Using Stem Suppletion for Semantic Reconstruction: The Case of Indo-European Modals and East Baltic Future Tense Formations

As is well known, pie possessed several distinct sigmatic formations with modal or future-like semantics. The paper deals with two sigmatic formations which must be reconstructed for pie and obviously possessed a similar semantic value. First: a full grade - si̯e/o -formation which is attested in Indo-Iranian, Continental Celtic and Balto-Slavonic; and second, an athematic - s -formation which is attested in Italic and in the Eastern branch of Baltic. The diverging morphology of these formations implies that they originally also differed in their semantics. The problem is that both formations are reflected as simple future tense in all daughterlanguages which preserved them. How-ever, it seems possible to detect the original semantic difference between these formations by using the evidence of the only ie branch which preserved both formations side by side, i.e. Baltic. The paper investigates the morphology of the sigmatic future tense in dialects of Lithuanian and Latvian and shows that for the common prehistory of East Baltic dialects a secondary conflation of originally independent pie formations— - si̯e/o -formation and - s -formation—in one single paradigm must be assumed. The par-ticulardistributionofbothformationswithintheunifiedparadigmofProto-East-Baltic makes it possible to obtain information on the lost semantic difference between them. Possible traces of the - si̯e/o -formation in the only recorded West Baltic language, Old Prussian, seem to confirm the conclusions drawn on the basis of the East Baltic evidence.


Introduction
We know that Proto-Indo-European possessed several distinct sigmatic formations with modal or future-like semantics. In particular, the evidence of Indo-Iranian and Celtic makes it possible to reconstruct a reduplicated thematic -se/o-formation with zero-grade of the root. Cf.
(2) Greek ἕλκω 'to drag' → 1sg. ἕλξω σπένδω 'to donate' 1sg. σπείσω Old Irish 1sg. tíagu 'to walk' → 1sg. abs. tíassu (< Proto-Celt *tei̯ x-sū < pre-Proto-Celt *tei̯ gh-sō) In Old Irish this formation functions as subjunctive mood. In Greek it is a future tense which may be, again, secondary.2 In my paper I will discuss two further well-known sigmatic formations which must be reconstructed for pie and obviously possessed a similar semantic value. The first is a thematic -si̯ e/o-formation with full grade of the root. Cf.
Whether or not the divergent morphology of these pie sigmatic modals was accompanied by diverging semantics is a disputed matter. We know that in the conjugation systems of ie daughter languages, differences in morphology do not necessarily imply functional differences. For instance, no detectable functional difference has so far been observed between descendants of pie sigmatic and asigmatic aorists in Indo-Iranian, Greek, Celtic or Slavonic. In these daughter languages, the distribution of the two aorist types is determined lexically and, to a lesser degree, by the phonology of the verbal roots. A comparable situation may be assumed for the common prehistory of the languages, i.e. for late pie. Concerning especially the sigmatic modals, a similar position is taken by Jasanoff (2003: 135) and Villanueva Svensson (2012). These scholars assume that all sigmatic formations in question functioned as desideratives, while their distribution in the lexicon of late pie was determined by structural properties of the particular verbal paradigms (in that reduplicated presents paired with reduplicated -se/o-desideratives, etc.).
Indo-European Linguistics 2 (2014) 42-72 I think that the evidence of the ie daughter languages which attest sigmatic modals or future tense formations rather speaks in favour of differences not only in their morphology, but also in their semantics. Several daughter languages of pie preserve more than one sigmatic modal formation. The distribution of these sigmatic modals with different morphology is nowhere reminiscent of the situation with sigmatic and asigmatic aorists. For instance, Old Irish preserves the pie reduplicated -se/o-desiderative as a future tense and the full grade -se/o-formation as subjunctive mood, i.e. with a clear difference in function. In Sanskrit, the reduplicated -se/o-desiderative and the full grade -si̯ e/o-formation are frequently formed from the same verbal root.3 For this reason, it seems probable that also in the case of the two sigmatic modals or future tense formations to be discussed in the present paper-the full grade -si̯ e/o-formation and the athematic -s-formation-an original difference in function should be assumed. The problem is that both formations are reflected as a simple future tense in all daughter-languages which preserve them. How can the original semantic difference be detected?
This seems possible by using the evidence of the only ie branch which attests both formations, i.e. Baltic. In this paper I will investigate the morphology of the sigmatic future tense in dialects of Lithuanian and Latvian. I will show that for the common prehistory of East Baltic dialects, a secondary conflation of both pie formations in one single paradigm must probably be assumed. I will demonstrate that the particular distribution of the formerly independent formations within the unified paradigm of Proto-East-Baltic makes it possible to obtain information on the lost semantic difference between them. Possible traces of the -si̯ e/o-formation in the only recorded West Baltic language, Old Prussian, might confirm the conclusions reached on the basis of the East Baltic evidence.

The East Baltic Future Tense Formation
The East Baltic future tense formation is derived from the infinitive of the verb (whose ending is Lith -ti, Latv -t) by adding a suffixal -s-to its stem.
pirk-ti pirk-s pìrk-t pìrk-s 'to buy' kél-ti kél-s cel-t cel-s 'to lift' saký-ti saký-s sacî-t sacî-s 'to say' Despite more than 150 years of research into the grammatical systems of East Baltic languages, the latter sentence still seems to contain all the securely established facts about the morphology of their future tense. Nearly everything else in the morphology of this formation is, especially from the diachronic point of view, a matter of controversy between different scholars and traditions.
This unsatisfactory situation seems to have arisen due to the following facts. First, the East Baltic dialects, especially on the Lithuanian side of the border, often differ considerably from each other in their future tense inflection. Second, the inflectional paradigm of the East Baltic future tense mostly comprises two or more different inflectional stems in different combinations. In the following paragraphs I will introduce the most basic facts about the inflection of the future tense in dialects of Lithuanian and Latvian. I will also try to draw some basic conclusions which, as I hope, will help to establish how the East Baltic future tense was probably inflected in the common prehistory of these languages, i.e. in Proto-East-Baltic and Proto-Baltic times. After this is accomplished, one can try to compare the Proto-Baltic situation with what is known about related formations in the other branches of Indo-European.
perka-me pirko-me peŗka-m pìrkã-m kẽlia-me k̇e lė-me ceļa-m cêlã-m sãko-me sãkė-me sakã-m sacĩjã-m The first fact one immediately notices when dealing with the Lithuanian future tense is that its inflectional paradigm comprises at least two different stems.4 The most frequent form, the 3fut., is an athematic inflectional form without a vocalic ending in any dialect. Cf.
(7) Lith buti → 3fut. bus or bùs 'to be' dúoti → 3fut. dúos or duõs 'to give' eĩti → 3fut. eĩs 'to walk' In verbs with a monosyllabic stem, the shortening of acute monophthongs (cf. buti → bùs, gýti → gìs 'to recover' , ḋe ti → dès 'to put' , jóti → jàs 'to ride') and metatony in acute diphthongs (cf. dúoti → duõs, líeti → liẽs 'to rain') in the eastern High Lithuanian dialects conclusively show that the 3fut. never contained a vocalic ending, being monosyllabic and hence athematic at least since Proto-Baltic times. The corresponding 1sg.fut. is nearly everywhere in Lithuanian busiu, dúosiu, eĩsiu, which means it is formed like a present stem of the i̯ e/o-or the i-class. The most prominent deviation from this symbiosis of two obviously different inflectional stems is the pattern 1sg.fut. busu, dúosu, eĩsu~3fut. bus, dúos, eĩs in many Low Lithuanian dialects (roughly between Šilutė, Šilalė, Kelmė in the South and Plungė, Telšiai, Akmenė in the north). However, this particular pattern is most probably a recent innovation of these dialects, having arisen on the model of the most prominent pattern of present stem inflection, which emerged through a comparatively recent apocope of short vowels at the end of a word. Cf.

(9)
High Lith Low Lith dìrbti 1sg.prs. dìrbu 3prs. dìrba 3prs. dìrb 'to work' mirti 1sg.prs. mìrštu 3prs. mìršta 3prs. mìršt 'to die' gélbėti 1sg.prs. gélbu 3prs. gélba 3prs. gélb 'to help' The major point of controversy concerning the inflection of the Lithuanian future tense is the original shape of the dual and plural inflectional forms. The problem is that all inflectional patterns which are attested in Lithuanian dialects can be explained as secondary. Clearly secondary are inflections of the type 1pl.fut. busam, dúosam, eĩsam, which are found beside the innovative Indo-European Linguistics 2 (2014) 42-72 1sg.fut. busu, dúosu, eĩsu only. The plural forms of this type must have emerged in the same way, following the same prominent class of present stems.
Presumably recent creations are also inflections like 1pl.fut. busiam, dúosiam, eĩsiam, which are widespread in the northern part of Low Lithuanian but hardly occur outside of this area. The pattern 1sg.fut. busiu, dúosiu, eĩsiũ 1pl.fut. busiam, dúosiam, eĩsiam may be basically following the inflection of i̯ e/o-presents. Note that the assumed analogy could only work after the stress retraction, which is a characteristic feature of the dialects in question. Cf.6 (10) High To be plausible, this analogy would have to be assumed already for the time before the loss of word-final short vowels in Low Lithuanian and probably also before the shortening and metatony in High Lithuanian monosyllables, but this is chronologically not impossible. It is also conceivable that inflections like 1pl.fut. busim(e), dúosim(e) have been only recently created as a match for the 1sg.fut. busiu, dúosiu. The model for this development could be provided by the prominent inflectional class of i-presents; the required analogy would be equally trivial.
A principled decision as to which scenario is more plausible and hence to be preferred seems impossible. What is possible, however, is a decision on a slightly different point. The assumption that 1pl.fut. busme, dúosme are more original than busim(e), dúosim(e) seems to be more economical than vice versa. This conclusion is based on the following well-known fact.
In all classes of Lithuanian present stems, the corresponding participle is always derived from the same stem as the inflections which constitute the paradigm. The participle of the Lithuanian future tense always has the shape busiąs, dúosiąs, i.e. it is formed like a i̯ e/o-present.7 The -i̯ e/o-stem underlying Indo-European Linguistics 2 (2014) 42-72 the participle can be identified with the stem of the 1sg.fut. busiu, dúosiu. If one now assumes that 1pl.fut. busme, dúosme are more original than busim(e), dúosim(e), one arrives at a Proto-Lithuanian paradigm consisting of two different stems, a -i̯ e/o-stem (1sg.fut. busiu and prcl.fut. busiąs) and an athematic stem (3.fut. bùs and 1pl.fut. busme). By contrast, if one assumes that 1pl.fut. busim(e), dúosim(e) are more original than busme, dúosme, one is operating with three different stems somehow secondarily united in a single paradigm. It is clear that the former assumption is more economical than the latter.

The Future Tense in Latvian
Let us now turn to Latvian, the only other Baltic language with a future tense formation of the given type. At first glance, the future tense paradigm attested in the most conservative varieties of Latvian seems to support the antiquity of the Lithuanian pattern with i-stem inflections in the dual and plural (cf. Endzelin 1923: 656-665).
The Latvian pattern is basically the same as in the southern dialects of Lithuanian, which exhibit However, the Latvian future tense participle is also formed in a way otherwise only possible for i̯ e/o-presents (cf. Endzelin 1923: 723-725 Exactly as in Lithuanian, the deviating morphology of the future tense participle can only be explained by assuming that at least one inflectional form in the future tense paradigm originally inflected according to the pattern of the i̯ e/o-presents. Since Lith 3fut. forms such as bùs, eĩs and duõs cannot be explained as recent creations either, the most economical hypothesis about the common prehistory of Lithuanian and Latvian would be, again, a historical analysis operating only with a i̯ e/o-formation (1sg.fut. Lith busiu, Latv bûšu and prcl.fut. Lith busiąs, Latv bûšus) and an athematic formation (3fut. Lith bùs, Latv bûs).
To be sure, this conclusion is only true if Latvian i-stem inflections such as 1pl. bûsim, iêsim, duôsim can be plausibly explained as secondary creations arisen independently of their Lithuanian counterparts busim(e), eĩsim(e), dúosim(e). However, such a scenario seems perfectly possible and plausible. In Latvian, the inflectional pattern 1sg.fut. bûšu, iêšu, duôšu~3fut. bûs, iês, duôs (with š in the 1sg. vs. s in the 3rd person) exhibits a morphophonological alternation of the type which must be assumed for the inflection of inherited i-presents and is not found anywhere else in the inflectional system. Cf. Latvian ziêdêt 'to flourish' and dzìrdêt 'to hear' in different dialects of Vidzeme: It is easy to imagine that this morphophonological alternation-originally present, for different reasons, both in the paradigm of i-presents and in the future tense-led to the creation of secondary i-inflections also in the future tense paradigm:

The Future Tense in Proto-East-Baltic
Thus, the evidence of Lithuanian and Latvian dialects is most economically explained by the following reconstruction of the future tense inflection in the common prehistory of these languages, i.e. in Proto-East-Baltic:9 (21) Proto-East-Baltic sg.
*bˈus, *dˈọ s prcl. fut. *bˈusi̯ añt-, *dˈọ si̯ añt-This reconstruction suggests that in the inflectional paradigm of the East Baltic future tense two different verbal stems with suffixal s have secondarily merged. One stem ended in Proto-Baltic *-s-. It is reflected in the dual and plural forms, in the 2sg. and in the third person (the grey cells in the paradigm given in 21 above). The other stem ended in Proto-Baltic *-si̯ a-(the black cells). This stem contributed the 1sg. and the participle.

6
The Traditional Approach The proposed reconstruction of the Proto-East-Baltic future tense paradigm is very similar to and, at the same time, very different from the more traditional approach, on which cf. most recently Jasanoff (1978Jasanoff ( : 104-107, 2003 and Villanueva Svensson (2010: 218-222, 2012. Like the present author, these scholars also try to account for the East Baltic future tense inflection on the basis of only 9 Whether diphthongisation of Proto-East-Balt *ọ , *ẹ into uo, ie (so in High Lithuanian and Central Latvian) preceded or followed the disintegration of Proto-East-Baltic is immaterial for the present discussion. The 2sg. probably ended in Proto-East-Balt *-sẹ , cf. the 2sg. of athematic presents such as Lith eisì, desì, dúosi from eĩti, 3prs. eĩti 'to walk' , ḋe ti, 3prs. dẽsti 'to put' and dúoti, 3prs. dúosti 'to give' . The double *ss which should be expected at the boundary between the stem and this ending might have been simplified to a single *s very early. Cf. Lith 2sg.prs. esì from buti, 3prs. ẽsti 'to be' where the simplification seems be of pie date already because it has to be assumed also for Skt ási, OAv ahī, Gk εἶ and Old Church Slavonic jesi.
Indo-European Linguistics 2 (2014) 42-72 two different stem formations which cannot be explained as arisen secondarily within East Baltic itself and must therefore be inherited from more ancient times. These formations are the athematic sigmatic formation reflected in the 3fut. such as Lith bùs, duõs, Latv bûs, duôs and the i̯ e/o-formation of the participle such as Lith busiąs, dúosiąs, Latv bûšus, duôšus. The fundamental difference between the reconstruction suggested in section 5 and the more traditional approach lies in the traditional attempt to explain all finite forms of the East Baltic future tense on the basis of the athematic formation alone. To achieve this, Jasanoff and Villanueva Svensson propose a "Narten"-type paradigm with the 3pl. ending in pie *-s-n̥ ti > Proto-Balto-Slav *-s-inti (cf. similarly Schmalstieg 1958, Endzelīns 1971. It is assumed that this hypothetical Proto-Balto-Slav *-s-inti was secondarily reanalysed as *-si-nti, which led to a spread of *i as a new stem formative throughout the plural and dual. The new i-plural allegedly generated a new 1sg.fut. of the type Lith busiu, dúosiu, Latv bûšu, duôšu by adopting the model of the ordinary East Baltic i-presents.10 In my view, this traditional theory of the East Baltic future tense inflection is, first, unnecessarily complex and, second, unconvincing in terms of Baltic historical morphology. The theory is too complex because it requires too much morphological analogy to explain the attested inflectional forms. Note that this approach necessarily separates the athematic dual and plural inflections like Lith 1pl.fut. busme, dúosme from the corresponding athematic 3fut. bùs, duõs. While the latter is thought of as reflecting the original athematic future tense directly, the former are believed to be a secondary replacement for more ancient busim(e), dúosim(e) which themselves replaced athematic inflections of the type Lith 1pl.fut. busme, dúosme in Proto-Baltic times. Much less analogy is required if one is prepared to explain the morphology of the East Baltic 1sg.fut. such as Lith busiu, dúosiu, Latv bûšu, duôšu simply by reference to the Proto-Baltic i̯ e/o-formation preserved in the corresponding participle.
The traditional theory is unconvincing in terms of Baltic historical morphology because it is not in harmony with the facts about the morphological behaviour of Baltic verbs, which are already securely established in the field. 10 Note that the reconstruction of "Narten" inflection for the East Baltic future tense formation is only seemingly corroborated by the full grade of such forms as, for instance, 1pl.fut. eĩs-me or 2pl.fut. eĩs-te from eĩti 'to walk' in dialects of Lithuanian. East Baltic has two verbs which can be safely assumed to have possessed a 3pl. pie *-n̥ ti > Proto-Balto-Slav *-inti. These are the reduplicated athematic presents Lith 1sg.prs. dúomi, 3prs. dúosti 'to give' and 1sg.prs. demì, 3prs. dẽsti 'to put' . Their 3pl.prs. ended in pie *-n̥ ti, cf. Skt dádati 'give' , dádhati 'put' and Old Church Slavonic dadętъ 'give' . In East Baltic, these presents clearly generalised the plural stem allomorph in the whole paradigm. However, they do not exhibit the assumed reanalysis of the inherited 3pl. ending and subsequent spread of *i, not even in their plural (cf. 1pl.prs. dúome, demè and 2pl.prs. dúoste in Lithuanian texts from the 16th c.)11 The only Baltic verb Jasanoff (1978: 105-107) can use as evidence supporting the plausibility of the assumed spread of *i from Proto-Balto-Slav 3pl.prs. *-inti (< pie *-n̥ ti) is Old Prussian waist 'to know' , the Baltic counterpart of the old perfect ocs věděti (1sg.prs. vědě or, secondarily, věmь). In Old Prussian, the athematic 2sg.prs. waissei (cf. ocs věsi) is accompanied by the i-forms 1pl.prs. waidimai (instead of something like ocs věmъ) and 2pl.prs. waiditi (instead of something like ocs věste). Jasanoff assumes that the i of these Old Prussian plural forms spread from the 3pl.prs. in Proto-Balto-Slav *-inti, cf. ocs vědętъ. However, the assumed spread of i is not the only possible and not the most plausible explanation for OPr 1pl.prs. waidimai etc. Due to its stative meaning 'to know' , the verb seems to have acquired a second stem in Proto-Balto-Slav *-ē-, cf. ocs 3sg.aor. vědě (secondarily also in the infinitive vědě-ti). If the unattested preterite of OPr waist was originally built with *ē too, the stative semantics of the verb would have nearly predicted an i-present, cf. the well-known pattern ocs 3sg.aor. mьně~3sg.prs. mьni-tъ, Lith inf. miṅe -ti~3prs. mìni 'to think' etc. Thus, the peculiar inflection of OPr waist can be explained differently, without any conflict with facts known about the behaviour of verbs with a 3pl.prs. in Proto-Balto-Slav *-inti in Baltic.
Indo-European Linguistics 2 (2014) 42-72 more economical and more in line with facts already securely established about the Baltic conjugation.

The Proto-Indo-European Roots of the East Baltic Future Tense
As already mentioned above, both reconstructed future tense formations of East Baltic possess parallels outside of the Baltic branch of Indo-European. The athematic future ending in Proto-Baltic *-s-(for instance in Proto-East-Baltic 3fut. *bˈus, *dˈọ s) is perfectly matched by the -s-future attested in the Sabellian subbranch of Italic.12 Cf.
(22) Umbrian 3sg.imp.II ee-tu → 3sg.fut. ee-s-t 'to walk' 3sg.imp.II fer-tu → 3sg.fut. fere-s-t 'to carry' 3sg.imp.II -peha-tu → 3sg.fut. peha-s-t 'to atone' A perfect match of the thematic future constructed with Proto-Baltic *-si̯ a-(such as in Proto-East-Baltic 1sg.fut. *bˈusi̯ ọ , *dˈọ si̯ ọ ) is found in the -si̯ a-future of Indo-Iranian.13 Cf. Consequently, we have to assume that the formations reflected in the East Baltic future tense paradigm probably existed already in the common prehistory of East Baltic and Sabellian (in the case of the athematic formation) or East Baltic and Indo-Iranian (as far as the thematic formation is concerned). This means that both formations must be inherited from pie, the common ancestor of these branches. The different morphology of the formations indicates that originally they must have differed also in their functions. However, this functional difference is not directly observable in the relevant daughterlanguages. All relatives of East Baltic that show reflexes of the old future tense formations have preserved only one of these, whereas the other has disappeared without a trace. Synchronically both formations, the athematic Sabellian and the thematic Indo-Iranian, seem to be the unmarked future tense formation of these languages.
The only possibility to establish the lost functional difference is to make use of the internal evidence of East Baltic, where the distribution of both ancient futures within the unified paradigm is strikingly asymmetric. The thematic formation with Proto-Balt *-si̯ a-only contributes the 1sg., whereas the athematic future with Proto-Balt *-s-provides all the remaining inflectional forms. It is obvious that the reason for this anomaly has to be sought in some functional difference between originally independent formations. Thus, our aim is to find a functional pattern that motivates a split between the 1sg. and the rest of the paradigm in the inflection of verbs with suitable semantics.

The East Baltic Future Tense Paradigm from a Typological Perspective
The first thing one learns from typological surveys of future tenses is that they are often secondary, comparatively young formations. Many languages display future tenses which have evolved from constructions with non-future semantics. It is possible to draw up a typology of secondary future tenses.14 The best evidence for crosslinguistically common paths of development from non-future to future is provided by languages which make use of grammaticalised auxiliary verbs whose original semantics are often directly attested in texts from earlier stages of the relevant languages. The following types of sec-14 Cf. Ultan (1978), Fleischman (1982), Bybee et al. (1987Bybee et al. ( , 1989Bybee et al. ( : 90-94, 1991, Heine & Kuteva (2002); a further type is described by Botne (1998 (24) a secondary future tense with auxiliary 'to become' and the infinitive of the main verb, i.e. 'he becomes cooking' → 'he will cook' (for instance in Russian and Polish, cf. most recently Wiemer 2011: 745-746 with references); b secondary future tense with verbs of movement such as 'come' or 'go' and the infinitive of the main verb, i.e. 'he is going to cook' → 'he will cook' (for instance in Basque, cf. Ortiz de Urbina 2003: 293-294); c secondary near future with auxiliary 'to say' and the main verb in serial construction in several Bantu languages, i.e. 'he says he cooks' → 'he will cook immediately' (cf. Botne 1998); d secondary ingressive future with auxiliary 'to take' and the infinitive of the main verb in Ukrainian, i.e. 'he takes cooking' → 'he will start cooking' (cf. Danylenko 2010Danylenko , 2012; e secondary future tense with the infinitive of the main verb and the modals 'want' , i.e. 'he wants to cook' → 'he will cook' (for instance in Serbian and Croatian, cf. Wiemer 2011: 744-745) or 'must' , i.e. 'he must cook' → 'he will cook' (for instance in English, cf. most recently Arnovick 1990, Wischer 2009).
Due to their sigmatic formatives, the future tense formations of East Baltic are obviously morphologically related to such modals as the Indo-Iranian desiderative or Old Irish subjunctive (cf. 1 and 2 above). This seems to exclude the paths of semantic development described in (24) a to d as potential etymological sources. At the same time, relatedness to desideratives and subjunctives speaks in favour of (24) e, i.e. for evolution from some ancient modal formations with semantics of desire or obligation. One should therefore look for modal formations which, first, are capable of developing into an unmarked future tense and, second, allow for a paradigmatic split of the given type. The latter condition requires that the semantics of the modal provide a natural dividing line between the 1sg. and the rest of the paradigm. This requirement would be satisfied, for instance, by an imperative mood where the 1sg. slot often remains unfilled because one rarely directs a request towards oneself. However, an imperative mood is hardly capable of developing into a future tense.16 It follows that the athematic component of the Proto-East-Baltic future tense paradigm can hardly reflect a more ancient imperative. 16 As it seems, such a development has not been described in a language with a documented history. The only potential exception known to me is the employment of the imperative mood as a future tense in Itelmen (cf. Volodin 1976: 250-251 for a detailed account). However, this is not the only special feature of the Itelmen imperative. When negated, the inflectional forms of the imperative function exclusively as a future tense, and this is at the same time the only possibility to form a negated future in Itelmen (Volodin 1976: 276-277). Prohibitions, on the other hand, are expressed in an entirely different manner. This seems to suggest that the Itelmen imperative should be considered an old future Indo-European Linguistics 2 (2014) 42-72 Fortunately, the imperative mood is not the only modal with a special treatment of the 1sg. but a member of a whole class of related formations. In functional terms, modals with semantics of desire or obligation can be divided into two groups.17 The first group is constituted by modals with so-called "subjectoriented" modality. This kind of modality defines the relation between the action expressed in the verb and its subject. To this group belong such modals as, for instance, the desiderative (i.e. 'he wants to cook'). The second group is constituted by modals with so-called "speaker-oriented" modality, which refers to the relation between the action and the speaker of the sentence. This latter group includes, beside the imperative, such modals as, for instance, the hortative (i.e. 'let us cook') or jussive (i.e. 'I want him to cook').
In a speaker-oriented modal, the 1sg. necessarily has a special status because here the speaker and the subject are identical. From this it follows that the 1sg. of a speaker-oriented formation expressing desire has roughly the same semantics as the 1sg. of a corresponding subject-oriented modal (i.e. 'I want me to cook' vs. 'I want to cook').
If one now assumes that the pie athematic -s-formation originally had subject-oriented desideratival semantics whereas the -si̯ e/o-formation was a speaker-oriented jussive, both the fact of their conflation in Proto-East-Baltic and the particular configuration of the new paradigm become understandable. Due to its more expressive semantics ('I want me to cook (at last)!'), the 1sg. of the inherited jussive may have secondarily replaced the less expressive but semantically very close 1sg. of the desiderative ('I want to cook') in its paradigm. At the same time, the different semantics of the remaining paradigmatic forms in said formations (cf. in the 3rd person 'I want him to cook' vs. 'he wants to cook') would have precluded mixing anywhere else in the inflection. The last step in the evolution of the two pie formations in the late prehistory of Proto-East-Baltic would be the typologically unremarkable development of the morphologically modified desiderative into a future tense and the loss of the jussive. tense which has developed a secondary imperative meaning in non-negated clauses. After the rise of a more recent future tense (with a desiderative marker, cf. Volodin 1976: 217-219), the temporal usage of the inherited future tense formation would have become limited to negative contexts. The use of a future tense as imperative, typologically far less remarkable, is also attested in the neighbouring Chukchee (cf. Nedjalkov 1992: 51-52). Thus, the case of Itelmen probably presents an example of the reverse development of a future tense into imperative mood. 17 Cf. Bybee et al. (1994: 177-181) and especially Bybee & Fleischman (1995). hill Indo-European Linguistics 2 (2014) 42-72

The 3rd Person Imperative in Old Prussian
The proposed reconstruction of the prehistory of the Proto-East-Baltic future tense paradigm is potentially directly confirmed by the evidence of the West Baltic language Old Prussian. Old Prussian texts might preserve a formation which may be directly equated with the -si̯ e/o-component of the East Baltic future tense. The semantics of this Old Prussian formation, attested in the 3rd person only, seem to confirm the hypothetically assumed speaker-oriented jussive semantics of the -si̯ e/o-formation in Proto-Baltic times.
The relevant Old Prussian formation may be called "3rd person imperative".18 Its single inflectional form is based on the infinitive, like the East Baltic future tense, and ends in OPr -sei which may be also spelled ⟨-se⟩, ⟨-si⟩ or ⟨-sai⟩. Cf.
The semantics of the Old Prussian 3rd person imperative may be demonstrated by the following occurrences in the Old Prussian translations of Luther's Catechism: (28) Old Prussian German
Unfortunately, the proposed analysis of the Old Prussian 3rd person imperative is not the only possible explanation of this formation. Imperatives like OPr boū-sei, dā-sei etc. are also explainable as descending from the other sigmatic formation of the East Baltic future tense, the pie athematic -s-formation.21 In Italic, the athematic -s-formation was capable of forming its own moods. Cf. the formation of the future-based imperfect subjunctive in Sabellian The origin of OPr -eī-remains unclear. However, the overall similarity of the whole formation to the pie athematic optative (note the zero grade of the root) seems to suggest that this marker somehow reflects pie *-i̯ éh1-~*-ih1-in such optatives as, for instance, Since the second constituent of the East Baltic future tense paradigm, the -s-formation, was athematic, its optative should have been derived in the same way. This means that Old Prussian 3rd person imperative forms such as boūsei, dāsei etc. are also explainable as descending from the Proto-Baltic optative of the athematic -s-formation reflected in 3fut. Lith bùs, duõs, Latv bûs, duôs etc. It cannot be decided with certainty which of the proposed explanations, the 3rd person indicative of the -si̯ e/o-formation or the 3rd person optative of the athematic -s-formation, reflects the real prehistory of the Old Prussian 3rd person imperative in -sei. This formation therefore remains only potentially relevant to the main issue of the present paper, the paradigmatic structure and original semantics of the East Baltic future tense.

10
The Participle Two points remain to be addressed. First, the participle of the East Baltic future tense has the same morphology as the 1sg. The evolution of a jussive participle into a future tense participle becomes plausible as soon as one takes into consideration the pragmatic dimension of its use. Clauses with finite forms of a jussive formation, such as I want you to marry me or I want him to guard the house, have the function of direct or indirect requests or demands. These requests or demands can hardly be reinterpreted as anything else but orders or, perhaps, permissions. By contrast, noun phrases with jussive participles as their dependent constituent do not contain a demand, but merely inform the hearer about the desires and expectations of the speaker. For this reason, such noun phrases can be easily understood and then also reinterpreted by the hearer as statements about the expected future of their heads. For instance, noun phrases which should be translated as girl I want to marry me and dog I want to guard the house can be easily understood as girl who (probably) will marry me and dog which (probably) will guard the house.
The fact that the -si̯ e/o-jussive participle was generalised in Proto-East-Baltic, ousting the participle of the desiderative -s-formation, is not surprising. Due to its semantics (desires and expectations of the speaker, who is of course always human), the jussive participle was more universally applicable to nouns of different semantics and thus probably also more frequently used than the desiderative participle. Cf. the noun phrases given in (37), where the jussive participle is equally adequate with a human head noun, a non-human animate and a non-animate head noun.
(37) 'girl I want to marry me' → 'who will marry me' 'dog I want to guard the house' → 'which will guard the house' 'crops I want to grow fast' → 'which will grow fast' By contrast, the desiderative participle was undoubtedly applicable to humans, but only sometimes to non-human animates and hardly ever to non-animate nouns, cf. (38) (38) 'girl who wants to marry me' → 'who will marry me' ? 'dog which wants to guard the house' *'crops which want to grow fast' The Indo-European -si̯ e/o-formations outside of Baltic are usually interpreted as instances of an unmarked future tense. It is, however, a well known fact hill Indo-European Linguistics 2 (2014) 42-72 that most occurrences of this formation in the most ancient Indo-Iranian texts (such as the R̥ g-Veda) are participles.23 For these participles, the same semantic evolution can be assumed as was just proposed for the prehistory of East Baltic. The rare finite forms might be explained as recent back-formations to the participles. This is equally valid for Gk κείω, which is the future tense of κεῖμαι 'to lie' and has been plausibly identified by Hollifield (1981: 173-188) as the single remnant of the pie full grade -si̯ e/o-formation in Greek. The participle κείοντ-is attested already in the Iliad, whereas the finite forms of the verb do not occur until the Odyssey and therefore may be recent.
The single remnant of the -si̯ e/o-formation in Slavonic is the Church Slavonic participle byšǫšt-'which will be in future' .24 Here, an evolution into an unmarked future tense need not be assumed.
In Celtic, the clearest attestation of the pie -si̯ e/o-formation is 1sg. pissíu-mí in the Gaulish inscription from Chamalières.25 This verb form belongs to the inflection of Proto-Celt *kwis-e/o-'to see' (cf. Gaul 3sg.imp. ap-pisetu from Thiaucourt and Old Irish ad-cí 'to see').26 The phrase exops pissíu-mí 'being weak-eyed …' is the last in a sequence of wishes, being preceded by two phrases with imperatives. The verb form might, therefore, be equally well interpreted as future tense 'I will see (again)' or as jussive 'I want me to see (again)' .

Conclusions
An in-depth investigation of the East Baltic future tense leads to the following conclusions: (a) the inflectional paradigm of the Proto-East-Baltic future tense formation was composed of two originally independent sigmatic formations of pie ancestry, an athematic -s-formation and a thematic -si̯ e/o-formation; (b) the asymmetric structure of the inflectional paradigm of the Proto-East-Baltic future tense can be accounted for if one ascribes subject-oriented desideratival modality to the former sigmatic formation and speakeroriented jussival modality to the latter; 23 Cf. for Sanskrit Macdonell (1910: 386-387), for Avestan Kellens (1984: 160-162). 24 Cf. most recently Birnbaum (1995) and Hill (2004: 104-106) with references. 25 Cf. Schumacher (2004: 58-59). 26 Cf. the references given in Hill (2004: 107-109).
Indo-European Linguistics 2 (2014) 42-72 (c) this reconstruction of the original semantics of both formations is compatible with the relevant data from outside of Baltic and helps to explain the preference for participles which is observed for the Indo-Iranian descendant of the -si̯ e/o-formation.