THE DATIVE AND INSTRUMENTAL DUAL IN EAST BALTIC

. The paper deals with two case-forms of East Baltic nominals, the dative and the instrumental dual. It summarises what is already known about these case-forms from Lithuanian and Latvian dialects as well as from Old Lithuanian sources. It is demonstrated that these data imply a particular reconstruction of the dual inflectional endings in these cases for Proto-Baltic. Subsequently, the paper shows that etymological matches of the reconstructed Proto-Baltic endings of the dative and instrumental dual may be attested in two further branches of Indo-European. Such endings are found in Gaulish, which is a member of the Celtic branch, and in Ancient Greek.


Introduction
The East Baltic languages Lithuanian and Latvian constitute a sub-branch of Baltic whose only other sub-branch, traditionally called West Baltic, is represented by now extinct Old Prussian. The Baltic languages belong to the Balto-Slavonic branch of the Indo-European language family. This means that Proto-Baltic, i.e. the common parent language of East and West Baltic, must have developed from the same proto-language that gave birth to Proto-Slavonic. In other words, Proto-Baltic and Proto-Slavonic were sisters whose mother, Proto-Balto-Slavonic, originated as a dialect of an even more ancient proto-language, Proto-Indo-European.
As is well known, the evidence of several ancient Indo-European (henceforth IE) languages implies that in Proto-Indo-European (henceforth PIE) not just two but three numbers were distinguished in the inflection of nouns, pronouns, adjectives and numerals. The most ancient Indo-Iranian languages (Vedic Sanskrit, Old Persian and Avestan), as well as Ancient Greek, Old Irish from the Celtic branch of IE and, finally, Gothic and Old Norse from its Germanic branch all possess at least remnants of a special number category clearly distinct from the singular and the plural. This additional number is traditionally called the dual. The original function of the dual in nouns must have been to indicate that specifically two items were meant, as opposed to just one in the singular and more than two in the plural. Accordingly, in deictic pronouns and adjectives the use of the dual served agreement with the relevant pairs of nouns. In personal pronouns the dual expressed the idea of 'we two' and 'you two'. In the domain of numerals special dual case-forms were used for inflecting 'two'.
The PIE dual has been well preserved in Balto-Slavonic nominals. It is amply attested in all medieval and is still in use in several modern Slavonic languages. As for Baltic, the dual of nouns and adjectives was frequent in Old Lithuanian texts from the 16 th and 17 th c. and, in dialects, remained in use at least until late into the 20 th c. (cf. K a z l a u s k a s 1968, 127-132; Z i n ke v i č i u s 1966, 203-207; 1980, 179-180). Accordingly, the dual caseforms of nominals are routinely treated in many Lithuanian grammars since K l e i n (1653; 1654), cf. S c h l e i c h e r (1856), K u r s c h a t (1876), O t r ę b s k i (1956,(81)(82)(83)(84)(85), S e n n (1966) and others.
In the present paper I intend to clarify the origin and to reconstruct the development of two dual case-forms in Lithuanian and, accordingly, in all of Baltic. These case-forms are the dative and the instrumental dual. To achieve this goal, I will start (section 3 below) from what is known about the dative and the instrumental dual of nominals from Lithuanian and Latvian dialects as well as from the Old Lithuanian sources. I will demonstrate that these data imply a particular reconstruction of inflectional endings in these two cases for the Proto-Baltic times. Subsequently (section 4), I will demonstrate that reflexes of these endings may be attested in at least two further branches of IE and therefore may be as old as PIE. The last section of the paper (section 5) will summarise the results and sketch the problems which remain to be addressed in the future.
However, before turning to the dative and instrumental dual of Lithuanian specifically some preliminary remarks on the dual inflection of nominals are in order.

Preliminaries on dual case-forms of nominals in Indo-European
At least in the IE language family, the dual inflection of nominals exhibits properties which are not found in the other two numbers. The first of these properties is the inclination of the dual inflection of nominals towards a secondary merger with the plural. This abandoning of the inherited distinction between the dual and the plural numbers of nominals can be complete, i.e. affect the entire system, or only partial. For instance, hardly any trace of the PIE dual is found in Latin despite the fact that Latin text records start as early as the middle of the 1 st millennium BC. By contrast, in the Germanic branch of IE, the written records of which begin many centuries later, the dual is preserved but only in personal pronouns. In Old Irish, which belongs to the Celtic branch and is attested since the early Middle Ages, the inherited dual case-forms of nouns are only used after the numeral 'two' while the adjectives and pronouns lack a dual.
However, the tendency to abandon the distinction between the dual and the plural does not mean that the inherited dual case-forms of nominals are necessarily lost, being replaced by their pluralic counterparts in the whole system. Sometimes it is the inherited plural case-forms which are replaced by their dualic peers that just assumed the new function of designating plurality of any kind. So, in dialects of Russian the inherited instrumental plural of pronouns and feminine nouns may appear replaced by descendants of the Old Russian dative-instrumental dual (cf. B o r kov s k i j, K u z n e c ov 1965, 220).
The second property of the dual inflection of nominals in IE is its inclination towards case-form syncretism. No IE language makes a distinction between the nominative and the accusative dual in nouns of any gender. At the same time these two cases are mostly kept distinct in the singular and plural of gendered, i.e. masculine or feminine, nouns. The same is true for the genitive and locative of Vedic Sanskrit, dative and instrumental of Old Church Slavonic, genitive and dative of Ancient Greek. This case syncretism in the dual may be partly inherited from PIE but seems also to be partly due to recent developments. The latter is strongly indicated, for instance, by the fact that the genitive and locative dual are syncretic in Vedic Sanskrit but kept distinct in the closely related Avestan (cf. H o f f m a n n, Fo r s s m a n 2004, 115).
As will become clear in the following section, both cross-linguistic tendencies of the nominal dual inflection help to correctly understand the situation found in the East Baltic languages Lithuanian and Latvian.

Dative and instrumental dual in Lithuanian and their Latvian counterparts
In contemporary standard Lithuanian, the inherited dual inflection of nominals is preserved only in two words, in the pronoun 'both' and in the numeral 'two'. In these two words both the dative and the instrumental dual end in -m in both genders. However, the dative dual abíem, dvíem and the instrumental dual abiẽm, dviẽm clearly differ in their prosody. Grammatical descriptions written in the 19 th c., when the dual of nouns was still in use in many Lithuanian dialects, as well as works on 20 th c. dialects confirm the same prosodic contrast also for nouns and adjectives, i.e. dat.du. galvóm vs. inst.du. galvõm etc. (cf. especially K u r s c h a t 1876; Z i n ke v i č i u s 1966, 205).
It follows that in the common prehistory of Lithuanian and Latvian the dative and the instrumental dual were prosodically different. Lithuanian late 16 th c. texts demonstrate that this prosodic contrast was originally accompanied by a difference in the segmental composition of the endings. This is shown by Daukša's writings (Catechism from 1595 and Postil from 1599) where the dative dual always ends in -m (cf. 1a) while the instrumental dual may end in -m (1b) or -mi (1c). The vacillation between -mi and -m in the instrumental dual can be either explained as an early stage of the apocope, which ultimately led to modern Lithuanian abiẽm, dviẽm etc., or as a secondary generalisation of the inherited dative dual for both cases. The former explanation is clearly corroborated by the inflection of personal pronouns where in dialects the dative plural mùm, jùm is in contrast with the instrumental plural mumì, jumì. Those dialects which also preserve the inherited dative and instrumental plural as respectively mùms, jùms and mumìs, jumìs (cf. B r ug m a n 1882, 302-303 on Garliava and, more comprehensively, LKA 3, 80-81) demonstrate that the shorter variants mùm, jùm and mumì, jumì most probably reflect the corresponding dual case forms. It follows that prior to the beginning of the written records in most dialects of Lithuanian the dative dual ended in -m while the instrumental dual ended in -mi, the vowel of which was probably preserved better in disyllabic forms (such as Daukša's dviemi and dialectal mumì, jumì).
What kind of vowel had been lost in the dative dual abíem, dvíem etc. and dialectal jùm, mùm before writing in Lithuanian started in the 16 th c. cannot be established with absolute certainty. It is, however, probable that in the personal pronouns a range of conservative Lithuanian dialects preserved this vowel until the 20 th c. In the East Aukštaitian dialect of Šeduva the dative plural of mẽs is attested as (probably unstressed) mum beside stressed mùma The dialect case-forms of personal pronouns mùma, jùma can hardly be a recent innovation. In the grammar of 18 th -20 th c. Lithuanian there is no model on which such case-forms might have developed anew. Moreover, the use of mùma, jùma in both the dative and the genitive case seems to indicate antiquity (cf. already B e z z e n b e r ge r 1877, 148-149). As is well known, the East Baltic (and Slavonic) genitive case is partly reflecting the more ancient case ablative (cf. S t a ng 1966, 181; E n d z el ī n s 1971, 134; Pe t i t 2010, 13 etc.). The dative vs. genitive syncretism is unusual in Lithuanian but, as 3 Whether such dative plural forms as mùmu, seserìmu, vaikãmu etc. in Dieveniškės (cf. A r u m a a 1930, 61-62) also result from a more ancient dative dual (as suggested in A r u m a a 1933, 65-78), remains unclear. In this dialect the instrumental plural lacks the expected -s as well, cf. sù dukterimì dvíem, but nevertheless probably reflects the inherited plural, cf. taĩs arkliaĩs etc. in the a-stems. This seems to suggest that both -mu in the dative and -mi in the instrumental plural descend from respectively -mus and -mis which only recently lost their -s (perhaps originally in seserìmu, dukterimì and other forms with more than three syllables). In case this assumption is wrong and mùmu, seserìmu etc. indeed descend from the dative dual, such forms in -mu can still be explained as a recent creation based on the inherited (but recently abandoned) dative plural in -mus (which is widespread in 16 th -17 th c. sources). The model would have been provided by the instrumental, where -mi in the dual beside -mis in the plural would have been inherited (cf. Z i n k e v i č i u s 1966, 206; 1980, 198; K a z l a u s k a s 1968, 172; 1970, 91). The situation in the other southern East Aukštaitian dialects is even less clear than in Dieveniškės. In Lazūnai (cf. A r u m a a 1930, 69, V i d u g i r i s 2014, [117][118][119][120][121][122][123][124][125][126][127][128][129][130][131][176][177][178][179][180][181][182][183][184][185][186][187][188][189][190][191], both the dative and the instrumental plural equally end in -m or -mi but contrast prosodically, i.e. dat.pl. galvóm, mùmi, akìmi vs. inst.pl. galvõm, mumì, akimì. Dat.pl. jíemi, arkliám vs. inst.pl. jaĩs, arkliaĩs etc. in the a-stems again point to the plural rather than dual as the historical source of the relevant case-forms. Similar systems are attested in several other local dialects (cf. K a z l a u s k a s 1968, 169). already stated above, a syncretism between the cases dative and ablative in the dual would be unsurprising on its more ancient stage (cf. especially the situation in Vedic Sanskrit and Avestan). It follows that Lith mùma, jùma, which are attested since the 18 th c., most probably reflect a more ancient syncretic dative-ablative dual of personal pronouns. It seems logical to assume the same ending -ma also for the dative dual of all other nominals, i.e. abíem, dvíem etc. The preservation of the word-final vowel in mùma, jùma and its early loss in abíem, dvíem etc. would closely resemble the situation in the instrumental dual already described above, i.e. mumì, jumì vs. abiẽm, dviẽm etc. as early as in many 16 th c. texts.
Taken together, the prosodic contrast between the dative dual abíem, dvíem etc. and the instrumental dual abiẽm, dviẽm as well as the segmental contrast between -ma in the dative dual and -mi in the instrumental dual imply that these case forms arose in the way shown in (2a). The difference in the place of stress (indicated by ˈ preceding the stressed vowel) on the Proto-Baltic level of reconstruction is somewhat unexpected from the perspective of the Indo-European grammar where both case-forms should be stressed the same. However, Hirt's law, established on completely independent grounds (cf. 2b), helps to explain this deviation. 4 (2 In the following section, I will demonstrate that these endings of the dative and instrumental dual, reconstructed on solely East Baltic data, probably possessed exact counterparts in two other branches of IE and therefore, were indeed inherited from PIE.

Dative and instrumental dual outside of Baltic
The suggested reconstruction of the relevant case-endings as Proto-Baltic dative dual *-ma (> OLith -ma, -m, Latv -m) and instrumental dual *-mī́ (> OLith -mi, -m, Latv -m) presupposes respectively PIE *-mo and *-mih. In order to establish whether comparable endings are attested in other branches of IE, one has to take into consideration the well-known fact that the relevant case-endings may begin with *m only in Balto-Slavonic and Germanic. In all other branches preserving the inherited endings of the relevant cases, these endings always begin with a reflex of PIE *bʰ, cf. Skt instr.pl. -bhis, Lat dat.-abl.pl. -bus etc. 5 As soon as this is acknowledged, potential etymological counterparts of the Proto-Baltic dative dual in *-ma and instrumental dual in *-mī́ outside of Baltic become apparent.

Dative and instrumental dual in Celtic
The first branch of IE which seems to have possessed such endings is Celtic. Special dual case-forms of nouns are preserved in Old Irish, the written records of which begin around the year 700 AD. In Old Irish the IE cases dative, instrumental and ablative secondarily merged in all three numbers. The outcome of this merger, which is traditionally called "dative", ends in palatalised -b (written <-ib>) in both dual and plural numbers, i.e. OIr túatha-ib (pronounced /tˈuaθǝ-β'/) of túath 'tribe' etc. This presupposes a Proto-Celtic ending or two distinct Proto-Celtic endings beginning with *b followed by a front vowel. Possible reconstructions are Proto-Celt *-bis (cf. now S t ü b e r 2017, 1204), which can be the match of Skt inst.pl. -bhis, and Proto-Celt *-bī, which would equate OLith instr.du. -mi. It is tempting to assume that the Old Irish dative plural reflected the former and its dual counterpart the latter. However, all by itself Old Irish provides no way to establish that originally the endings in question were indeed different and distributed in this particular way.
Fortunately, three distinct Celtic languages are attested already in antiquity, i.e. many centuries earlier. The most ancient of them is Lepontic which is documented in ca. 140 inscriptions from Northern Italy, written in a variety of the Etruscan alphabet and dating from the late 7 th to late 2 nd c. BC (cf. recently S t i f t e r 2020). The so-called Celtiberian, once spoken in the north of the Iberian Peninsula, left some 200 inscriptions in a variety of the Iberian script, dating from the 2 nd -1 st c. BC (cf. MLH 4,. Finally, the last "continental" Celtic language Gaulish is attested firstly, in a few inscriptions from Northern Italy, written in a variety of the Etruscan alphabet and dating from the 2 nd c. BC; secondly in some 280 inscriptions in Greek letters from Southern France, written in the 2 nd -1 st c. BC; thirdly in ca. 150 inscriptions in Latin script from different parts of France, dating from the 1 st c. BC to the late 5 th c. AD (cf. again S t i f t e r 2012, 523-527; 2020).
These early attested Celtic languages demonstrate that the cases dative and instrumental were still kept distinct in the Proto-Celtic non-singular nominal inflection. The ending of the dative plural is attested as Lepontic -pos (cf. L e j e u n e 1971, 96-104) and Celtiberian -bos (cf. MLH 4, 400). It must be accordingly reconstructed as Proto-Celt *-bos, a direct match of Latin dat.-abl.pl. -bus (cf. S t ü b e r 2017, 1205). The instrumental plural does not appear to be securely attested in Lepontic or Celtiberian. However, the dative plural of these "continental" Celtic languages and its Old Irish counterpart clearly cannot reflect the same case-form. Since the Old Irish dative plural can only reflect the PIE instrumental and its Lepontic and Celtiberian counterpart only the PIE dative, we have to assume that in Proto-Celtic these two cases were still distinguished at least in the plural. Now, in the third "continental" Celtic language, i.e. in Gaulish, both the dative and the instrumental plural appear to be securely attested (cf. again S t ü b e r 2017, 1204-1205). The clearest cases are collected in (3) where the taxonomy of attestations follows RIG 1; 2(1) and 2(2). 6 The dative function of Gaulish -βο, -bo is clearly established by the context. The instrumental function of Gaulish -bi, -be is disputed (cf. E s k a 2003, 105-115 for a discussion). It is possible that in late Gaulish both cases merged into a new syncretic dative ending in -bi like in Old Irish. Nevertheless, the difference between the case formed with Gaulish -βο, -bo on the one hand and the case bearing Gaulish -bi, -be on the other cannot be explained in any other way than as reflecting the inherited difference between the dative and the instrumental. The only peculiar feature of Gaulish -βο, -bo in the dative and -bi, -be in the instrumental is the lack of -s which one expects in the plural of these cases because of both internal Celtic evidence and external comparison. As already stated above, the former is provided by Lep -pos, Celtib -bos and OIr dat.pl. -ib presupposing Proto-Celt *-bis, the latter by Skt instr.pl.
-bhis and Lat dat.-abl.pl. -bus. This unexpected lack of -s in Gaul -βο, -bo and -bi, -be is traditionally explained by a recent loss. However, as recently stated by S t i f t e r (2012, 533), this hypothetical loss cannot be due to a sound change because word-final -s is always preserved in Gaulish in all other positions. For this reason, it is sometimes assumed that Gaul -βο, -bo in the dative somehow lost its -s "under formal influence from originally s-less" instrumental (S t i f t e r 2012, 533, similarly S t ü b e r 2017, 1205 who speaks of a "contamination"). However, an originally s-less instrumental plural can hardly be assumed for Proto-Celtic. Both Celtic itself (OIr dat.pl. -ib < Proto-Celt *-bis) and its relatives (Skt -bhis) attest for PIE *-bʰis. 7 The problem is easily resolved if one assumes that Gaulish -βο, -bo in the dative and -bi, -be in the instrumental do not continue the plural endings of these respective cases but rather their dual endings. Since the dual was preserved in Celtic until Old Irish times, it can be assumed that Proto-Celtic possessed a dative dual in *-bo (> Gaul -βο, -bo) as well as an instrumental dual in *-bī (> Gaul -bi, -be, OIr dat.du. -ib). 8 These endings would be exact matches of their Proto-Baltic counterparts *-ma (> OLith -ma, -m, Latv -m) and *-mī́ (> OLith -mi, -m, Latv -m) reconstructed above.
As for the functions of the Gaulish endings, none of the inscriptions containing the relevant case-forms seems to unambiguously require the plural to the exclusion of the dual. The mothers of the towns Glanum and Nemausus, who are mentioned in the dedications G-64 and G-203, seem to have been local deities. In how many divine mothers the inhabitants believed, is unclear. How many smiths, wells, fathers and sisters were meant in L-13, G-183, L-15 and L-6, is equally difficult to establish. 9 However, even if it were clear that Gaulish -βο, -bo and -bi, -be were used in the plural at the time of the creation of the inscriptions, it would still be possible to assume that in Gaulish (though not in Lepontic, Celtiberian or Old Irish) the inherited dative and instrumental dual were secondarily generalised to represent both non-singular numbers. Similar changes in dialects of Latvian and Russian (cf. section 2 above) show that this would be typologically unremarkable.

Instrumental dual in Greek
The second branch of IE which seems to preserve an instrumental dual ending directly matching Proto-Baltic *-mī́ (> OLith -mi, -m, Latv -m) is Greek. Like in Celtic, in this branch the case-endings which begin with m in Baltic and Slavonic always start with a reflex of PIE *bʰ, i.e. with Proto-Gk *pʰ. This is shown by the instrumental-locative plural ending <-pi> in the most ancient Mycenaean dialect (late 14 th -13 th c. BC) and by the dative singular ending -φι in Homeric texts (late 9 th c. BC). 10 The former -cf. <e-ka-ma-pi>, <te-u-ke-pi> of Gk ἔχμα 'holdfast', τεῦχος 'armour' etc.seems to reflect Proto-Gk *-pʰis, the counterpart of Skt instr.pl. -bhis. The latter -cf. κεφαλῆ-φι, στήθεσ-φι of κεφαλή 'head', στῆθος 'chest' etc.of the nasalising dative dib, deib 'two' of the neuter gender in the whole paradigm of the numeral. In the neuter, the numeral is nasalising in all cases, i.e. also in the nominativeaccusative and in the genitive (which must be itself due to the well-known use of nasalisation as neuter gender marker in Old Irish, cf. S o m m e r 1912, 140). 9 The noun gobed-bi 'smiths' in L-13 is accompanied by dugiíonti=ío, probably 'who worship', which is clearly a plural form. However, the dual has been lost in verbs in all of Celtic, including Old Irish where dual subjects require a predicate in the plural. The same can be assumed for Gaulish and perhaps already for Proto-Celtic. 10