2. The state of the art

As is well known, since the 1970s scholars have been debating about the emergence of tense-aspect morphology in language acquisition. Most previous studies in the domain of language acquisition have given prominence to the interaction between inherent verb meaning and tense-aspect markers in child language (see Li & Shirai Reference Li and Shirai2000: 5). To give just a brief outline of the psycholinguistic literature on this topic, it is worth mentioning early studies such as those by Bronckart & Sinclair (Reference Bronckart and Sinclair1973) or Antinucci & Miller (Reference Antinucci and Miller1976), whose claims are at the basis of the so-called Defective Tense Hypothesis (Weist, Wysocka, Witkowska-Stadnik, Buczowska & Konieczna Reference Weist, Wysocka, Ewa Buczowska and Konieczna1984; see also Li & Shirai Reference Li and Shirai2000: 40). According to these authors, children at an early stage in their growth lack the category of tense, for they have not yet developed the notion of temporal deixis (cf. Shirai & Andersen Reference Shirai and Andersen1995: 745 for some discussion). When they do begin to use past tense inflection, it is only with telic verbs. These verbs express an observable action or event viewed from the perspective of its end point, with the result that tense distinctions turn out to be redundant and mostly dependent on lexical aspect distinctions. A more nuanced and less strong version of this theory is represented by the so-called Aspect Hypothesis, which was first formulated by Shirai (Reference Shirai1991), Shirai & Andersen (Reference Shirai and Andersen1995: 746) and then re-proposed by Li & Shirai (Reference Li and Shirai2000). These authors base their argument on a prototype account of tense-aspect acquisition and point out that children acquire tense distinctions gradually, usually starting with an initial association between past tense (and perfective) morphology on the one hand and those prototypical verbs characterized by [+telic], [+punctual] and [+result] features on the other. Then, at a later stage, children also begin extending the past tense (and perfective) morphology to non-prototypical verbs, i.e. those verbs lacking one of these features. More precisely, the (perfective) past marking is first applied to achievement and accomplishment verbs, for, as [+telic], they represent the prototype of the temporal category of past; only later the (perfective) past marking can be extended to activity and state verbs, which, as [−telic], are less prototypical. Therefore, unlike the previous and absolute Defective Tense Hypothesis, the prototype model advances the view that not all past tense inflections are associated with [+telic] verbs, but rather that there is a gradual extension from an early stage, which shows a predominance of past morphology attached to achievement and accomplishment verbs, toward a later stage, which exhibits a development to non-prototype past for activity and state verbs as well.

The point to be emphasized here is that in these theories a prominent role is assigned to lexical aspect, i.e. to the inherent semantic meaning of the verb. In fact, there is a strong interaction both between lexical and morphological aspect, and between these two and tense inflection: on the one hand, perfective morphology, i.e. grammatical aspect, turns out to be prototypically associated with telic verbs, i.e. with lexical aspect, and on the other, the prototype of the category past is similar to that of the category perfective,Footnote ³ so that both tense and aspect morphology prove to be very strictly interrelated with lexical aspect. And because the acquisition of tense-aspect morphology proves to be dependent on the Aktionsart (i.e. lexical aspect) features of the verb, it is not surprising that this interrelatedness, which fits well with a lexicalist approach, is in line with recent theories on language evolution, according to which ‘the origin of all grammatical morphemes (function words, inflections) is in lexical stems’ (Hurford Reference Hurford, Christiansen and Kirby2003: 53; see also Heine & Kuteva Reference Heine, Kuteva and Wray2002: 394).Footnote ⁴ It is noteworthy that these results are also strikingly in line with the neurological data discussed by Ziegeler (Reference Ziegeler1997: 235), who, citing Givón (Reference Givón1995), underlines the independent but parallel processes in diachronic and ontogenetic grammaticalization. She emphasizes how lexical sources, realized as complete lexemes or identifiable only as prototypical lexical properties, have been found to be at the origin of the development of grammatical morphology across time. The neurological order of development would demonstrate that grammar, which develops only after a lexicon has emerged and stabilized, is highly affected by lexical semantics.Footnote ⁵

I will now argue that the inflectional marking of tense in the Proto-Indo-European verbal system turns out to be dependent on the lexical features of the verbal stems. As an anonymous reviewer has pointed out, the interrelation between tense inflection and inherent features of the verb would in fact support Bybee et al. (Reference Bybee, Perkins and Pagliuca1994) hypothesis, according to which perfective aspect emerges with telic verbs, imperfective aspect with atelic verbs. This hypothesis underlines once again the fundamental role which the inherent lexical features of verbs play in determining the emergence of inflectional categories like grammatical aspect. In this paper I do not take grammatical aspect into consideration, but the data I have examined fully confirms the primary role of the verbal root's lexical features in emergent inflectional paradigms. In particular, the PIE grammatical past tense is a more recent acquisition in the evolutionary path of emergent paradigms, as also demonstrated by Bybee et al. (Reference Bybee, Perkins and Pagliuca1994: 92) and confirmed by data presented in this study. These latter will show that the PIE past tense marker is a recent gram that gradually came to be applied to all verbs regardless of their lexical aspect, mostly after the PIE typological shift to a nominative–accusative stage where tense distinctions became more important than the earlier aspectual ones (see section 7). Furthermore, the precedence of lexical aspect in respect of tense distinctions is also in line with Bybee's semantic relevance hierarchy, according to which, being the most relevant category to the verb, aspect is likely to be lexicalized rather than grammaticalized within a paradigm (Bybee Reference Bybee1985: 21–22). Tense was firstly expressed within the inflectional paradigm of the protolanguage by means of different sets of (personal) endings. As will be specified in the following sections, the early PIE verbal system was characterized by a binary distinction between root–stem formations with [+telic] and root–stem formations with [−telic] feature, added with personal endings only. However, an originally temporal and locative adverbial particle gradually coalesced with verbs as a suffix added to the personal endings in order to mark the present tense. The important fact to be noticed is that this suffix was initially added only to those root–stem formations with an inherent [−telic] feature, from which a present tense stem could subsequently be derived. On the contrary, from root–stems with a [+telic] feature, present tense forms were impossible and in fact are unattested (see Strunk Reference Strunk1994: 420). Therefore, [+telic] root formations without this suffix began automatically to indicate past tense.

Furthermore, it is worth indicating what kind of theoretical contribution Indo-European studies could provide to the general linguistic debate over the classification of verbal aspectual classes. In fact, verbs are traditionally classified as telic or atelic according to the action or event they express, respectively with or without an inherent endpoint (see, among others, Bybee Reference Bybee1985: 100). After the early well-known Vendlerian four-class distinction, however, it has been pointed out that aspectual class is a compositional property (Verkuyl Reference Verkuyl1972). Many scholars have argued that aspect is not only a semantic but mostly a syntactic property, which ‘pertains to VPs and whole sentences, and not to verbs as such’ (Slabakova Reference Slabakova2001: 23). Following Verkuyl's hypothesis, verbs are even analyzed as complex structures built up upon a completely neutral root with no information about its argument structure (see Slabakova Reference Slabakova2001: 40). The aspectual class of a verb is often difficult to classify since it depends on the verb's combination with arguments, adverbials, etc., which can coerce one and the same verb into denoting different situation types. However, this does not prevent us from considering a basic prototypical lexical classification as fundamental.Footnote ⁶ In other words, a verb is initially encoded in the lexicon of a language with a prototypical aspectual entry, even if its [±telic] trait can be modified depending on the context.

In this paper I will argue that it is possible to classify verbs according to their main prototypical [±telic] lexical aspect feature.Footnote ⁷ Determining this latter feature makes it possible to account in a more satisfactory way for the different semantic and syntactic combinations in which a verb occurs, as well as how and why its morphological paradigm is structured as it is. To be precise, it is the kind of forms and their distribution within the morphological paradigm of a verb which helps us to determine its lexical (a)telicity. Thus, as will be seen from the data to be presented in this paper, in cases where a verb is inflected in a paradigm as a root formation only in the aorist stem while its present stem is a derived formation (i.e. affixed, reduplicated, etc.), this can only be explained if the verbal root is considered to be prototypically [+telic]. Conversely, when a verb is inflected as a root formation only in the present stem while its aorist stem is a derived (and therefore a more recent) formation, it means that the verbal root is prototypically [−telic] (see also Strunk Reference Strunk1994: 419). Moreover, the analysis of the contexts in which the verb occurs will confirm the morphological data as well. This kind of correspondence is proved to be valid for the PIE language, as I will try to demonstrate both on the basis of morphological parameters (drawing on the verb's non-random paradigm structure) and on the basis of syntactic–semantic parameters (drawing on the sentence context). Following this approach, traditional Indo-European studies of the tense-aspect system have tended to reconstruct one unitary basic prototypical lexical (a)telicity feature for a given verb, regardless of the many different meanings which this verb may assume.Footnote ⁸ In fact, semantic change due to different contexts could have obscured the relationship between the original lexical aspect of the root and the structure of its morphological paradigm (see Clackson Reference Clackson2007: 135).

A historical perspective: ie *e- as a past tense marker in Vedic Sanskrit and Homeric Greek

In light of the above survey of some of the influential literature on the emergence of tense-aspect morphology from an ontogenetic perspective, one might wonder if a historical analysis can help us find traces of a similar evolution path. In other words, what can historical linguistics say about the emergence of a tense-aspect system in a proto-language? Tracing back to the most ancient articulated verbal systems that we can reconstruct in the history of language evolution, we can investigate how the Proto-Indo-European verbal system might have been organized at the earliest stages. Comparative Indo-European studies generally agree with the thesis according to which the first verbal system of the proto-language was originally characterized by a ‘non-inflected state’ (see, among others, Lehmann Reference Lehmann2002: 167). More precisely, many traditional Indo-Europeanists reject the old hypothesis which was formed in order to explain the rich inherited inflectional system proper to the attested historical languages (see Drinka Reference Drinka and Andersen1993: 143 for discussion) and according to which Proto-Indo-European would have been characterized by a complex and articulated morphology. On the contrary, attention to residual forms attested in earlier stages of Indo-European languages such as Homeric Greek and Vedic Sanskrit (known also as ‘Old Indic’), provides insights into the earliest possible strata of the proto-language, at a time before a more complex verbal morphology had been stabilized (see also Lehmann Reference Lehmann2002: 172). Therefore, drawing on the comparative method and the procedure of examining residual archaisms in both the Rigveda and the Homeric poems, this paper will focus on sketching the distribution of primitive verbal forms in the proto-language and then trying to account for this distribution. These earlier forms consisted of verbs with broad meaning, which were still inflectionally underspecified as regards verbal categories such as tense or mood. The only kind of categorial differentiation which could be ascribed to these verbs was lexical in nature, in the sense that there existed verb stems indicating actions or events with an inherent endpoint [+telic], and verb stems indicating actions or events without an inherent endpoint [−telic]. At this stage it was the pragmatic context that played the fundamental role in distinguishing the specific meaning of each verb, which could indifferently refer to present or past actions, to statements as well as to prohibitions, to gnomic or atemporal events, such as those referring to habitual (timeless) activities of gods (West Reference West1989: 135). Only in a later stage were inflectional elements gradually added to these verbal forms. Among such inflectional elements, which were originally deictic particles or adverbs, it is possible to reconstruct a past tense marker *e-, the so-called augment, which was common to both Vedic Sanskrit and Homeric Greek (as well as Armenian and Phrygian). This prefix *e-, which meant illic et tunc, i.e. ‘there and then’, came to characterize the distal tense as opposed to the proximal tense, which in turn was already expressed by the suffix *-i meaning hic et nunc, i.e. ‘here and now’. As will be more fully explained in the next section, this past tense marker *e- is a more recent trait, which came to be added to all verbs irrespective of their lexical feature, in order to mark the past tense in the late Indo-European language. The residual augmentless verbs thus testify to a previous PIE stage which is still preserved in the history of languages such as Vedic Sanskrit and Homeric Greek. These augmentless verbs, which co-exist with the augmented forms only in the earliest literature of both Sanskrit and Greek, have traditionally been considered as an optional usage proper to poetry (see section 5). They are instead essential in order to reconstruct the steps of the inflectional path that was followed by the evolving PIE verbal system.

The distribution of the unaugmented forms in Homeric Greek

On the basis of the comparison between the formations above, it is generally agreed that the IE prefix augment *e-, inherited both by Gr. ἐ- e- and Skr. a-, is a sign of more recent verbal forms with past tense distinctions. There were no grammatical tense distinctions at the earliest stages of the proto-language (see, among others, Stempel Reference Stempel1999: 42, Lehmann Reference Lehmann2002: 173).Footnote ¹⁴ One might think that the correspondence between augmentless forms in Vedic Sanskrit and Homeric Greek could be an innovation proper only to those two languages, and hence not ascribable to a more ancient Proto-Indo-European stage. However, even if the augment is traditionally considered a more recent innovating dialectal trait shared by Sanskrit and Greek, the augmentless forms of injunctive must have assumed a past meaning already in a pan-Indo-European (pre-Sanskrit and pre-Greek) stage. In fact, when the injunctive endings began to take the suffix *-i in order to indicate the present tense, all the injunctive forms without this suffix began automatically to indicate the past tense, i.e. the non-present tense. As pointed out by Blumenthal (Reference Blumenthal1974: 71), ‘the separate IE particle *e, which became the augment, came into use as a contrastive mark of pastness marking the past tenses’. However, initially, ‘the augment as a mark of pastness is strictly unnecessary, as can be seen from Hittite, where endings alone mark time’ (Blumenthal Reference Blumenthal1974: 71). As remarked in the previous sections, at the early PIE stage there existed verbal forms supplied with personal endings only, thereby forming an archaic, earliest system characterized by a binary distinction between root–stem formations with a [+telic] feature and root–stem formations with a [−telic] feature. Then the originally temporal and locative adverbial particle *-i gradually coalesced with verbs as a suffix added to the personal endings (which then became 1sg *-mi, 2sg *-si, 3sg *-ti, 3pl *-nti from original 1sg *-m, 2sg *-s, 3sg *-t, 3pl *-nt) in order to mark a present action or event taking place ‘here and now’. At first, the new endings marked with *-i were added only to those root–stem formations with an inherent [−telic] feature, such as the PIE root *es- ‘be’ (PIE 3sg *es-ti, Ved. ásti, Gr. ἔστι ésti, etc. ‘is’), from which it was possible to derive a present tense stem. On the contrary, from a root–stem with a [+telic] feature, such as the PIE root *g ^wā- ‘come’ (Ved. ágāt, Gr. ἔβη ébē, etc. ‘came’), present tense forms such as PIE 3sg *g ^wā-ti were impossible; they were never coined and are unattested (see Strunk Reference Strunk1994: 420).

By this time, all root–stem formations without the present tense marker *-i had come to indicate past tense. Therefore, the prefixed augment *e- turns out to be a more recent trait which marks the past tense redundantly, a trait shared only by a specific dialectal Indo-European area. This fact is not in conflict with the hypothesis here proposed, according to which the correspondence between the Vedic Sanskrit and Homeric Greek augmentless forms goes back to an earlier Proto-Indo-European stage: personal endings endowed with the suffix *-i are in fact common to most of the attested Indo-European languages, testifying to a system where present stem was marked in opposition to an unmarked past tense. As I will show in the next section, the data argues that the inflectional tense morphology of the verbal system of both Sanskrit and Greek was constructed originally on the basis of the lexical–semantic value of the bare verbal root. In fact, since initially there was no difference between the personal endings of the present and the aorist stems, it was the inherent lexical aspect of the underlying root which governed the assignment of a given verb to the emergent system of aorist or present (see also Clackson Reference Clackson2007: 133). Furthermore, the distribution of the residual augmentless verbs in Homeric Greek sheds light on the fate of primitive injunctive forms; in Classical Greek such forms disappear and the augment is obligatorily prefixed in the past tense, both imperfect and aorist. In fact, however, the distribution of the augment in Homeric Greek turns out to be not random, but is perfectly in line with the distribution of injunctive forms in the Rigveda. This correspondence thus permits us to reconstruct a primitive stage which can be ascribed to the proto-language.

In greater detail, it can be seen that in contexts where a present injunctive is expected in the Rigveda, Homeric Greek exhibits an unaugmented imperfect and never (or very rarely) an aorist, as can be seen in the examples below:

1. (2) tám bhara-t súbhtaṃ crv índuḥ Order bear-inj greatly.esteemed dear juice (Soma) ‘the Soma-juice holds the dear Order greatly esteemed’

   (Sanskrit – RV. IX 97,24d)

1. (3) τὴν δὲ πρóσω φέρ-ε κμα, θέμωσε δὲ χέρσoν ἱκέσθαιtèn dè prósō phér-e kúma, thémōse dè chérson hikésthaiit part ahead bear-impf wave drive then shore reach‘the wave bore it (the ship) ahead, then it drove it to reach the shore’

   (Old Greek – Od. IX, 542)

From a morphological point of view, the Vedic present injunctive 3sgbharat in (2) corresponds formally to the Homeric Greek augmentless imperfect indicative 3sg φέρε phére in (3), both forms deriving from the IE root *bher-.

In a similar way, in contexts like those where an aorist injunctive would be expected in the Rigveda, Homeric Greek shows an unaugmented aorist and never (or very rarely) an imperfect, as can be seen in (4) and (5):

1. (4) órvd gávām mat jānatgā-t enclosure cows mother knowing come-inj ‘the Cows' Mother comes from the cowshed knowing (the road)’

   (Sanskrit – RV. V 45,2b)

1. (5) ατὰρ ὁ βῆ Λυκίηνδε θεῶν πˈ ἀμύμoνι πoμπῇ autàr ho bê Lukíēnde theôn hup' amúmoni pompêthen he come-aor Lycia.to gods under glorious guidance ‘then he came to Lycia under the glorious guidance of the gods’

   (Old Greek – Il. VI, 171)

In (4) and (5), the augmentless aorist of the same verb come occurs as the Vedic injunctive gāt and the Homeric indicative βῆ bê, both forms being derived from the IE root *g ^wā-.

The important datum to be underlined here is that this correspondence in the distribution of verbal forms reflects exactly the lexical features of the Proto-Indo-European verbal roots. When the lexical aspect feature of the root is [−telic], then most of the injunctives of the corresponding verbs in the Rigveda are found in the ‘present injunctive’ (never or very rarely in the aorist injunctive) and the corresponding augmentless preterites in Homeric Greek are imperfect (never or very rarely aorist). On the contrary, when the lexical aspect feature of the root is [+telic], then most of the injunctives of the corresponding verbs in the Rigveda appear in the ‘aorist injunctive’ (never or very rarely in the present injunctive) and the corresponding augmentless preterites in Homeric Greek turn out to be in the aorist (never or very rarely in the present). It should also be noted that, unlike the Homeric language, in the later stage represented by Classical Greek all morphological distinctions, both of tense and aspect, are uniformly distributed throughout the verbal system, regardless of the lexical aspect features of the root.Footnote ¹⁵

As stated above, tense distinctions are not expected in the earliest stages; the tense specification present or aorist added by scholars to the injunctive are merely the widely accepted conventional terms used to designate those injunctive forms which would evolve into the present or the aorist system at a later stage of Sanskrit verbal morphology (by means of the adjunction of the suffix *-i and the prefix augment *e- respectively). It goes without saying that all the augmented imperfect or aorist forms found in the Iliad and the Odyssey represent a more recent stage with respect to the unaugmented imperfect and aorist forms derived from the same verb. Clearly, then, the correlation between telicity and grammatical marking underwent a progressive weakening. It is well known that unaugmented verbal forms have disappeared in Classical Greek. By that time the structure of verbal paradigms was already stabilized with a regular tense distinction and the past tense was always marked with the augment.

Analyzing the data

Having sketched an outline of the main features of the topic discussed in the paper, let us now turn our attention to the data. This section will examine in detail the Proto-Indo-European verbal conjugation, tracing it back to those attested early Vedic and Greek forms which consisted only of the so-called reconstructed ‘secondary personal endings’ *-m, *-s, *-t, *-nt Footnote ¹⁶ added to the lexical stems from which the present and aorist tenses would subsequently be derived through the addition of specific affixes. The aim will be to better understand the kind of relationship existing between the Vedic injunctive forms on the one hand and the corresponding Homeric Greek preterite forms without temporal augment on the other. These latter, which represent forms of an ancient Proto-Indo-European injunctive as well (see, among others, West Reference West1989: 135), have traditionally been explained as occurring randomly and optionally throughout the Homeric poems or in accordance with the so-called μετρικὴ ἀνάγκη metrikè anánkē, i.e. ‘metrical necessity’, which would have induced the poet to incorporate the prefixed augment according to metrical considerations proper to poetry.Footnote ¹⁷ As Hoffmann (Reference Hoffmann1967: 145) points out, the same trait of optionality has often been alleged for Vedic language as well. To the contrary, the analysis of our data will show instead that the distribution of such augmentless forms in Vedic Sanskrit and Homeric Greek indicates a precise rule linked to the lexical aspect of the original verbal roots.

The corpus examined here consists of all occurrences of Vedic injunctives which have a reliable correspondence with augmentless preterites in Homeric Greek, taking into account only those injunctives which have a non-modal (indicative) sense and ruling out those having a modal (subjunctive, optative or imperative) sense.Footnote ¹⁸ The texts considered for Vedic Sanskrit are the Rigveda and the Atharvaveda; the injunctive is most common in the former, less common in the latter and it definitively disappears from the later language, i.e. Classical Sanskrit. As for Greek, I have examined the Homeric poems Iliad and Odyssey, and Homeric Hymns.Footnote ¹⁹

Table 1 displays the data by comparing on the same line the number of occurrences in Vedic and Greek. The table is organized as follows. The first column on the left is for Proto-Indo-European roots reconstructed by comparing Vedic and Homeric Greek verbs, including their meaning and Aktionsart [±telic], i.e. the inherent lexical aspect of the root.Footnote ²⁰ The second column on the same line gives the corresponding inherited Vedic Sanskrit root, and the next column is for both present (pres.) and aorist (aor.) injunctives; this is followed, when appropriate, by relevant remarks or observations, and, finally, by the number of occurrences in the Vedas, i.e. Rigveda and Atharvaveda. In order to more easily visualize the data in the table, only a single injunctive form is given in each row. I have chosen, where possible, the most reliable for reconstruction, the less marked, the most ancient form (it goes without saying that, for instance, the root aorist is older than reduplicated or other extended forms of the aorist), which also shows the better correspondence with the equivalent Homeric Greek preterite.

Table 1

Distribution of occurrences of Vedic injunctives and Homeric Greek preterites.

^a As regards the two Greek verbs of the *-mi conjugation examined here, i.e. δίδωμι and τίθημι, it should be noted that in order to try to reconstruct an earlier stage it is important to consider the most ancient forms of the aorist, i.e. the athematic forms without -κ-; this affix, albeit very ancient, is in fact a more recent development.

As with the Vedic occurrences, I display on the right side of table 1 first the Greek verbs (regularly given as indicative present 1 singular) derived from the Proto-Indo-European root which is indicated in the same row, then the augmentless preterite forms both imperfect (impf.) and aorist (aor.), followed by any relevant observations and finally, the number of occurrences. In the last three columns I have added the corresponding augmented forms of the same Homeric verbs, followed at times by remarks and finally by the number of occurrences, in order to compare thoroughly the non-random quantitative distribution of the augment in the verbal paradigm.Footnote ²¹ Again, also for the Greek part, only a single form of each verb is given per row, in order to provide a concise overview of the data. These single verbal forms have been chosen following the same criteria as for the Vedic forms.

The remarks are intended to clarify some important issues concerning the analyzed forms. More precisely, often some residual features have been indicated which are well known to be characteristic of previous stages. These include (i) occurrences of the ancient category of dual, which tends to disappear in the later language;Footnote ²² (ii) the presence of root formations, which are more ancient than the derived, reduplicated or affixed stems; (iii) the existence of media tantum; (iv) formulaic expressions; and so on. Formations are also given which indicate a more advanced stage of development; representative examples are derived (e.g. causative, desiderative, denominative, etc.), reduplicated or affixed stems (e.g. -na- and -κ- stems for Vedic and Homeric Greek, respectively, and Greek prefixed prepositional stems, among others). Furthermore, it is sometimes specified if an inflected form is attested in those texts which are traditionally considered to be more recent, such as the Atharvaveda with respect to the Rigveda, or the Odyssey with respect to the Iliad, and the Homeric Hymns with respect to both Homeric poems.

Finally, the 31 PIE roots are ordered alternating the [+telic] and [−telic] and according to the most reliable correspondence between derived injunctives, taking into consideration parameters such as the period of the occurrences, the regularity of forms, and so on.