Coordination strategies in Messapic

Messapic, like many ancient Indo-European languages, shows evidence for the existence of more than one coordinating conjunction. Alongside an inherited Indo-Euro-pean clitic = ti , comparable to Lat. = que or Gk. =τε, Messapic has also grammaticalised an additional coordinator anda from a lexical source. Although the two conjunctions are found in some strikingly similar contexts, this paper argues that they also show noteworthy functional differences, which can plausibly be contextualised against a broader cross-linguistic understanding of how novel coordinators grammaticalise. In this way, the typological study of coordination has the potential to contribute further insights into this fascinating—and as yet insufficiently understood—epigraphic language.


Coordination in Messapic and Indo-European
The history of coordinative conjunction in the Indo-European languages is widely characterised by the replacement of the old Proto-Indo-European enclitic conjunction *=kʷe by innovative forms that grammaticalised from lexical items (Goldstein 2019: 3). Classical Greek can serve as an example of this phenomenon. A new conjunction καί, ultimately deriving from a PIE adverbial element *km̥ t 'along with, besides' (Beekes 2010: 615), is in the process of replacing inherited =τε < *=kʷe in the documented history of the language. A clear diachronic trend can be observed in this regard, with καί being only about 2.3 times as frequent as =τε in Herodotus, but becoming an overwhelming 38.3 pitts 10.1163/22125892-0100110020 | Indo-European Linguistics (2022) 1-18 times as frequent in the New Testament.1 And even where =τε is used, it is frequently used in combination with καί (in the construction x τε καὶ y).
Although the evolution of new coordinating conjunctions happens independently in the various branches of Indo-European, we find these forms following remarkably similar grammaticalisation paths, with lexical source meanings such as 'also' , 'besides' , or 'in addition' regularly and recurrently becoming these incipient conjunctions. In the documented history of Classical Latin, for instance, an innovated et < *eti 'also' is in the course of supplanting inherited =que in much the same way as in Greek (Torrego 2009: 454;Dupraz 2014: 138). Similarly, in the Sabellic languages a cognate of Lat. enim, originally an adverbialised accusative pronoun meaning 'in addition to this, besides' , has supplied the novel coordinator attested in Osc. íním, Pael. inim, Umb. enem and completely ousted any inherited clitic conjunction (Untermann 2000: 343-344). In this regard, therefore, the linguistically diverse landscape of ancient Italy reflects the general evolutionary trends of ancient Indo-European.
No exception is Messapic, a poorly attested Indo-European language once spoken in the heel of Italy. Today, this language presents the usual interpretative difficulties of a fragmentarily-attested epigraphic language. However, unmistakeable reflexes of *=kʷe can be identified in a clitic =ti or =θi, which occurs in a number of contexts where it serves to coordinate lists of names, objects, or circumstances (e.g., litanθi notanθi palanθi in MLM 22 Ro).2 At the same time, a new coordinator anda(ti) has been grammaticalised for apparently very similar functions (e.g., [i]assetian andati atan in MLM 13 Ro) (Marchesini 2009: 153;Matzinger 2019: 98). Like its cousins in Greek, Latin, and Sabellic, this new form probably derived from a lexical source, perhaps from a particle *n̥ do 'towards, in addition' (Dunkel 2014: 153;De Simone 1988: 360-362).
The examples given above for =ti and anda both show the coordination of accusative common nouns as direct objects. It is instructive to note, therefore, that these coordinators are used in active and ostensibly free variation, much like, for example, =que and et in early Latin, =τε and καί in Homeric Greek, or =ca and utá in Vedic Sanskrit. Given that the grammaticalisation of conjunctions in Indo-European languages follows a predictable path, it is tempting to use data from the history and usage of other Indo-European conjunctions to cast further light on Messapic. Which syntactic or semantic factors can explain the distribution of two competing coordinators in languages that are in the process of grammaticalising an innovative lexical conjunction? If Messapic 1 The statistics were calculated using the PROIEL corpus (Eckhoff et al. 2018 is one such language, can our theoretical understanding of this process contribute to the interpretation of this as yet insufficiently understood corpus? And conversely, does the Messapic case have the potential to add to our store of typological knowledge on the means whereby languages grammaticalise new coordinators?

Methodology and data collection
To examine the evolution of coordination strategies in Messapic, it is useful first to aggregate all instances of coordination attested in the corpus. For this purpose, an initial dataset was created using the morphological information for Messapic in the Corpus of the Epigraphy of the Italian Peninsula in the 1st Millennium bce (CEIPoM) database, which contains linguistic information on the epigraphic corpora of a range of languages spoken in ancient Italy (Pitts 2022), and then manually rechecked against the documentation of the published Messapic texts. Fortunately, the small size of the corpus which makes this quality check imperative at same time also makes it feasible, and CEIPoM did miss a small number of possible coordinators in obscure contexts. I believe the list given below to be exhaustive. The inherited clitic =ti surfaces in various forms, including the orthographic variant =θi and also an assimilated variant =si after sibilants. The innovated anda, on the other hand, also occurs as andati or andaθi, where the conjunction is itself redundantly recharacterised with the clitic =ti. This can be compared with Gaul. etic, which shows the same pattern of a conjunction *eti reinforced by *=kʷe (Matzinger 2019: 98;Dunkel 2014: 263). Despite the fact that such redundant coordination incorporates the inherited form, it nonetheless represents a 'heavier' and more lexicalised form of coordination and should therefore be counted as an instance of anda, not of =ti. A comparable case can be found in Lat. atque, which also univerbates a lexical item with the inherited clitic, but shows the behaviour of a relatively weakly grammaticalised conjunction (Pinkster 2021: 628). This ties in with the well-established theoretical maxim that it is constructions, not individual morphemes, which grammaticalise (Bybee, Perkins, & Pagliuca 1994: 11). For the purposes of this paper, therefore, andati is considered a variant of anda.
The following is a full list of potential instances of Messapic coordination that are relevant to the research question of this paper. The list strives to be maximalist, including a number of cases where the context or even the presence of the conjunction itself is unclear. The passages are provided with their reference in Monumenta linguae Messapicae and in the order in which pitts 10.1163/22125892-0100110020 | Indo-European Linguistics (2022) 1-18 they occur in this corpus. The text of the relevant Messapic inscriptions is reproduced from the interpretative linguistic text ('Testo ii') in MLM, except that the conjunction under discussion is put in bold type. A linguistic interpretation is given based on the relevant literature (cf. Krahe 1955;Parlangèli 1960;Santoro 1982;De Simone 1988;Matzinger 2019; see further references below).
klohi zis θotoria marta pido vastei basta veinan aran in daranθoa vasti … inθi trigonošoa … inθi reššorišoa … inθi vastima … inθi ardannoa … The first line of this inscription, after the introductory formula klohi zis 'hear, o Zeus' , is usually interpreted as 'Totoria Marta gave (pido) her own field (acc. veinan aran) to the city of Vaste (dat. vastei basta).' Followed by a list of coordinated adverbials indicating the circumstances of this gift, each with a head noun in the locative case. These may refer either to locations (e.g., 'in the senate …') or, perhaps more plausibly, functions (e.g., 'in the senatorship …') (Matzinger 2019: 139-145). In all cases except for the first coordinand, the coordinator =ti is enclitic to the preposition.
(3) MLM 1 Bas (3rd century bce) … inθi reššorišoa kazareihi šonetθihi otθeihiθi dazohonnihi … This phrase has exactly the same syntactic structure as the preceding, with two coordinated genitive proper nouns. The meaning of reššorišoa is unknown.

(4) MLM 5 Bas (3rd century bce) oššo[vas?] no hazav[a(s)θi?]
This is a dubious occurrence of =ti, as the coordinator is not legible and its presence must be inferred. If correct, however, it may coordinate two genitive nouns used as predicate nominals after the verbal particle no. A translation might run 'I belong to Oššo and Hazava' .
(5) MLM 14 Bas (3rd century bce) anda[ This could be an instance of anda, but all further context is unclear.
This phrase is preceded by an introductory formula (klaohi zis), while what follows is obscure. Here anda seems to be coordinating two locative adverbials, 'in the city and in the deranθoa' . The preceding denθ«ava»n might be an accusative noun (De Simone & Marchesini 2002: ii 110) or a preterite verb in the third person plural (Parlangèli 1960: 301).
The assimilated =si < =ti at the beginning of this sentence seems to be appended to a nominative form t..laihas. Although the function of this form is unclear, it is plausible to assume that =ti is coordinating either an entire sentence, or its nominative subjects.
The clitic conjunction follows a genitive proper noun, which it is presumably coordinating as an attribute.
The coordinator is followed by a locative designation of a place or function, followed by a sequence of genitive proper nouns. This immediately follows the preceding section, whereas tai ma kos clearly introduces a new sentence (see following occurrence). This leaves only an obscure section relevant to the interpretation of =ti, but it may be noted that the conjunction appears to follow an accusative ending -en, which also occurs earlier in the passage. It is possible, then, that two accusative objects are being joined.
(16) MLM 3 Car (c) (3rd century bce) tai ma kos teimnatataneespalana.ndai anda daranθoa dazetθes haštorrihi anda dazihi … The syntagm tai ma kos is elsewhere combined with a (possibly subjunctive) verb to express a prohibition (cf. Matzinger 2019: 100). After daranθoa, which is probably a locative adverbial, a lengthy sequence of proper nouns in the nominative and genitive follows.  (Krahe 1955: 34). In this case, dokihi could represent the first element within the coordinated phrase, and a skeletal (albeit speculative) interpretation might be 'of Dokes, whose … he verb-s, and of Kretas Korθes, …' .  Bakkum 2009: 169). In this case, the conjunction joins two accusative objects-iassetian and hezzan-of the verb apistaθi 'dedicates' . As usual, little can be said about their semantics, but these are presumably common nouns.
(26) MLM 6 Ro (no date given) rania vallana θeotori andiraho nai klatan intaθi vateihasθi niligaves The verb (ni)ligaves is attested several times in the Messapic corpus (Matzinger 2019: 78). It is probable that intaθi is also a third person singular verb-rather than, for instance, some form with a clitic conjunction =ti-as this allows a straightforward syntactic interpretation of the preceding context, involving a subject rania vallana, an indirect object θeotori andiraho and a direct object klatan (De Simone 1988: 384;Poetto 1997: 788;Matzinger 2019: 68-69). If so, two clauses are coordinated here by =ti, and the conjunction is enclitic to the subject vateihas of the second sentence.
(28) MLM 10 Ro (no date given) ]azasθi niligaves A nominative subject is here followed by a verb niligaves, with no further context. Since the verb is probably singular, =ti is more likely to be coordinating two clauses than two subjects. The structure of the first part of this inscription is relatively clear, with a subject ]sterinnes, probably a proper noun, in the nominative case, and a sequence of coordinated accusative common nouns litan, notan, and palan representing the direct object (De Simone 1988: 405). Nothing specific can be said about the meaning of these words, but it is worth noting that, unlike in 1 Bas, the conjunction =ti is here repeated on every coordinated element, including the first.

(33) MLM 32 Ur (3rd century bce) ]ti stahan aprodita[---]a [d]azimaihi melan[---]los in karmatθoa de[..]antoθi
The word preceding =θi might be a proper noun with a nominative in -o (< *-ō-n), as attested for oššo or poollo. In this case, it could be one of the nominative subjects of stahan aprodita ('they placed for Aphrodite') earlier in the inscription (Santoro 1982: ii 65). case. In second place by frequency stands the coordination of locative adverbials, which often specify the location or term of office in or during which the event described by the inscription took place. And finally, we find a not insignificant handful of cases where arguments or even entire clauses are coordinated. The table below shows the inscription reference for all potential cases of each of these functions, adding a bracketed question mark (?) for occurrences that are dubious or can be analysed differently. Note that the same occurrence may occur more than once in this

Discussion
The Messapic data presented in the previous section indicates that =ti and anda, to a considerable extent, occupy the same semantic space. They can both coordinate genitive proper nouns, locative circumstances, and arguments in the nominative and accusative. At the same time, their distribution does not appear to be susceptible to a diachronic or diatopic explanation: anda and =ti are found in inscriptions from the same time periods and with the same geographic provenance (e.g., the Grotta della Poesia), and not infrequently occur in the exact same inscription (1 Br, 3 Car, 10 Uz, 17 Ve). This suggests that-as in many other early Indo-European languages-these two conjunctions coexisted, and that the variation between them requires a functional explanation. Indeed, we do observe some patterns that, at least in the absence of more detailed information, are interesting to take at face value. For example, =ti appears proportionately more common for the coordination of genitive proper nouns, and it is exclusively used in the clear cases of clausal coordination. Although the statistical resolution remains low, these are potential patterns which can be fruitfully be compared with cross-linguistic parallels.
As was noted in the first section of this paper, a number of other early Indo-European languages also provide evidence for a situation in which a more recently grammaticalised lexical item (analogous to anda) competes with an inherited clitic conjunction (=ti). Since grammaticalisation tends to follow typologically predictable pathways, the analysis of coordinative conjunction in these languages may exemplify some of the distinctions we might expect to observe in a comparable case such as Messapic, and thus help inform the interpretation of this more poorly attested epigraphic language.
Such distinctions can be identified both on a syntactic and on a semantic level. In the first place, more strongly grammaticalised elements tend to have narrower syntactic scope (Lehmann 2015: 152-157). For instance, =τε in Homeric Greek tends to join elements at the level of the word or phrase, while καί predominates for sentential coordination (Klein 1992: 12). A similar pattern holds for Lat. =que and et and becomes increasingly prominent in later Latin literature (Pinkster 2021: 625). On a semantic level, a more strongly grammaticalised conjunction should display bleaching towards more neutral coordination, and away from any potential adverbial source meanings such as also or besides. Again, we see this in Classical Latin, where et can still be used adverbially, while =que cannot (or very rarely; see Torrego 2009: 457), and also in the adverbial usage of lexical conjunctions in Greek (καί) and Vedic Sanskrit (utá) (Klein 1992: 19).
The semantic bleaching of a coordination may also be reflected in the conceptual similarity and symmetry of the coordinands. Thus, a conjunction may join unrelated elements, but it may also join elements which are in some way associated with each other in the real world, such as 'husband and wife' or 'sun and moon' (Haspelmath 2004: 13;Viti 2008: 46). In early Indo-European languages, this function is particularly associated with the inherited coordinator (Viti 2008: 47;Torrego 2009: 457;Pinkster 2021: 621). When the link between the coordinands is clear from our encyclopaedic knowledge of reality, the coordinator itself has little semantic work left to do, and it is not surprising that we find this function taken by a strongly grammaticalised coordinator with limited semantic content of its own. Similarly, a more strongly grammaticalised conjunction will tend to join elements symmetrically, in such a way that the order of elements is not linguistically relevant and has no implications in terms of pitts 10.1163/22125892-0100110020 | Indo-European Linguistics (2022) 1-18 causality or focalisation. Viti (2008: 55) contrasts the sentence 'John studied much and passed the exam' (where causality is strongly implied) with the symmetric 'John cooks pancakes and Mary prepares the coffee' , where the order of the coordinands can be inverted without significantly altering the meaning, and argues that this distinction is also reflected in the use of early Indo-European clitic conjunctions (including in Vedic Sanskrit, Greek, and Latin).
At a first examination, the Messapic data may seem to belie some of these expectations. Thus, all the clear and potential examples of coordinated clauses use =ti, not anda, which is contrary to the expectation that highly grammaticalised morphemes should have narrowed syntactic scope. Conversely, however, it does have parallels in early epigraphic Latin. For instance, the closing formula idemque probavit 'the same person approved it' is invariably joined to what precedes with =que. It is possible to hypothesise that this archaism reflects the semantic bleaching of the inherited conjunction before it acquired the arbitrary syntactic restrictions that would characterise it in its later and more moribund stages. Arguably, if =que was semantically empty, but otherwise free to coordinate clauses, we would expect it to occur in situations where asyndeton-the most semantically empty form of coordination par excellence-is an acceptable alternative. It is instructive to note that the Oscan equivalent of the idemque probavit formula, ísídum prúfatted, is without exception appended asyndetically to what precedes.5 After all, the act of commissioning a project and approving it afterwards are conceptually inseparable. Coordination in this case has little semantic context and is adequately marked by the relatively empty coordinator =que.
In Messapic, the clearly analysable cases of clausal coordination involve the verb niligaves, and only one of them has sufficient context for the identification of the first element of the coordination. This appears to involve a dedicatory verb with a meaning closely related to that of niligaves itself (De Simone 1988: 392). The full sentence rania vallana θeotori andiraho nai klatan intaθi vateihasθi niligaves could be translated along the lines of 'Rania Vallana dedicates the klatan to Tautor Andirahias indeed, and Vateihas placed it' (see discussion and references in previous section). In other words, although the precise denotational meaning of the two verbal acts is unclear, they clearly occur within the context of the same votive act and are consequently connected by =ti. Elsewhere, Messapic attests asyndeton in a comparable context. MLM 5 Ro, for 5 The formula occurs 20 times in the Oscan corpus. Explicit coordination is found a few times in the structurally divergent schema x deded íním prúfatted 'somebody had this made and approved it' , where two coordinated predicates share the same subject (data from CEIPoM 1.2).
instance, connects three dedicatory verbs apistaθi … topido … niligaves 'dedicates … gave … placed' without any explicit coordinator. So, although the use of =ti to coordinate clauses counterindicates a narrowing of its syntactic scope, its apparent functional equivalence with comparable asyndetic constructions is nonetheless in line with the theoretical expectation of semantic bleaching in a highly grammaticalised conjunction and has arguable parallels in archaic Latin. This observation also ties in well with the Messapic tendency to join lists of proper nouns with =ti. In many cases, the epigraphic corpora of ancient Italy preferentially employ asyndetic coordination with proper nouns,6 and this is in accordance with the diagnostics of semantically empty coordination described above. Enumerations of proper nouns are highly symmetric lists of which the internal order is linguistically irrelevant. In fact, in many contexts, such internal structure could serve to create unwanted salience, and thus have implicational overtones that someone commissioning an inscription would wish to avoid. A case can be made that Messapic itself also attests asyndeton in a comparable context (cf. the possible enumeration of theonyms in formulae such as klohi venas zis in MLM 3 Car, 8 Ro, perhaps 5 Ro). Constituents involving common nouns, on the other hand, are free to be coordinated more explicitly, whether by the use of a lexical coordinator such as anda(ti), or by the emphatic repetition of the clitic on the first coordinand, as in litanθi notanθi palanθi. The tendency for =ti to occur in lists of proper nouns is, therefore, precisely in line with the theoretical expectations outlined previously, and once again is also reflected in its apparent functional equivalence with asyndeton.

Conclusion
In conclusion, therefore, the close functional similarities of =ti and anda notwithstanding, =ti preferentially occurs in contexts where we may reasonably assume that the semantic content of the coordinator is minimal: in the coordination of clauses representing two inseparable parts of a dedicatory act and in the enumeration of proper nouns. In this regard, Messapic shows interesting resemblances to other early Indo-European languages, perhaps particularly to pitts 10.1163/22125892-0100110020 | Indo-European Linguistics (2022) 1-18 the situation for Latin at the same point in time. The semantic bleaching of the clitic conjunction is in evidence in both languages, whereas the syntactic restrictions of classical and later Latin are only in the process of emerging, with Lat. =que becoming increasingly rare for clausal coordination as one passes from Plautus through Cicero to Tacitus (Pinkster 2021: 625). By this point, Messapic was long extinct, but we can but speculate on the course =ti and anda would have taken over the subsequent centuries. It is likely that the similarities between other early Indo-European languages and our fleeting glimpse into the history of Messapic can be ascribed to the predictable path that cross-linguistically common grammaticalisations tend to follow. This paper has attempted to show that, studied in the light of typologically informed grammaticalisation pathways, the Messapic corpus can yield useful insights and is rich and varied enough to be studied in its own right. Moreover, based on the data presented here, the case of Messapic can-at least tentatively-be added to the list of documented languages that show evidence for the common Indo-European grammaticalisation cycle of coordinators. In addition, the functional differences between =ti and anda have the potential to help with the interpretation of those parts of the Messapic corpus which as yet remain obscure, as well as with any new inscriptions that may be published in the future.