In rows (c) (f), (g) and (j) of Table 1 the vowels present in the past-tense forms disappear in the present-tense forms. Since Jakobson (1948), who hypothesized that the longer form of the verbal stem is always the underlying one (see also Lightner 1965; Melvold 1989, etc.), this has been assumed to be due to a general vowel-before-vowel deletion process:

In some cases, however, vowel deletion appears to fail. One such instance is the nominalizing suffix in (4) followed by a vocalic case-number suffix:

Two takes on the origin of the intervocalic glide in (4) are possible. One possibility is that the underlying form of the suffix is -ij-. The other is that the hiatus is resolved by the epenthesis of the homorganic glide [j]. While it is likely that the latter hypothesis is correct for this case, in other environments there is no glide epenthesis, as will now be shown.

Two suffixes in Table 1 ([a]/[aj] in row (b) and [e]/[ej] in row (e)) surface as simple vowels before a consonantal suffix and as the same vowels followed by the glide [j] before a vocalic suffix. The natural question is which form is underlying.

Following Jakobson (1948), the longer form has been assumed to be underlying (Lightner 1965; Melvold 1989, etc.), with the glide deleted before a consonant: the three athematic verbs with a stem-final [w] (živú/žilá ‘live.pres.1sg/past.fsg’) support this view. A natural consequence of this view is the distinction between rows (b) and (e), where the surface [a] and [e] in the past correspond to [aj] and [ej] in the present, and rows (c) and (j), where the surface [a] and [e] of the past-tense forms are deleted in the present.

The opposite view is taken by Garde (1972), who argues that Russian has a productive hiatus-resolving process of glide-insertion that is triggered not only in verbal morphology but also in loanword adaptation, e.g., creating lakéj ‘lackey’ and gerój ‘hero’ (from the French laquais [lakɛ] and héros [eʁo] respectively). This process would also naturally account for (4). The same view on the genesis of [j] in verbs is defended by Itkin (2007).

Both approaches have their advantages and disadvantages. As Garde points out, it is artificial to draw the distinction between rows (b) and (e), on the one hand, and rows (c) and (j), on the other, by complicating the underlying representations of the more productive patterns with an underlying glide that would be deleted before the consonantal past-tense suffix. Garde also lists a number of cases where [j] surfaces before a consonant. While some of them constitute either underived environments (root-internally), post-cyclic junctures (e.g., pojtí ‘to start going’, from po-id-ti), or an underlying initial yer in a suffix that surfaces as consonantal, two cases support his claim: the superlative/elative suffix -ejš- (which might also be viewed as a combination of two comparative suffixes, the productive -ej- and the unproductive -š-, for which there is no independent motivation for a yer before [š]), and the nominalizing suffix -stv- (e.g., zlodéjstvo ‘villainy’, from zlodéj ‘villain’), where an initial yer is motivated only historically (though Melvold (1989:139ff.) argues for an initial segmentally unspecified timing slot there).

Conversely, the three athematic w-final verbs žitʲ/živú ‘live.inf/1sg’, slɨtʲ/slɨvú ‘be known as.inf/1sg’ and plɨtʲ/plɨvú ‘swim.inf/1sg’ would be difficult to analyze synchronically as cases of glide-insertion. Furthermore, glide-insertion has to be assumed to fail in rows (c) and (j) in Table 1, for the suffixes -nu- (rows (f) and (g)) and with the secondary imperfective suffix -ɨw-, which means that the choice, be it for glide-insertion or glide-deletion, has to be lexically based (see Itkin (2008:147) for further counterexamples).3

As the choice between these two approaches is far beyond the scope of this paper, I will not attempt to resolve it here and continue to refer to these two suffixes as -aj- and -ej- to facilitate their differentiation from the superficially similar -a- (row (c)) and -e- (row (j)) and base no argumentation on their underlying form.4

If glide-epenthesis is postulated, e.g., in (4), something should block it in cases of transitive softening, a.k.a. iotation, or transitive palatalization (Jakobson 1929; Meillet 1934; Kortlandt 1994; Townsend and Janda 1996, inter alii; see Halle 1963; Lightner 1972; Coats and Lightner 1975; Bethin 1992; Brown 1998 for generativist analyses). Transitive softening is a type of Slavic consonant mutation known to arise from a [Cj] cluster in a prevocalic position. The output is different for different consonants: in Russian, velar and alveolar obstruents change into the post-alveolar /š, ž, č/, a labial followed by [j] turns into a palatalized [lʲ], j remains unchanged, and all other sonorants alternate with a corresponding palatalized form:5

Except for row (c), Table 3 illustrates transitive softening for second-conjugation verbs where the thematic suffix (visible as -e- or -i- before a consonantal suffix) turns into a glide before the vowel [u] of the 1sg (see Bethin (1992:281)).6 The fact that this hiatus is not resolved by an epenthetic glide suggests either that the glide in (4) is indeed underlying or that sometimes glide insertion fails.

To conclude, there is evidence in Russian both for and against postulating an underlying glide in rows (b) and (e) in Table 1 and the conditions under which a front vowel turns into a glide before another vowel are unclear. Since mostly specific instances of these phenomena are of relevance here, I will attempt to stay neutral in all cases where a difference in analysis would entail a difference in the underlying form.

In this section I will argue that the morphosyntactic status of -a-/-i- suffix can be tentatively established from the fact that it appears after verbalizing suffixes, the productive -ow- and the unproductive -ot-. While the former is usually regarded as denominal, the next section 3.1.2 will show that the issue is more complex.

The large (ca. 20 strong) sub-class of sound-emission verbs with a stem ending in [ot] (e.g., xoxotátʲ ‘to laugh loudly’, klokotátʲ ‘to gargle’, topotátʲ ‘to stamp’, etc.) or [et] (e.g., lepetátʲ ‘to babble’, skrežetátʲ ‘to gnash’, etc., see Itkin (2007:204–205) on the “vowel harmony” determining the realization of the suffix) suggests that -ot- is a meaningful (though non-productive) verbalizer. While for most verbs in this category no meaningful root can be identified before -ot-, at least the verbs vorkotátʲ ‘to grumble’, topotátʲ ‘to stamp’ and trepetátʲ ‘to tremble’ can be argued to be built on the roots -vork-, -top- and -trep-, given the corresponding verbs vorkovátʲ ‘to coo’, tópatʲ ‘to stamp, tramp’ and trepátʲ ‘to pull, flutter’.

While Bernshteyn (1974:61–62), who mentions the use of -ot- in other Slavic languages and calls it onomatopoeic, treats it as nominal, given the existence of such nouns as xóxot ‘loud laughter’ or tópot ‘tramp’, I think this is incorrect for at least the current stage of Russian. Firstly, all nouns in -ot- denote nomina actionis, and the derivation of action nouns by theme truncation is very well attested for unquestionable verbs (e.g., prixód ‘arrival’ from ‘to come’; see Itkin (2013) for second-conjugation e-verbs). Secondly, all -ot- nouns have a corresponding verb.9 Crucially, if -ot- is a verbalizer, the suffix -a-/-i- cannot be one, and, given that it does not seem to have any identifiable semantic import, the question arises if it can correspond to aspect (the functional head Asp) or theme.

The answer is provided by the suffix -ow-. As (5) shows, this suffix surfaces as [ov] in the past (before the thematic suffix [a]) and as [u] followed by the surface glide [j] in the present:10

While the Russian grammatical tradition treats -owa-/-uj- as a single complex thematic suffix (see, e.g., Es’kova 1989 and Itkin 2007:125, cf. also Jabłońska 2004 and Svenonius 2004a for Polish), Melvold (1989) argues that it should be regarded as the combination of the verbalizing suffix -ow- with the -a-/-i- theme. As the existence of this theme is independently motivated, it follows that the glide in the present tense arises from its -i- allomorph:

If this analysis of the [owa]/[uj] allomorphy is correct, the question arises once again what the roles of the two suffixes are. I will now show that the aspectual properties of -ow- suggest that it is either a verbalizer or Asp, whereas -a/i- must be concluded to be a theme in the sense of Aronoff (1994) and Oltra Massuet (2000).

While the vast majority of unprefixed verbs in Russian are imperfective, many unprefixed verbs with the suffix -ow- can behave as imperfective or perfective (on Russian biaspectuality see Gladney 1982; Chertkova and Chang 1998; Janda 2007; Itkin 2014; Zinova and Filip 2015; Piperski 2018, among others):

Since no underived -a-/-i- verb is biaspectual,11 while many -ow- verbs are, biaspectuality must come from -ow- rather than from -a-/-i-, which excludes the possibility that -a-/-i- corresponds to Asp.

Further support for this comes from the fact that the suffix -a/i- has no recognizable semantic contribution. Gardiner (1979) tentatively suggests that it introduces an “Extension” feature, which amounts to volitionality: “The actor in directional verbs appears as purposeful, as directing the process in a certain direction” (p. 386). However, for at least the verbs sniskátʲ ‘to gain (arch.)’ and alkátʲ ‘to crave’ (as well as stradátʲ ‘to suffer’ and dvígatʲ ‘to move’, which have somewhat archaic variants with transitive softening in the present, stráždu ‘suffer.1sg’ and dvížu ‘move.1sg’), this is incorrect. Gardiner’s tentative generalization further breaks down with the suffix -ow- taking the -a-/-i- theme and deriving such verbs as brézgovatʲ ‘to be squeamish (about)’ (which is stative) or vibrírovatʲ ‘to vibrate’ (which applies to inanimate actors).

It might seem that a verbalizer need not have a clear semantic contribution beyond an event argument. If so, one can imagine an alternative take on -ow-, built on the analysis of Russian deverbal nominalization by Babby (1997) and Pazelskaya and Tatevosov (2008), who propose that Russian deverbal nominalization involves two steps: deverbalization (creating a nominal, [–V]) followed by nominalization (deriving a noun [–V, +N]). One can similarly hypothesize that denominalization (creating a [–N] node) must occur before verbalization (turning it into [–N, +V]) and that the function of -ow- is that of a denominalizer while -a/i- is verbalizing. Indeed, the verbal suffix -ow- would seem to be homophonous with both the genitive plural suffix -ow- (surface allomorphs [ov] and [ej]) and the adjective-forming suffix -ow- (surface [ov]), so a unification would seem desirable (see Simonović and Mišmaš 2020 for a proposal).

Setting aside the fact that neither the genitive plural nor the adjectival [ow] alternate with [u], further evidence against this view comes from athematic action nouns, as in (8). Since such nominalizations show that stems in -ow- are already verbal, -a/i- alone cannot be treated as a verbalizer.

A further issue with the denominalization hypothesis is that it is not clear what this process would mean from the compositional point of view. Finally, the appearance of -ow- in verbs derived from loanword roots, such as (6) and those discussed in the next section, casts further doubt on the intuition that -ow- is denominal.

Russian, like many other languages, is subject to affix pleonasm (Gardani 2015) in loanword integration. The status of -a-/-i- as a theme could be contested on the basis of the fact (9) that with loanword roots the suffix -ow- is often preceded by the sequences -iz-, -ir-, and -iz-ir-, which I will refer to as stem augments. The question arises which of these suffixes correspond to v, and -ow- is obviously no longer the only candidate for this status.12

The nature of stem augments is unclear. On the one hand, they are not obligatory: as shown by some randomly chosen verbs (10) from a modern slang dictionary (Nikitina 2003), -ow- is highly productive without any augments, even with borrowed stems (10d).

Secondly, a simple search in Zaliznjak (1977) shows that, unlike in Serbo-Croatian (Simonović 2015) or in Romanian (Bleotu 2019), stem augments cannot be used without being followed by -ow- (which in turn requires the theme -a-/-i-). While the obligatoriness of -ow- with stem augments is equally well compatible with its status as the actual verbalizer or as Asp, the behavior of semelfactive correlates of augmented -ow- verbs argues that (a) the augments have no semantic contribution and (b) the suffix -ow- is more likely to encode Asp.

As examples in (11) show, -ir- can be either retained (11a) or dropped (11b) and Google search for the two possible semelfactives from the verb sterilizovátʲ ‘to sterilize’ yields both variants (11c). If the augments represented unquestionable verbalizers, they would not be omissible.13

The complementary distribution of -ow- and -nu- with systematic concurrent differences in Aktionsart suggests that they should be treated the same, despite the fact that the former is often denominal while the latter is deverbal, and makes it even less likely that -a-/-i- is a verbalizer. I conclude that the thematic suffix -a/i-, given its morphosyntactic position (after verbalizing suffixes) and its lack of meaning, as well as the fact that it has no effect on aspect and therefore cannot be regarded as Asp, must be a theme in the Oltra-Massuet’s sense.14

The discussion in section 3.1 shows how difficult it is to pinpoint the contribution of a theme, but also to distinguish it from a verbalizer. The productive denominal suffix -ow- cannot be assigned an obvious meaning, its contribution is further obscured by stem augments, and the putatively non-verbal status of the stem it combines with is cast into doubt by the fact that it may alternate with the semelfactive -nu-. Yet these very facts contrast -ow- with other thematic suffixes arguing against a unified analysis, since, as I will presently show, some of them will pattern closer to -ow- and others, closer to -a-/-i-.

The first vowel of semelfactive allomorph -anu- could be hypothesized to correspond to one of the -a- thematic suffixes (rows (b-d) in Table 1). If correct, this hypothesis would suggest that (a) in Russian, as in Romance, a verb can have a thematic suffix between the lexical stem and the aspectual/verbalizing suffix, and (b) if the -nu- allomorph is taken to be preceded by a null theme, athematic verbs should be analyzed as containing a null theme. There are, however, arguments against this view.

Firstly, the accentual properties of -anu- (which is dominant and accented on [u]) do not seem to follow from the accentual properties of the semelfactive suffix -nu- (which is accented but not dominant) or of any of the -a- themes, none of which is dominant: the -a-/-Ø- theme (row (c) in Table 1) is unaccented, whereas the gliding -aj- theme (row (b)) and the transitive softening -a-/-i- theme (row (d)) are accented (Garde 1998:334). The minimal pairs below involving the two suffixes with the same roots illustrate the different positions of the main stress:

Secondly, the imperfective aspectual pairs of verbs formed with the suffix -anu- need not involve the thematic [a], e.g., gazanútʲ ‘to step on the gas’ vs. gazovátʲ (-ow-a-, imperfective), dolbanútʲ ‘to hit’ vs. dolbítʲ (-i-, imperfective), skrebanútʲ ‘to scrape’ vs. skrestí (athematic imperfective). Conversely, as Kuznetsova and Makarova (2012) point out, if -nu- could combine with a thematic a-verb, why does it not combine with any other thematic suffixes?21

It can be argued (see Garde 1998:366) that the -nu- sequence represents not one morpheme, but two: that v is exponed by -n- and -u- is the theme that it selects for. Evidence for this can be drawn from the observation that in at least two verbs (obmanútʲ ‘to cheat’ and mínutʲ ‘to pass’), -n- is synchronically part of the stem (cf. obmán ‘a lie’), which makes more likely the possibility that -u- could be a theme, albeit with a limited distribution. A further advantage of this view would be a simple explanation of the mechanism by which -nu- is deleted in the past tense: for some verbs the theme -u- would have a phonologically null allomorph in the past and -n- would be deleted before the past-tense suffix -l- by an independently motivated rule of Russian deleting stem-final dental consonants in this environment (even though nasals are usually not subject to it).

Three objections can be raised to this proposal. Firstly, it requires non-local allomorphy: the choice of the zero allomorph in the past would have to be determined by the root across the putative suffix -n-. Given, however, that -nu-deletion is made more likely by the presence of a prefix (Garde 1998:368, Es’kova 2011b, Nesset and Makarova (2012)), this objection could be regarded as minor.

The second objection is that the thematic suffix -u- would be selected by just two morphemes, which also happen to be homophonous (and maybe by verbs like mínutʲ ‘to pass’ and obmanútʲ ‘to cheat’, for which a degemination account is also possible). Given that the theme -o- (row (h) in Table 1, see section 3.1) is selected by just five roots, this objection is also quite minor.

Thirdly, if -u- were assumed to be the shared thematic suffix, the difference between the pre-accenting mutative suffix and the accented semelfactive suffix illustrated in (16)–(19) would have to be counterintuitively attributed to the consonants rather than the vowels.

Given that no observable gain is provided by the more complex bimorphemic view, I leave open the question whether it is desirable (cf. Božič 2015 reaching the same conclusion for the Slovenian cognate -ne-/-ni- and Štarkl et al. 2022 for the opposite one for Slovenian and BCMS).22 Irrespective of whether -nu- is bimorphemic, its semantic import (semelfactive for the productive accented -nu- and stative for almost all others) means that there is no reason to regard -nu- (or -n-) as a theme, which clearly differentiates it from the -a-/-i- suffix.

As discussed in section 3.1.2, the complementary distribution of the semelfactive -nu- and the denominal -ow- (21) suggests that the two should be treated the same, and the clear semantic profile shared by mutative -nu- verbs suggests that all of them should be viewed as v.

Yet the fact that neither of the two -nu- suffixes can survive in the secondary imperfective (14b), (15b), even though, like all thematic suffixes, both can combine with Aktionsart prefixes (14a), (15a),23 can also mean that they and the secondary imperfective suffix -ɨw- occupy the same syntactic head. Markman (2008) proposes, based on their complementary distribution, that the semelfactive -nu- and the secondary imperfective -ɨw- are both merged in a verb-selecting light v, as shown in (22).24 (Note that Markman does not discuss -ow- and does not address either the complementary distribution of -ow- and -nu-, or the compatibility of -ɨw- and -ow-.)

In the framework used by Markman roots combining with -nu- (and -ɨw-) are already verbal, so these suffixes are not category-changing (a view that is non-controversial for -ɨw- but can be contested for -nu-). The situation is less clear for the mutative -nu-: the roots it combines with are mostly bound ones in the sense that the adjectives that they form are often also derived (see fn. 19). Yet the fact that both the semelfactive and the mutative -nu- form semantically uniform classes makes them more similar to the denominal -ow- (section 3.1) than to -a/i- and further demonstrates that the thematic suffixes in Table 1 do not form a syntactically uniform group (pace Svenonius 2004a, b, arguing that the semelfactive -nu- is merged in the same v as the theme).

The final piece of evidence against treating the semelfactive -nu- as a verbalizer comes from the fact that -nu- can combine with verbalizing suffixes: the sound-emission suffix -ot- (section 3.1.1) and the de-onomatopoeic suffix -k- (see Itkin 2007:205):

The incompatibility of -nu- with -ow- cannot therefore be derived from the assumption that they compete for the same syntactic position (v) and makes it more likely that -nu- and the secondary imperfective -iw- are both in Asp (but see Tatevosov 2013 for arguments against treating -ɨw- as the locus of semantic imperfectivity). If such were shown to be the case, it would further support the claim that thematic suffixes cannot all be treated the same.

First-conjugation verbs with the vowel [a] before the past-tense suffix fall into three different classes in function of how they look in the present: the -aj- theme (row (b) in Table 1, (52a)), the -Ø-/-a- theme (row (c) in Table 1, (52b)), and the -i-/-a- theme (row (d) in Table 1, (52c)):

All three types show no evidence of the thematic suffix in the secondary imperfective:

(53) and (54) are inconclusive: due to hiatus resolution, even if present, the vowel [a] would be deleted before the vocalic secondary imperfective suffix. (55), however, is more telling. If the underlying representation of this thematic suffix is -aj-, it would not give rise to hiatus and its surface absence from the secondary imperfective would argue that it is not present there underlyingly. Conversely, if the underlying representation of this suffix is -a-, the question arises why the hiatus created by the thematic suffix -a- and the secondary imperfective suffix -ɨw- is not resolved by the insertion of [j] (as in the present tense) or by the insertion of [w] (as in secondary imperfectives of second-conjugation e-verbs, deadjectival -ej- verbs, or verbs with the roots -da[j]- ‘give’, -zna[j]- ‘know’, and -sta[n]- ‘become’, illustrated in (56)).

However, the disappearance of the thematic suffix -aj- from the secondary imperfective has been given yet another explanation.

As discussed in section 4.1, several attempts have been made to account for the allomorphy of the secondary imperfective suffix. One such strategy can also explain the lack of the thematic suffix -aj- in the secondary imperfective by assimilating -ɨw- to -aj-. Coats (1974), Feinberg (1980) and Enguehard (2015; 2017) propose that the -ɨw- allomorph of the secondary imperfective suffix and the thematic suffix -aj- share the underlying representation, which is -va- for Enguehard (2017) and -aj- for the other three works. To explain how the underlying -aj- is turned into -ɨw-, Coats (1974) and Feinberg (1980) rely on a stipulative readjustment rule, while Enguehard (2015; 2017) offers a phonological derivation that also explains the concomitant [o]/[a] change in the verbal root ((38), (53)) by the reassociation of the first (thematic) [a]:33

Setting aside the fact that these analyses lose the intuition that [v] is epenthetic, what is the semantics of both instances of -aj- (-va- in (57))? Suppose first that the first -aj- (putatively turning into -ɨv-) is a verbalizer, i.e., contributes some category-changing semantics.34 In this case, however, the second instance of -aj- cannot be identical to it (as imperfectivization does not change the meaning of the verb or its category).35

Two possible alternatives are that -aj- corresponds to contentful aspectual (imperfective) heads in both positions (58a) or to themes, i.e., meaningless exponents on v and Asp (58b); in this latter case Enguehard’s (2017) floating [I] can be the exponent of the secondary imperfective suffix. While neither of these structures leads to interpretational issues, the question then arises why the -ɨv- suffix also appears with the second-conjugation suffix -i- (38)–(39).

Indeed, if the suffix -i- in (38) is the second-conjugation aspect head, as in (58a), or its theme, as in (58b), then only one (higher) -aj- is expected and no -ɨw- is predicted to appear, contrary to fact. If the suffix -i- is a verbalizer and what turns into -ɨw- in the secondary imperfective is aspect (58c) or theme (58d), Enguehard’s (2017) underlying -va- can no longer be maintained (or we would find the incorrect *kormivɨvatʲ in the surface representation):

However, an underlying -aj- as the underlying representation for the surface [ɨv] after Coats (1974), Feinberg (1980) and Enguehard (2015) is also not unproblematic when primary imperfective i-verbs (36)–(37) are considered. If their structure is as illustrated in (58e) or (58f), why is the underlying -aj- not pronounced?

Even supposing that an answer to this question can be found, a major difference between the thematic suffix -i- and of the thematic suffix -aj- would still persist: given that [ɨv] appears in secondary imperfectives of i-verbs (58e-f), even if [ɨv] is an allomorph of -aj-, the status of the thematic suffix -i- ends up different from that of the thematic suffix -aj-.

From my standpoint, therefore, two most empirically and theoretically adequate accounts of the interaction between the secondary imperfective and thematic suffixes can be proposed. The first one would be that while -i- is a verbalizer and therefore must be present in the secondary imperfective, -aj- isn’t and therefore must not. The second one (maintaining the hypothesis that the two suffixes have the same morphosyntactic status) is that the underlying representation of the thematic suffix -aj- is -a-, which is deleted before the vocalic secondary imperfective suffix -ɨw-, triggers glide insertion before the vocalic present-tense suffix -e- and potentially causes the deletion of the vowel in the following passive past participle suffix -en- (see fn. 4):

While (59c) is needed to account for the behavior of all [a] thematic suffixes before the PPP suffix, to maintain the hypothesis that the thematic suffix -aj- is present in the secondary imperfective just like the thematic suffix -i-, the contrast between (59a) and (59b) needs to be explained. Importantly, it is not just the lack of glide insertion that needs to be explained: an alternative would be the use of the -w- (surface [v]) secondary imperfective allomorph, which is regularly used with the deadjectival suffix -ej- (50) and occasionally, with the unproductive second-conjugation thematic suffix -e- (35). In the absence of such an explanation, the two thematic suffixes cannot have the same morphosyntactic status,36 and since -aj- appears after the secondary imperfective suffix, it can be reasonably assumed to be merged in a higher position.

Additional evidence for treating -aj- as a high suffix comes from the small, closed class of unprefixed perfective verbs. While most of them take the thematic suffix -i- (e.g., rešitʲ ‘to decide’), there are also unprefixed perfectives without a thematic suffix (e.g., pástʲ ‘to fall’), with the suffix -a-/-i- (e.g., obʲazátʲ ‘to oblige’) and with the suffix -ej- (e.g., odolétʲ ‘to overcome’). Yet the only verb in this class with the thematic suffix -aj- (pojmátʲ ‘to catch’) can be analyzed as made perfective by a prefix (added to a cranberry root). Since -aj- is the only thematic suffix that can follow the secondary imperfective suffix, the hypothesis that it is too high to also precede -ɨw- or to yield unprefixed perfectives seems more reasonable than an attempt to explain what forces it to be absent in those environments.

To summarize, there is no a priori evidence that the thematic suffixes -a-/-i-, -aj- (whatever its underlying representation), and -a- are present before the secondary imperfective suffix -ɨw-. They clearly differ in this respect from the thematic suffix -ej-, which is always retained in the secondary imperfective (50), as well as from -i- (which is sometimes absent) or -e- (for which, as argued in section 4.2, there is evidence for its presence in the secondary imperfective and no evidence for its absence there). These distributional facts seem to suggest that thematic suffixes differ also in their ability to appear in the secondary imperfective, which brings up the question of the semelfactive (14) and the mutative (15) -nu- suffixes, which, as mentioned in section 3.2, are also obligatorily absent from the secondary imperfective.

Starting with the easier case of semelfactives, nearly all semelfactive verbs have imperfective variants with a different verbalizer or thematic suffix (60)–(61) and it is those that are used to form their secondary imperfectives (in -ɨw-).

The aspectual contribution of the semelfactive -nu- is unquestionable. Irrespective of whether it and the imperfective -ɨw- are both in v, as Markman (2008) argues, or in Asp, as suggested by Gribanova (2013) and supported by the evidence from -ot- and -k- verbs (section 3.2.3),37 they would be predicted to be in complementary distribution.

Stems combining with the mutative -nu- suffix form their secondary imperfectives only with the zero allomorph (15), just like first-conjugation [ej]-verbs discussed in section 3.3, and the mutative -nu- suffix is never present:

The absence of a verbalizer in the secondary imperfective is unexpected: an aspectual suffix cannot combine with a non-verbal stem. The alternative that the thematic suffix -aj- functions as a verbalizer (or a theme on a null verbalizer) with a missing unprefixed verb should also be discarded: prefixed primary imperfectives are systematically perfective.

Should it therefore be concluded that the presence or absence of a given thematic suffix in the secondary imperfective is not by itself an indication of its morpho-syntactic status? If the answer is yes, then the different behavior of the second-conjugation suffix -i- and the first-conjugation suffix -aj- says nothing about their status as a verbalizer.

Alternative explanations are, however, available for the mutative -nu-. Firstly, since many mutative verbs lose this suffix in the past tense (13) for no obvious reason, the explanation could lie in very superficial allomorphy. Secondly, it can be argued that the stative semantics of -nu- is somehow incompatible with the secondary imperfective (see fn. 25 for a semantic distinction between nu-verbs and ej-verbs). Yet a third alternative is that the mutative -nu- is also a kind of Asp, which would place it in complementary distribution with the secondary imperfective suffix. No definite conclusion is therefore to be drawn from this suffix.