Ad-Phrasal Affixes and Suspended Affixation*

ABSTRACT

Suspended Affixation (SA) in Turkish exhibits apparently complex patterns of acceptability. This paper argues that SA results when an ad-phrasal affix takes a coordinate structure complement. I argue that some affixes are not ad-phrasal and that some phrase types cannot be coordinated. These two factors account for the bulk of the observed restrictions on SA in Turkish.

1. The phenomenon

Suspended affixation (SA) is a widespread phenomenon in which affixes appended to the rightmost conjunct in a coordinate structure have scope over all the conjuncts.  It is particularly robust in Turkish, where it is found in both nominal and clausal domains.

In the nominal domain, the relevant suffixes are the plural, the possessive suffix, which marks person and number of the possessor, and case.1

(1) N – PL – POSS – CASE

Any one of the three can optionally be suspended. This is clear from the interpretations of (2) and (3) (see the translations), and from the syntactic contexts in which (4) can occur:

(2) kedi ve köpek-ler kedi-ler ve köpek-ler cat AND dog-PL "cats and dogs"

(3) kedi ve köpeğ-im kedi-m ve kopeğ-im cat AND dog-POSS1SG "my cat and dog"

(4) kedi ve köpeğ-i kedi-yi ve kopeğ-i cat AND dog-ACC "the cat and dog (ACC)"

Further, all three can be suspended:2

* Versions of this paper were presented at the International Conference on Turkish Linguistics (ICTL 14), at the 2008 LSA meeting in Chicago, to local audiences at Berkeley, UC Santa Cruz, and Boğaziçi University, and at the Cornell Suspended Affixation Workshop. I would like to thank the audiences at those presentations for feedback and suggestions. Special thanks are due to Jaklin Kornfilt for many fruitful discussions and help with grammaticality judgments, and to Judith Aissen for substantial help with framing the issues and making it possible for the paper to be written.1 See section 4 for a more articulated analysis of the Possessive.

1

123456789

1011121314151617181920212223242526272829303132333435363738

12345678

9

(5) kedi ve köpek-ler-im-i cat AND dog-PL-POSS1SG-ACC "my cats and dogs (ACC)"

SA is also found in the clausal domain. The examples in (6-8) show conjoined predicate adjectives and nominals, where subject agreement surfaces only on the rightmost predicate, but scopes over the conjoined structure.

(6) güzel ve zengin-im beautiful AND rich-1SG "I am beautiful and rich"

(7) güzel ve zengin-0-di-m beautiful AND rich-COP-PST-1SG "I was beautiful and rich"

(8) köy-ün en zengin adam-ı ve bölge-nin en ünlü hırsız-ı-y-dı-m village-GEN most rich man-POSS and district-GEN most famous thief-POSS-COP-PST-1SG "I was the village's richest man and the district's most famous thief"

Although not apparent in (7), the rightmost predicate in (7) and (8) carries other suffixes which in fact precede agreement. One is the copula, which is unpronounced in (7) but surfaces as –y when it follows a vowel (8, 9b). The other is tense, as in (7), (8), and (9b).

(9) a. hasta-yım sick-1SG

``I am sick.’’

b. hasta-y-dı-m sick-COP-PST-1SG ``I was sick.”

This is not the place to argue it, but it is fairly clear that there is no copula in the case of untensed predicates, as in (6) and (9). If there were a copula there, we would expect it to surface as /y/ in the second-person forms, but no /y/ appears: hasta-sınız, *hasta-y-sınız.

2 Orgun (1995a,b) observes a puzzling restriction on suspended affixation in the nominal domain: POSS (and later) affixes cannot be suspended if PL is present and not suspended:

(i) *kedi-ler ve köpek-ler-im-i

Inkelas and Orgun (1995) present an explanation of this restriction in terms of a theory of level ordering, which I am going to take as orthogonal to the concerns of the present paper.

2

3940414243444546474849505152535455565758596061626364656667686970717273747576

1011121314151617

18

The general template for predicate adjectives and nominals then is (10), where the boxed material can be suspended in coordination.3

(10)a. PRED – AGR1

b. PRED – COP – TNS – AGR2

SA is also found in verb-based forms which are suffixed with an aspect suffix (ASP). Aspect suffixes form participles which morphologically occupy the same position as PRED in (10). Thus, we find SA in (11) and (12), where the verb in each conjunct is suffixed with the progressive suffix (-Iyor) or the future suffix (-EcEk). The second participle in each case carries the copula (unpronounced), as well as tense and agreement. Tense and agreement are found only on the rightmost participle, but scope over both.

(11) çocuk-lar koş-uş-uyor ve gürültü yap-ıyor-0-lar-dı children run-RECIP-PROG AND noise make-PROG-COP-PL-PST "the children were running around and making noise"

(12) deniz-e gid-ecek, güneş-te kızar-tıl-acak, ve eğle-n-ecek-0-ti-k sea-DAT go-FUT sun-LOC roast-CAUS-PASS-FUT AND enjoy-RFL-FUT-COP-PAST-1PL "We were going to go to the sea, get roasted in the sun, and enjoy ourselves"

The general template here is given in (13), where again it is the boxed material which can be suspended.4

(13) [V+ASP] – COP – TNS – AGR2

In the verbal domain, the copula plays a key role, as it is only verbs which contain the copula that participate in SA. Verbs without an ASP suffix take tense (and agreement) directly, not mediated by the copula, and these do not permit suspension of tense (nor of agreement, see §4

3 AGR1 is a suffix from what some scholars call the "z-paradigm": the set of (apparent) agreement suffixes that attach to non-verbal predicates; AGR2 is a suffix from the "k-paradigm", which is a set of (actual) agreement markers that attach to verbal stems. I follow Good and Yu (2005) in assuming that the AGR1 forms are not morphological suffixes but pronominal enclitics. (10)b is in fact an oversimplification, because any suffix that follows the copula can be suspended along with it:

(i) aptal veya çocuk-0-muş-çasına "As if he was a simpleton or a child"simpleton or child-COP-EV-ASIF

4 This, too, is a simplification (see footnote 3). Any affixes that attach outside the copula can be suspended along with it:

(i) plaj-a gid-ecek ve voley oyna-yacak-0-mış-çasına "As if he was going to go to the beach and play volleyball"

3

7778798081828384858687888990919293949596979899

100101102103104105106107108

1920212223242526272829303132

33

below). Thus, while (14) (without suspension) is fine, (15) without tense (or agreement) is ill-formed:

(14) deniz-e git-ti-m ve eğlen-di-m sea-DAT go-PAST-1SG and enjoy-PAST-1SG "I went to the beach and enjoyed myself" (15) *deniz-e git ve eğlen-di-m

That these verbs do not contain the copula is clear when the verb stem is vowel-final: the form is anla-dı-m "understand-past-1sg", not *anla-y-dı-m.

Agreement (AGR2) alone cannot be suspended either:

(16) deniz-e git-ti ve eğlen-di-m

Example (16) is grammatical, but can only mean "He went to the sea and I had a good time". Similarly, (17) is ungrammatical on the intended reading given, though it is grammatical with the meaning "He went to the sea, he got roasted in the sun, and we enjoyed ourselves".

(17) *deniz-e git-ti, güneş-te kızar-t-ıl-dı, ve eğle-n-ecek-0-ti-k. sea-DAT go-PST sun-LOC roast-CAUS-PASS-PST AND enjoy-RFL-FUT-COP-PAST-1PL Intended: "We were going to go to the sea, get roasted in the sun, and enjoy ourselves"

Finally (for now), the ASP suffixes are never suspendable:

(18) *deniz-e git, güneş-te kızar-t-ıl, ve eğle-n-ecek-0-ti-k. sea-DAT go sun-LOC roast-CAUS-PASS AND enjoy-RFL-FUT-COP-PAST-1PL

The aim of the present paper is to develop an analysis that will make sense of this rather complex array of facts, and a few others which will surface as the investigation proceeds.

2. Previous accounts

Kornfilt (1996) proposed that SA arises when the copula takes a coordinate structure as complement, e.g. an AP, ASPP or other non-verbal phrase types. Being itself a clitic, the copula suffixes to the rightmost word of that complement. Tense (and Agreement) suffixes attached to the copula then scope over the coordinate complement, yielding SA.

(19) [[XP & XP]XP]+COP+TNS+AGR

4

109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152

34

Kornfilt’s account then is a fundamentally syntactic one which seeks to derive the properties of SA from syntactic coordination together with otherwise independently motivated principles governing the cliticization of the copula.

Kabak (2007) has criticized this analysis on the grounds that [a] it overgenerates, because it does not account for the ungrammaticality of (20) where the material to which the copula attaches does not participate in SA:

(20) *ev-imiz-i sat-sa ve bir dükkan al-sa-y-dı-k house-POSS1PL-ACC sell-COND AND a shop buy-COND-COP-PAST-1PL "If we sold our house and bought a shop"

and [b], it undergenerates, because it has nothing to say about SA in nominals:

(21) kedi ve köpek-ler-im-i cat AND dog-PL-POSS1SG-ACC "my cats and dogs (ACC)"

In essence, Kabak argues that the presence of the copula is neither a necessary nor a sufficient condition for SA. He assumes SA is found generally in coordinate structures, thereby allowing it in the nominal domain. To account for failures of SA, Kabak does not appeal to constraints on syntactic coordination, but rather develops a morphological account. In particular, he proposes a condition on SA that the final word in non-final conjuncts must be a "morphological word": a combination of root and affixes the last of which can legitimately be terminal in a word. The bare V 'git' in (18) is legitimate only as an imperative. In that case, Kabak assumes that there is an obligatory but inaudible imperative suffix, which is incompatible with the SA context in (18).

I will argue that while Kabak's proposal appears to account for a wide range of data, it leaves some puzzles about SA unexplained. I will suggest that Kornfilt's syntactic proposal is fundamentally correct but needs to be generalized. Once we understand why it works where it does, it will extend naturally to the nominal paradigm.

3. Facts to be accounted for

Before turning to that proposal, I summarize the basic facts of Turkish SA that need to be accounted for:

(a) SA only occurs in coordinate structures (in general, the coordinated elements are phrases, not words)

(b) The overt affixes are always to the right (realized as suffixes to the final word of the final conjunct)

(c) Derivational affixes can never be suspended (Kabak 2007). For example, the agentive suffix –cI cannot be suspended in (22b):

5

153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198

35

(22) a. eski-ci-ler ve naylon-cu-lar junk-AGT-PL AND plastic-AGT-PL

`` sellers of junk and sellers of plastic’’ b. *eski ve naylon-cu-lar junk AND plastic-AGT-PL

Intended: ``sellers of junk and sellers of plastic’’Okay: ``junk and sellers of plastic’’

The same is true in the verbal domain (see Kabak 2007 for extended discussion).

(d) In nominals, the PL, POSS, and CASE affixes can be suspended

(e) In verbs, the post-predicate affixes following the copula can be suspended (along with accompanying agreement), but no other verbal affixes

(f) Agreement affixes can never be suspended, unless attached to another suspended affix

The fact that agreement affixes by themselves cannot be suspended has been mentioned (cf. (16) in section 1). This is illustrated by (23), which is ungrammatical under an SA interpretation (of course it can mean “he went to the sea and I enjoyed myself).

(23) *deniz-e git-ti ve eğlen-di-m sea-DAT go-PAST and enjoy-PAST-1SG Intended interpretation: "I went to the beach and enjoyed myself"

This follows from Kornfilt’s account, since agreement can only be suspended when it is suspended along with the copula and the rightmost verb in (23) does not involve the copula. In Kabak’s account, this follows from the morphological requirement that every verb end in an agreement affix (whether pronounced or not). Thus the initial verb in (23) must end in 3sg agreement (ø), conflicting with the intended interpretation. However, an alternative explanation for the absence of SA effects in (23) is that the syntactic conditions which license SA are simply not met. I develop such an account in the next section.

4. Proposal

My proposal makes two basic assumptions:

(A1) Suspended affixes are ad-phrasal (i.e. they are heads in the syntax which take phrasal complements)

(A2) The phrases they attach to are coordinatable (not all phrases are).

6

199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242

36

This proposal adopts Kornfilt’s essential insight, namely that SA involves coordination of phrases. However, it generalizes her proposal so that any ad-phrasal head which takes a coordinatable complement is predicted to participate in SA.

It extends immediately then to SA in nominals. As long as each of the suffixes in (1) is associated with a syntactic head and as long as the complement of each of those heads can be coordinated, the patterns in (2)-(5) are predicted.5

In order to account for exceptions to SA, the strategy must be to show that one (or both) of the assumptions A1, A2 is not satisfied.

Consider, for example, the fact that derivational affixes cannot be suspended. The standard DM assumption is that all affixes are heads in the syntax; but even if this is accepted, it seems fairly clear that derivational affixes in general do not take phrasal complements, but rather take other heads as complements. I assume, in effect, that the affixes that we regard as derivational are not, in general, ad-phrasal.6

A perhaps more interesting case is that of agreement affixes, which we saw above can never be suspended on verbs. (16) is repeated below.

5 For evidence that affixes can be heads in the syntax and take phrasal complements, see Hankamer (2004), where it is argued that the relational suffix –ki is an ad-phrasal affix, combining with case-phrase (KP)-sized phrases to make larger phrases:

(i) [[o yüksek raf-ta]-ki mavi kitap-lar-da]-ki-ler-de that high shelf-LOC-ki blue book-PL-LOC-ki-PL-LOC 'in the ones in the blue books on that high shelf'

As indicated in the translation, the interpretation requires a structure as indicated by the brackets, which indicates that the suffix –ki combines with a KP. If –ki is an ad-phrasal affix, it seems plausible that every affix that attaches outside –ki is also ad-phrasal. These affixes include (as partially demonstrated in (i)) the nominal affixes POSS, PL, and CASE; and also the copula and everything that can follow the copula (i.e. all the post-predicate morphology dealt with in Kornfilt's analysis):

(ii) mavi ev-ler-de-ki-ler-0-di-kblue house-PL-LOC-ki-PL-COP-PST-1PL"We were the ones in the blue houses."

6 An interesting set of questions is lurking here, which I do not have space to explore. One question is whether roots can be coordinated, e.g. can [biber ve tuz]+luk be used to mean a shaker or holder for salt and pepper? I have gotten mixed judgments. But it is clear that if biber ve tuzluk exists, it cannot mean the same thing as biberlik ve tuzluk ("salt shaker and pepper shaker (two objects)"). This raises the question of what we mean by "suspended affixation": in all of the canonical examples discussed in the literature, we find both [A conj B]+Af and [A+Af] conj [B+Af], and the two are synonymous. If one of the variants is absent, or synonymy fails, would we still regard [A conj B]+Af as a case of suspended affixation?

7

243244245246247248249250251252253254255256257258259260

37383940414243444546474849505152535455565758596061626364

65

(16) deniz-e git-ti ve eğlen-di-m sea-DAT go-PAST and enjoy-PAST-1SG OK in the interpretation: "He went to the sea and I enjoyed myself."NOT OK in the interpretation: "I went to the sea and enjoyed myself"

Following e.g. Embick 2010, I assume that agreement affixes are not themselves syntactic heads, but are dissociated morphemes which are inserted post-syntactically. An agreement affix therefore cannot directly, by itself participate in SA. In order to (appear to) participate in SA, two conditions must be satisfied: first, the affix must itself be or be attached (post-syntactically) to an element which is a syntactic head; and second, that head must take a coordinatible complement.

Looking at (16), the coordinated elements are TP’s, since tense is present in each conjunct. Assuming that subject agreement spells out on T, there is no way for it to spell out only on the head of the right-hand TP conjunct without also spelling out on the head of the left-hand one. As long as agreement heads no syntactic phrase which takes TP as its complement, there simply is no way to derive (16) with the SA interpretation under present assumptions.

In contrast, (possessor) agreement can be suspended on nouns. Following Kornfilt 1985, I assume that possessed nominals are DP’s, that the possessor occupies Spec, DP, and that morphology associated with possession may spell out on the head of that DP, i.e on D. In Turkish, possessive morphology includes a possessive morpheme (-I) as well as agreement morphology. Examples like (3) with SA then have the structure shown in (20), where the suspended affix is associated with a syntactic head (D) and that head takes a coordinatable complement (NP).

(24) [ [NP & NP] D ]DP

The possibility of suspending possessor agreement on nouns then is entirely expected on this account, as is the impossibility of suspending subject agreement on verbs.

The direct analogue to (3) in the verbal domain is not (16) then, but (15), with a coordinate VP complement to T:

(15) *deniz-e git ve eğlen-di-m.

As noted earlier, Kabak derives the ungrammaticality of (15) from his condition on SA: namely that omission of affixal material must leave the left conjunct ending in a ``morphological’’ word. Pursuing a purely syntactic approach to SA, I will suggest below that the problem with (15) is that VPs cannot be coordinated in Turkish.

5. Coordinatability

5.1. Different coordinators

8

261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306

66

The relevance of coordinatibility to SA can be demonstrated in a striking way in nominal syntax. There are two morphemes which conjoin nominals in Turkish: ve and ile. ve is a general coordinator which conjoins like phrases of various types, including NP, DP, and TP. Ile conjoins only nominals. Yet even in the domain of nominal conjunction there are differences between ve and ile.

Either coordinator permits suspension of a CASE suffix alone:

(25) göz-lüğ-üm ve kalem-im-i eye-THING-POSS1SG AND pencil-POSS1SG-ACC "my glasses and pencil (ACC)"

(26) göz-lüğ-üm ile kalem-im-i eye-THING-POSS1SG AND pencil-POSS1SG-ACC "my glasses and pencil (ACC)"

However, only ve permits suspension of the possessive. In (27), the possessive scopes over both conjuncts; in (28), it does not.

(27) göz-lük ve kalem-im-i eye-THING AND pencil-POSS1SG-ACC "my glasses and pencil (ACC)"

(28) göz-lük ile kalem-im-i eye-THING AND pencil-POSS1SG-ACC "the glasses and my pencil (ACC)" not: my glasses and my pencil (ACC)

Continuing to assume that the possessive morpheme is D, (25)-(26) must involve coordination of DP’s, while (27) must involve coordination of NP’s. Both are possible with ve, but apparently only DP coordination is possible with ile. That is, the impossibility of (28) on the suspended reading follows if ile is restricted to coordinating DPs. This is quite plausible since this difference between ve and ile has an obvious historic source: while ve is a general coordinator, ile is derived from a postposition which took a DP complement (alternatively: it is syntactically still a postposition which takes a DP complement).

This case makes clear that an analysis of SA which puts coordination at the center requires an account of what can coordinate. While a full treatment is beyond the scope of this paper, the next two sections consider several contexts in which SA is not possible: coordination of unlike complements to the copula (§5.2) and coordination of certain nominalized verbal phrases (§5.3). In both cases, assumption A1 appears to be satisfied: the suspended affix is associated with a complement-taking head in the syntax. Following the logic of the present approach, we will explore the possibility then that A2 (coordinatability) is not satisfied. The discussion of §5.3 lends some support to the suggestion that VP’s do not coordinate in Turkish.

5.2 Coordinated predicates must match in type

9

307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352

67

We saw earlier that the copula takes complements of various sorts, including AP’s and AspP’s:

(29) hasta-y-dı-m ve öl-ecek-0-ti-m sick-COP-PST-1SG AND die-FUT-COP-PST-1SG "I was sick and I was going to die"

Interestingly, the SA construction is not an alternative to (29):

(30) *hasta ve öl-ecek-0-ti-m sick AND die-FUT-COP-PST-1SG TRYING TO MEAN: "I was sick and going to die"

In order to derive (30), AP and AspP would have to form a coordinate complement to the copula. Apparently this is impossible. Further, the condition that only like phrases coordinate extends further: AspP’s headed by distinct Asp heads also cannot participate in SA:

(31) *çürük yemek ye-diğ-im için hasta ol-uyor ve doktor-a gid-eceğ-im rotten food eat-PART-POSS1SG for sick get-PRES AND doctor-DAT go-FUT-1SG TRYING TO MEAN: "I am getting sick because I ate rotten food and I will go to the doctor"

Thus in addition to the ability of ad-phrasal affixes to attach to coordinated phrases, we need constraints on what phrases may be coordinated. While clauses containing unlike predicates may be coordinated, it seems that predicates containing unlike aspect suffixes may not be coordinated directly.

5.3. Infinitival constructions

The two infinitival constructions, the one in –mE and the one in –mEk, are coordinatable, and participate in (nominal) SA.

Examples (32-34) show the –mE infinitival in one of its common constructions, as the left member of a indefinite izafet group; examples (33-34) show that it is in general an infinitival phrase, and not just a verb, that enters into this construction.

(32) çalış-ma yer-i çalış-abil-me etiket-i work-MA place-POSS3SG work-ABL-MA license-POSS3SG work place ability-to-work license

(33) on-un-la gör-üş-ebil-me fırsat-ım ol-ma-dı he-GEN-WITH see-RECIP-ABL-MA opportunity-POSS1SG be-NEG-PST “I had no chance to converse with him”

(34) mektub-u oku-ma fırsat-ım ol-ma-dı letter-ACC read-MA opportunity-POSS1SG be-NEG-PST

10

353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398

68

“I didn't get an opportunity to read the letter”

Not surprisingly, an attempt to put a coordination of –mE phrases into this construction leads to grammaticality:

(35) on-un-la gör-üş-ebil-me ve sorun-u tartış-ma fırsat-ım ol-ma-dı he-GEN-WITH see-RECIP-ABL-MA and question-ACC discuss-MA opportunity-POSS1SG be-NEG-PST “I had no chance to converse with him and discuss the matter”

(36) Avrupa'-yı böl-me ve toplum-umuz-un iç barış-ı-nı tehlike-ye at-ma risk-i-ne gir-iyor-lar Europe-ACC split up-ME and society-POSS1PL-GEN inside peace-POSS1SG-ACC danger-DAT throw-ME risk-POSS3SG-DAT enter-PROG-3PL "They run the risk of splitting up Europe and putting the inner peace of our society in danger."

Unsurprisingly now, such a coordination leads to legitimate SA:

(37) Diğer iki program-ı biz-im çoğalt-ma ve siz-ler-e gönder-e-me-me-miz-in sebeb-i7

other two program-ACC we-GEN duplicate-ME and you-PL-DAT send-POT-NEG-ME-POSS1PL-GEN reason-POSS3SG "the reason for our not being able to duplicate the other two programs and send them to you"

Here the Possessive (plus Case), which as we have seen can be suspended in nominal constructions, can also be suspended when its complement is a –mE phrase. Similar behavior is exhibited by the –mEk infinitival:

(38) bu şehir-de yaşa-mak ve o şehir-de öl-mek-ten kork-uyor-um this city-LOC live-INF AND that city-LOC die-INF-ABL fear-PROG-1SG "I'm afraid of living in this city, and (of) dying in that city"

The infinitive phrase is "bigger" than the light infinitive phrase: it occupies the place of a DP (as seen by the fact that the only nominal affix that can attach to it is a case affix), while the light infinitive phrase is a complement of D (assuming the possessive morpheme is a D, following Kornfilt (1985)). Following Abney (1987), we might assume the light infinitive phrase is a VP, or something about the size of a VP. Of course it's not really a VP, because it's headed by –mE. Other affixes, all also presumably heads of phrases, that can intervene between a V and –mE are the Abilitative, the Negative, the (Negative) Potential, the Passive, the Reciprocal, and the Reflexive. None of these can be suspended, and no affix that attaches directly to (a phrase headed by) one of these can be suspended either. I propose that this is simply because there is no way to coordinate

7 There is an interesting issue with the scope of the negative in this example, which I am not going to try to explore.

11

399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440

6970

71

any of those phrases. For convenience, let us call phrases headed by V or any of the affixes named above, by a rather obvious convention, "verbal phrases".8

5.4 Question: Why should it be impossible to coordinate verbal phrases?

I will venture a conjecture here that, for want of time and space, I cannot develop as fully as it deserves; but I will attempt to show that it is at least plausible that the non-coordinatability of these phrases is a consequence of an architectural feature of the grammar. It is hard to produce independent evidence for the non-coordinatability of verbal phrases, because (so far as I know) they do not occur in any contexts other than as complements to the various affixes that they combine with. These include the ASP set, the infinitivals, the Tense/Conditional pair, and a range of adverbial and participial affixes. Without such a continuation, a verbal phrase is ungrammatical (unless interpreted as a rude imperative, in which case we might assume the existence of a silent imperative affix). If a Verb is required to raise at least as high as one of these heads, we will have an account of why verbal phrases are never found except in combination with one of them.9 We will at the same time have an account of why verbal phrases

8 This is perhaps as good a place as any to point out that the suffix –(y)Ip is not a coordinator, but rather an adverbializer. This suffix, which can attach to any verbal phrase, produces structures that superficially look like SA:

(i) ceket-im-i al-ıp gid-er-im jacket-1SGPOSS-ACC take-IP go-AOR-AGR2 "I'll take my jacket and go"

There are several reasons why –(y)Ip cannot be a coordinator. One is that the understood subject of the –(y)Ip phrase (a) can never be overt and (b) must be understood as bound by the subject of the matrix phrase. Another is that restoring the "suspended" affixes does not always lead to the same interpretation:

(ii) gel-ip gid-en-ler (iii) gel-en-ler ve gid-en-ler come-IP go-EN-PL come-EN-PL and go-EN-PL "those who come and go" "those who come and those who go"

(From Lewis 1967, p.178.)

Most tellingly, extraction from the main clause does not yield a coordinate structure violation:

(iii) Araba-yı, mayo-m-u giy-ip yıka-mak iste-di-mCar-ACC bathing-suit-POSS1SG put on-IP wash want-PST-1SG"The car, I wanted to put on my bathing suit and wash."

(iv) *Araba-yı, mayo-m-u giy-mek ve yıka-mak iste-di-m Car-ACC bathing-suit-POSS1SG and put on-INF wash want-PST-1SG

"The car, I wanted to put on my bathing suit and wash."

9 Something like this is suggested in Zanon (2014).

12

441442443444445446447448449450451452453454455456

72737475767778798081828384858687888990919293949596979899

100101

102

(with unlike verbs) can't be coordinated. The verb in each conjunct would be trapped inside the coordinate structure, unable to raise to the higher head.10 This would give substance to the insight of Kabak (2007) that the final word in the nonfinal conjunct of an SA construction has to be a legitimate word.

6. A Remaining Puzzle Now let us re-consider Kabak's problematic example:

(20) *ev-imiz-i sat-sa ve bir dükkan al-sa-y-dı-k house-POSS1PL-ACC sell-COND AND a shop buy-COND-COP?-PAST-1PL "If we sold our house and bought a shop"

(20') ev-imiz-i sat-sa-y-dı-k ve bir dükkan al-sa-y-dı-k house-POSS1PL-ACC sell-COND-COP?-PST-1PL AND a shop buy-COND-COP?-PAST-1PL "If we sold our house and if we bought a shop"

We see the same effect with the –dI+ydI combination:

(39) ev-e git-ti-y-di-k (ve) erken yat-tı-y-dı-k house-DAT go-PST-COP?-PST-1PL (and) early go-to-bed-PST-COP?-PST-1PL "We had gone home and we had gone to bed early." (40) ev-e git-ti (ve) erken yat-tı-y-dı-k house-DAT go-PST (and) early go-to-bed-PST-COP?-PST-1PL "He went home and we had gone to bed early."

Example (40) cannot mean the same thing as (39), i.e. SA is impossible here.

I have glossed the /-y-/ in these examples as COP? because I am not convinced that it is a copula. It certainly is not the same copula as we have found in many other examples, starting with (7), which I will henceforth call the real copula. The real copula is like a defective verb, and can be followed by only three suffixes: -dI (PST), sE (CND), and mIş (EV). The pseudocopula, if it is an affix at all, can only be followed by –dI and –sE. A verb form can contain both the real copula and the pseudocopula:

(41) gid-ecek-0-ti-y-se go-FUT-COP-PST-COP?-CND

10 If the verbs in the conjuncts were identical, then of course across-the-board raising should be possible, but it would not lead to suspended affixation. It is plausibly something like this that leads to "backward gapping" (cf. Ross 1970) in Turkish:

(i) Hasan yumurta-yı (ve) Orhan patlıcan-ı ye-di Hasan egg-ACC and Orhan eggplant-ACC eat-PST "Hasan ate the egg and Orhan the eggplant."

13

457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495

103104105106107108109

110

"If he was going to go"

And the pseudocopula, unlike the real copula, can occur outside agreement (both (42) and (43) are grammatical):

(42) gid-ecek-0-ti-y-se-m go-FUT-COP-PST-COP?-CND-1SG "If I was going to go"

(43) gid-ecek-0-ti-m-0-se go-FUT-COP-PST-1SG-COP?-CND "If I was going to go"

This latter property opens the possibility of an interesting experiment. Let's see what happens when we try to construct a SA situation with forms like (43):

(44) *çarşı-ya gid-ecek-0-ti-m veya ev-de kal-acak-0-tı-m-0-sa market-DAT go-FUT-COP-PST-1SG or house-LOC stay-FUT-COP-PST-1SG-COP?-CND "If I was going to go to the market or was going to stay home"

As seen in (44), SA is just as bad when agreement is inside the pseudocopula as when it is outside it (20, 40). So the reason for this badness cannot be dependent on whether the last word of the left conjunct is a legitimate word or not.

Unfortunately, my proposal does not explain these facts either. The pseudocopula, whatever it is, does seem to be the kind of thing that takes phrasal complements; and we have every reason to believe that phrases headed by –dI and –sE are coordinatable. I will have to leave this as an unsolved mystery. 7. Conclusion

I have argued that Turkish SA is, as Kornfilt suggested, the natural consequence of Ad-phrasal affixation interacting with coordination of the phrases to which the affix adjoins. This accounts for a wide array of facts, some of them otherwise unexpected, if we assume that only a limited inventory of phrases can be coordinated. It does not account for the failure of SA stranding the pseudocopula, and it looks like an explanation of that will have to await a better understanding of SA or the pseudocopula.

REFERENCES

Abney (1987) The Noun Phrase in its Sentential Aspect. MIT Diss.

14

496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541

111

Embick (2010) Localism vs. Globalism in Morphology and Phonology. MIT Press.

Good and Yu (2005) Morphosyntax of two Turkish subject pronominal paradigms.In Heggie and Ordonez (eds.), Clitic and Affix Combinations: TheoreticalPerspectives. 315-341.

Halle & Marantz (1993) Distributed Morphology and the Pieces of Inflection.In Hale and Keyser, eds., The View from Building 20. 111-176.

Hankamer, Jorge (2004) "Why there are two ki's in Turkish", in Imer and Dogan, eds., Current Research in Turkish Linguistics, Eastern Mediterranean University Press, 13-25.

Hankamer, Jorge (2004) "An ad-phrasal affix in Turkish", in Csirmaz, Lee, andWalter, eds., Proceedings of the Workshop on Altaic Formal Linguistics I,MIT Working Papers in Linguistics 46:289-299.

Inkelas, Sharon, and Orhan Orgun (1995) "Level Ordering and Economy in the Lexical Phonology of Turkish". Language 71.4:763-793.

Kabak, Baris (2007) "Turkish suspended affixation". Linguistics 45-2, 311-347.

Kornfilt, Jaklin (1985) Case Marking, Agreement, and Empty Categories in Turkish. Harvard University dissertation.

Kornfilt, Jaklin (1996) "On some copular clitics in Turkish". In ZAS Papers in Linguistics 6, Alexiadou et al., eds., 96-114. Berlin, Zentrum fur Allgemeine Sprachwissenschaft.

Lewis, G.L. (1967) Turkish Grammar. Oxford University Press.

Orgun, Cemil Orhan (1995) Flat vs. branching morphological structures: the case of suspended affixation. In Proceedings of the Berkeley Linguistics Society 21, Adlers, et al., eds., Berkeley Linguistics Society: 252:261.

Ross, John R. (1970) "Gapping and the Order of Constituents", in K. Bierwisch and M. Heidolph, eds., Progress in Linguistics. Mouton, The Hague.

Yu and Good (2000) Morphosyntax of two Turkish subject pronominal paradigms. Proceedings of the North Eastern Linguistics Society 30. Hirotani et al., eds. 759-773.

Zanon, Ksenia (2014) On the status of TP in Turkish. Studies in Polish Linguistics, vol. 9, issue 3, pp. 163-201.

15

542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587

112