Auxiliary verb constructions and clitic placement: Evidence from Tsakonian

The Tsakonian clitic system possesses a clitic auxiliary with the same syntactic and prosodic properties as the object clitic pronouns with which it may cluster preverbally or postverbally. The clitics of the two Tsakonian subdialects (Peloponnesian Tsakonian and Propontis Tsakonian) differ typologically since the latter has second position clitics but the former does not. It is shown here that Peloponnesian Tsakonian clitics do not simply constitute a mixed system in a state of transition between the inherited Medieval Greek enclitics and SMG proclitics, because of certain peculiarities they show. In particular, circumclitics and split clitics have arisen, and second position clitics are retained not as free variations but as elements whose placement depends on strict prosodic and/or syntactic conditions.


liosis
Journal of Greek Linguistics 17 (2017) 37-72 being deficient at one or more levels of linguistic analysis, and especially at the level of prosody (cf. Anderson's 2011 definition of clitics as "prosodically deficient elements").1 (For an introduction to clitics, see, for example, Zwicky 1977, Caink 2006, Spencer & Luís 2012; for a review of the field and key approaches, see Klavans 1995, Riemsdijk 1999, Gerlach & Grijzenhout 2000 Discussion of the subject has usually been confined to clitic pronouns (ClPrs), in particular to object pronominal clitics, and discussion of clitics in Greek has been no exception. This is one of the best studied, but most controversial issues of Greek linguistics: are the Greek ClPrs syntactic or morphological elements? Do they belong to a sui generis grammatical category that combines word-like and affix-like characteristics?2 (For a comprehensive review of the bibliography on the nature of pronominal clitics and their structural position in Greek, see Ralli 2006.) Answering these questions usually requires recourse to phonological observations and dialect data, and has consequences for the typologies that have been proposed for these elements, as well as for our understanding of the diachronic development of clitic pronouns in the history of the Greek language.

Typologies of the Greek clitics and their historical development
The two best-known typologies based on the structural position of ClPrs are those of Condoravdi & Kiparsky (2001(ck)) and of Revithiadou & Spyropoulos (2006, 2008(rs)). Both propose a tripartite distinction: a) In the first group, which includes the eastern Greek dialects in general and Late Medieval Greek (MedG), clitics are always enclitic. According to ck's terminology, they constitute maximal projections adjoined to a phrasal projection without requiring a prosodic host of a particular syntactic category, while according to the terminology of rs, they are second position clitics (2P), in the sense that they always appear in second place within the prosodic constituent headed 1 Given that the subject of this paper is an examination of object clitic pronouns and clitic forms of the copular auxiliary, the term 'clitics' (Cls) refers to both these categories, unless otherwise stated. 2 Of course, these questions apply crosslinguistically, not only to Greek. The difficulties inherent in the attempt to define and identify exactly what constitutes a clitic have recently led some linguists to describe the word as an umbrella term that takes in elements which cannot be easily categorized either as affixes or ordinary function words (Zwicky (1994:xiii); Spencer & Luís (2012:149)), with all the attendant theoretical implications. Haspelmath (2015) proposes replacing the traditional tripartite taxonomy free word~clitics~affixes with a polar opposition plenimorphs~minimorphs, and the further distinction between "weakly coalescent minimorphs" (= clitics) and "highly coalescent minimorphs" (= affixes). liosis Journal of Greek Linguistics 17 (2017) 37-72 peculiarities are due to its different origin: it is the only modern variety of Greek that has its origins not in the Hellenistic Koine but rather in the ancient Laconian dialect (see Tzitzilis 2014;Tzitzilis forthc.), and shows innovations that came about dialect-internally or through contact.3 Therefore, Tsakonian may be expected to possess characteristics which could shed new light on the issues surrounding the Greek clitics. This paper is based on this hypothesis.

1.2
The case of Tsakonian Tsakonian is spoken today in a small mountainous area on the south-eastern edge of the region of Arcadia in the Peloponnese. Besides Peloponnesian Tsakonian (PelTs), which is divided into a northern and a southern branch, there was another, quite distinct subdialect, Propontis Tsakonian (PrTs), which was spoken, according to Costakis (1951;1979), by Tsakonian migrants and their descendants from the 18th to the 20th century near Gönen in north-western Asia Minor, Turkey (see below, Figure 1). After the Greco-Turkish war of 1919-1922 and the exchange of populations between Greece and Turkey, the speakers of PrTs were scattered throughout various areas of Greece and later shifted to smg.
In what follows I will focus on the syntactic, prosodic, and statistical analysis of pronominal clitics and the clitic auxiliary in auxiliary verb constructions (avcs) in the two subdialects.4 Journal of Greek Linguistics 17 (2017) 37- 72   table 1 The paradigm of pronominal clitics in the two subdialects according to Pernot 1934:186-188, Costakis 1951:175-176, and Costakis 1986-1987 1 2 3 Clitic pronouns in Tsakonian The Tsakonian system of object clitic pronouns is, as shown in Table 1, caseless: it marks only person and number distinctions, the distinction between direct and indirect object being made based on the context of the discourse.
The following examples show that PrTs has typical 2P clitics, as in the neighbouring dialects of north-western Asia Minor and the eastern Greek dialects in general: more problems when we are dealing with languages that have clitics whose position is independent of the morphological category of their hosts, for example many Slavic languages and varieties of Greek, than in cases where clitics generally attach to a host of a particular category, e.g. [+v], as, for example, in the Romance languages (Gerlach & Grijzenhout 2000). c) In Modern and Medieval Greek, in contrast to Ancient Greek, there is generally no V-to-C movement (Drachman & Klidi 1992;ck 2001. d) Two purely syntactic analyses of postverbal clitics have been proposed for the Cypriot dialect in particular (Terzi 1999;Agouraki 2001). According to Terzi, in the absence of any other licenser, the verb moves to the highest position (= MoodP) in order to license clitics. The problem with this analysis is, according to ck (2001:7-8), that it lumps together as syntactic licensers some extremely disparate elements (i.e. functional heads, such as negation, modal particles, and complementizers, as well as nonfunctional heads, such as preverbal focus phrases, and wh-phrases). e) Confining ourselves to purely syntactic factors provides even less satisfactory results in the case of the circumcliticization pattern of Tsakonian (see section 4.1.3). f) Taking account of prosodic factors allows us to propose an economical and unified description of the syntactic status and structure of the 2P and non-2P clitics of Tsakonian. (2) PrTs 2P clitics a. ma fut t do2sg parakalés ask.1sg.sbjv 'I will ask you' (Costakis 1957:96) ta the ríza root s its 'I cut its root in half ' (Costakis 1957:104) c. θorákai saw.3pl ɲ do3sg ton the aγó hare 'They saw the hare' (Costakis 1957:118) The clitic appears preverbally when clause-initial position is occupied by a "function word"5 (2a) or a focused constituent (2b), and postverbally when clause-initial position is occupied by the verb (2c). Generally, in dialects with 2P clitics, the original position of the clitics is preverbal, according to ck: they adjoin to tp, the first functional projection over the vp, whose head the verb moves to in order to license its tense and subject-agreement features. This is shown in (3): Postverbal position is thus the special case, in the sense that, as ck (2001:7) put it, "if there is no available prosodic host to their left, they encliticize onto the adjacent word on their right by prosodic inversion (Halpern 1995)".
As is generally proposed for second position clitic systems, the clitic prosodically subcategorizes for a word on its left within the same cp. For example, in (2a) the clitic [t(i)] encliticizes onto the function word (future particle) [ma], and in (2b)  barrier between the clitic pronoun which precedes and the verb that follows; otherwise, the deletion of [i] would be unmotivated.
In contrast, in Peloponnesian Tsakonian, we seem to have typical non-2P clitics, as in the neighbouring Peloponnesian dialects, in the western dialects more generally, and also in smg: (4) PelTs non-2P clitics a. θa fut nam io1pl alí tell.3sg.sbjv 'S/he will tell us' (Liosis 2007) b. ɲ do3sg enéndze brought.3sg 'S/he brought it' (Liosis 2007) c. δíte give.2pl.imp mí io1sg ɲi do3sg 'Give it to me!' (Liosis 2007) Here, clitics are always phonologically dependent on the verb, as can be seen from the elision of the final vowel preceding the initial vowel of the verb form in (4a) ([námu] > [nam]) and (4b) ([ɲi > ɲ]), and from the rhythmic foot [mí ɲi] in (4c), which results from postlexical stress that developed in order to avoid violating the three-syllable rule. This rule, which applies throughout the history of Greek and in many of its dialects, forbids the stress from falling further left than the third syllable from the end, and its domain is exclusively the prosodic word; cf. smg [arˈniθika] (= denied.1sg) but [arniˈθikame] (= denied.1pl), not *[arˈniθikame]. If the two clitics in (4c) did not belong to the prosodic word of the verb, the creation of the foot and the development of postlexical stress would be unjustified.
Clitics in PelTs, as in smg, are generally proclitic on the verb (4a, 4b), but enclitic on an imperative (4c). According to ck (2001:16), the original syntactic position of non-2P clitics is preverbal, i.e. the same as that of 2P clitics, with the difference that "they do not attach syntactically to tp [but] they are word-level affixes […] that is to say, they attach to words, forming larger words". However, their postverbal position after an imperative is not the result of prosodic inversion, as with postverbal 2P clitics, but rather that of verb movement; according to Philippaki-Warburton (1998), the clause structure of smg can be described by the following sequence: As mentioned, tp is the functional projection where clitics are adjoined. Since the imperative is the only affixal mood in Modern Greek, the overt movement of the verb to the mood functional head is required. This movement operation leaves the clitics behind (in the tp), thus causing the reverse order. It is clear that the same clause structure is found in Tsakonian: In conclusion, the two subdialects differ typologically as regards the position and the prosodic properties of the clitics, a fact which presumably has some connection with contacts with neighboring dialects in the geographical areas where each subdialect is (or was) spoken ( Figure 1). Given the diachronic development of 2P clitics into non-2P clitics, it is PrTs that preserves the older stage. Therefore, it is likely that PrTs preserved this stage due to the support it received from the neighboring dialects, which also possessed 2P clitics, and/or that PelTs adopted the new syntactic and phonological pattern due to the influence of the clitics in its neighbouring dialects, which had already developed into non-2P.
Journal of Greek Linguistics 17 (2017) 37-72 figure 1 The area where PelTs is spoken, and the area where PrTs was spoken until early 20th century

Clitic auxiliaries in Tsakonian
A basic characteristic of the Tsakonian verbal system is the periphrastic formation of the present, the imperfect, and, in PrTs, an aorist that is derived from older perfect constructions. These tenses are formed with the auxiliary 'to be' and the participle of the main / lexical verb.6 As shown in the following table (Table 2), in both the subdialects the auxiliary remains fully finite within an table 2 The Tsakonian clitic auxiliary within an avc according to Pernot 1934:216, Costakis 1951:179-180, and Costakis 1986-1987  avc, even in PrTs, where, despite the extensive analogical influences between the two numbers and the two tenses, the auxiliary remains marked at least for person. According to Heine (1993:70), "Auxiliary is an item covering some range of uses along the Verb-to-TAM chain", or according to Anderson (2006:4), "an item on the lexical verb-functional affix continuum". Schematically, the historical path of grammaticalization of an auxiliary, which includes the semantic, syntactic and phonological changes it undergoes, is described in (7) (Anderson, 2006:334): (7) lexical verb > auxiliary > affix This schema coincides broadly with the well-known four-part 'cline of grammaticality' of Hopper & Traugott (2003:7) shown in (8): (8) content item > grammatical word > clitic > inflectional affix A combination of (7) and (8) gives us the general implicational (also probabilistic, historical etc.) hierarchy in (9), where an auxiliary is a grammatical word and a cliticized auxiliary is considered to represent an intermediate stage between an auxiliary and an affix (for the term 'cliticization of an auxiliary' within the framework of grammaticalization theory, see Heine (1993:55-56)): Journal of Greek Linguistics 17 (2017) 37-72 (9) lexical verb > auxiliary > clitic auxiliary > affix Given that sequences (7-9) represent a continuum between a completely lexical/syntactic element and a completely grammatical/functional element, it is obvious that the divisions we make are essentially methodological in nature, and that the intermediate grey zones could theoretically be divided ad infinitum, cf., for example, Bolinger's (1980:297) "quasi auxiliary" as an intermediate stage between a lexical verb and an auxiliary, or Zwicky's (1977) distinction between 'simple' and 'special' clitics. Special clitics, which constitute an intermediate stage between simple clitics and inflection (cf. Nübling 1992), are etymologically opaque, and their syntactic position may differ from that of simple clitics. An important distinction for the purposes of this paper is that between an auxiliary and a clitic auxiliary (ClAux), as it will be shown first that the Tsakonian copula plays the role of an auxiliary in the construction of verbal periphrases, and then that it has developed into a clitic element.

3.1
The Tsakonian copula as an auxiliary verb The Tsakonian copula displays all the properties of a typical auxiliary within an avc, as shown in (10) The Tsakonian avcs are characterized by semantic non-compositionality of the integral parts (= the total meaning is not the sum of the individual meanings of the participle and the copula) and syntactic unity, i.e. rigid order (although this differs between the subdialects) and linear adjacency of the auxiliary and its coreferential non-finite / participial complement. (For the criteria that have been proposed for the identification of an element as an auxiliary more generally, see Steele 1978, Ramat 1987, Heine (1993, among others; for the criteria that help to decide whether a structure that typically involves a verb and a type of complement constitutes a periphrasis, see Bertinetto 1990 and Vincent 1991, among others.) More precisely, based on all the available material for Tsakonian (see section 4.1.1 below), there are few or no constituents of the clause that can intervene liosis Journal of Greek Linguistics 17 (2017) 37-72 between the auxiliary and the participle: if such an element is present, it is usually a discourse marker / discourse connective (11a), and certainly not a subject or a (non-pronominal) object (11b, 11c): In fact, all the examples where the auxiliary is separated from the participle come from relatively old material, and are confined exclusively to PelTs, such as (11a) from Deffner, which implies that these represent an earlier stage in the formation of the avc, where the syntactic links between its members may have been somewhat weaker. For contemporary speakers of the dialect, utterances such as (11a) are ungrammatical.
The linear proximity and syntactic unity of the components of the present and imperfect avcs contrast sharply with the weaker syntactic bonds between the elements that make up a perfect avc.7 As seen in (12), in this dialect the perfect is formed with the periphrastic present of the auxiliary 'to have' and a verbal adjective in [ˈ-tos], preceding which any constituent of the clause may intervene, in (12a) the object clitic pronoun and the subject, in (12b) the object:  Costakis 1986Costakis -1987 The difference between the auxiliary of PrTs and that of PelTs is not only syntactic (this will be discussed below), but also typological. According to the crosslinguistic typology of Anderson (2006:21-27), taking as variable the head of inflection (= morphosyntactic locus of inflection), we may distinguish 5 types of avcs: 1) Aux-headed, 2) Lex-headed, 3) Split, 4) Doubled, 5) Split / Doubled. Clearly then, PrTs follows the split pattern, i.e. the auxiliary is marked for person only, while the participle is marked for all the other categories. In other words, the auxiliary in PrTs has been grammaticalized almost to the point where it operates solely as a person marker (with the partial exception of the 3rd person clitic, which shows optional tense marking, but is more often omitted entirely). PelTs follows the split/doubled pattern, i.e. person and tense are marked only on the auxiliary, aspect and gender only on the participle, while number is marked on both.

3.2
The Tsakonian auxiliary as a clitic The first argument that the Tsakonian auxiliary has the status of a clitic is historical in nature. As demonstrated convincingly by Tzitzilis 2014, periphrases with 'to be' were already an established characteristic of the Lakonian of Hesychius, dating back to the 5th century ad; that is to say, being inherited in Tsakonian, these periphrases have a long history, much longer than that of the future periphrases with 'want' + subjunctive and the perfect periphrases with 'have' + infinitive, which emerged in Late MedG, probably between the 10th and 15th centuries (see, for example, Markopoulos 2009). Therefore, there was much more time available for the auxiliary 'to be' to move towards more grammatical forms, although of course this is all relative.
Second, the forms of the auxiliary which appear in avcs have generally lost their lexical stress, and are phonologically reduced compared to the forms of the copula. This is particularly apparent in PrTs. The full form of the copula has a stressed initial /i/, as in (13), but this is obligatorily deleted in all persons of the auxiliary, e.g. /íma/ > /ma/, /ísa/ > /sa/ etc. (Cf.  (PrTs, Costakis 1957:114) In PelTs, the full form and the stress of the copula may be preserved in slow and careful speech, but we more commonly find unstressed monosyllabic forms without the initial and / or final vowel, as in (14), where the remaining [ŋ] < /n/ is the result of assimilation with the following velar consonant.
'She stays there' (PelTs, Charalambopoulos 1980:146) Third, in typological terms, the order lexical verb~auxiliary, as found in PrTs and, as we shall see, in a few cases in PelTs too, is predominant in ov languages, while the order auxiliary~lexical verb predominates in vo languages. This is a crosslinguistic observation that goes back to Greenberg (1963:67), and has been widely adopted in the literature; see, for example, Steele (1978:42), Heine (1993:24). Therefore, the auxiliary has the same linear relationship with the lexical verb as the lexical verb does with the object, and the auxiliary is the phrasal head in avcs (cf. Anderson 2006:23). The problem is that Tsakonian is not an ov language, but rather resembles smg as regards word order. This can be seen from the following example, where ena xorjáta is the snp and to γáiδare s is the onp: Thus we would expect the auxiliary to precede, rather than follow, the lexical verb. The fact that it appears in a syntactic position which is unexpected for its grammatical category constitutes evidence that the auxiliary in Tsakonian is already a deficient, cliticized element. This leads to an even stronger argument in favor of the clitic status of the Tsakonian auxiliary. It is based on the simple idea that if clitics are clitics, they should have similar properties, regardless of the grammatical category to which they originally belonged. In other words, the forms of the auxiliary are clitics, specifically subject agreement clitics, because they obey syntactic and phonological restrictions analogous to those which govern the behavior of Object ClPrs.
Journal of Greek Linguistics 17 (2017) 37-72 Indeed, in the following examples, it can be seen that the ClAuxs not only conform to the clitic pattern in each subdialect (2P clitics in PrTs, non-2P clitics in PelTs), but also form clusters with ClPrs involving a fixed relative order:  (Costakis 1986(Costakis -1987 (Costakis 1986(Costakis -1987 In PrTs, a ClAux is enclitic to a function word (fw) / focused constituent (FoC) in initial position, as in (16a), or to a verb occupying this position, as in (16b), and always precedes the ClPr. In PelTs the ClAux is proclitic to a non-imperative verb regardless of whether the initial position is occupied (17a) or not (17b), and always follows the ClPr. In other words, the linear relationship between the auxiliary and the clitic pronoun is the same whether they are in preverbal or postverbal position.
In order to understand the difference in the relative positioning of ClAux and ClPr in the two subdialects, we must take into account the phenomenon of clitic climbing. Uncontroversially, the auxiliary is the phrasal head and its complement (in this case the participle) is the semantic head of the avc. However, the ClPr, which was a semantic argument of the participial complement (Cmp), has become a syntactic argument of the auxiliary, and must always appear adjacent to it. According to Joseph 1990a, this is the (optional) moment in the history of a construction with an auxiliary when the structure ceases to be biclausal and becomes monoclausal, a process described as 'clause union' . The following schema represents the diachronic coevolution of monoclausal avcs and clitics in Tsakonian: As has previously been mentioned, the original syntactic order of the auxiliary and its complement was Aux-Cmp, because Tsakonian is (neutrally) a vo language. As can be seen from the first line of the diagram, in the presence of a fw/FoC, the 2P clitic appeared immediately following the fw/FoC and immediately preceding Aux (i). In the absence of a fw/FoC, the 2P clitic appeared immediately following Aux by means of prosodic inversion (ii).8 With the ongoing grammaticalization of each of the two structures, the semantic center of the periphrasis was transferred to the complement / lexical verb, while the auxiliary became a clitic and clustered with the clitic pronoun (iiivii). This meant that it could no longer by itself host the clitic pronoun, and was thus forced to encliticize (with the attached ClP) onto the adjacent word to its left, namely the fw/FoC (iii, v), or, in the absence of a fw/FoC, that to its right, namely the V, through prosodic inversion (iv, vi). This is the point at which the two subdialects diverged. In PelTs, as long as 2P clitics were still present, the structures (iii) and (iv) predominated, while in PrTs it was the structures (v) and (vi). In other words, in each dialect the two members of the clitic cluster preserved the linear order relative to one another that had obtained prior to prosodic inversion. When PelTs adopted the standard model of preverbal clitics, the order Cl-Aux-V was established even when there was no preceding fw/FoC (vii).9 8 Alternatively, as one reviewer proposes, given that the first line of Diagram 18 describes a proto-Tsakonian stage closer to ancient Greek, the order Aux-Cl-V could have resulted from the movement of Aux to C in the absence of any XP (= FW/FoC) in first position. However, this interpretation cannot explain the next stages. For example, if the order XP-Aux-Cl-V in PrTs is directly derived from the original order XP-Cl-Aux-Cmp, Aux cannot have moved higher than Cl, because the initial position is occupied by XP. 9 It is worth noting that Bulgarian has participial structures for the non-confirmative aorist using the clitic auxiliary 'to be' . The word order in these structures in standard Bulgarian and its southern dialects is exactly comparable to that in PrTs, that is, according to Franks & King (2000:61), the auxiliary precedes the ClPr and the two elements can appear together in either preverbal or postverbal position, e.g. Bulg. dal si gi~PrTs δokó sa s (= gave.m.sg you.were Journal of Greek Linguistics 17 (2017) 37-72 In PrTs the 3rd person ClPrs ( ɲ(i) and s(i)) tend to merge with the preceding Aux in an idiosyncratic way, i.e. through special phonological processes not attested elsewhere in the morphophonology of the dialect and unexpected from the point of view of its affixal morphology. This results in the emergence of 'opaque forms' (for the term, see Bonet 1995:607) such as /tiɲ/ in example (19) 'Because he made him a lord' (Costakis 1957:108) The same phenomenon is observed when the 3rd person ClPrs are preceded by a modal particle. In (20a) the ClPr ɲi has merged with the 1st person future marker ma, and in (20b) with the subjunctive marker na: (20) PrTs opaque 3rd person clitics (= particle + ClPr) a. miɲ fut.do3sg fáme eat.3pl.sbjv

(< /na ɲi/)
'Let's look for it' (Costakis 1957:106) In (21) the phonological rule is given that describes the formation of opaque clitics in this subdialect, which is reminiscent of that found in some Romance languages (cf. In a still more grammaticalized form, with a higher degree of phonological reduction or coalescence, the final palatal consonant /ɲ/ of the forms /miɲ/, /tiɲ/ etc. may be deleted, as in (22)

(/na ɲi/ > /niɲ/)
'that I cut it' (Costakis 1957:112) The phonological behavior of ClPrs in PrTs, especially when preceded by a ClAux or a function word, supports the ultimate formulation of ck 2004 that clitics in MedG, and generally in all the Greek varieties with 2P clitics, are enclitic.

PelTs clitic auxiliaries and pronouns in a "transitional" stage
Let us now return to PelTs. As already stated, this subdialect seems to follow the smg model as regards clitic placement: a clitic (auxiliary or pronoun) must always precede the non-imperative verb. However, the three following examples give us food for thought: 10 /ma/ is the 1st person future marker (cf. example (20a)) or the homonymous auxiliary of the same person, /sa/ and /ta/ are the 2nd and 3rd person auxiliaries respectively, /θa/ is the future particle of the remaining persons, and /na/ is the subjunctive marker.
Journal of Greek Linguistics 17 (2017)  tsh what épaθíka suffered.1sg 'I will tell you what I suffered' (Charalambopoulos 1980:151) What these three examples have in common is that (some of) the clitics, paradoxically, appear in postverbal position. That is, (a) is a polar question with the verb in initial position, and the ClPr ɲ(i) clusters with the ClAux eme,11 while (b) constitutes the only case of split cliticization of such a cluster: the ClAux is on the left and the ClPr on the right of the verb. In (c), the element nd is a reflex of the postverbal ClPr ɲúmu, and resembles a dummy pronoun: it occupies the preverbal slot, as expected following the functional element θa, but does not have a referent; as will be shown below, the two items form a single unit, a circumclitic on the verb. (Cf. Joseph (1990b:178) who, following Pernot (1934:189), considers that in examples such as (23c) we have "anomalous nonagreeing 'doubling' of the plural weak pronoun [which] reveals a semantic idiosyncrasy (in the form of a null interpretation) and a syntactic idiosyncrasy (in the form of an irregular sort of agreement pattern)".) The prosodic properties of postverbal clitics in PelTs do not differ from the properties of clitics that follow imperative verbs (see example (4c) and associated discussion).12 Thus they are incorporated into the prosodic word of the verb, as can be seen from the development of rhythmic stress in the following example to prevent violations of the three-syllable rule: 11 One reviewer draws a parallel between the syntactic pattern of (23a) and the Portuguese future tense, where the enclitic auxiliary has been reanalyzed as "inflectional material, so that the pattern involves mesoclisis of the pronoun". Based on the mobility of the Tsakonian auxiliary, which can appear in preverbal or postverbal position, we can exclude the interpretation that it constitutes an inflectional ending, and therefore also that which would describe ClPr as a mesoclitic. 12 The only exceptions are the disyllabic stressed clitics námu and ɲúmu, which are discussed in section 4.  (Pernot 1934:187) The easy interpretation of the "irregular" cases is that they are residual or transitional structures that arose during the development of the clitic system of PelTs from 2P to non-2P. This idea has already been proposed for Tsakonian by Pernot (1934:189), and similar cases have been recorded in the bibliography for other dialects: for example, Janse (1997) notes sporadic cases of split cliticization of indirect and direct object ClPrs in Cappadocian, as in example (25a), which he explains as the result of the presence of the 2P clitics pattern combined with a tendency for generalization of the postverbal syntax of clitics as in Pontic. Also well known is the negative dʒó in the dialect of Farasa (example (25b)), "which preserved the ancient syntax of its cognate uk" (ck 2004:180, fn. 4), just as in example (23b) with the (also cognate) ó-of PelTs. "Ancient syntax" here means that uk seems not to draw the clitic into preverbal position like all the other function words. rs 2006 also note the existence in the dialect of Mariupol of a few examples corresponding to the syntax of (23c), such as (25c), "where both copies of the clitic may surface".13 (25) 13 Note that du, the more archaic of the two cognates to and du of the 3rd person clitic in example (25c), occupies postverbal position. The same phenomenon of simultaneous preverbal and postverbal syntax of clitics in Marioupolitika, but this time with repetition of the 1st person pl form mas, is also described by Kissilier et al. (2009:400-401).
Journal of Greek Linguistics 17 (2017) 37-72 I assume that explaining the properties of the clitic system of PelTs as simply the result of its transitional nature is probably an oversimplification. A basic characteristic of transitional systems is fluidity and instability between old and new structures, which are usually interpreted as free variations. The statistical analysis below shows that we do indeed find such variations, e.g. examples of clitics which are sometimes procliticized and sometimes encliticized onto verbs in initial position; however, these are determined by underlying factors at work, which lead to new regularities.

4.1
Statistical analysis of PelTs Cls 4.1.1 Raw data, variables, and values The material available for PelTs covers a period of approximately a century and a half, from the mid-19th century until today. I have made a corpus of the most representative texts, and divided it into three chronological phases: 1) 1870-1925 (Oikonomou 1870(Oikonomou , 1888Scutt 1913Scutt -1914Deffner 1921Deffner , 1923aDeffner , 1923bDeffner , 1926, 2) 1925(Stratigis c. 1927[Houpis 2000]; Costakis 1951Costakis , 1983Costakis , 1986Costakis -1987Charalambopoulos 1980), 3) 1980-today (Liosis 2007). This division serves methodological purposes, but is also based on linguistic and sociolinguistic criteria. For example, in the first period, the negative particle ό-is completely absent from the non-periphrastic tenses (the particle δen is used instead), and there are no circumverbal clitics. The third period, according to Liosis (2007), is characterized by intense contact with smg, and by the emergence of semi-speakers. For the purpose of this investigation, the material from the third period consists exclusively of audio recordings of fluent speakers of the dialect.
I counted the total number of instances of clitic use in each period, and defined three variables: 1) the position of the clitic, with three values: a) postverbal, b) circumverbal, c) preverbal; 2) the position of the verb, with two values: a) initial, b) non-initial; 3) the type of sentence, with two values: a) statement, b) question.

4.1.2
Clitic placement and verbal position: yes/no questions As ck (2001,2004) have shown, the only real syntactic factor that determines the placement of Greek clitics in 2P systems is the position of the verb: if the verb is initial, the clitic follows; otherwise, it precedes. I consider that a preposed subject or any other topicalized / left-dislocated constituent does not constitute part of the clause, and therefore in these cases the verb remains in initial position in its core clause. The results of the frequency test as regards the relationship between clitic placement and verbal position are shown in Table 3:  All cases of split cliticization and circumverbal clitics have been excluded from the table and analysed separately. Having taken this into account, the conclusion that emerges from the table is that in PelTs, clitics are predominantly non-2P, but there are a few cases where the syntax of the clitic following an initial verb is preserved.14 These appear to be residual structures. Here is an example from Deffner (1926:40)  The frequency of these postverbal clitics decreases (from 12.4 % in the first period, to 7.4% in the second, to 4.8% in the third), but they do not totally disappear, and this fact has some significance: if we analyse the postverbal clitics more closely, we will see that it is not just a matter of free variation.
In PelTs, in contrast to all other varieties of Greek, the type of sentence seems to be a critical factor for determining the position of the clitic. Qualitative analysis of the material allows us to form the hypothesis that interrogative clauses constitute an intonational environment for preservation of the 2P clitics pattern. In the following example, (a) and (b) differ only as regards the intonation contour, but in (a) the clitic cluster precedes the initial verb, while in (b) it follows. The verb in the interrogative clause in (c) is also considered to be in initial position, since, as previously noted, topicalized elements (here the object, as shown by the resumptive pronoun s(i)) are considered to be clauseexternal. Thus, in this example too, the postverbal position of ClAux is what we would expect based on our analysis. In contrast, in the last two examples the initial position in the interrogative clause is occupied: by a function word (i.e. the interrogative particle mí) in (d) and by a focused word in (e); therefore, the clitic precedes the verb.  (Costakis 1986(Costakis -1987 d. mí q nd do2sg erékai found.3pl i the pónu? pains 'Did the pains maybe find you?' (Costakis 1986(Costakis -1987  shout.ptcp.f.sg 'Was it me who was shouting?' (Costakis 1986(Costakis -1987 Statistical analysis of the data confirms the hypothesis. Consider for each period the crosstabulation of the two variables (cl.pl by verb. pos.) in Table 4, which are controlled by the type of clause.
As in Table 3, there are hardly any examples of postverbal clitics with verbs in non-initial position (but see fn. 14). The percentage of instances where the clitic follows the initial verb in an interrogative clause is impressively high, especially in the first two periods (85.7% and 94.1%, respectively) while in statements it is extremely low: actually, it is only in the first period that this constitutes a legitimate choice (almost one in ten cases (9.7%)), while in the two subsequent periods the percentage of postverbal clitics is less than 3 % (1.9 % and 2.9 %, respectively) and their use can be considered a relic. The extreme resilience of the 2P clitics pattern in an interrogative environment shows that by the 2nd half of the 19th century, a special syntactic model had already taken shape, justifying the statement of Pernot (1934:222) that, despite the fact that the clitic usually precedes the verb, "l'interrogation demande l'inversion". The noticeable fall in the percentage of postverbal clitics in interrogative clauses in the third period (from 94.1% to 58%) is probably due to the increased influence of smg on Journal of Greek Linguistics 17 (2017) 37-72 PelTs in the last few decades, which has been confirmed at all levels of linguistic analysis (Liosis 2007) and ultimately leads to the total dominance of the non-2P clitic model.

4.1.2
Clitic placement and verbal position: split cliticization Of all the problematic cases, the easiest to interpret is probably the split syntax of clitics in the environment of negative avcs. This is the only case of split cliticization of a ClAux + Clpr cluster, which is canonical in the sense that there is no variation here; in all negative sentences with an avc, the ClAux is always on the left and the ClPr always on the right of the verb. See the following examples from MedG (28a), the dialect of Farasa ((28b), cf. example (25b)), and Tsakonian (28c), where, as previously mentioned, the presence of a cognate of the ag negative οὐκ forces the ClPrs to appear in postverbal position:  [pronoun]". However, if this is the case, why does the Tsakonian ClAux remain preverbal? The answer is that the negative ó-in Tsakonian is not a free element, but rather a bound morpheme incorporated into the auxiliary (cf. Joseph (1990b:175-176), who argues in favor of the affixal status of the negative ó-on the basis of its idiosyncratic form). It is therefore preferable to speak of a single, unified form: the negative ClAux. (Cf. Anderson (2006:33-34), who refers to the existence of negative auxiliaries in whole families of languages, such as Uralic, Tungusic etc.) This is shown by the fact that ó-appears only in the periphrastic tenses, while in the synthetic tenses (e.g. the aorist) we find the negative δέ(ν), as in the following example from Deffner (1921:162), where the preverbal position of the clitic is the expected one: kakó harm 'He didn't do her any harm' Given that negation in Tsakonian, as in Greek in general, is a functional projection above / to the left of TnsP (cf. (6)), the only available position for the negative ClAux is preceding the verb, and this is the reason for its separation from the ClPr.
In the two periods following Deffner, in southern PelTs we find quite a few examples of analogical extension of the use of the negative ό-to the aorist, but only when the clause does not contain ClPrs (30a), or contains only postverbal ClPrs (30b), otherwise δέ(ν) is chosen (30c).15 (30) Aorist and negation in modern PelTs a. ókáne neg.came.3sg 'He didn't come' (Charalambopoulos 1980:156) b. óδútse neg.gave.3sg námu us ts h ípta nothing 'He didn't give us anything' (Liosis 2007) c. kanéna noone δé neg ɲ do3sg orátse saw.3sg 'Nobody saw her' (Charalambopoulos 1980:148) 15 A side effect of the new situation with parallel use of the two negative particles in PelTs is the presence of the hybrid form δó (< δé + ó), e.g. δó ɲ oráka (= not him saw.1sg) 'I didn't see him' (Costakis 1986(Costakis -1987. Cf. also Joseph (1990b:175-176 (Costakis 1983:64) The preverbal parts m(i)= and nd(i)= cannot appear alone, as in this case they would denote the singular of the person in question. The same is true of the postverbal parts =námu and =ɲúmu: these are in the correct number, but, as mentioned above, modern PelTs does not possess enclitics, at least in nonimperative affirmative clauses. Thus we are obliged to recognise these elements as circumclitics. Note that in (31b) the ClAux is combined with the proclitic part of the circumclitic pronoun.
In the material, it is possible to track the emergence and establishment of the circumclitics m(i)=…=námu and nd(i)=…=ɲúmu in the system of ClPrs. In the first period, when the postverbal clitics are still quite well preserved, cases of circumcliticization are completely absent. In the second period, however, we find hybrid examples (i.e. with 2P and non-2P clitics), such as (32), where the whole preverbal form is repeated after the verb through hypercorrection, even though the clause-initial position is occupied: námu io1pl 'They did this and that to us' (Costakis 1986(Costakis -1987 At the next stage, alliteration with a preceding particle and especially na, as in (33), causes haplological reduction of the preverbal pronoun to mu:17 16 The form ɲímu is the northern PelTs variant of southern PelTs ɲúmu. 17 Moreover, the presence of the modal particle na, which is proclitic on the verb, and other such particles onto which the clitic was originally encliticized, has been proposed to be the cause of the prosodic reanalysis of the clitics as proclitic on the verb that took place in the medieval period (Horrocks 2010:277-278;rs 2008 (Liosis 2007) Further reduction of the preverbal part, as the deletion of the final -u in (34), is also due to sandhi phenomena between the clitic and the verb which follows it (cf. namu > nam in (32)): alíere tell.2sg.sbjv námu io1pl 'that you tell us' (Liosis 2007) The form m < mu is now homonymous with the 1sg. m < mi (cf. 17b), and is reanalysed as such. This can be seen in (35), where the final -i of the preverbal part is pronounced because the verb starts with a consonant: give.2sg.sbjv námu io1pl 'What will you give us?' (Liosis 2007) Further proof that the first part of the clitic has undergone reanalysis is the fact that in the 2nd person it is replaced analogically with the preverbal form nd(i) of the 2nd person singular (see 31b), which cannot of course be derived from the enclitic form ɲúmu of the 2nd person plural.
The above description is confirmed by statistical analysis of the 1st and 2nd person plural ClPrs, as shown below, in Table 5. During the first period, I counted five instances where 1st and 2nd person plural ClPrs appear with a verb in initial position, of which only one was postverbal and the others preverbal, which is consistent with the usual behavior of ClPrs in this period (cf. Table 3). With verbs in non-initial position, all instances were in preverbal position. However, in the next period, the variable clitic placement now has three values, as circumclitics begin to appear, constituting 40 % of instances with verbs in non-initial position. Thus the syntactic environment with the verb in non-initial position was the starting point for the spread of circumclitics. Today, circumverbal placement of 1st and 2nd person plural ClPrs is the reduction of the clitic námu > mu in PelTs is entirely comparable with the development of the same form in PrTs, where it is preverbal, e.g. ja na mu pároi (= to sbjv us take.3pl (Costakis 1986(Costakis -1987) '(in order) for them to take us' , cf. δókai nám δulía (= gave.3pl us work (Costakis 1986(Costakis -1987:420)) 'they gave us work' .
Journal of Greek Linguistics 17 (2017) 37-72 in (36), námu is, according to Selkirk's (1995) typology of host-clitic relations, a prosodic word clitic, which simply adjoins to the phonological phrase, in contrast to the proclitic cluster m-ec (< mi eci), which adjoins to the prosodic word of the verb, as shown by the deletion of the final i of the auxiliary eci before the initial vowel of the verb. This kind of left / right prosodic asymmetry, which in our case is simultaneous, is predicted according to the templates for prosodic analysis proposed by Selkirk (1995) and applied in several varieties of Greek by rs (2006) and Revithiadou (2008). The interpretation we must seek for the formation of circumclitics is historical. The PelTs disyllabic / stressed clitics appear fossilized within the framework of a postverbal syntactic pattern that has ceased to be productive. This is what allows them to be 'corrected' by the pleonastic use of a true, unstressed proclitic element, with which ClAuxs may cluster.21 The result is the formation of discontinuous clitics that attach to both edges of their host.

Conclusions
The main peculiarity of the Tsakonian clitic system is the presence of a cliticized auxiliary, which shares the syntactic and prosodic properties of object clitic pronouns and may cluster with them in preverbal or postverbal position. There is a major typological difference between the clitics of the two subdialects: PrTs possesses a 2P clitic system while PelTs does not. However, since there are so many cases where the PelTs clitics deviate from the model that would have them proclitic on the verb, the superficial impression that this subdialect possesses typical non-2P clitics is clearly misleading. Furthermore, the PelTs clitic system is not simply a mixed system in a state of transition between the inherited enclitic model of MedG and the proclitic model of smg, because liosis Journal of Greek Linguistics 17 (2017) 37-72 certain changes have taken place in it, such as the emergence of the circumclitic pattern, which seem to deviate from a non-2P clitic system rather than inclining towards the establishment of such a system. Specifically, the PelTs clitics are parametrized based on the interrelation of one syntactic factor, the position of the verb, and two phonological factors, the pitch contour of the utterance and the independent stress of disyllabic clitics. So, the final descriptive formulation for the PelTs clitics (based on ck's own formulation for the MedG clitics) is as follows: a) A clitic (Aux or unstressed Pr) is postverbal if the verb is initial in its core interrogative clause. b) Otherwise it is preverbal. c) A stressed Clpr is circumverbal.
Circumcliticization of ClPrs in Tsakonian is a phenomenon that is unique not only in Greek and its varieties, but also in the other Balkan and Indo-European languages. As far as I am aware, it is confined to exotic languages such as Sm'algyax, an indigenous polysynthetic language of British Columbia, Canada (Mulder & Sellers 2010).
One final observation: in many works on the subject, the proclitic or enclitic status of pronominal clitics is often defined based on their prosodic and syntactic behavior in relation to auxiliaries in perfect or future periphrases (Mackridge 1993:338;Pappas 2004:60-70;ck 2001:17-19, 29-31, among others). Experience with the auxiliary in Tsakonian shows that drawing such conclusions is, to say the least, premature if the auxiliary has not first been accurately located on the verb~functional affix continuum.