2.1 Verbalizers

In Distributed Morphology (DM), verbal stem-forming morphology is usually equated with the verbalizing projection v, which may be associated with different (Aktionsart-related) “flavors”, e.g., become, do, cause (Folli & Harley 2005; 2007; Harley 2005; 2009; Alexiadou & Lohndal 2017; Panagiotidis et al. 2017; etc.). Like other categorizers, v mediates between the root and higher functional projections. In the verbal domain, alternations relating to argument and event structure, such as the causative alternation, passivization, etc., are thought to result from the interaction of v with the higher, external argument-introducing projection Voice and other argument structure-related functional projections (Alexiadou & Anagnostopoulou 2004; Schäfer 2008; 2017; Alexiadou & Doron 2012; Alexiadou 2013; Harley 2013; 2017; Alexiadou et al. 2015; Kastner 2020; etc.). The standardly assumed “verbal spine” with its functional projections is given in (1).

Elements like the vocalic conjugational class markers in Spanish, (2), which categorize roots as (different inflectional classes of) verbs, look like good candidates for overt markers of the verbalizing projection v in (1): They “verbalize” roots without contributing argument structure-, Voice- or Tense-related information; and the choice of inflectional class marker (e.g., first, second, or third conjugation) is lexically determined by the root.

However, Oltra-Massuet (1999) and Oltra-Massuet & Arregi (2005) argue that forms like (3) show that the Spanish verbal spine actually has several positions that are spelled out as vocalic “themes”.

According to them, the vocalic themes in (3) do not spell out the functional heads v and T but adjoin to them postsyntactically/at MS (“a Morphological Structure level (…) between Spell-Out and Phonology”, Oltra-Massuet & Arregi 2005: 44) due to a morphological well-formedness condition, given in (4).

The evidence for this analysis comes from suffixes that are argued to be segmentable into a functional head plus a theme vowel, such as the ones in (5).

In the following, these will be referred to as “complex thematic suffixes”, as opposed to “simple thematic suffixes” that consist only of the theme vowel, as in (2). Fábregas (2017) actually takes a different approach than Oltra-Massuet & Arregi (2005) and argues that complex thematic suffixes as those in (5) are allomorphs of the simple theme vowels in (2) and, like them, spell out v. Fábregas provides three arguments in favor of this analysis: 1) Suffixes like those in (5) never occur without the suffix-final theme vowel (e.g., -ific- never occurs without -a-, etc.), 2) they never attach to light verbs, which is expected if they are light verbs/v’s themselves, and 3) complex thematic suffixes never vary in the theme vowel they select, i.e., -ific- always inflects with the theme vowel of the first conjugation, -ec- with that of the second conjugation, etc. Deciding between these two analyses is not the goal of this article, so complex thematic suffixes will be cited with a bracketed hyphen in the following (e.g., Span. -ific(-)a-). However, apart from the formal question (are theme vowels heads of or adjuncts to functional projections?) there is also the question of whether theme vowels have specific functions or “meaning” (however abstract), or whether they are “ornamental”, semantically vacuous elements—“pure morphology, form without meaning” (Aronoff 1994: 47), as generally argued in the DM literature (e.g., Oltra-Massuet 1999; Oltra-Massuet & Arregi 2005; Embick 2010; Calabrese 2015a; 2020). In the following, we will approach this problem from a diachronic perspective, based on data from the history of Greek.

2.2 Diachrony of verbal “themes”

The study of the diachrony of verbal stem-forming morphology is relatively well represented in the grammaticalization literature, especially where morphology expressing tense, aspect, and mood is concerned (e.g., the Germanic “weak preterite” dental suffix, the Latin imperfect suffix -bā̆-, the Romance synthetic future from infinitive + have, etc.). However, the diachrony of purely categorizing or “first phase” synthetic, derivational morphology (including theme vowels) has received less attention, especially from a theoretical perspective. Some relevant examples of diachronic reanalysis and resegmentation (e.g., “affix telescoping”) are discussed in Haspelmath (1995); (cf. also Rainer 2015) from a descriptive, functionalist perspective, but without explicitly treating the diachrony of theme vowels. And while some verbal themes are generally acknowledged to be diachronically related to, or even descended from, nominal morphology (for example, the Latin 1^st conjugation marker (*)-ā-, Leumann 1977: 545; Bertocci 2017; Weiss 2020: 423), the exact circumstances of their reanalysis from nominal to verbal (and, in an Oltra-Massuet-style approach, from syntactic head to postsyntactic adjunct) are rarely explicitly stated, and the synchronic analysis is complicated by the fact that these morphemes are usually used in primary and in derived environments at the same synchronic stage of the language (cf. de Vaan 2012 on deverbal ā-verbs in Latin).

In this paper I will assume that one way of creating new synthetic verbalizing morphology is via reanalysis of originally nominal or adjectival morphology in the context of denominal and deadjectival verb formation (see Grestenberger Forthcoming for a more detailed discussion of the theoretical assumptions behind this approach). An example is given in Table 1, illustrating the development of the Modern Greek (MG) verbalizer -ev- from Ancient Greek verbs in -eú-ō. These AG verbs were originally denominal verbs derived from agent nouns in -eú-. In this context, the originally nominal suffix was reanalyzed as part of the verbal domain and subsequently became a productive verbalizer in MG, where it can select nouns, adjectives, adverbs and loanwords.²

This reanalysis is descriptively “upwards” in that an element of the lower nominal domain was reanalyzed as part of the higher verbal domain in this particular context (i.e., denominal verb formation), as illustrated in (6).

This is reminiscent of other types of “Upwards Reanalysis” (UR, Roberts & Roussou 2003) that have been discussed in the literature, such as the reanalysis of lexical verbs as modal verbs (e.g., Roberts & Roussou 2003; Fischer 2003; Cournane 2014), or, more broadly and not restricted to the verbal domain, changes that fall under the “Late Merge Principle” (LMP, van Gelderen 2004; 2008; 2011; etc.) by which an element in a lower domain becomes reanalyzed as base-generated higher up in the structure (essentially an updated version of “Merge-over-Move”). Economy principles such as these have long had their place in the formal literature on syntactic change (cf. also Biberauer 2019; Biberauer & Roberts 2017), and from the perspective of a non-lexicalist framework like DM, it is not surprising that we should see the same principles at work in the diachronic reanalysis of synthetic derivational morphemes, as the difference between “words” and “morphemes” is epiphenomenal in DM. As argued in Grestenberger Forthcoming, such cycles seem to include an n > v cycle, in which nominal morphology gives rise to (different types of) verbalizers, and an a > v cycle, in which adjectival morphology gives rise to (different types of) verbalizers, especially those associated with the causative/factitive alternation. By the same mechanism, verbalizers themselves can also move “upwards” and become reanalyzed as Voice, Aspect, or Tense morphology. Possible examples include the *-ske/o-verbalizer in Greek discussed in Section 3.3, which has turned into an aspectual or maybe a tense marker in Ionic Greek (cf. Ittzés 2008; Ringe & Eska 2013: 168–170) and the Greek suffix -th(ē)- which seems to have been reanalyzed from an inchoative/stative verbalizing suffix in AG (García Ramón 2014; Grestenberger 2021) to a Voice and/or Aspect marker in MG (Christopoulos & Petrosino 2018; Alexiadou 2021).

In fact, the same type of reanalysis gave rise to a number of other MG verbalizers that Panagiotidis et al. (2017) discuss, such as -iz-, -(i)az-, -on-, -ar-, -en-. However, it is unclear whether “vocalic elements” or theme vowels can be treated parallel to these. Spyropoulos et al. (2015) and Panagiotidis et al. (2017) argue that the vocalic themes of the MG second conjugation (-a-, -i-) are indeed reflexes of the same v-position as -ev-, specifically, that they spell out an abstract vocalic “slot” -□_v which “in second-conjugation verbs is the morphological exponent of a categorizing verbal head v” (Panagiotidis et al. 2017: 34). Assuming, then, that at least some theme vowels are verb(alizer)s, we hence expect them to 1) have the same diachronic development as other verbalizers and 2) to behave synchronically like other verbalizers. Concerning 1), this means that nominal and adjectival stem-forming morphology is a possible diachronic source of theme vowels. Concerning 2), we need clear diagnostics for distinguishing verbalizers from other elements in the “verbal spine” in order to compare theme vowels to other types of (primary) verbal stem-forming morphology. We return to a discussion of these diagnostics in section 3.2.

In the following, we are going to compare thematic and athematic verbs in Ancient Greek from this perspective and examine the synchronic distribution and function of the different kinds of AG theme vowels and verbalizers, arguing that AG verbal stem-forming morphology inherited from Proto-Indo-European (PIE) spells out v, not Asp, and that new verbalizers are grammaticalizing in AG through a → v and n → v reanalysis.

3.1 Thematic vs. athematic verbs

AG roots combine with stem-forming affixes to form different types of verbal stems: The present stem, usually characterized as [-pfv], which can occur with past tense endings (imperfect) or nonpast tense endings (present), the aorist stem, which is [+pfv] and only takes the past endings (except for the aorist subjunctive, which takes the present/nonpast endings), the future stem, which takes the nonpast endings and is underspecified for perfectivity, and the perfect stem, whose aspectual properties are notoriously difficult to pin down and which takes nonpast (perfect) and past endings (pluperfect). The synthetic perfect is analyzed as Asp[-aor, +perf] by Reed (2014) and as Asp[perfect] by Schreiner 2021. I follow the latter here, but note that the analysis of the feature(s) of the perfect does not affect the following discussion.³ What is important is that the stem-forming morphology of the perfect, whatever its featural content, is in complementary distribution with that of the present and the aorist.

These stems can moreover combine with different moods, namely subjunctive (present, aorist, perfect), optative (present, future, aorist, perfect), and imperative (present, aorist).

Voice (active/nonactive) is expressed on the finite verbal endings together with Person, Number, and Tense. Specifically, the active and nonactive (“middle”) sets of endings are analyzed here as allomorphs conditioned by the feature [±D] (i.e., presence/absence of an external argument DP) on Voice (Grestenberger 2016; 2018; 2021, following, e.g., Alexiadou 2013; Alexiadou et al. 2015; Schäfer 2017 for MG; Kastner & Zu 2017 for Latin; Kastner 2020 for Hebrew; etc.).

The perfective “passive aorist” suffix -thē̆- (older allomorph -ē̆-) was originally a verbalizer that formed stative and inchoative verbs (Jasanoff 2004; Tronci 2005; García Ramón 2014), but seems to have been reanalyzed as a Voice/Asp[pfv] marker on the way to Modern Greek (Christopoulos & Petrosino 2018; Alexiadou 2021). This suffix will not be important to the following discussion. I will also ignore the “augment”, a prefix e- analyzed as T[+past] by Reed (2014), but still optional as a marker of the [past] tenses (imperfect, aorist, pluperfect) in Archaic Greek.

The following tree summarizes the structure of AG verbs and its features, leaving aside the projections pertaining to Mood for now. Note that Voice is missing in certain verbs (unaccusative activa tantum and the passive aorist, Grestenberger 2021) and is therefore given in brackets.

The question that arises now is where in this structure verbal theme vowels are spelled out and what their relationship to other types of verbal stem-forming morphology is. Consider a (synchronically) simple thematic verb like lū́ō ‘I release’ (tr.), whose partial paradigm for the imperfective (present) stem is given in Table 2. The partial paradigm of the aorist active (indicative, subjunctive, and optative) and perfect active (indicative) are given in Table 3. The bolded elements are those that can and will be considered “theme vowels” in the following. For expository reasons I use the Classical Greek forms in the following tables.

Columns (a)–(b) in Table 2 show the imperfective indicative forms of a simple thematic verb, in which the theme vowel alternates between -o- in the first person and third person plural, and -e- everywhere else (including the dual); the same distribution holds in the nonactive indicative forms (not illustrated here for reasons of space). However, this is not the only thematic element we can identify. For example, the theme vowel seems to be invariant -o- in the present optative (column d), and -a- in the so-called “alphathematic” inflection of the sigmatic aorist indicative (column a) of Table 3) and the the perfect indicative (column d) of Table 3), as well as in certain reduplicated aorists, the “kappa-aorist”, and the pluperfect. The “long-vowel subjunctive” in column c. of Table 2 is originally composed of theme vowel + subjunctive marker, which happens to be identical to the theme vowel,⁴ resulting in the corresponding long vowels. The long-vowel subjunctive was extended to athematic formations as well, in which the modal suffix was originally a short vowel *-e/o- (again, identical to the simple indicative theme vowel), so I will treat it as simplex morpheme in the following.

While the stem forms seen so far all contain elements that are usually considered theme vowels, AG also has “athematic” verb classes without any theme vowels. Table 4 gives the partial paradigm of an athematic present with the suffix -nū̆-, Table 5 its aorist and perfect forms (corresponding to the thematic ones in Table 3). Verbalizing/stem-forming morphemes are given in bold.

For the s(a)-aorist and kappa-perfect in Tables 3 and 5, I assume that the distribution of the allomorphs is -sa-/-ka- before consonants and -s-/-k- before vowels,⁵ and hence treat these classes as “athematic” in the following, in the sense that they lack the alternating theme-vowel -e/o- of, e.g., simple thematic verbs as in Tables 2 and 7 (though recall that this type of aorist/perfect inflection is traditionally called “alphathematic”). This means that the segmentation of the aorist optativ allomorph is -ai-, parallel to the reanalyzed imperfective optative suffix -oi-, which was extended from the thematic conjugation to athematic verbs where it replaced the older athematic optative.⁶ For reasons of space, the following discussion will focus on the comparison of the present and aorist indicative forms (see Zukoff 2017 and Schreiner 2021 for recent analyses of the Greek perfect), while further motivation of the segmentation of the modal markers (and their changes between Homeric and Classical Greek, which are significantly more pronounced than in the indicative) will have to be deferred.

In the classes seen so far, the stem-forming suffixes -nū̆- and -s(a)- in Tables 3, 4, 5 seem to be in complementary distribution with the theme vowel in Table 2: both are linearly immediately adjacent to the root (Table 6) and both can cause root allomorphy, specifically ablaut alternations and suppletion (Table 7). In the following, I use the 1^st person plural as citation form, since this tends to be the most transparent form of the paradigm in terms of morphophonology. For the athematic root present (eĩ-mi ‘I go’, 1pl. i-men) and the reduplicated athematic present (tí-thē-mi ‘I place’, 1pl. tí-the-men) in Table 6, column a., I assume that the verbal stem-forming morpheme is (linearly) zero, but that it triggers reduplication in the latter type which is then expressed as a prefix to the root (with reduplicating vowel -i-). This is compatible with Schreiner (2021)’s analysis of the perfect reduplicant, namely that “the phonological instantiation of an Aspect node specified for [perfect] has the shape RED/e/-” (Schreiner 2021: 12), except that I analyze the node in question as v (more below). However, adding “Ø” in the (linear) representation of these reduplicated forms will be useful for comparing athematic reduplicated verb stems with thematic reduplicated ones (such as reduplicated -ske/o- presents or reduplicated aorists), where by assumption the node that triggers reduplication is spelled out overtly (or: receives a theme-vowel adjunct in the sense of Oltra-Massuet 1999 etc.). That is, reduplicated athematic tí-the-Ø-men ‘set, place’ in Table 6 is the athematic equivalent of reduplicated thematic forms such as mí-mn-o-men ‘stay, remain’ in column b., as deík-nu-men is the athematic equivalent of its later thematic replacement deik-nú-o-men, etc.⁸

Table 7 illustrates root allomorphy in different types of thematic stems. Most commonly, root allomorphy in AG manifests as ablaut alternation in which the root vowel changes in terms of quality (place of articulation, as in rows b–c) or quantity (as in row a) depending on the morphosyntactic environment (e.g., “present” vs. “aorist” stem). This allomorphy is locally conditioned, outward-looking, and morphosyntactically conditioned in the sense of, e.g., Embick 2010; 2012; 2015. It is local in the sense that it requires strict adjacency: Morphemes to the right of the theme cannot condition any changes in the root, as shown in the presents in rows b. vs. d: middle derk-ó-metha in b. and active pheúg-o-men in d. have the same root ablaut (so-called “e-grade”). That is, neither the phonology nor the morphosyntactic features of Voice (active vs. middle/nonactive) can influence the shape of the root across the intervening theme, and the same is true for even higher projections such as Mood or Tense/Agr.⁹ That the allomorphy is outward-looking and morphosyntactically conditioned is illustrated by the present vs. aorist forms in rows b.–d.: Both are “simple thematic” forms, but the present stem-forming theme triggers e-grade of the root, while the aorist theme triggers so-called zero grade of the root. This shows that the conditioning factor is the morphosyntactic feature of the functional head spelled out by the theme rather than its phonology.¹⁰ The same holds for themes that cause quantitative modifications of the root (vowel), as in row a: In the present, the theme vowel diachronically reflects a complex thematic suffix *-j(-)e/o-, with the glide assimilating to give a geminate, and the aorist is diachronically an s-aorist formed with the same suffix -s(a)- as in Table 5, with the loss of -s- causing compensatory lengthening of the root vowel (written <ei>).¹¹ Finally, row e. and possibly also row f. show instances of root suppletion: The root that means ‘slay’ in row e. has the allomorphs then (/t^hɛn/), thein (/t^heːn/; with compensatory lengthening of the root vowel in, e.g., the present stem) vs. pha(n)/phn (in, e.g., the reduplicated aorist and the verbal adjective). Historically, both allomorphs go back to the same root , but synchronically they may as well be treated as suppletive pairs (cf. Kölligan 2007: 360–1).¹² The root variants manth/math in row f. could likewise be treated as suppletive allomorphs or via a morphologically conditioned infixation rule (see section 3.5 on “complex thematic” suffixes, including “nasal-infixing” -an(-)e/o-).

Synchronically, some or all of these root alternations could be implemented as morphonological readjustment rules or as root suppletion (i.e., as “late insertion” of different root shapes depending on the morphosyntactic environment, cf. Haugen & Siddiqi 2013)—I will remain agnostic on the exact implementation in the following. The important point is that this allomorphy is strictly local, outward-looking, and morphosyntactically (rather than phonologically) conditioned. Moreover, both thematic elements such as the e/o-theme vowel and athematic elements such as the stem-forming morphemes -nū̆-, -s(a)-, -(k)a-, etc., condition these particular ablaut grades on the adjacent roots (cf. row b for an example of archaic “o-grade” of the root in the perfect).

To summarize, in comparing the basic averbos of selected thematic and athematic verb classes, we have arrived at a preliminary segmentation of AG verbs as spelling out √-v/Asp-Mod-T/Agr. As mentioned above, (active/nonactive) Voice is expressed together with T/Agr on the endings—there are several ways in which this can be formally implemented, which will not concern us here.¹³ The important take-away is the observation that thematic and athematic suffixes (including complex thematic suffixes; see section 3.5 for a more detailed discussion of these) are in complementary distribution in the sense that they compete for the same position and cause the same types of locally conditioned root allomorphy. We can now turn to the question of whether -nū̆-, -s(a)-, -o-, -an(-)e/o- etc. spell out v, Asp, or a “span” consisting (minimally) of both.

3.2 Diagnostics for v-status

The forms discussed in the previous section suggest that thematic present and aorist suffixes occupy the same structural position as the inherited athematic verbal stem-forming suffixes -nū̆-, -s(a)-, etc., but it is less clear what that position is, exactly. Do these affixes spell out v or Asp, or do they adjoin to one of these projections? Answering these questions is related to a general problem of the older Indo-European (IE) languages, but specifically relevant to Ancient Greek and (Vedic) Sanskrit, namely the fact that their verbal stem-forming morphemes seem to combine properties of inner (“lexical”) aspect/”Aktionsart”¹⁴ markers (for example, related to transitivity or boundedness) and of “outer”/syntactic aspect, i.e., perfectivity. The standard analysis is that these stem-forming elements (including the theme vowels of the simple thematic present and aorist) were originally (in pre-Proto-Indo-European, that is, before the reconstructable ancestor stage of the attested Indo-European languages) Aktionsart/lexical aspect markers that were reanalyzed as temporal-aspectual markers. They are therefore usually termed “tense-aspect stems” (and “present” vs. “aorist” stem instead of the more accurate “imperfective” and “perfective” stem), cf., e.g., Hoffmann 1970; Rix 1986; Strunk 1994; and standard textbooks and handbooks such as LIV²; Fortson 2010; Meier-Brügger 2010; etc. (though see Hollenbaugh 2018; 2021 for a more recent analysis of aspect in Vedic, Greek, and Proto-Indo-European that departs from the traditional view¹⁵). Formal accounts that adopt this analysis and treat verbal stem-forming morphemes explicitly as realizations of Asp include Reed 2014; Calabrese 2019; 2020; and Schreiner 2021. Diachronically, this development would be an instance of UR as described in section 2.2, with v-elements becoming reanalyzed as ±pfv elements and a “more functional” (semantically bleached, strictly “synsem”) meaning. But this is not quite how these affixes behave in AG (or Sanskrit, for that matter), where they have properties that indicate that they are positioned adjacent to the root (cf. again Tables 6, 7) and are semantically and morphonologically idiosyncratic in ways that are generally associated with “inner”, categorizing, rather than “outer”, functional projections in the DM-tradition of, e.g., Marantz 2001; 2007; Arad 2003; Anagnostopoulou & Samioti 2014. The goal of the following sections is therefore to provide more specific arguments in favor of analyzing these elements as v- rather than Asp-morphemes. For reasons of space, this is not the place for an in-depth engagement with previous formal Asp-accounts (especially Calabrese 2019 and Schreiner 2021), though I plan to return to this elsewhere. Rather, the goal is to provide arguments that have not received much attention so far and that favor the v-analysis of these Greek verbal stem-forming elements, including (simple and complex) thematic stems.

Besides these two formal options (realization of v vs. Asp), there are other analytical options that have been proposed in the literature. For example, it is possible that the verbal stem-forming elements discussed in the previous section actually spell out a span (“a contiguous sequence of heads in a head-complement relation”, Svenonius 2016: 205) consisting of v-(Voice-)Asp (see Merchant 2015 for a Spanning account of the MG verb) or a “rebracketed” v/(Voice/)/Asp head (Christopoulos & Petrosino 2018 propose that MG has a rebracketed [Voice,Asp] head), rather than just (contextually conditioned allomorphs of) v or Asp. These four possibilities are summarized in (8).

In the following I will provide arguments in favor of the latter analysis, fully aware that some of them are compatible with more than one option in (8), or even with all four. First, we need clear diagnostics for identifying verbalizing morphology. The following diagnostics are based on the discussion in Spyropoulos et al. 2015 and Panagiotidis et al. 2017 on identifying MG verbalizers and the diagnostics in Harley 2017 for distinguishing between “Voice-bundling” and “Voice-splitting” languages—that is, languages that “bundle” the properties of v and the agent-introducing projection Voice, and languages in which these are two separate projections.

According to Panagiotidis et al. (2017), these diagnostics pick out the MG verbalizers that are historically reanalyzed from nominal suffixes + verbalizer (i.e., -ev-, -iz-, -(i)az-, -on-, -ar-, -en-) as well as the “theme vowels” of second conjugation verbs like aγapó ‘love’ (-a-, -i- <η>, pres.ind. aγapáo, pfv. past aγápisa, etc.), which as we will see in section 4 are also diachronically denominal.

Applying these now to the Ancient Greek situation, (9a) is not particularly controversial and easily verified for AG: Voice and Tense morphology are realized separately from verbal stem-forming morphology, as we saw in the previous section. Moreover, there is no close relationship between agent introduction and verbal stem-forming morphology, suggesting that agents are introduced separately by Voice (thus fulfilling one of Harley 2017’s criteria for a Voice-splitting language). Whether v and Asp are also separate projections still remains to be seen.

(9b) is more controversial because it depends on whether one conceives of categorizing morphology as purely syntactically categorizing (thus essentially doing what ±N, ±V and similar categorial features were doing in classical G&B approaches), without being associated with Aktionsart or valency semantics such as causative, factitive, inchoative, etc., as Oltra-Massuet (1999) and Oltra-Massuet & Arregi (2005) argue. Panagiotidis et al. (2017) seem to reach the same conclusion (“…the verbalizing suffixes do not encode Aktionsart in the sense that they are neutral with respect to inner aspect”, Panagiotidis et al. 2017: 37), but then also state that “the same verbalizing suffix can create verbs with a variety of argument structures” (Panagiotidis et al. 2017: 39), presumably meant to indicate that there is no one-to-one correlation between a particular verbalizing morpheme and a particular argument structure type, unlike with “high” (syntactic) causatives (v-selecting causative morphology, Harley 2017) and ‘high” passives (i.e., Voice-selecting passive morphology in the sense of Bruening 2013; Alexiadou et al. 2015; Schäfer 2017, etc.). But this still does not exclude the possibility that there are subregularities and correlations with specific Aktionsart types: There is mounting evidence that such correlations can predict the distribution of thematic vowels and other conjugational class markers in, e.g., French (Kastner & Martin 2020; 2021) and Serbo-Croatian (Kovačević et al. 2021; Mišmaš & Simonović 2021; Milosavljević & Arsenijević 2022) to some extent.

This is true for AG verbal stem-forming morphology as well, at least at the oldest stage. Among the generalizations that have been noted in the literature concerning the function of the verbalizers discussed here are, among others, that -nū̆- and -s(a)- tend to form alternating transitive verbs (cf. Bertocci 2013 on Greek nasal presents in general), -ane/o- is found with verbs of perception and psych verbs and is characterized as transitive/terminative by van de Laar (2000: 345), -ske/o- is found with verbs of motion (van de Laar 2000: 364), etc.,¹⁷ though there are naturally exceptions to such generalizations. Moreover, roots with seemingly similar meaning may use different verbalizers to express the same argument structure, and one and the same root may use different verbalizers for different (alternating) contexts, e.g., √pag/pēg ‘thick, fixed, solid’ forms both an inchoative -ē- (-η-) aorist (‘became fixed, solid’) and a transitive-causative sigmatic aorist (‘cause to be fixed, solidify’) in Homeric Greek. These examples can be multiplied (see again van de Laar 2000 for a detailed study of such “individual verbal systems” in Greek). The AG verbal stem-forming morphemes therefore fulfill both (9b) and (9c), as well as (9d) because the selectional relationship seems to be root-specific: Thus alternating √pag/pēg uses the ē-aorist to form its perfective inchoative, while alternating √or ‘rise’ uses a middle root aorist (3sg. ō̃rto) to express the same meaning. This close relationship between root and verbal stem-forming morpheme holds for the morphophonological side as well: As we already saw in Table 7, verbal stem-forming morphology in turn determines the shape and ablaut grade of the root.

Finally, certain AG verbal stem-forming morphemes productively derive denominal and deadjectival verbs, and this is crucially the case both for the ones associated with the present system ((*)-je/o-) and those associated with the aorist system (-s(a)-; -(th)ē-). This argument has so far been neglected in the discussion of their status, and we will therefore return to it in section 4.

Taken together, these diagnostics point towards v-status of the Greek verbal stem-forming suffixes, both the athematic and the simple thematic ones (section 3.5 provides arguments that complex thematic suffixes also pattern the same way). Nevertheless they do not unambiguously tease apart v and Asp—a Spanning or Rebracketing account is thus still in the picture.

However, if the order of functional projections given in (1) is indeed cross-linguistically invariant, then evidence for a separate v and Voice should also imply a separate v and Asp, since Voice would intervene. At least in some versions of Spanning, the intervening alternating features of Voice would then exclude the possibility of treating v-Voice-Asp as a span.¹⁸ Harley (2017) provides the following criteria for Voice-splitting languages, in which v and Voice are separate functional projections:

(10a) is difficult to apply to AG, which has been argued to have the same type of syncretic “middle” Voice head that is realized with nonactive morphology in anticausative, reflexive, reciprocal, and passive contexts as MG does (Alexiadou & Doron 2012; Alexiadou 2013; Alexiadou et al. 2015; Schäfer 2017; on AG Grestenberger 2016; 2021). A designated passive suffix -(th)ē- does develop eventually, but behaves synchronically like a verbalizer rather than a Voice marker in (archaic) AG (Tronci 2005; García Ramón 2014; Grestenberger 2021) and thus cannot be argued to be a “high”, Voice-selecting Pass head (in the sense of Bruening 2013). On the other hand, alternating causative and agentive-transitive verbs as in Table 8, which share the same verbal stem-forming morpheme, but take active endings in the causative/transitive and nonactive endings in the inchoative/passive alternants show that the introduction of the external argument takes place independently from verbal stem-forming morphology, unlike in Voice-bundling languages. These verb classes could then be taken as evidence for the criteria in (10a) and (10c).

(10b) could also apply to AG, provided that the dativus (in)commodi is to be interpreted as a high applicative. While there is no designated applicative verbal suffix in AG, animate datives of convenience/inconvenience (benefactives/malefactives) like in (11), which can modify transitive and intransitive unergative predicates, could be interpreted in this way and thus fulfill (10b). However, further study of these constructions is needed to confirm this analysis.

To summarize, the AG verbal stem-forming suffixes fulfill all of the criteria for v-status in (9) (with a question mark over their relationship with Aktionsart, 9b), and Harley’s diagnostics for Voice-bundling vs. Voice-splitting languages suggest that AG belongs to the latter, though this is based on a preliminary assessment. However, in order to make a stronger case for v-status of these elements, we need additional, ideally unambiguous evidence of their co-occurrence with genuine aspectual morphology relating to outer/syntactic aspect. The implicit assumption so far is that Asp is in fact not expressed overtly in most contexts in AG, effectively the opposite of approaches in which Asp is overt and v is covert and/or “pruned” (e.g., Calabrese 2019). But if it were possible to find contexts in which Asp can be independently shown to be realized overtly and in which it combines with verbal-stem forming morphemes, this would make a stronger case for the latter’s status as verbalizers. This aspectual morphology would be expected to be morphophonologically and semantically more regular than “inner” morphology as it expresses a limited set of morphosyntactic features (presumably only ±pfv vel sim.) and to be compatible with many, if not all of the putative v-elements discussed in this section.

There are two environments that have hithero been neglected in this discussion and that could potentially fulfill these criteria: the epic/Ionic “iterative preterit” and the participles and verbal adjectives of the stems discussed above.

3.3 Additional arguments for v, I: the Ionic iterative preterit

The so-called “Ionic iterative preterit” (cf. Brugmann 1902–03; Schwyzer 1939: 710ff.; Chantraine 1984: 226f.; Ittzés 2008; Kimball 2014) is particularly productive in (epic) Ionic Greek and is formed with a complex thematic suffix -sk(-)e/o- which selects verbal stems and does not co-occur with the augment. That is, unlike all the other verbalizers discussed so far, which select and are linearly adjacent to the root and restricted to either the present or the aorist (or the perfect) stem, the iterative preterit suffix -sk(-)e/o- can select present and aorist stems (no forms from the perfect stem are attested), rather than just the root. Together with its semantics, which can be characterized as progressive or iterative, this makes it a plausible candidate for actually spelling out Asp unlike the verbal stem-forming suffixes discussed so far. Table 9 gives some examples of Ionic iteratives and their respective bases.

These examples show that iterative -sk(-)e/o- selects verbal stems rather than roots,¹⁹ independent of whether the stem is athematic (rows a–b) or thematic (rows c-d). Rows a. and b. illustrate iterative preterits from an athematic reduplicated and a nū̆-present, respectively. Rows c. and d. illustrate iterative preterits from a simplex thematic present and a simple thematic aorist, respectively (crucially from one and the same root; see Kimball 2014 for a discussion of the semantics of iterative preterits from present vs. aorist stems). All in all, this suffix is therefore a good candidate for a genuine aspectual suffix.²⁰

3.4 Additional arguments for v, II: nonfinite forms

So far we have concentrated on theme vowels and verbalizers in finite verb forms. But in nonfinite “deverbal” forms we see quite clearly that some suffixes select stems (minimally √-v), while others select roots. The “verbal adjective” in -t- (-t-ó- m./n., t-ḗ f.) selects roots and forms state-denoting adjectives, as illustrated in Table 10.

As these examples show, -t- treats thematic and athematic verbs alike in attaching to the root and ignoring any verbal stem-forming material, including theme vowels as in column c. Assuming that this stem-forming material realizes v, this affix can therefore be used as a diagnostic for the presence vs. absence of v—more precisely, it seems to be in complementary distribution with verbalizers. (13a) gives the proposed structure of -t-verbal adjectives, based on the analysis of their Modern Greek descendants by Alexiadou et al. (2015), who argue that -t- selects (event-denoting) roots and spells out stative aspect (thus also Anagnostopoulou 2003; Anagnostopoulou & Samioti 2014; Grestenberger 2018; Calabrese 2020; etc.). Given that it only selects the root to the exclusion of any verbal stem-forming morphology, the examples in Table 10 suggest the same structure for AG verbal adjectives in -t-. In addition, AG -t- is also able to form possessive adjectives from nouns (cf. Schwyzer 1939: 503), as in the examples in (12). The presumed structure for these is given in (13b).

The verbal adjectives in -t- differ from the participial suffixes -nt- (pres./aor. active participle), - (w)ō̆s/-uia (m.-n./f. perf.act. participle) and -men-o/ē- (nonactive/“middle” participle): These always select the verbal stem—contrast the participial forms in Table 11 with the “verbal adjectives” in Table 10.

The fact that these participles contain verbal stem-forming and Voice (act./mid.) morphology and can be construed transitively with direct objects (if the corresponding finite forms are transitive, that is) suggest that these participial suffixes contain more structure than the -t-forms, as illustrated in (14) (following Grestenberger 2018; 2020, as well as Alexiadou et al. 2015 on Modern Greek and Embick 2000 on Latin participles).

Crucially, just like in (13), the participial suffixes are analyzed as overt allomorphs of the aspectual head Asp, with the Spell-Out conditions on the active and nonactive participle suffixes given in (15).²²

If this is the correct synchronic analysis, it means in effect that Asp is realized as zero (and/or is “pruned”, Embick 2010; 2015, etc.) in all finite forms (with the exception of iterative/progressive “Ionic” Asp, section 3.3), and only surfaces overtly in the participial system. More precisely, the Spell-Out rule for Asp in finite forms must be as in (16), to ensure that the participial allomorphs do not surface in the infinitival forms (assuming the latter spell out nonfinite T).

An anonymous reviewer has criticized that this seems incompatible with the handbook view that present participles (used as adjuncts, or “circumstantially”, that is) express simultaneity with respect to the event described by the main verb, while aorist participles express anteriority of the event with respect to that of the main verb (e.g., Rijksbaron 2002: 117; though not uniformly, cf. Napoli 2017: 116). But as long as the participles in these contexts are not specified for Tense independently of the main verb (on which there is widespread agreement, e.g., Smyth & Messing 1956: 457; Napoli 2006; 2017; Pompei 2012; Buijs 2013), the analysis above holds assuming that the Spell-Out of Asp is syncretic—that is, there are no separate allomorphs for Asp[+pfv] and Asp[-pfv] (or however one wants to characterize syntactic aspect in Greek at this stage, cf. footnote 8 above), either in finite or in non-finite contexts (again with the exception of the particular “flavor” of Asp spelled out by the Ionic iterative preterit suffix). This parallels the analysis by Embick (2000) of the Latin participial suffixes -nt- and -t-(-s-) as unspecified for the features of Voice, despite their traditional labels as “active” and “passive” participles, respectively, and hence as compatible with both “active” and “passive” syntactic contexts (as in the widely referenced active, transitive deponent participles in -t(us), cf. Grestenberger 2017; 2018). Similarly, the realization of Asp as -nt- vs. -men- in (15) would be sensitive to the adjacent Voice head, but not to the features of Asp itself (with the exception of the perfect active allomorph, which is more highly specified than active/Elsewhere -nt-; cf. Grestenberger 2022 for a more detailed discussion).

To summarize, assuming that the Ionic iterative preterit suffix -sk(-)e/o- and the participial suffixes -nt- and -men- are overt allomorphs of Asp, and given that these co-occur with verbal stem-forming morphology, we conclude that the latter must spell out something lower in the structure. This, together with its adjacency to the root, its root-specific/lexical selection-based distribution and its ability to trigger root ablaut, points towards verbalizing v.

Note that an interesting prediction follows from the discussion of the two overt Asp morphemes in this section: If both of them compete for the same structural position, we do not actually expect Ionic iterative preterits to form participles, i.e., we do not expect *-sk(-)o-nt- and *-sk(-)o-men-participles to be formed to these preterits. This is indeed the case: No participle is attested for any of the iterative preterits in Homer and Herodotus (the main sources of attestation of this type).²³

Before turning to denominal verbs, we will briefly look at complex thematic suffixes in the next section.

3.5 Complex thematic suffixes

To recapitulate briefly, “complex thematic” suffixes appear to consist of a verbalizing affix + a theme vowel and thus structurally pose the same problem as the complex Spanish verbalizers in ex. (5), e.g., clas-ific(-)a- ‘to classify’. The point of this section is not to weigh in on whether these are (essentially monomorphemic) allomorphs of v (as proposed by Fábregas 2017) or v + thematic adjuncts (Oltra-Massuet 1999; Oltra-Massuet & Arregi 2005, followed by much of the DM literature), as the diagnostics introduced in the previous section, in particular verbal adjective and participle formation, do not help distinguish between these two analyses. They do show, however, that complex thematic suffixes are treated like simple thematic suffixes and athematic affixes in being able to co-occur with Asp morphology (the participial suffixes and the Ionic iterative preterit), but not with morphology that selects the root, like the verbal adjective-forming suffix -t-. The following tables illustrate this for the complex thematic present stem suffixes -n(-)e/o-, -an(-)e/o-, and -(i)sk(-)e/o- (on which see also footnote 23). The first two classes actually form simple thematic aorists, whose stem-forming suffix has the same distribution as its complex thematic counterpart, as shown in Tables 12, 13. Table 14 exemplifies two -(i)sk(-)e/o-verbs without and one with reduplication in the present stem. The finite aorist forms are given as unaugmented.

Stems with complex thematic suffixes (including the ones in Table 14) are also able to form Ionic iterative preterits, as illustrated in Table 15.

3.6 Interim summary

The main point of this section was to show that the distribution of the theme vowel -e/o- of simple thematic verbs corresponds to that of primary athematic verbalizers, and that both spell out the verbalizing head v (or an adjunct to it, in the case of the theme vowel), rather than Asp. Evidence for this comes from their semantics (i.e., their association with particular verb classes such as motion verbs or verbs of perception, or with telic/terminative inner aspect, cf. section 3.2), their adjacency to the root, their idiosyncratic root-selection, their ability to trigger root-specific alternations, and their ability to co-occur with elements that have been independently argued to spell out Asp (participial suffixes, the Ionic iterative preterite suffix). Moreover, complex thematic suffixes have the same distribution as simple thematic suffixes, in that they show up in all finite forms and participles, but not in the verbal adjective (i.e., in root-selecting formations), and in the iterative preterit.

Having established these properties of primary verbalizers and verbal theme vowels, the next section compares the behavior of synchronic denominal verbs. Some of these, too, contain elements that could be interpreted as theme vowels. However, we will see that these do not pattern with the primary verbalizers/theme vowels according to the diagnostics established in this section, and should hence not be analyzed as verbalizing morphemes synchronically.