Grammatica 42. Sabel, Joachim 2002 A minimalist analysis of syntactic islands. The Linguistic Review 19:3, 271-315. Schütze, Carson T. 2001 On the nature of default case. Syntax 4: 205–238. Sezer, Engin 1991 Issues in Turkish Syntax. Ph.D. dissertation, Harvard University. Sezer, Engin 2002 Yaﬂayan Türkçe [Living Turkish]. Unpublished ms., Harvard University; version 4.2. Stowell, Tim 1981 Origins of phrase structure. Ph.D. dissertation, MIT. Szucsich, Luka 2002 Nominale Adverbiale im Russischen: Syntax, Semantik und Informationsstruktur. Munich: Otto Sagner. Tura, Sabahat Sansa 1986 Definiteness and referentiality in Turkish nonverbal sentences. In: Dan I. Slobin and Karl Zimmer (eds.), Studies in Turkish Linguistics, 165–194. Amsterdam /Philadelphia: John Benjamins. Underhill, Robert 1972 Turkish participles. Linguistic Inquiry 3: 87–99. Subject case in Turkish nominalized clauses 215 Watanabe, Akira 1996 Nominative-Genitive conversion and Agreement in Japanese: A Cross-linguistic perspective. Journal of East Asian Linguistics 5:373–410. Williams, Edwin 1994 Thematic Structure in Syntax. Cambridge, Mass.: The MIT Press.

Log In

Subject Case in Turkish nominalized clauses

Subject Case in Turkish nominalized clauses

Related Papers

RELATED PAPERS