ACTA ORIENTALIA

EDIDERUNT

SOCIETATES ORIENTALES DANICA FENNICA NORVEGIA SVECIA

CURANTIBUS LEIF LITTRUP, HAVNIÆ HEIKKI PALVA, HELSINGIÆ ASKO PARPOLA, HELSINGIÆ TORBJÖRN LODÉN, HOLMIÆ SIEGFRIED LIENHARD, HOLMIÆ SAPHINAZ AMAL NAGUIB, OSLO PER KVÆRNE, OSLO WOLFGANG-E. SCHARLIPP, HAVNIÆ

REDIGENDA CURAVIT CLAUS PETER ZOLLER

LXXIII

Contents

ARTICLES

S. GANESHRAM: Daśāvatāras in Tamil bhakti literature and programme of sculptures in Vijayanagara-Nāyaka art ...... 1 OTARED HAIDAR: The poetics the Iraqi War: Between Discursive Conflicts and Diasporic Discourse ...... 17 VIRGINIE PREVOST: Mağmāğ et les sept savants : la création du Dīwān al-‘azzāba ...... 35 R.K.K. RAJARAJAN: Antiquity of the divyakṣetras in Pāṇḍināḍu ... 59 JULIAN RENTZSCH: The evolution of Turkic modal verbs ...... 105 STELLA SANDAHL: The seven oceans in the Purāṇas and elsewhere ...... 151 DMITRY SHLAPENTOKH: Afanasiy Nikitin’s Journey Beyond the Three Seas: An Orthodox Russian in Medieval India ...... 173 STEFAN BOJOWALD: Zum ägyptischen Lautwandel zwischen „a“ und Gutturalen ...... 193

BOOK REVIEWS

BASSIOUNEY, REEM (ed.). Arabic and the Media: Linguistic Analyses and Applications, reviewed by Torkel Lindquist ...... 207 VIBEKE BØRDAHL AND MARGARET B. WAN (eds.). The Interplay of the Oral and the Written in Chinese Popular Literature, reviewed by Christopher Rosemeier ...... 213 Wörterbuch der tibetischen Schriftsprache, 16 fascicles (to be continued), reviewed by Per Kvaerne ...... 216

Acta Orientalia 2012: 73, 105–150. Copyright © 2012 Printed in India – all rights reserved ACTA ORIENTALIA ISSN 0001-6483

The evolution of Turkic modal verbs

Julian Rentzsch Szeged

Abstract

This paper investigates the mechanisms underlying the grammaticalization of Turkic modal auxiliary verbs. Special attention is paid to the linking device between main verb and auxiliary, i.e. the segment that corresponds to the infinitive in many Standard Average European modal constructions. Turkic modal constructions display a considerable variation regarding this linking device, which, however, can be reduced to a set of seven structural types, the origin and development of which will be investigated. It will be argued that interacting analogous mechanisms that build on formal and semantic principles contribute to the evolution and distribution of these types.

Keywords: Auxiliary verbs, root modality, grammaticalization, variation, analogy, .

1. Full verbs and auxiliary verbs

Turkic modal constructions are predominantly morphosyntactic units consisting of a main verb and an auxiliary segment, which may be of 106 Julian Rentzsch either nominal or verbal origin. This paper focuses on auxiliary verbs, although some constructions that entail a nominal auxiliary will be mentioned as well. The constructions to be discussed in the present article mainly pertain to the semantic domains of root possibility and willingness. As root modality does not constitute a closed formal class in the Turkic languages but resorts to morphosyntactic patterns that occur in other semantic domains as well, a non-modal type of construction that displays a similar behavior as the modal constructions to be investigated will also be considered in order to provide a broader formal and semantic context. This non-modal type is instantiated by a construction that denotes inception, i.e. ‘to begin’. All Turkic modal auxiliary verbs originate in full verbs. The lexical properties of full verbs include, besides their semantic value, specific syntactic characteristics, namely their argument structure, which implies specific government patterns. In the event that a full verb assumes auxiliary functions, in other words if it combines with another verb, some strategy is needed to link the main verb to the auxiliary. One option that is available in the Turkic languages is to build a nominal form from the main verb, i.e. a verbal noun, and to integrate the resulting nominal form into the original government pattern of the auxiliary. This strategy is in fact frequently found in the Turkic languages, but it is by no means the only one. Other widespread strategies include converb constructions, a shift in the government pattern and subjunctive uses of voluntative or optative forms. Hence, auxiliarization of an original full verb necessarily involves the choice of some linking device. This choice may be different in individual linguistic varieties, and some varieties may even permit several options, but in any event it is conventional in any given variety. Conventionalization of lexicosyntactic and morphosyntactic patterns is tantamount to grammaticalization (cf. Himmelmann 2004: 31). 1 Needless to say, the selection is not

1 For the purposes of this paper, grammaticalization is thought to entail all kinds of conventionalization processes that produce (simple or complex) grammatical items, i.e. developments from lexical items to grammatical ones, from lexico-grammatical combinations to grammatical items and from grammatical items to more grammatical ones. The starting point for every conventionalization process (lexicalization, idiomaticization and grammaticalization) will be labeled arbitrarity, which here Evolution of Turkic Modal Verbs 107 completely random but makes use of components that are already available in the language. On the other hand, the available components may change some of their morphological and syntactic properties. The following discussion pays special attention to the linking device between main and auxiliary verb, identifying the main types available in the Turkic languages, elaborating on their diachronic evolution and showing their approximate synchronic distribution within genetic and areal groups. Setting aside the details, the Turkic languages can be subdivided into four major branches – Southwest Turkic (Oghuz), Southeast Turkic (Karluk-Uyghur), Northwest Turkic (Kipchak) and South Siberian Turkic – plus three languages that cannot be directly linked to one of these branches – Yakut (and closely related Dolgan), Khalaj and Chuvash. For the purposes of this paper, a rough diachronic division into (8th–12th c.), Middle Turkic (13th–19th c.) and Modern Turkic (since 20th c.) will be sufficient.

2. Old Turkic full verb prototypes and their use in auxiliary constructions

Almost all Turkic modal auxiliary verbs may function as full verbs as well, and many of them are attested either as full verbs or as verbs with both full verb and auxiliary functions in Old Turkic, i.e. either in the Turkic Runic inscriptions of Siberia and Mongolia or in . Old Turkic verbs that became involved in Turkic modal constructions include u- ‘to be capable’ (Clauson 1972: 2), bil- ‘to know’ (Clauson 1972: 330) and bol- ‘to become’ (Clauson 1972: 331) for the semantic domain of possibility and küse- ‘to wish, desire, long for’ (Clauson 1972: 749), tile- ‘to seek’ (Clauson 1972: 492) and iste- ‘to seek, pursue’ (Clauson 1972: 243) for the semantic domain of willingness. Inception as the non-modal category consulted for comparative purposes is expressed by bašla- ‘to begin, to lead’ (Clauson 1972: 381–382), which is originally a denominal verb from baš ‘head, beginning’. The present study will focus on these seven signifies the spontaneous combination of lexical and grammatical items without a pre- existing model. 108 Julian Rentzsch verbs. All of them are attested as full verbs, so their argument structures and government patterns can be identified for the Old Turkic stage. Bol- as a verb denoting a change of state does not govern any (marked) case; its arguments (maximally2 two) are in the direct case, which is morphologically unmarked in Turkic. All other verbs mentioned could take direct objects in Old Turkic, which appear in the accusative case or, if unspecific, in the unmarked direct case:

(1) Šimnu küč-iŋe qop-uγ u-γay. NP power-POSS.3.DAT all-ACC be.capable-FUT ‘By the power of Ahriman he will be capable of everything.’ (Old Uyghur, M II 5, 10–11; cited from Clauson 1972: 2)

(2) Eki yiltiz-ig üč öð-ki nom-uγ bil-t-imiz. two roots-ACC three time-REL doctrine-ACC know-ASP-1.PL ‘We knew the two roots and the doctrine of the three times.’ (Old Uyghur, Xwāstvānift L 158–159)3

(3) Burxan qut-ïn küse-deči bodisatv-lar maxasatv-lar Buddha fortune-POSS.3.ACC desire-PTCP Bodhisattva-PL Mahāsattva-PL ‘Bodhisattvas and Mahāsattvas, which desire the Buddhahood’ (Old Uyghur, Altun Yaroq P1.02.02.r14–15)

(4) Alqu-dïn sïŋar nom-uγ tile-yü all-ABL direction Dharma-ACC seek-CV ‘Seeking the Dharma everywhere’ (Old Uyghur, BT13.19.A.1.l.A01–02)

(5) Ïraq-ta iste-d-i tözün-ler iz-in. far-LOC seek-ASP-3 Ārya-PL trace-POSS.3.ACC ‘In the distance he sought the traces of the Āryas.’ (Old Uyghur, Xuanzang Biography VII: 1465–1466)

2 Argument slots can be left empty in Turkic. 3 If not stated differently, Old Turkic examples are cited according to the VATEC database (see sources). The transcription has been adapted to the principles of Johanson & Csató (1998: XVIII–XIX). Evolution of Turkic Modal Verbs 109

(6) Öz-üŋe öŋ iš-in bašla-γïl. self-POSS.2.SG.DAT front work-POSS.3.ACC begin-IMP.SG ‘Begin with the work which confronts you.’ (Old Uyghur, TT I 150; cited from Clauson 1972: 382)

Some of these verbs are already attested in auxiliary functions in Old Turkic. Root possibility is commonly expressed by the auxiliary u-, which combines with the main verb by means of a converb (gerund), which is suffixed to the main verb stem: . In Runic Turkic, the converb is -(y)V (see ex. 7), whereas both -(y)V and -GAlI occur in Old Uyghur, the latter form being much more common (ex. 8). Another possibility marker that is very frequent in Old Uyghur (though nonexistent on Runic monuments) consists of the converb in - GAlI and the auxiliary bol- (ex. 9).4 It seems to be confined to participant-external possibility (for the terminology, see van der Auwera & Plungian 1998). Additionally, there are rare early occurrences of -(y)V bil- as a marker of participant-internal possibility (ex. 10).5

(7) Yaγï bol-up ėt-in-ü yara-t-un-u enemy become-CV organize-REFL-CV be.suitable-CAUS-REFL-CV u-ma-duq yana ičik-miš. can-NEG-PTCP again submit-ASP ‘Having become enemies [of the Chinese Qaγan], they who were unable to organize themselves submitted again.’ (Runic Turkic, Köl Tėgin East 10; Berta 2004: 145)

4 von Gabain (1941 [31974]: 127) and Erdal (2004: 259) report the existence of -(y)V bol- as a possibility marker besides -GAlI bol- and cite the example ör-ü bol-maz ‘it is impossible to rise’. However, not a single occurrence of this item can be traced anywhere in the VATEC corpus, and it seems doubtful whether the sole example cited by the two scholars is significant. The claim that -(y)V bol- is “the normal positive counterpart of -U uma- [...] in Uyghur” (Erdal 2004: 259) is not supported by the data. 5 A single occurrence of bil- with another converb as a marker of possibility is even found in the Toñuquq inscription, one of the oldest Turkic texts (726 A.D.): tė-yin bil- mez ermiš ‘one obviously cannot say’ (Toñuquq 5–6, Berta 2004: 46). 110 Julian Rentzsch

(8) Belgürtme et’öz-üg belgürt-geli u-yur. manifestation body-ACC make.manifest-CV can-ASP ‘He can emanate the Nirmāṇakāya.’ (Old Uyghur, Altun Yaroq B01.25.v05–06)

(9) Ol yol-ča bar-ïp arïtï sansar-dïn oz-γalï DEM way-EQU go-CV completely Saṃsāra-ABL escape-CV bol-maz. become-NEG.ASP ‘Walking on that path it is completely impossible to escape from Saṃsāra.’ (Old Uyghur, Maitrisimit 0.08.r29–30)

(10) Bir-ig išle-t-ü bil-mek-i üze ink.brush-ACC work-CAUS-CV know-VN-POSS.3 above ‘As he manages to use the ink brush’ (Old Uyghur, Xuanzang Biography VIII: 1042)

As regards volitive modality (Palmer 22001: 76–78), the auxiliary that is usually employed in Old Turkic is küse-, which follows the -GAlI converb of the main verb (11). The other Old Turkic verbs that developed into widespread volitive auxiliaries, tile- and iste-, do not seem to occur with the auxiliary function in Old Turkic.6

(11) Yėrtinčü-te eðgü qïlïnč-ta yorï-γalï küse-ser world-LOC good deed-LOC walk-CV wish-COND ‘If they wish to dwell in the world in good Karma’ (Old Uyghur, Altun Yaroq P1.02.19.v01–02)

6 A possible exception is iste- in the passage that has been reconstructed as Til[e-geli] iste-ser tüpker-geli bol-maz töz-in tüp-in ‘If one wants to seek it one cannot fathom its root and its base’ (Old Uyghur, Xuanzang Biography VII: 0044–0045). It seems more likely, however, that the reconstructed part should be read Til[e-ser] iste-ser ‘If one seeks and searches’. Tile- and iste- often co-occur in identical form as a hendiadyoin, in which case both members function as full verbs. Evolution of Turkic Modal Verbs 111

Inception is expressed by the auxiliaries bašla- (transitive) and bašlan- (intransitive), which combine with the main verb by means of the converb either in -(y)V (12) or in -GAlI (13).

(12) Ač-maq tupul-maq-lïγ edrem öz-in ač-ïl-u open-VN penetrate-VN- virtue self-INS open-PASS-CV DNN bašla-n-tï. begin-REFL-ASP ‘The virtue of opening and penetrating started to unfold by itself.’ (Old Uyghur, Xuanzang Biography VIII: 0519–0521)

(13) Yoog šastr nomla-γalï bašla-mïš. yoga śāstra preach-CV begin-ASP ‘He has begun to preach the Yogācārya-śāstra.’ (Old Uyghur, Xuanzang Biography III: 0815)

Despite rare possible examples to the contrary (Schlüter 2010), grammaticalization is almost always irreversible (Haspelmath 1999). We may therefore confidently infer that with those Old Turkic verbs that occur both as full verbs and as auxiliary verbs, the full verb function predates the auxiliary function, even if the auxiliary function is significantly more often attested than the full verb function, as is the case with u-. This idea will be pursued throughout the article, hence the combinatory properties of the full verbs will be considered prior to those of the auxiliaries. Moreover, the Old Turkic stage, inhomogeneous as it is due to its temporal extension and its dialectal diversity, will serve as the background against which younger data will be evaluated. Therefore, the situation reflected in the Old Turkic data will be considered the “original” one, from which other situations “deviate”, although we cannot be sure of the state of affairs in equally old but unattested Turkic varieties. This modus operandi may seem somewhat incautious and dissatisfying, but it is unavoidable as we must rely on data that are actually available. The data reveal that two converbs, -(y)V and -GAlI, connect main verb and auxiliary. Neither converb is restricted to auxiliary constructions. Both can occur independently to form adverbial clauses. In this case, -(y)V denotes the intraterminal (imperfective) 112 Julian Rentzsch aspect (von Gabain 1941 [31974]: 123; for the terminology, see Johanson 1971, 2000), while -GAlI has a purposive meaning (‘in order to’). The distribution of these two converbs varies between the individual auxiliaries: while some auxiliaries allow only one converb, others can combine with either of the two. The distribution is shown in Table 1. The data for u- from Runic Turkic and Old Uyghur might suggest that -(y)V is the older formative in auxiliary constructions, but this remains speculative absent data for other auxiliaries. Not at all attested in Old Turkic are constructions involving verbal nouns (with the possible exception of

Table 1. Auxiliary verbs and linking devices in Old Uyghur.

u- bil- bol- küse- bašla(n)- ‘to be capable’ ‘to know’ ‘to become’ ‘to want’ ‘to begin’ -(y)V x x x -GAlI x x x x

3. Later stages

Some of the Old Turkic auxiliary constructions survive in individual later Turkic varieties, others have been replaced by diverging structures. Modern Uyghur preserves -GAlI bol- (with slight phonetic changes in the converb segment) as a marker of participant-external possibility (14) and -GAlI bašla- as a marker of inception (15). The converb in -(y)V developed into -(y)A or -A during the Middle Turkic era,7 but apart from this change some of the auxiliary constructions that employ this converb continue to exist. A reflex of *-(y)V u- survives in the Modern Turkish marker of negative possibility - (y)AmA-, e.g. oku-yama-dım ‘I could not read it’ (Clauson 1972: 2), while *-(y)V bil- became quite popular in the Middle Turkic era and is today mainly attested in Western Oghuz as well as in some Kipchak

7 The niceties of phonetic developments in the individual Turkic varieties as well as the morphophonological variation that is triggered by stem features will generally be ignored in this paper except where relevant to the present topic. Evolution of Turkic Modal Verbs 113 languages, including Tatar and Kazakh, from which ex. (16) is cited.8 -A bašla- occurs in the Kipchak group (e.g. Tatar, Bashkir, Kazakh and Kirghiz) as well as in Uzbek (the Karluk-Uyghur group). Example (17) is from Uzbek. Besides these combinations, which continue an old type of formation, *-(y)V has been a very productive linking device in various auxiliary constructions, especially actionality operators (Aktionsarten), some of which evolved further into aspect markers (see e.g. Johanson 1995), but it occurs in some innovative modal constructions as well, especially in the highly prolific possibility marker -A al-, the auxiliary segment of which originally denoted ‘to take’ (Clauson 1972: 124). The first attestations of -A al- are found in the Codex Cumanicus (Kipchak Turkic, early 14th c., see ex. 18). This marker was also very common in Chaghatay Turkic (15th–19th c.). Today it is the general marker of possibility in Southeast Turkic and a large number of (see ex. 16 for a Kazakh example). A very marginal appearance of -(y)A in a volitive construction is found in a Middle Azerbaijani transcription text from Isfahan (17th c.), where it combines with the verb iste- ‘to seek’ as auxiliary (19).

(14) Bu aptobus-qa oltur-idiγan-lar nahayiti köp, öčret-te DEM bus-DAT sit.down-PTCP-PL extremely many row-LOC tur-mi-sa bėlit-i-ni al-γili bol-ma-ydu tėxi. stand-NEG-COND ticket-POSS.3-ACC take-CV become-NEG-ASP.3 even ‘Those who want to travel on this coach are extremely numerous; one cannot even obtain a ticket without standing in line.’ (Modern Uyghur, Šal 2006: 51)

(15) Eti-si u-lar mėn-i yene soraq qil-γili bašli-di. next.day-POSS.3 DEM-PL I-ACC again question do-CV begin-ASP ‘The next day, they started interrogating me again.’ (Modern Uyghur, web-1)

8 In Tatar and Kazakh, this marker denotes participant-internal possibility, while in Western Oghuz it covers participant-internal, participant-external, deontic and (in specific constellations) even epistemic modality. 114 Julian Rentzsch

(16) “Sawat-ïŋ bar ma?” dep sura-ydï. “Žaz-a literacy-POSS.2.SG existent Q QUOT ask-ASP.3 write-CV bil-e-siŋ be?” “Sawat-ïm bar, žaz-a known-ASP-2.SG Q literacy-POSS.1.SG existent write-CV bil-e-min,” de-ydi Bäkir. Öz-i orta mektep-te know-ASP-1.SG say-ASP.3 NP self-POSS.3 middle school-LOC sabaq ber-etin muγalim eken. “Orïsša ma, qazaqša ma?” class give-PTCP teacher evid Russian Q Kazakh Q “Azdap orïsša da žaz-a al-a-mïn.” a.little Russian too write-CV take-ASP-1.SG ‘“Are you literate?”, he asks. “Can you write?” “I am literate, I can write”, says Bäkir. He is a middle school teacher. “Russian or Kazakh?” “I can also write a little Russian.”’ (Kazakh, web-2)

(17) Hali yigirma-ga ham bor-ma-g‘an=dir, mo‘ylab-i yet 20-DAT too go-NEG-ASP=EPIST moustache-POSS.3 ham endi-gina chiq-a boshla-g‘an. too now-just come.out-CV begin-ASP ‘He will not even be twenty; his moustache has just begun to grow.’ (Uzbek, Qodiriy 1926 [1994]: 32)

(18) Ol ilan boy-na iamanlich et-se, DEM serpent body-POSS.3.DAT evil do-COND ǵan-ina et-e al-mas. soul-POSS.3.DAT do-CV take-ASP ‘Even if the serpent does harm to the body, it cannot do harm to the soul.’ (Middle Kipchak, Codex Cumanicus 125; Kuun 1880: 168)

(19) Gandi giouab et-oub dee-di gud-é iste-menem s/he answer do-CV say-ASP go-CV want-NEG.ASP.1.SG on-dan sonra puchman ol-oub gueti. dem-ABL after repentent become-CV go.ASP ‘He answered: I do not want to go. Later, he repented and went.’ (Middle Azerbaijani, AZR 94, 14–15; Mt 21.29) Evolution of Turkic Modal Verbs 115

Besides the two converbs mentioned so far, some other converbs play a role as linking devices in modal auxiliary constructions, particularly the converb in -(V)p, which combines with bil- ‘to know’ (Turkmen), bol- ‘to become’ (Chaghatay and a large number of modern languages) and al- ‘to take’ (Shor) to designate various types of possibility, with iste- ‘to seek’ (rarely in Chaghatay) to designate willingness and with bašla- ‘to begin’ (Noghay) to mark inception. Also, the conditional (Old Turkic -sAr, most later Turkic varieties -sA) is involved in some modal constructions, e.g. together with bol- ‘to become’, as a marker of participant-external and deontic possibility. Auxiliary constructions with -(V)p and -sA become increasingly popular from the Middle Turkic period onward, although some even occur in Old Turkic (see Section 8). In the Middle Turkic era, a new type of modal construction emerges in which the modal value materializes in a matrix clause while the action to be modified by the modal value appears in a subordinate clause. The verb in the subordinate clause takes a mood suffix, typically optative, voluntative or conditional,9 which functions as a subjunctive. The modal value of the matrix clause is either encoded by nominal elements, such as mumkin ‘possible’, imkān ‘possibility’ (both borrowed from Arabic) or *kergek ‘necessary, necessity’, or by verbal elements, such as tile- or iste- ‘to want’. Example (20) is cited from Chaghatay, example (21) from Anatolian Oghuz (both texts from the 16th c.). The underlying structure of both examples is completely identical, with a finite form of the verb tile- in the matrix clause, a complementizer ki(m)10 and an optative form with the subjunctive function in the subordinate clause.11 It was partly

9 Examples (from the Chaghatay Baburnama) for different mood forms: optative: Mumkin ėmes ėdi kim alarγa bėril-gey ‘It could not be given to them’ (51b10–11); voluntative: Kėrek kim ... čiq-sun-lar ‘They must set out’ (360b1–2); and conditional: Kėrek kim ... kėl-se-ler ėdi ‘They should have come’ (185b14). 10 Ki is formally identical to the Persian complementizer, while kim has been grammaticalized from the Turkic interrogative word ‘who’. In some Middle Turkic texts, such as the Dede Qorqud Oγuznameleri (from which ex. 21 is cited), kim and ki are interchangeable. 11 The same construction exists with the auxiliary iste- as well (although not with identical geographic distribution), e.g. Dirse Xān iste-d-i kim oγlanǰuγ-ï-nuŋ üst-ine güvle-yüp düš-e-ydi ‘Dirse Khan wanted to cast himself upon his little son’ (Middle Oghuz, Dede Qorqud 13b3–4; Tezcan & Boeschoten 2001: 41) from the same source as ex. (21). 116 Julian Rentzsch contact with Persian, which possesses similar structures, that gave rise to this type of construction (on other factors see Section 8).

(20) Tile-dük kim bu qiš ‘āriyat-i bėr-gey. wish-ASP.1.PL COMP DEM winter loan-POSS.3 give-OPT.3 ‘We wished that he would lend it to us for this winter.’ (Chaghatay, Baburnama 59a4–5)

(21) Deli beg dile-d-i ki Dede-yi depe-re čal-a. lunatic nobleman wish-ASP-3 COMP np-ACC head-DIR hit-OPT.3 ‘The lunatic nobleman wanted to hit Dede Qorqud on the head.’ (Middle Oghuz, Dede Qorqud 44a9–10; Tezcan & Boeschoten 2001: 75)

In this sort of construction, the subject of the wish-verb need not be identical to the subject of the subordinate verb (in ex. 20 it is different, in ex. 21 it is identical). Consequently, it encompasses not only the relation expressed by the English construction to want to, but a more broad volitive relation Si wants Si/j to X. While this construction type survives in areas with intensive Turkic-Iranian contact (Kıral 2001; Stein 2010), there is also a structure that displays a stronger degree of integration and involves other auxiliary verbs besides tile- and iste-. This structure dispenses with the complementizer ki(m) and features an auxiliary verb in a finite form and a full verb in a mood form, which serves as a subjunctive. This type of construction is typical of Turkic varieties in the Balkans and adjacent areas (Johanson in print),12 e.g. Gagauz and the Western Rumelian Turkish dialect of the city of Vidin in Bulgaria (henceforward Vidin Turkish), from which examples (22) to (24) are cited.13 The subvariant with iste- (in which the semantics of the auxiliary and the linking segment are most easily reconcilable) has also made its way into the less formal registers of Standard Turkish, where the auxiliary segment always follows the main verb, while in

12 The loss of the complementizer is also frequent in the Turkic varieties of Iran, but the range of auxiliaries does not seem to have expanded (cf. Kıral 2001; Stein 2010). 13 The text was recorded between 1931 and 1938 (Németh 1965: 11). Evolution of Turkic Modal Verbs 117

Vidin Turkish the word order is flexible.14 While the subjunctive function in the periphrastic constructions with a complementizer could be performed by various mood forms (see above), the Balkan structure developed paradigms with predictable mood forms in the individual persons (e.g. 1st and 3rd person: voluntative, 2nd person: optative in the Standard Turkish paradigm; see Rentzsch 2010: 218).

(22) Ben yi-yeyim sen-i iste-y-im. I eat-VOL.1.SG you-ACC want-ASP-1.SG ‘I want to eat you.’ (Vidin Turkish, Németh 1965: 251)

(23) Bu hayvan-nar oqu-sun bil-ir=mi? DEM animal-PL read-VOL.3 know-ASP=Q ‘Can these animals read?’ (Vidin Turkish, Németh 1965: 185).

(24) Qïs bašla-r aγla-sïn. girl begin-ASP cry-VOL.3 ‘The girl starts to cry.’ (Vidin Turkish, Németh 1965: 144)

A completely different strategy of linking main verb and auxiliary involves verbal nouns. Verbal nouns have been common derivational tools during all documented stages of Turkic. Word forms that consist of a verb stem and a verbal noun suffix can fill any syntactic slot that is suitable for nouns. Verb stems usually preserve their argument structure, with the common exception that subjects frequently appear in the genitive instead of the unmarked case. (In this case, the verbal noun phrase is re-analyzed as a nominal phrase with its double marking [dependent→genitive, head→possessive] features.) Given that verbal nouns are suitable for any nominal slot, they qualify for all kinds of arbitrary and conventionalized constructions, including modal ones. Strikingly, in Old Turkic, verbal nouns are hardly ever

14 Original Turkic word order principles prefer the auxiliary segment to follow the main verb. This tendency is often violated in varieties under strong contact influence, an issue that is irrelevant to the present discussion. 118 Julian Rentzsch used in modal constructions with verbal auxiliaries. They frequently occur, however, in constructions with nominal auxiliaries such as kergek ‘necessary, necessity’ (25), where a selection of four different verbal nouns can be found in Old Uyghur alone, namely -mAK, -sVK, -GU and -mIš (Erdal 2004: 526–527). In terms of syntactic function, the verbal noun phrase acts as the subject and the nominal auxiliary as a predicate, thus reflecting a perfectly normal Turkic sentence structure without any formal change.

(25) Bir ay čaxšapat tut-maq kergek erti. one month commandment keep-VN necessary PST ‘It was necessary to observe the One-Month Commandment.’ (Old Uyghur, Xwāstvānift L 274–275)

Although virtually unattested in Old Turkic, employing verbal nouns as a linking device between main and auxiliary verb is a most natural choice, which is actually extremely common almost everywhere in the Turkic world since early Middle Turkic times. Very often, the verbal noun phrase is integrated into the matrix structure according to the original government features of the auxiliary. Although deviations can be observed very early (see below), it is very likely that this is the initial stage of the linking strategy that uses verbal nouns. Example (26) from Middle Oghuz instantiates a possibility marker comprising the verbal noun -mAK in the unmarked case and the auxiliary verb ol- (<*bol-) ‘to become’, which can roughly be rendered ‘X-ing is (not) possible’.15 The verbal noun phrase occupies the subject slot in this construction, thus complying with the usual argument structure of (b)ol-.

(26) Oγul, ṣabāḥ var-ub öylen gel-mek ol-maz, son morning go-CV noon come-VN become-NEG.ASP.3 öylen var-ub axšam gel-mek ol-maz. noon go-CV evening come-VN become-NEG.ASP.3

15 Whether or not the auxiliary verb is negated is irrelevant to the present study, although it may become structurally relevant in some constructions. (In some languages, for instance, -(V)p bol- exists only in negated form and *-(y)V u- survives in Turkish only as a marker of negative possibility.) Evolution of Turkic Modal Verbs 119

‘Son, one cannot simply leave in the morning and return at noon, or leave at noon and return in the evening.’ (Middle Oghuz, Dede Qorqud 88a4–5; Tezcan & Boeschoten 2001: 124)

Except bol-, all the full verbs mentioned here that grammaticalized to auxiliaries originally governed the accusative, and most of them (except u-, which is almost completely obsolete) appear in auxiliary constructions with a verbal noun in the accusative as well. In Uzbek and Turkish, for instance, there are markers of possibility of the type (ex. 27–28) and volitive markers of the type (ex. 29–30). 16 Bašla- ‘to begin’ is occasionally attested with a verbal noun in the accusative as well (see ex. (31) from Kazakh). In volitive constructions the pattern can be slightly varied in order to express the wish that the action be performed by a different subject (Si wants Sj to X). This is realized by expressing the subject of the action (i.e. Sj) with possessive suffixes, which are inserted between the verbal noun and the accusative suffix: . This is a very widespread strategy. Occasionally, this structure is also employed in same-subject wishes (Si wants Si to X = Si wants to X); see ex. (32) from Yakut (with another auxiliary baγar- ). Here, the double marking of the personal referent renders a redundancy on the one hand but unifies the patterns for same-subject and different-subject wishes (Si wants Si/j to X) on the other, thus covering the same broad volitive relation as the subjunctive pattern (cf. ex. 20–22 above). The pattern can serve as the starting point for further grammaticalization as in the Tuvan same-subject volitive construction shown in ex. (33), which contains a fossilized third person possessive marker irrespective of the person of the auxiliary. (The auxiliary küze- in this construction is etymologically identical to the wish-verb küse- already familiar from Old Turkic and preserves its original government pattern.)

(27) Ana, o‘g‘l-im, biz-ning xalq-ning hol-i-ga behold son-POSS.1.SG our people-GEN condition-POSS.3-DAT

16 The semantic niceties that distinguish these markers from other, more frequently used markers of the same modal umbrella categories will be commented on below (Section 8). 120 Julian Rentzsch

yig‘la-sh-ni ham bil-ma-y-san, kul-ish-ni ham! cry-VN-ACC too know-NEG-ASP-2.SG laugh-VN-ACC too ‘Behold, my son, you can neither cry about the state of our people nor laugh!’ (Uzbek, Qodiriy 1926 [1994]: 37)

(28) Tek gerek-en bekle-me-yi bil-mek. single be.necessary-PTCP wait-VN-ACC know-VN ‘The only necessary thing is to be able to wait.’ (Turkish, Şafak 2009: 226)

(29) Kumushbibi shu choq-g‘a-cha qara-ma-g‘an va NP DEM time-DAT-LIM look-NEG-ASP and qara-sh-ni ham tila-ma-gan edi. look-VN-ACC too wish-NEG-ASP PST ‘Kumushbibi had not looked at him until now and had not wanted to do so.’ (Uzbek, Qodiriy 1926 [1994]: 60)

(30) Ne-ler ol-acağ-ın-ı bil-meden ve bil-me-yi what-PL become-VN-POSS.3-ACC know-NEG.CV and know-VN-ACC iste-meden var güc-üm=le ilerle-d-im. want-NEG.CV available power-POSS.1.SG=with move.on-ASP-1.SG ‘I moved on with all the power I had, not knowing what would happen and not wanting to know.’ (Turkish, Şafak 2009: 118)

(31) Antropov gol soγ-uw-dï basta-d-ï. NP goal shoot-VN-ACC begin-ASP-3 ‘Antropov has started to score.’ (Kazakh, web-3)

(32) Aččïk ïït-alïax-pïn baγar-bap-pïn. hungry send-ITER.VN-POSS.1.ACC want-NEG.INTRA-1.SG ‘I do not want to send them away hungry.’ (Yakut, Matthew 15.32) Evolution of Turkic Modal Verbs 121

(33) Ekiriir-in küze-p=tur sen be? recover.VN-POSS.3.ACC want-INTRA 2.SG Q ‘Do you want to recover?’ (Tuvan, John 5.6)

A variant of this strategy that seems to be particular to volitive expressions displays the verbal noun not in the accusative, but in the unmarked case. This variant conforms to a Turkic syntactic rule that assigns the unmarked case to unspecific direct objects (cf. Johanson 1977). Some of the verbal nouns that are part of this construction are frequently labeled “infinitive” in the Turcology literature (especially - mAK: Brockelmann 1954: 254; Clauson 1972: XLIV), but they lack the modal semantic component of what Haspelmath calls the infinitive (cf. Haspelmath 1989: 288). -mAK, an item that is attested since the Old Turkic stage, is a pure nominalizer (sometimes also said to form “abstract” nouns: von Gabain 1941 [31974]: 73). The other verbal noun found in such constructions, -(V)r, originally denoted the intraterminal aspect (“imperfectivity”, “present tense”, “aorist” etc.; cf. von Gabain 1941 [31974]: 74, 111–112) and is roughly comparable in meaning and function to the English gerund in -ing. Example (34) from 16th-century Western Oghuz displays the verbal noun in -mAK with the auxiliary iste- (originally ‘to seek’), while example (35) from Tuvan shows the verbal noun in -(V)r with the auxiliary boda- (‘to think’, a borrowing from Mongolic). Although traceable in Turkic varieties of very remote geographic areas and of different genetic branches, this type of volitive construction is only rarely grammaticalized in the Turkic languages.

(34) Qarγa quzγun qan gör-üp oγlan-uŋ üstine crow raven blood see-CV boy-GEN on qon-maq iste-r-idi. settle-VN want-ASP-PST ‘The crows and ravens saw blood and wanted to land on the boy.’ (Middle Oghuz, Dede Qorqud 15a9; Tezcan & Boeschoten 2001: 43)

122 Julian Rentzsch

(35) Olar Ooŋ-bile čuγaala-ž-ïr bodaan čüve-dir. they s/he-GEN-with speak-COOP-VN think.PTCP PTCL-PTCL ‘They wanted to speak to him.’ (Tuvan, Matthew 12.46)

4. The dative shift

Another development that affects auxiliary constructions with verbal nouns as linking devices can be interpreted as a mutation of the types and . The derived construction deviates from the primary ones in that it displays the verbal noun not in the accusative or the unmarked case, but in the dative case. This development, which will be labeled the dative shift in this paper,17 is attested since the early Middle Turkic era. The earliest occurrences that I am aware of are from the 14th-century Kipchak Turkic text, Codex Cumanicus, where there are volitive constructions with the auxiliary tile- and the linking device -mAGA (a verbal noun in -mA and the dative in -GA, ex. 36). At least since the 17th century, the same type (with the auxiliary iste-) is also attested in Oghuz Turkic, e.g. in Istanbul Turkish (ex. 37) and Azerbaijani (ex. 38).18

(36) Tile-r-sen bil-ma-ga, söve-r-sen tengir-ni ge want-ASP-2.SG know-VN-DAT love-ASP-2.SG god-ACC or söu-mes-sen, sor-gil sen-ing congl-ni, söv-er-mi love-NEG.ASP-2.SG ask-IMP.SG you-GEN heart-ACC love-ASP-Q tengri-ni. god-ACC ‘You want to know whether you love God or whether you do not love him. So ask your heart whether it loves God.’ (Middle Kipchak, Codex Cumanicus 123, Kuun 1880: 162–163)

(37) Gair kimße-ler-den ßor-mag-a ißte-mem

17 Needless to say, this phenomenon is completely unrelated to the so-called dative shift in English. 18 In Oghuz Turkic, the segment -mAGA (which looks identical to the Kipchak one) has to be analyzed as the verbal noun in -mAK and the dative in -(y)A. Evolution of Turkic Modal Verbs 123

other somebody-PL-ABL ask-VN-DAT want-NEG.ASP.1.SG ‘I don’t want to ask anybody else.’ (Middle Ottoman, Hazai 1973: 156)

(38) Iste-mez=di youda-nun ulke-ci-na guez-mag-a want-NEG.ASP=PST Judea-GEN land-POSS.3.SG-DAT wander-VN-DAT chunki gifout-lar iste-r-ler=idi oz-i-ni uldur-mag-a. because Jew-PL want-ASP-PL=PST self-POSS.3.SG-ACC kill-VN-DAT ‘He did not want to go to the land of Judea because the Jews wanted to kill him.’ (Middle Azerbaijani, AZR 165, Jn 7.1)

The same type (partly with other verbal nouns) continues to exist in several modern languages such as Tatar (Kipchak branch, see ex. 39, this example is from the Mišär dialect19), Vidin Turkish (Oghuz branch, see ex. 40) and Uzbek (Karluk-Uyghur branch, see ex. 4120). Compare particularly ex. (41) with its twin construction with the accusative (ex. 29), which is taken from the same text. Both types are clearly free variants in this text.

(39) Bu-lar min-ĕ tanï-r-γa tile-mi-ler. dem-PL I-ACC know-VN-DAT want-NEG.ASP-PL ‘They don’t want to recognize me.’ (Mišär Tatar, Kakuk 1996: 106)

(40) Čoǰuγ-um-i al-ïp Kja:be-ye git-meg-e iste-y-im. son-POSS.1.SG-ACC take-CV Kaaba-DAT go-VN-DAT want-ASP-1.SG ‘I want to go to the Kaaba with my son21.’

19 The text was recorded by Ignác Kúnos between 1915 and 1917; see Kakuk (1996: I). 20The Uzbek example is taken from a novel from the 1920s, i.e. a time when the standardization of the was less advanced than today. The variant with the dative is not considered acceptable in Modern Standard Uzbek. 21 While čoǰuq is the normal word for ‘child’ in many Turkish varieties, it apparently means ‘boy, son’ in this dialect. It stands in clear contrast to qïz ‘daughter, girl’ in the context of the passage cited. The neutral word for ‘child’ in this dialect is maqsïm (from Arabic ma‘ṣūm ‘innocent’; see Németh 1965: 384, 395). 124 Julian Rentzsch

(Vidin Turkish, Németh 1965: 206)

(41) Bu maktub-da o‘z ot-im-ni yoz-ish-g‘a DEM letter-LOC own name-POSS.1.SG-ACC write-VN-DAT ham tila-ma-d-im. too wish-NEG-ASP-1.SG ‘I also did not want to mention my name in this letter.’ (Uzbek, Qodiriy 1926 [1994]: 117)

The dative shift is not confined to volitive constructions but occurs with the auxiliaries bil- ‘to know’, bol- ‘to become’ and bašla- ‘to begin’ as well. Expressions of possibility consisting of a verbal noun in the dative and the auxiliary bil- are attested in Middle Ottoman and Middle Azerbaijani (see ex. (42) and (43) from the 17th century), as well as in Vidin Turkish (ex. 44) and (pre-standardized) Uzbek (ex. 45). Compare particularly the Uzbek examples (45) and (27), as well as the Turkish examples (42) and (44) on the one hand and (28) on the other, where there is no perceivable difference in meaning between and .

(42) Arpa.ßu bu vilajet-ler-de jap-mag-a bil-mez-ler barley.water DEM province-PL-LOC make-VN-DAT know-NEG.ASP-PL ‘They can’t make beer here (i.e. They don’t know how to make beer here).’ (Middle Ottoman, Hazai 1973: 68)

(43) Nechun bil-menem chimdi ca-a tabe why know-NEG.ASP.1.SG now you-DAT following ol-mag-a become-VN-DAT ‘Why can I not follow you now?’ (Middle Azerbaijani, AZR 198; Jn 13.37)

(44) Oqu-maγ-a bil-iy-se, hatim read-VN-DAT know-ASP-COND complete.reading.of.the.Qur’an bašla-r, oqu-r. Evolution of Turkic Modal Verbs 125

begin-ASP read-ASP ‘If he can read, he starts reading the Qur’an, and he reads.’ (Vidin Turkish, Németh 1965: 343)

(45) Bu xabar-dan quvon-ish-g‘a=da bil-mas va DEM news-ABL be.glad-VN-DAT=too know-NEG.ASP and xafalan-ish-g‘a ham yo‘l top-mas edi. be.sad-VN-DAT too way find-NEG.ASP PST ‘He was unable to be glad about this news, but also did not find a way to be sad.’ (Uzbek, Qodiriy 1926 [1994]: 227)

Several Kipchak languages, such as Kumyk, Noghay (46), Kazakh (47) and Kirghiz, demonstrate expressions of possibility of the type . Outside the Kipchak group, the same type is also attested in Modern Uyghur (Karluk-Uyghur group, ex. (48)). The verbal nouns that surface in these constructions are -MA in Noghay, - Uw/-UU in Kazakh and Kirghiz and -(I)š in Uyghur. The non-shifted model that seems to underlie this construction entails an unmarked verbal noun (cf. ex. (26) above).

(46) Ol bu qullïq-tï et-pe-ge bol-ayaq. DEM DEM work-ACC do-VN-DAT become-FUT ‘S/he will be able to do this work.’ (Noghay, Baskakov 1956: 310)

(47) Bul oqu-w zal-ï-nda barlïq qažetti DEM read-VN hall-POSS.3.SG-LOC all necessary kitap-tar-dï oqu-w-ïŋïz-γa bol-adï. book-PL-ACC read-VN-POSS.2.PL-DAT become-ASP.3.SG ‘In this reading room, you can read all the necessary books.’ (Kazakh, Sulejmenova et al. 1997: 58)

(48) Šeher-de bir qača duγ-ni beš mo-din bir yüen-gi-če city-LOC one bowl Lassi-ACC five Máo-ABL one Yuán-DAT-LIM ič-i-miz, čöl dep ikki koy-γa sat-sa-ŋlar 126 Julian Rentzsch

drink-ASP-1-PL desert say-CV two Kuài-DAT sell-COND-2.PL čüšin-iš-ke bol-atti, üč yüen bolsa understand-VN-DAT become-ASP.PST.3.SG three Yuán top bek qimmet=ken, insap soda qil-iŋlar. very expensive=EVID fairness trade do-IMP.2.PL ‘In the city, we drink a bowl of Lassi for five Máo to one Yuán; if you sell it for two Kuài because we are in the desert one could understand it, but three Yuán is very expensive, please do fair trade.’ (Modern Uyghur, Šal 2006: 54)

In the Oghuz group, the auxiliary bašla- ‘to begin’ also combines with verbal nouns in the dative. This construction is very pervasive in this branch and has even changed the government pattern of the corresponding full verb, which takes nominal complements in the dative instead of the accusative. The following example is from Vidin Turkish, but the same type is found in Gagauz, Standard Turkish, Azerbaijani and Turkmen as well. Bašla- can also combine with a verbal noun in the dative in Modern Uyghur (alternative to the combination with the converb -GAlI, see ex. 15), but in this language bašla- maintains its original government features in full verb functions.

(49) Qïs bekle-r gjȧl-sin, yoq, bašla-r aγla-maγ-a. girl wait-ASP come-VOL.3 unavailable begin-ASP cry-VN-DAT ‘The girl waits for him to come, but he does not, she starts crying.’ (Vidin Turkish, Németh 1965: 133)

Like the subjunctive strategy already mentioned, the dative shift represents a fundamental deviation from grammatical behavior normally expected in the Turkic languages. Converbs are attested as linking devices in auxiliary constructions since the oldest layers of Turkic, and verbal nouns in the accusative and the unmarked case do not violate the grammatical principles of arbitrary combinations of full verbs and nominal arguments. In contrast, a dative shift violates existing government rules in the Turkic languages. Possible reasons for this phenomenon as well as the synchronic distribution and some diachronic issues will be discussed below (Section 9). Evolution of Turkic Modal Verbs 127

5. The “infinitive”

As a result of a phonetic reduction of verbal nouns in the dative, a few Turkic languages display genuine “infinitives”, i.e. synchronically unanalyzable non-finite modal forms, one of the main tasks of which is to form complements to matrix predicates (cf. Haspelmath 1989: 288–289).22 Chuvash has a so-called infinitive in -mA (Pavlov 1965: 260–264), which Clark describes as “a purposive adverbial form, indicating the goal or result of the action expressed in a second verb” (Clark 1998: 447). The same morpheme is found in Karaim and Kumyk. This morpheme, albeit identical in form to a deverbal noun - mA that is attested in Old Turkic (von Gabain 1941 [31974]: 72, 78) and has become a fully productive verbal noun in some languages (cf. ex. 28 and 30 from Turkish and ex. 36 from the Codex Cumanicus), is strikingly different in meaning and function. The verbal noun in -mA is clearly a nominal element and can take possessive suffixes and case markers, while the infinitive in -mA is an uninflectable adverbial element. Etymologically, the infinitive in -mA is a verbal noun in the dative (*-mAGA, which represents the verbal noun -mAK with the dative -A in Oghuz, and the verbal noun -mA with the dative -GA in some non-Oghuz varieties). This origin is particularly transparent in those languages where -mA and -mAGA constitute free variants (e.g. Koman, Kumyk). With its adverbial characteristics, the infinitive can be classified as a converb (more on that issue below). Example (50) from Chuvash displays the infinitive in an expression of possibility with the auxiliary verb pultar- ‘to bring about’ (originally the causative of *bol- ‘to become’). Besides as a linking device in auxiliary constructions, the Chuvash infinitive can also function as a predicate in purposive adverbial clauses (i.e. as a purposive converb). Example (51) demonstrates that this morpheme occurs as early as in the Codex Cumanicus (14th c.), where it alternates freely with -mAGA (cf. ex. 36). Intermediate variants between -mAGA and -mA are also attested in several Turkic varieties, such as -mAA in Gagauz.

22 Obviously, Haspelmath’s conception of the infinitive covers only a subset of items that are labeled thus in the literature (albeit a very prominent one); cf. items like Turkic -mAK, which does not demonstrate any modal semantic component and is still labeled infinitive in the literature. 128 Julian Rentzsch

(50) Tură śak čul-senčen te Avraam ači-sem God DEM strone-PL.ABL too NP child.POSS.3-PL tu-ma pultar-at’. do-INF can-INTRA.3.SG ‘God is able to create children unto Abraham even from these stones.’ (Chuvash, Matthew 3.9)

(51) Iazuk-le kizi kim tile-r kensi iazuch-in ayt-ma sin-DNN person COMP wish-ASP own sin-POSS.3.ACC say-INF ‘A sinful person who wishes to confess his sins’ (Middle Kipchak, Codex Cumanicus 125, Kuun 1880: 167)

6. Excursion: Variation of linking devices with nominal auxiliaries

Developments that parallel those identified with verbal auxiliaries occur with nominal auxiliaries as well. This option will not be explored exhaustively here; only a few examples that are in concord with the observations made so far will be cited. Subject-predicate structures with a nominal predicate serve as the prototype for these constructions: the main verb appears with a verbal noun suffix that is unmarked for case and functions as the subject of the sentence. The auxiliary is a nominal item (noun or adjective) and functions as the sentence predicate (X-ing is a necessity, X-ing is possible, etc.). This type of construction is normal in Old Turkic (see ex. 25 above) and found virtually everywhere in the Turkic world. However, deviations from this origin are attested as early as in Old Uyghur, where the nominal auxiliary kergek ‘necessary, necessity’ occasionally combines with the converb -GAlI, which is suffixed to the main verb (52), thus rendering a construction that closely resembles the most frequent verbal auxiliary construction type in Old Uyghur (see Section 2).

(52) Bu nom erdini-g bošγun-γalï tut-γalï oqï-γalï oqï-t-γalï DEM teaching jewel-ACC learn-CV hold-CV read-CV read-CAUS-CV biti-geli biti-t-geli ögret-ig qïl-γalï kergek. write-CV write-CAUS-CV teach-DVN do-CV necessary ‘It is necessary to learn this Dharmaratna, to keep it in mind, to read it Evolution of Turkic Modal Verbs 129

and have others read it, to write it and have others write it and to teach it.’ (Old Uyghur, Altun Yaroq P1.00.12.v15–17)

*Kergek is also attested with subjunctives such as the voluntative in the Middle Azerbaijani example (53), and with a verbal noun in the dative as in the Kumyk (Kipchak Turkic, Caucasus) example (54):

(53) Ciz-un nour-unz guerek beila ichieuk you.PL-GEN light-POSS.2.PL necessary so light ver-cun kalk yug-un-da. give-VOL.3 people front-POSS.3-LOC ‘Your light shall shine before the people.’ (Middle Azerbaijani, AZR 11; Mt 5, 16)

(54) Al-ïn-aγan tüšüm-nü tiyišli baga-sïn-a sat-maq take-PASS-PTCP harvest-ACC necessary price-POSS.3-DAT sell-VN učun imkanlïq-lar boldur-ma-γa gerek-biz. for opportunity-PL create-VN-DAT necessary-1.PL ‘We have to create opportunities to sell the harvest being gathered at the necessary price.’ (Kumyk, WEB-4)

Nominal auxiliaries other than *kergek may also display deviations from the original subject-predicate pattern in individual Turkic varieties. In the Modern Uyghur example (55), the adjective mumkin ‘possible’ (borrowed from Arabic through Persian) combines with a verbal noun in the dative. The option to do so is extremely striking from the common Turkic perspective. The original subject-predicate structure (i.e. yėyiš mumkin) is equally possible in Uyghur.23

23 Example (55) is complicated by the fact that the auxiliary construction forms part of a subordinate clause (an object clause). This, however, is irrelevant. In Modern Uyghur, virtually all constellations are possible. Both and are attested in finite and subordinate clauses alike, with as an additional variant in subordinate nominal clauses (e.g. yė-yiš-niŋ mumkin bol-ma-ydiγanliq-i). I am unable to say whether all these attested options are equally acceptable to all speakers. 130 Julian Rentzsch

(55) Yolučilar γudira-š-ti=yu, bu yer-din bašqa traveler-PL grumble-COOP-ASP=and DEM place-ABL other yer-de tamaq yė-yiš-ke mumkin bol-ma-ydiγanliq-i-ni place-LOC food eat-VN-DAT possible be-NEG-VN-POSS.3-ACC hės qil-ip leγmen-ni yė-yiš-ke bašli-di. feel-CV lā.miàn-ACC eat-VN-DAT begin-ASP.3 ‘The travelers grumbled, but realizing that it was not possible to eat at another place, they started to eat the Leghmen.’ (Modern Uyghur, Šal 2006: 55)

Thus, practically all types of auxiliary constructions that occur with auxiliary verbs are attested with auxiliary nouns as well, albeit much less frequently. Note that the starting point for nominal and verbal auxiliary constructions (which survives in many varieties) is fundamentally different, the former typifying subject-predicate structures, the latter verb-complement structures.

7. Survey of abstract structural types

From the data presented above, the following abstract types of constructions can be identified:

AUX-1 AUX-2 AUX-3 AUX-4 AUX-5 AUX-6 AUX-7

Note that except in AUX-2, where the auxiliary always precedes the complement clause, the order of the constituents separated by the symbol <+> may vary according to language-specific rules or even be free in some varieties. The order chosen for the schematic representation reflects the most typical Turkic word order properties, in which auxiliaries follow the main verb. Evolution of Turkic Modal Verbs 131

AUX-4 is heterogeneous in that it comprises complements (verbal nouns) that formally represent either subjects (AUX-4A) or objects (AUX-4B). The structures of AUX-2 and AUX-3 closely resemble one another; the latter very much looks like a derivation of the former (but see below). AUX-6 clearly derives from AUX-4 and AUX-5, and AUX-7 from AUX-6. AUX-5 is the only type that is intrinsically incompatible with nominal auxiliaries, as Turkic nouns do not usually govern the accusative. Table 2 shows the seven auxiliary verbs central to this paper and the structural types in which they occur. The findings are based on the data available to me. Some more combinations might be found in further sources but they seem not to be documented in the literature. Especially AUX-7, which is just a phonetically reduced variant of AUX-6, shows potential for the retrieval of more combinations from additional text corpora.

Table 2. Auxiliary verbs and structural types.

AUX-1 AUX-2 AUX-3 AUX-4 AUX-5 AUX-6 AUX-7 bašla- x x x x bil- x x x x bol- x x x (x)24 iste- x x x x x x küse- x x tile- x x x x x u- x

The survey shows that the most prolific types are AUX-1, AUX-5 and AUX-6. (Note, however, that the table does not address frequency and dissemination; it merely shows which combinations are attested.) As shown above, aux-1 represents the usual strategy of linking auxiliary verbs to main verbs in Old Turkic and AUX-5 reflects the normal government rules of the underlying full verbs that are involved in the auxiliary constructions, so it is not astonishing that these types are very common. More striking is the versatility of AUX-6, as it violates the Turkic government rules. So there must be good reasons for this structure to have come into being at all (on this issue, see Section 9).

24 With the causative suffix in Chuvash. 132 Julian Rentzsch

8. Diachronic issues and distribution

Up to now, prominent types of auxiliary constructions have been identified on an empirical basis, resulting in a rudimentary taxonomy of the options that are available in the Turkic languages. In order to refine the picture, some diachronic, diatopic and frequential information needs to be added. It has already been mentioned that AUX-1 is the dominant type for linking main verbs to auxiliary verbs in Old Turkic, so it is not astonishing that converbs constitute an important tool in the auxiliarization of former full verbs in many Turkic varieties. When it comes to the converbs actually used for auxiliary constructions, however, significant changes have occurred. The converb -GAlI, which was of eminent importance in Old Uyghur, is completely marginal in modern Turkic auxiliary constructions.25 The only modern Turkic language that utilizes this converb in auxiliary constructions is Modern Uyghur, where it is used in constructions with (A) bašla- ‘to begin’, (B) bol- ‘to become’, (C) qoy- ‘to put’, (D) tur- ‘to stand’ and (E) yat- ‘to lie’, which signify inception (A and D), participant- external possibility (B), permission (‘to let’, C) and imminence (‘to be about to X’, E) (see Friederich (2002: 208–209)). All these combinations except (C) are documented in Old Uyghur as well (see Section 2 and von Gabain 1941 [31974]: 127, 132), with a potentially different meaning for (D). (A) and (B) can alternatively be constructed with a verbal noun in the dative (aux-6). A possible explanation for the striking similarity between Old Uyghur and Modern Uyghur regarding the linking segment in auxiliary constructions could be that Modern Uyghur continues a linguistic tradition that was peculiar to the dialect(s) of the Old Uyghur tribes (as opposed to other tribes like the Türk, Kirghiz, Oghuz etc., who might have used the converb -(y)V instead; support for this possibility is provided by the data in the Runic inscriptions). On the other hand, we cannot rule out the possibility that -GAlI was common to several Old Turkic tribes and replaced by other linking devices in all languages except in the extreme southeast of the Turkic-

25 It does survive, however, as a predicator of adverbial clauses in many Turkic languages. In some languages, especially Oghuz and South Siberian Turkic, it has undergone a semantic development from purposive (‘in order to’) to abtemporal (‘since’), which makes it less suitable for auxiliary constructions. Evolution of Turkic Modal Verbs 133 speaking area. Both the mutual relation of Old and Modern Uyghur and the dialect situation in Old Turkic remain unresolved problems.26 What can be stated, though, is that in Karakhanid Turkic (Karluk, 11th c.), the usability of -GAlI in auxiliary constructions is radically reduced in comparison to Old Uyghur and does not occur with any of the auxiliaries mentioned so far (cf. Mansuroğlu 1959: 105).27 In Chaghatay (Eastern Middle Turkic), it is virtually confined to construction (C), which is amply attested e.g. in the Baburnama (16th c.). Whether this is the result of the loss and replacement of -GAlI in these environments or of the continuance of a different dialectal tradition is unclear. Except Modern Uyghur, modern Turkic languages employ other converbs in AUX-1. This may be the intraterminal converb -(y)V, which appears in Runic inscriptions (with the auxiliary u-) and is also amply attested in Old Uyghur auxiliary constructions (including modal ones with u- and bil- as well as many non-modal (actional etc.) constructions, such as those with bašla- and many others (cf. von Gabain 1941 [31974]: 129–132)). It is also frequent in Middle Turkic (Ottoman, Kipchak, and Chaghatay), partly in updated form (-A in Kipchak and Chaghatay), and appears both in old and younger constructions. As an independent marker, the converb -(y)V (and its younger variants) has lost much of its productivity in almost all modern Turkic languages. It hardly forms predicates of adverbial clauses any more, and in some languages it is completely confined to grammaticalized or lexicalized items.28 This trend can be increasingly observed since the early Middle Turkic era. This situation has led to the fact that particularly younger auxiliary constructions tend to resort

26 Uyghur scholars usually take a direct lineage from Old Uyghur to Modern Uyghur for granted, while Western Turcologists tend to view such a simple relation with scepticism (cf. Johanson 2003). It is at least possible that in spite of many foreign (e.g. Karluk) influences, Modern Uyghur preserves some features of an Old Uyghur stratum. For remarks on Old Turkic dialects and the problems involved in identifying them, see Erdal (2004: 11–18). 27 It does occur with the auxiliaries qal- ‘to remain’ and er- ‘to be’, the former denoting ‘almost to X’, the latter maybe ‘to intend’ (Erdal 2004: 250). 28 A comparatively widespread productive residue entails the double use of -(y)V with either two different verb stems or the same verb stem, which, however, is a conventionalized construction in its own right and not identical to the productive use of single -(y)V. 134 Julian Rentzsch to the converb in -(V)p instead.29 -(V)p was originally a postterminal (“perfective”) converb. It lost its postterminal value during the Middle Turkic era (Johanson 1990: 139–142) and developed into a highly versatile converb. The differing availability of the various converbs in stages when grammaticalized constructions were still morphologically transparent and especially during the formation of new auxiliary constructions has led to a considerable variation between individual Turkic languages with respect to the converb used, e.g. Azerbaijani -A bil- vs. Turkmen -(I)p bil- (possibility), Modern Uyghur -GAlI bol- vs. Altai -(I)p bol- (possibility), Modern Uyghur -GAlI bašla- vs. Uzbek - A bašla- vs. Noghay -(I)p bašla- (inception). Another converb that is available for auxiliary formations since the late Old Turkic stage is the conditional in -sA (< -sAr), which combines with bol- as well as with the nominal auxiliary kerek in Karakhanid (Erdal 2004: 259, 527). The conditional poses a classificational challange. It was probably originally a finite marker (i.e. a mood marker; cf. von Gabain 1941 [31974]: 187; Brockelmann 1954: 240) and maintains some marginal finite usages in all documented stages of Turkic. On the other hand, it is most commonly used to build dependent adverbial clauses, in which case it can often be exchanged with other, clearly non-finite adverbial markers (e.g. Turkish -dIğI takdirde, Uzbek -ganda etc.). Yet it also occurs in constructions where it assumes subjunctive-like functions and is interchangeable with optative or voluntative items (see above, footnote 9). For our present purposes, it is convenient to classify -sA bol- as aux-1 (analogous to -GAlI bol- and -(I)p bol-) and kėrek kim X- sA as aux-2 (analogous to kėrek kim X-sUn), keeping in mind that the conditional manifests hybrid properties that would allow us to classify -sA bol- as aux-3 as well.30 New types of auxiliary formation appear in early Middle Turkic times (early 14th c.) in volitive expressions. In Khorezmian Turkic,

29 A few auxiliary constructions (though no modal ones) were formed with -(V)p even in Old Turkic (see von Gabain (1941 [31974]: 131–132)). 30 However, classifying -sA bol- as AUX-3 would be in conflict with the assumed evolutionary link between AUX-2 and AUX-3, in which the former constitutes the antecedent of the latter. There is no empirical evidence to suggest that -sA bol- was preceded by a construction like *bol- kim X-sA. Hence, the classificational affiliation of auxiliary constructions that contain the conditional is ultimately historically motivated. Evolution of Turkic Modal Verbs 135

AUX-2 (ex. 56) alternates with AUX-4 (ex. 57), while AUX-6 (ex. 36) and AUX-7 (ex. 51) are attested in Koman.

(56) Rasūl ‘A.S. tile-d-i kim bu kursī-ni messenger peace.upon.him wish-ASP-3.SG COMP DEM seat-ACC kör-se, āyatu’l-kursī-niŋ savāb-ïn bil-se. see-COND verse.of.the.throne-GEN merit-POSS.3.ACC know-COND ‘The Prophet, peace be upon him, wished to see the throne and thus to know the merit of the Verse of the Throne.’ (Qiṣaṣ al-Anbiyā’ 218v8–9; Boeschoten & Vandamme & Tezcan 1995: 497)

(57) Yüz-üŋ-ni bir kör-mek tile-yür-men. face-POSS.2.SG-ACC one see-VN wish-ASP-1.SG ‘I want to see your face just once.’ (Qiṣaṣ al-Anbiyā’ 108v19; Boeschoten & Vandamme & Tezcan 1995: 220)

The development with the verbal auxiliary tile- is paralleled by a similar development with the nominal auxiliary kėrek ‘necessary’, which is equally attested in AUX-2 constructions in Khorezm Turkic and in AUX-6 constructions in Koman. In Chaghatay, the borrowed Arabic adjective mumkin ‘possible’ also appears in AUX-2 constructions. Remember that the nominal auxiliary kergek was already typically being constructed according to the AUX-4 type in Old Turkic. This strategy survived in the Middle Turkic era in Chaghatay and Ottoman (but not Koman) and continues to exist in many modern Turkic varieties as well. While AUX-2 is not attested in Old Turkic modal constructions, complex sentences involving a matrix predicate and a complementizer are quite frequently attested in Old Uyghur. The subordinate clauses are e.g. relative, purposive, causal and temporal clauses. The predicates of the subordinate clauses may have various inflectional forms such as aspect markers (indicatives) or the conditional in -sAr. Voluntatives are not yet common in these constructions in Old Uyghur. The complementizers that appear in these structures are originally question words such as kim ‘who’, qayu ‘which’, qačan ‘when’ and ne ‘what’ (von Gabain 1941 [31974]: 189–192). Of these, 136 Julian Rentzsch kim assumes eminent importance in Middle Turkic, developing to a highly frequent complementizer in Islamicized Turkic languages, often in free variance with the similar-looking Persian complementizer ki. The Old Uyghur structures with complementizers pave the way for Middle Turkic modal constructions of the AUX-2 type. In early Chaghatay (late 15th c.), the widespread AUX-2 type was already being supplemented by AUX-3, i.e. subjunctive constructions without a complementizer (58). This type is clearly younger and was initially less frequent. It became very popular in some Turkic varieties in Iran and Rumelia (Stein 2010: 243–245; Johanson in print). In the latter region, AUX-3 has expanded its scope to some non-volitive constructions as well (see ex. 23 and 24). AUX-2 on the other hand, although extremely common in Middle Turkic written documents, survives mainly in some spoken Turkic varieties that have had intensive contact with Persian, e.g. in Iran and .

(58) Iste-r-em alin qil-sa-m seǰde ve want-ASP-1.SG front do-COND-1.SG prostration and öp-se-m lab-iŋ. kiss-COND-1.SG lip-POSS.2.SG.ACC ‘I want to prostrate before you and to kiss your lips.’ (Chaghatay, Navā’ī, cited after Brockelmann 1954: 413)

As regards AUX-4, the first clear cases with a verbal auxiliary are attested in the Khorezmian Turkic construction -mAK tile- (see ex. 57 from a document finished in 1310). In this construction, the verbal noun fills the object slot of the auxiliary. The expression of possibility -mAK ol- (where the verbal noun fills the subject slot of the auxiliary) is attested in Middle Oghuz, e.g. in the Dede Qorqud Oγuznameleri from the 16th c. (ex. 26), but there might be earlier occurrences as well. Even in Old Uyghur there are examples of a verbal noun with a possessive suffix and bol- that might instantiate an initial stage of this construction. The interpretation in terms of possibility is not certain, however. In some contexts the same construction can only be interpreted as ‘X-ing happens’. In the example provided here (59), the Evolution of Turkic Modal Verbs 137 interpretation of bulmaqi bolur in terms of external possibility (‘it is possible to find’) seems suitable, but it is far from secure.

(59) Ančulayu oq arïγ süzük yėg bilge bilig-ig so.much EMP pure clear best wise knowledge-ACC bul-maq-ï bol-ur. find-VN-POSS.3 become-ASP ?‘To that degree one can find the pure and clear supreme wisdom.’ ?‘The discovery of the pure and clear supreme wisdom happens to that degree.’ (Old Uyghur, Altun Yaroq P1.02.22.v26)

As was stated before, AUX-4 was the dominant type of construction with the nominal auxiliary kergek in Old Turkic (in which the verbal noun formally functions as the sentence subject), and it is not inconceivable that the existence of this structure with a very frequent nominal auxiliary contributed to its spread with verbal auxiliaries. AUX-5 presupposes an auxiliary verb that can take a direct object in its non-auxiliary usages, which excludes bol- (and of course all nominal auxiliaries) from the set of potential candidates. It was shown that from among the auxiliaries on which this paper focuses, AUX-5 is in fact attested with all but bol- and the very early obsolete u-. On the other hand, given that combinations of verbal nouns in the accusative are amply attested in Old Turkic as the direct object of full verbs and given that AUX-5 neatly fits the grammatical principles of Turkic syntax, this construction is strikingly marginal in the Turkic languages. Not only is it hardly attested in pre-modern Turkic, but it is even quite infrequent in most modern languages. It is only widespread in different-subject wishes (Si wants Sj to X), where Si is indicated by a personal marker at the wish-verb and Sj by a possessive marker at the verbal noun. This is a highly common strategy in most major standard languages, not only with the verbs iste-, tile- and küse-, but also with other wish-verbs such as xwāhla- (originally a denominal verb from the Persian noun xwāh ‘wish’). Different-subject wishes can also be expressed by AUX-2 and AUX-3, which are particularly popular in Rumelian Turkish (AUX-3) and languages in intensive contact with 138 Julian Rentzsch

Persian (AUX-2), besides the use of some other strategies.31 Same- subject wishes (Si wants Si to X = Si wants to X) are much less frequently expressed by AUX-5; examples of languages that often resort to this strategy are Yakut, where the wish-verb baγar- combines with a verbal noun in the accusative (ex. 32) or, alternatively, with a converb (AUX-1), and Tuvan (ex. 33, auxiliary küze- < küse-).32 AUX- 5 in same-subject wishes occurs in languages like Uzbek (ex. 29), Modern Uyghur, Kazakh and Turkish (ex. 30) as well, but in these languages same-subject wishes are more commonly expressed by other strategies (AUX-4 in Turkish, suffixes in the other languages mentioned). In these languages, AUX-5 expresses a wish that is semantically marked or emphasized (‘even to have the wish to X’), often but not exclusively in negated form. AUX-5 is also rarely attested with the auxiliaries bil- ‘to know’ (ex. 27 and 28) and bašla- ‘to begin’ (ex. 31), but from a general Turkic perspective these are rare exceptions. enjoys a systematic status in Turkish and Uzbek in that it specifically denotes participant-internal possibility, supplementing the more general possibility markers -(y)Abil- (< -(y)V bil-, Turkish) and -A ol- (< -(y)V al-, Uzbek), which both typify the pattern aux-1. The variant with a verbal noun in the accusative is a recently grammaticalized renewal of possibility that resorts to a fully productive construction pattern. Instances of AUX-5 with bašla- are scarce compared with AUX-1 and AUX-6 and may be considered only weakly conventionalized though not entirely arbitrary. Generally speaking, while the underlying structure is very old, grammaticalized AUX-5 constructions seem to be quite recent formations that are not very widespread and in most cases manifest a rather weak degree of grammaticalization. AUX-6 is attested with the majority of those auxiliary verbs that have been chosen as the focus of this paper. The earliest attestations (Codex Cumanicus, 14th c.) of this pattern are with the auxiliary verb tile- ‘to wish’ as well as the nominal auxiliary kerek ‘necessary’. In the modern Turkic languages, AUX-6 is most frequently attested with

31 E.g. a mood form, which is followed by a quotative marker (typically a converb of *te- ‘to say) and a wish-verb. The quotative marker functions as a complementizer. There is a huge range of subtypes of this strategy. 32 These volitive constructions of Yakut and Tuvan display some additional characteristics that are mentioned in Section 3. Evolution of Turkic Modal Verbs 139 bašla- ‘to begin’ and bol- ‘to become’. Bašla- combines with a verbal noun in the dative in Western Oghuz33 (Standard Turkish, Standard Azerbaijani, dialects; some Western Oghuz varieties demonstrate aux- 3 instead or as a free variant) and Modern Uyghur (with -GAlI bašla- (aux-1) as an (older) alternative form).34 Bol- in the AUX-6 pattern is frequent in Kazakh, Kirghiz and several Western Kipchak languages as well as – probably as a result of contact with Kipchak varieties – Modern Uyghur (where it co-occurs with older -GAlI bol-). In addition to bašla- and bol-, AUX-6 appears erratically in various historical stages and regions: in Koman (Middle Kipchak) with tile-, in Middle Azerbaijani with tile-, iste- and bil-, and in Middle Ottoman with iste- and bil-. In Vidin Turkish, iste- and bil- occur in AUX-6 as well (alternatively to AUX-3; iste- occurs in AUX-4, too). Except with bašla-, AUX-6 has disappeared from the Turkish and Azerbaijani standard languages. The examples of tile- and bil- in aux- 6 that have been cited from an Uzbek novel from the 1920s (ex. 41 and 45) represent stray alternative variants of AUX-5. In Standard Uzbek, AUX-6 is entirely absent. In Khakas, the verbal noun -(I)r in the dative case (rendering -(I)rGA) appears in two volitive modal constructions (with the auxiliaries xïn- ‘to wish, to like’ and saγïn- ‘to think’, the latter corresponding to synonymous Old Turkic saqïn-, which is attested as an auxiliary with the converb -GAlI in Old Uyghur) as well as with the nominal auxiliary kirek ‘necessity’ (< kergek), thus fitting the pattern AUX-6. However, in Khakas (as well as the other Yenisei Turkic language Shor), -(I)rGA demonstrates a significantly broader versatility than verbal nouns in the dative in the other Turkic languages and is much more commonly used in non- grammaticalized environments to serve “infinitive” functions. The development that gives rise to this situation can certainly be seen in context with the grammaticalization of AUX-6 in other languages. To sum up, AUX-6 represents a pattern that is actually scarce in the Turkic language family as a whole but tends to appear very stubbornly across time and space. Auxiliary constructions involving a verbal noun in the dative usually fall into disuse shortly after their appearance. They constitute a tendency that surfaces again and again and are obviously still felt to violate the (written or unwritten) rules of

33 Eastern Oghuz (Turkmen) features -(I)p bašla- (AUX-1). 34 Closely related Uzbek displays -A bašla- (AUX-1). 140 Julian Rentzsch the grammar. Only in individual cases do AUX-6 constructions come to persevere, with clusters of several specimens in the extreme southwest (West Oghuz) and southeast (Modern Uyghur) of the Turkic area as well as, with serving in a broader context, Yenisei Turkic. Individual Turkic languages differ greatly in terms of which abstract structural types of auxiliary formation they permit and how these types correlate with the individual auxiliary verbs. Given the large number of living modern Turkic varieties and the innumerable documents in pre-modern Turkic varieties, which behave very differently regarding the phenomena studied here, only findings based on a small selection of varieties can be presented in this framework. From the Middle Turkic era, the variant of Chaghatay that is represented in the Baburnama (16th c., A) and Middle Azerbaijani as represented in text AZR (17th c., B) are considered. For Modern Turkic, the situation in Standard Turkish (C), Vidin Turkish (D), Uyghur (E), Uzbek (F), Kazakh (G) and Khakas (H) is presented. To recall auxiliary verbs and structural types in Old Uyghur, cf. Table 1. Note in particular the similarities and differences between Standard Turkish and Vidin Turkish, which are closely related genetically, and likewise between Chaghatay, Uzbek and Uyghur. Table 3 and table 4 reflect the same linguistic facts from different perspectives. Both tables show which combinations of auxiliary verbs and structural types are attested in the eight languages selected. Table 3 is arranged according to the seven structural types, while table 4 is organized by the individual languages. Remember that the tables display the situation in eight languages only, not in the complete language family. The auxiliary verb küse- is not attested in any of these eight Turkic varieties, nor is the AUX-7 structure.

Table 3. Distribution of auxiliary construction types.

AUX-1 AUX-2 AUX-3 AUX-4 AUX-5 AUX-6 AUX-7 bašla- AEFGH D FG BCDE bil- BCDGH D CF BDE bol- AEFGH CD G iste- AB BD CD CF BD küse- tile- A CEFG B Evolution of Turkic Modal Verbs 141 u- CD

Table 4. Survey of auxiliary construction types in individual Turkic languages.

A B C D E F G H bašla- 1 6 6 36 16 15 15 1 bil- 16 15 136 5 1 1 bol- 1 4 16 1 16 1 iste- 1 136 45 36 5 küse- tile- 2 6 5 5 5 5 u- 1 1

9. Motivating the linking devices

Several factors have contributed to the emergence of the seven auxiliary construction types that have been identified in this article. Conventionalization requires structures to build on, regardless of whether these structures are autochthonous or borrowed. In AUX-1, the oldest attested type, the origin in a more general (and still productive) clause linking strategy that connects a subordinate predicate to a superordinate one is still transparent. AUX-4 and AUX-5 integrate the main verb into nominal phrases which then comply with the syntactic rules of nominal elements. In AUX-4, the verbal noun either functions as the subject in a subject-predicate construction, which at the same time conforms to the predominant Old Turkic pattern for nominal auxiliaries, or it functions as the unspecific direct object of the auxiliary verb. Both options comply with the productive rules of Turkic syntax. AUX-5 adopts the productive strategy to form object clauses, with the main verb in nominal form in the object slot. AUX-2 is modeled on a structure in Persian, which is an important contact language for those Turkic varieties that feature this type. But even this type draws on components that were already available in Old Turkic, where (possibly under the influence of other contact languages) subordinate clauses could be introduced by question words, including kim ‘who’, which displays a similar shape as the Persian complementizer ki. Finite mood forms from the Old Turkic morphological inventory are employed in the subjunctive function in this type. AUX-3 is basically a remodeling of AUX-2 by elimination of 142 Julian Rentzsch the complementizer. In some Turkic varieties, word order has been restructured according to Turkic rules by moving the auxiliary segment to the end (obligatory in Standard Turkish, optional in Vidin Turkish). In some varieties, AUX-3 has spread from volitive constructions to other auxiliary constructions (Vidin Turkish). Contact with Balkan languages has given rise to the prominence of AUX-3 in the westernmost varieties of Turkish (Johanson in print). While AUX-6 and AUX-7 seem to lack a formal model within the Turkic languages at first sight, note that the dative of the verbal noun in -GU (i.e. - GUKA) is attested in Old Uyghur as forming purposive subordinate clauses (Erdal 2004: 490–491), that is adverbial clauses with purposive semantics. In this function,35 -GUKA is very similar to the converb -GAlI, the main formal difference being that -GUKA is synchronically analyzable while -GAlI is not. It is quite obvious that - GUKA paves the way for converbial usages of other verbal nouns with datives (-mAGA, -UwGA, 36 -(I)šKA and -(I)rGA), which then are available for grammaticalization processes just like the converb -GAlI in AUX-1. As Haspelmath has shown (1989), purposive action nouns are typical sources of infinitives. This finding provides us with a semantic stimulus for AUX-6, for in some of its usages is a purposive verbal noun. The purposive meaning of arises from the goal-oriented (directive) semantics of the dative case combined with a non-factual verbal noun (factual verbal nouns, which exist as well in Turkic, cannot be grammaticalized into purposive items). The semantic parallel between the segment in AUX-6 and the converb -GAlI in AUX-1 is striking: in (Old and Modern) Uyghur AUX-1 constructions, -GAlI is a suitable converb precisely because of the universal tendency to employ purposive items as “infinitives” in irrealis complement clauses, which in turn constitute the infrastructure for the auxiliarization processes leading to the pattern AUX-1 and AUX-6. Essentially, AUX-6 is merely a subtype of AUX-1.

35-GUKA serves another function as well, namely forming complements of verbs or postpositions that govern the dative. In this case, -GUKA is a free combination of a verbal noun with a case suffix (cf. Erdal 2004: 456–457; 360). 36 Note that the Kazakh form -UwGA (as well as its other Kipchak Turkic variants) is etymologically identical to Old Turkic -GUKA. Evolution of Turkic Modal Verbs 143

The semantic analogies in effect in Turkic auxiliary construction types go even further. The mood items that perform subjunctive functions in AUX-3 are the conditional, voluntatives and the optative. Voluntatives and optatives express an inclination of a conscious subject towards the action, i.e. the performance of the action (by whoever is indicated by the personal marking, which can be the first or third person with voluntatives and any person with the optative) is presented as desired, approved etc. 37 They (metaphorically) “aim” at the action and share a common hyperonymical notion with purposive items, which is goal-orientation. Using vol and opt as linking devices in auxiliary constructions can be regarded as an adaptation to AUX-1 on the basis of metonymy: vol and opt are used “in place of” a purposive item, drawing on the common hyperonym of goal-orientation. Hence, AUX-1, AUX-3 and AUX-6 (as well as AUX-7 as a more grammaticalized variant of AUX-6) can be said to build on semantically analogous principles. The seven types of auxiliary constructions with their different distribution across the Turkic languages not only typify individual constructions that evolved in diachrony; they also represent formal patterns that can spread within a broad class of loosely associated semantic notions that center on root modality, without being historically motivated. Most Turkic varieties use more that one linking strategy in modal and related constructions with the result that a given pattern can potentially be imposed on structures that are perceived as semantically related. This mechanism can be described as a formal analogy. E.g., AUX-3 was initially confined to volitive constructions, a situation that is still mirrored in Turkic varieties in Iran, but spread to expressions of possibility and inception in Vidin Turkish. While the precise historical evolution that led to the picture in Vidin Turkish is not entirely clear, it is striking that the correlation of AUX-3 and AUX- 6 to the auxiliary verbs is exactly identical, suggesting that AUX-3 and AUX-6 did not develop independently in this variety but were “harmonized” at a certain stage. In Modern Uyghur, AUX-6 was first introduced by semantic analogy to AUX-1 with the auxiliary verbs bol- and bašla- and spread to the auxiliary adjective mumkin by formal analogy to these.

37 The conditional seems to originally have carried a hypothetical (“irrealis”) meaning (‘suppose that’), which makes it suitable for subjunctive functions as well. 144 Julian Rentzsch

10. Summary

The grammaticalization of Turkic modal constructions evolves according to seven structural types that can be realized by different morphological material in individual Turkic varieties. The majority of structures are already predisposed in Old Turkic syntactic patterns, which are applied to auxiliary constructions. Specific language contact situations have contributed to the emergence and/or preferred use of some patterns (AUX-2, AUX-3, potentially AUX-6 and AUX-7). Some of the patterns are evolutionally interdependent. AUX-3 is derived from AUX-2, while AUX-7 is just a phonetically reduced variant of AUX-6. AUX-6 at least partly presupposes AUX-5 (as in the Uzbek examples of tile- and bil- with versus ) or AUX- 4 (as in Vidin Turkish iste- with versus ). Three of the types, AUX-1, AUX-3 and AUX-6, are semantically analogous and build on purposive linking devices. As non-factual verbal nouns in the dative are re-analyzable as purposive converbs (as long as they do not fill an argument slot of a verb or postposition), AUX-6 with its complex genesis is in effect a variant of AUX-1. In AUX-6, the wheel of grammaticalization has come full circle to AUX-1. “Infinitives” as they occur in AUX-7 are derived from a verbal noun in the dative. Formal and semantic analogy operates in concert with the effect that a given structural type can spread to other auxiliaries. A cluster of areal, genetic and historical factors results in a highly diversified picture in individual Turkic varieties as regards their auxiliary constructions.

Acknowledgement

The research for this paper was made possible by a grant from the Alexander von Humboldt Foundation.

Abbreviations

This article uses the standard abbreviations listed in the appendix to The Leipzig Glossing Rules with additions and modifications mentioned below. Evolution of Turkic Modal Verbs 145

Ø Zero ASP Aspect COOP Cooperative CV Converb DIR Directive DNN Denominal noun DVN Deverbal noun EMP Emphatic EPIST Epistemic EQU Equative EVID Evidential ITER Iterative LIM Limitative NP Proper noun OPT Optative PTCL Particle S Subject VB Verb VN Verbal noun VOL Voluntative X (represents an arbitrary verb stem)

Sources

AZR: Middle Azerbaijani transcription text, 17th c., cf. Johanson 1985. Baburnama: Chaghatay Turkic text, 16th c., ed. Beveridge 1905. Chuvash Gospel: Moskva: Institut Perevoda Biblii 2009, http://www.ibtrussia.org Qodiriy, Abdulla 1926/1929 [1994]. O‘tkan Kunlar/Mehrobdan Chayon. Toshkent: G‘afur G‘ulom nomidagi adabiyot va san’at nashriyoti. Şafak, Elif 2009. Aşk. İstanbul: Doğan Kitap. Šal, Qeyyum Abduqadir 2006. Šopur, yoluči, öčret. Qumul edebiyati 2006/1, 51–55. 146 Julian Rentzsch

Tuvan Gospel: Moskva: Institut Perevoda Biblii 2001, http://www.ibtrussia.org VATEC: Vorislamische Alttürkische Texte: Elektronisches Corpus, http://vatec2.fkidg1.uni-frankfurt.de/

WEB-1: http://forum.uyghuramerican.org/forum/showthread.php?6766- Men-Eng-Axerida-Esirge-Chushken-Idim (retrieved on August 12, 2010)

WEB -2: http://www.egemen.kz/9792.html (retrieved on September 24, 2010)

WEB-3: http://www.alaman.kz/?p=13357 (retrieved on March 3, 2011)

WEB-4: http://yoldash.etnosmi.ru/one_stat.php?id=7931 (retrieved on July 6, 2010) Yakut Gospel: Moskva: Institut Perevoda Biblii 2008, http://www.ibtrussia.org

References

Baskakov, Nikolaj Aleksandrovič 1956. Russko-nogajskij slovar’. Moskva: Gosudarstvennoe Izdatel’stvo Inostrannyx i Nacional’nyx Slovarej. Berta, Árpád 2004. Savamat jól halljátok. A türk és ujgur rovásírásos emlékek kritikai kiadása. Szeged: JATE Press. Beveridge, Annette S. (ed.) 1905. The Bábar-náma. Fac-simile. Leyden & London. Boeschoten, Hendrik E. & Vandamme, Marc & Tezcan, Semih (eds.) 1995. Al-Rabghūzī, The Stories of the Prophets. Qiṣaṣ al- Anbiyā’. An Eastern Turkish version. 2 vols. Leiden & New York & Köln: E.J.Brill. Brockelmann, C. 1954. Osttürkische Grammatik der islamischen Litteratursprachen Mittelasiens. Leiden: Brill. Evolution of Turkic Modal Verbs 147

Clark, Larry 1998. Chuvash. In: Johanson, Lars & Csató, Éva Á. (eds.). The Turkic languages. London & New York: Routledge, 434–452. Clauson, Gerard 1972. An etymological of pre-thirteenth- century Turkish. Oxford: Clarendon Press. Erdal, Marcel 2004. A grammar of Old Turkic. (Handbuch der Orientalistik 8/3.) Leiden & Boston: Brill. Friederich, Michael 2002. Uyghurisch Lehrbuch. In Zusammenarbeit mit Abdurishid Yakup. Wiesbaden: Dr. Ludwig Reichert Verlag. Haspelmath, Martin 1989. From purposive to infinitive – a universal path of grammaticalization. Folia Linguistica Historica 10/1–2, 287–310. Haspelmath, Martin 1999. Why is grammaticalization irreversible? Linguistics 37, 1043–1068. Hazai, György 1973. Das Osmanisch-Türkische im XVII. Jahrhundert. Untersuchungen an den Transkriptionstexten von Jakab Nagy de Harsány. Budapest: Akadémiai Kiadó. Himmelmann, Nikolaus P. 2004. Lexicalization and grammaticization: Opposite or orthogonal? In: Bisang, Walter & Himmelmann, Nikolaus P. & Wiemer, Björn (eds.). What makes grammaticalization? A look from its fringes and components. (Trends in Linguistics. Studies and Monographs 158.) Berlin & New York: Mouton de Gruyter, 21–42. Johanson, Lars 1971. Aspekt im Türkischen. Vorstudien zu einer Beschreibung des türkeitürkischen Aspektsystems. (Studia Turcica Upsaliensia 1.) Uppsala. ——— 1977. Bestimmtheit und Mitteilungsperspektive im türkischen Satz. Zeitschrift der Deutschen Morgenländischen Gesellschaft, Suppl. III: 2, 1186–1203. ——— 1985. Isfahan – Moskva – Uppsala. Kring några medelaserbeidjanska handskrifter och stationerna på deras väg. Svenska Forskningsinstitutet i Istanbul, Meddelanden 10/1985, 26–44. 148 Julian Rentzsch

——— 1990. Zur Postterminalität türkischer syndetischer Gerundien. Ural-Altaische Jahrbücher N.F. 9, 137–151. ——— 1995. Mehrdeutigkeit in der türkischen Verbalkomposition. In: Erdal, Marcel & Tezcan, Semih (eds.). Beläk bitig. Sprachstudien für Gerhard Doerfer zum 75. Geburtstag. (Turcologica 23.) Wiesbaden, 81–101. ——— 2000. Viewpoint operators in European languages. In: Östen Dahl (ed.). Tense and aspect in the languages of Europe. (EUROTYP 20/6.) Berlin & New York: Mouton de Gruyter, 27–187. ——— 2003. Old Uyghur, Eastern Turki, Modern Uyghur. Orientalia Suecana 51–52, 257–266. ——— (in print). Mood meets mood: Turkic versus Indo-European. In: Stolz Thomas & Vanhove, Martine & Otsuka, Hitomi & Urdze, Aina (eds.). Morphologies in contact. (Bremen Festival of Languages). Johanson, Lars & Csató, Éva Á. (eds.) 1998. The Turkic languages. London & New York: Routledge. Kakuk, Zsuzsa 1996. Mischärtatarische Texte mit Wörterverzeichnis. Aufgrund der Sammlung von Ignác Kunos herausgegeben. (Studia Uralo-Altaica 38.) Szeged. Kıral, Filiz 2001. Das gesprochene Aserbaidschanisch von Iran. Eine Studie zu den syntaktischen Einflüssen des Persischen. (Turcologica 43.) Wiesbaden: Harrassowitz. Kuun, Géza 1880. Codex Cumanicus bibliothecae ad templum divi Marci Venetiarum. Budapest: Editio Scientiarum Academiae Hungaricae. Mansuroğlu, Mecdut 1959. Das Karakhanidische. In: Deny, Jean & Grønbech, Kaare & Scheel, Helmuth & Togan, Zeki Velidi (eds.). Philologiae Turcicae Fundamenta. Vol. 1. Wiesbaden: Steiner, 87–112. Németh, Julius 1965. Die Türken von Vidin. Sprache, Folklore, Religion. Budapest: Akadémiai Kiadó. Palmer, F.R. 22001. Mood and modality. Cambridge. Evolution of Turkic Modal Verbs 149

Pavlov, Ivan Pavlovič 1965. Xal’xi čăvaš literatură čĕlxi. Morfologi. Šupaškar: Čăvaš ASSR Kĕneke Izdatel’stvi. Rentzsch, Julian 2010. Zur Modalität im Türkischen. In: Hendrik Boeschoten & Rentzsch, Julian (eds.). Turcology in Mainz/Turkologie in Mainz. (Turcologica 82.) Wiesbaden: Harrassowitz, 209–224. Schlüter, Julia 2010. To dare or not to: Is auxiliarization reversible? In: Iinden, An Van & Verstraete, Jean-Christophe & Davidse, Kristin (eds.). Formal evidence in grammaticalization research. Amsterdam & Philadelphia: John Benjamins, 289–326. Stein, Heidi 2010. Optativ versus Voluntativ-Imperativ in irantürkischen Texten (15./16. Jh.). In: Boeschoten, Hendrik & Rentzsch, Julian (eds.). Turcology in Mainz/Turkologie in Mainz. (Turcologica 82.) Wiesbaden: Harrassowitz, 230–255. Sulejmenova, Eleonora & Kadaševa, Karlygaš & Akanova, Dana 1997. Suxbat. Qazaq tili/Kazaxskij jazyk. Učebnyj kompleks. 2- oe izdanie. : Izdatel’skij Dom “Žibek Žolï”. Tezcan, Semih & Hendrik Boeschoten 2001. Dede Korkut Oğuznameleri. İstanbul: Yapı Kredi Yayınları. van der Auwera, Johan & Plungian, Vladimir A. 1998. Modality’s semantic map. Linguistic Typology 2, 79–124. von Gabain, Annemarie 31974 [1941]. Alttürkische Grammatik. Wiesbaden: Harrassowitz.