2.1 Secondary Imperfective

In Russian, verbal aspect is morphologically expressed. Perfective and imperfective verbs are derived as follows (see Karcevskij 1927: 87).³ First, unprefixed verbs are usually imperfective (first column of (2)) (Garde 1980: 379). Perfective verbs can be derived from these by prefixation (second column of (2)).⁴ Compared to its stem, the resulting perfective verb can have: i. the same lexical meaning (2a); or ii. a new lexical meaning (2b).

The perfective verbs with a new meaning (second column of (2b)) have no imperfective counterpart yet. As a consequence, a new imperfective form (called Secondary Imperfective) can be derived with one of the two imperfectivizing suffixes: -a or -iva (Garde 1980: 385–386). Secondary Imperfective verbs in (3a) are derived with -a; and Secondary Imperfective verbs in (3b) are derived with -iva (or its variant -yva).

I focus on the underlying representations of -a and -iva in Sections 2.1.1 and 2.1.2 respectively.

2.2 Long Adjectives

Before analyzing the status of -i- in -iva, I discuss another case of morphological item with no clear status, this time in the realm of adjectival inflection.

Russian shows two adjectival forms: Short Adjectives (SA) and Long Adjectives (LA).⁸ Schematically, Short Adjectives end with -(V) and (nominative) Long Adjectives end with -Vj(V), where V are vowels that vary depending on gender and number features. Examples are given in Table 1.

The phonetic realization of SA suffixes is illustrated in Table 2. The phonological representation of each suffix can be deduced from its realization in stressed context (i.e. where vowels are not reduced).

The realizations of LA suffixes are given in Table 3. The phonological representation of the vowels that precede -j- can generally be deduced from their realization in stressed context. The only exception is the vowel of the masculine suffix -ój/-ij: [ɨ] is not an expected realization of /o/ in unstressed context (Avanesov 1968: §13.2). Thus the representation of this suffix is ambiguous (I leave this issue aside for the moment). It is /-ój/ or /-ij/ depending on the accentual context. Except the case of masculine -ój/-ij, the vowels that precede -j- are identical to those found in SA suffixes (compare Table 2).

The representation of vowels that follow -j- (i.e. [ə] and [i]) is trickier to determine because these are never stressed. Due to vowel reduction, [ə] can be the realization of /o/ or /a/, and [i] can be the realization of /i/ or /e/ (see Avanesov 1968: 64–67, §18.3–18.5). Thus, the phonological quality of these vowels cannot be deduced from their realization. However, if we look to the accusative counterpart of -aja (e.g. nóv-uju ‘new-F.SG.ACC’), we can observe that both vowels in -aja are replaced by u. Given that -u is an accusative marker that always alternates with a nominative marker -a in adjectives and nouns (see Garde 1980 for paradigms), it can be supposed that the feminine suffix -aja is made of two -a markers (i.e. /-aja/). Inductively, it can now be assumed that the neuter suffix -oje/-eje is also made of two -o markers (i.e. /-ojo/), and the plural suffix -ije is made of two -i markers (i.e. /-iji/).⁹

In sum, the vowels that surround -j- are the same gender/number markers that are found in Short Adjectives. Thus, it is reasonable to parse -Vj(V) into -V₁-j-V₂, where -V₁- and -V₂ are gender/number markers and -j- is an LA marker (see Table 4 for illustration).¹⁰ Note that the masculine suffix is the only suffix that does not show the same marker on the left and on the right of -j-. I will return to this issue in Section 4.

My main concern is the morphological status of the LA marker -j-. Etymologically, -j- was a clitic expressing a definite value in Old Slavic (Vaillant 1948: §79). As such, it was followed by its own gender/number marker. But, in Modern Russian, the definite value of -j- disappeared and the long form of adjectives came to be unmarked (Comtet 1997: 136). In this context, it is dubious to analyze -j- as a clitic. As a consequence, the following question arises: how to account for the fact that a non-clitic -j- involves its own gender/number marker?

To conclude this section, I pointed out that -iva and LA suffixes can be parsed into -i-va and -V₁-j-V₂ respectively, with -i- and -j- having an enigmatic morphological status. The following sections aim to puzzle out this issue. I show that the insertion of -i- and -j- are motivated by the morphosyntactic context.

3.1 -j- and the Epenthesis Hypothesis (Garde 1972)

Garde (1972) proposes to analyze -j- as an epenthetic consonant. He bases his hypothesis on the following observation: hiatuses are never found in Russian endings (Garde 1972: 473). Thus, he argues that the occurrence of -j- in -aja (as well as in other endings) could be due to a repair mechanism in hiatus contexts ((7); Garde 1972: 375).

This process relies on phonological information. However, Garde (1972) argues that the occurrence of an epenthetic /j/ is not limited to hiatuses. As an example, /j/ also occurs in the final position of the masculine counterpart of (7), see (8). Here, the emergence of /j/ does not seem to result from a hiatus. Thus, Garde (1972: 378–380) needs to define a second condition for epenthesis, namely that /j/ also surfaces between a final vowel and a zero morpheme (we saw in Table 4 that -ój/-ij involves a -Ø marker). In other terms, Garde introduces morphological structure into his analysis.

Garde’s analysis implies that the occurrence of -j- can be predicted from its context. Thus, it should not be included in the lexical representation. But he also mentions (Garde 1972: 378) that an epenthesis conditioned by a zero marker does not result from phonology only. Something should be said about the relation between the motivation of -j- and the morphosyntactic structure. The issue is not developed in Garde’s analysis. In this paper, I propose that the insertion of -j- is motivated by an identity avoidance constraint. But before moving on to my proposal (Section 4), I address the hypothesis of Coats (1974) and Feinberg (1980) about the status of -i-.

3.2 -i- and the Reduplication Hypothesis (Coats 1974; Feinberg 1980)

Most previous proposals suppose that -i- is the realization of a new lexical item that distinguishes -iva from -(v)a: e.g. /-ɨv/ in Flier (1972) or /-ʊ/ in Matushansky (2009). These studies fundamentally assume that this component is not predictable.

A competing view was introduced in Coats (1974) and later developed in Feinberg (1980). These argue that -iva is a sequence of two -(v)a suffixes (Coats 1974: 35–36; Feinberg 1980: 149–152).¹¹

This analysis is based on the following observation: i. Secondary Imperfective verbs ending with -(v)a do not truncate the verbal suffix of the stem they are derived from (see (9a)); but ii. Secondary Imperfective verbs ending with -iva curiously do ((9b); Feinberg 1980: 148).¹²

Coats and Feinberg propose to make this morphological truncation unnecessary. They point out that there is no need for a new imperfectivizing suffix stored in the lexicon if we suppose that the imperfective verb in (9b) is built on the same model as verbs in (9a): i.e. -iva is the realization of -(v)a preceded by the suffix of the perfective stem -(v)a (see (10)).¹³

However, this hypothesis (though valuable) faces the following issue: some verbs ending with -iva are not derived from a perfective stem ending with -(v)a (see (11)). In this case, the authors assume an ad hoc reduplication of the imperfectivizing suffix -(v)a to account for the emergence of -iva (Coats 1974: 35–36; Feinberg 1980: 149).

For that reason, this analysis has been quite unpopular so far. In what follows, I introduce an argument from Distributed Morphology in support of the Reduplication Hypothesis. I show that a reduplication of -(v)a in verbs like in (11) is not that ad hoc. It is confirmed by an independent argument, even though the motivation of this process still needs to be fully elucidated.

3.3 An argument from Distributed Morphology (Enguehard 2015a; 2016)

This sub-section outlines an independent argument introduced in Enguehard (2015a, 2016) in support of the Reduplication Hypothesis.¹⁴

In Enguehard (2015a, 2016), I showed that the Reduplication Hypothesis is independently confirmed by a morphophonological operation involved in verbs ending with -iva. This operation consists of an a-mutation of last root vowels /o/ (12).¹⁵

After having argued that -iva is situated higher than v in the morphosyntactic tree (13),¹⁶ I pointed out that the o/a alternation conditioned by -iva is theoretically unexpected. Given that -iva is situated in the external domain (i.e. outside the first phase), PIC (v.1) implies that it cannot access the root (circles in (13) represent spheres of possible interaction). Accordingly, phonological operations are impossible between -iva and the root.

However, phonological operations are possible between: i. -iva and v; and ii. between v and the root. Thus an indirect interaction is possible between -iva and the root if we assume that v involves a material that: i. is phonologically deleted by -iva and; ii. triggers a modification of the root.¹⁷

A similar indirect interaction between -iva and the root was already proposed by Halle (1963: 369–371), Flier (1972: 239–240) and Coats (1974: 33) in order to account for perfective verbs ending with -i that show a palatalization of the last root consonant in the Secondary Imperfective form (14).

Following their analysis, the suffix -i of the perfective verbs in the first column of (14) is maintained in the underlying representation of their imperfective counterpart (second column of (14)). On the surface, -iva deletes -i, which is realized as a palatalization of the preceding consonant (15) (see also footnote 12).

I argue that the o/a alternation results from the same mechanism: -iva deletes a -(v)a suffix expressing v in (13). As a consequence, -(v)a is phonetically realized as an a-mutation of the root vowel /o/ (16) (for a detailed phonological derivation, see Enguehard 2015b; 2016: 595–598).

This analysis implies that all verbs showing an a-mutation of /o/ involve a deleted -(v)a in their underlying structure. In other words, the reduplication of -(v)a in the representation of verbs like s-práš-iva-t’ ‘to ask’ is confirmed by an independent argument.

It is important to mention that this assumption raises an issue about the representation of the palatalization in (15). In the classical analysis, it is assumed that the vowel deletion results from a constraint banning hiatuses (Jakobson 1948: 159–160; Svenonius 2004a: 180). However, if we assume that the underlying representation of s-práš-yva-t’ is /s-prosʲ-i-va-va-tʲ/, we should not expect any vowel deletion. Note that this issue concerns the representation of the verbal suffix -(v)a, not the Reduplication Hypothesis. If we assume (contra Garde 1972) that -(v)a is represented by /-aj/ (as in Coats 1974 and Feinberg 1980), then /s-prosʲ-i-aj-aj-tʲ/ correctly predicts s-práš-yva-t’. Nevertheless, I recall that the representation /-aj/ is not without problems. It requires a readjustment rule in order to derive the emergence of -v- (see Section 2.1.1). Similarly, it is possible to assume a readjustment rule to account for the unexpected vowel deletion before /-va/. Instead, I suggest a representational approach of this issue. I assume that -i and -(v)a share the same skeletal slot (framed in (17)).¹⁸ The suffix -i surfaces as a vowel (17a), except when it co-occurs with -(v)a (17b). In this case, due to space limitations, it links to the preceding consonant.¹⁹ This proposal is a representational equivalent – in terms of circularity – of the readjustment rule proposed in Flier (1972).

Now assuming the Reduplication Hypothesis proposed by Coats (1974) and Feinberg (1980), the occurrence of -i- turns out to be predictable: it appears in cases where -(v)a is reduplicated. This reduplication should not be analyzed as a surface (phonological) phenomenon (the analysis of Coats and Feinberg involves two underlying -(v)a suffixes), but as an underlying (morphosyntactic) configuration. In (18), it is represented with the repetition of two identical nodes (here: v).²⁰

But how to account for the emergence of -i-? It was proposed in Coats (1974: 34) and Feinberg (1980: 147–148) that -i- results from a phonological raising of the first vowel /a/. It is important to note that such a change is unusual in this context (/ava/ is perfectly well-formed, e.g. davát’ ‘give’). In order to support this unexpected phenomenon, Coats draws a parallel with the (previously mentioned) height contrast between stressed /ó/ and unstressed /i/ in -ój/-ij (see Table 3).²¹ The parallel proposed in Coats (1974) is intriguing, but the specific raising rule he proposed has not been explained or confirmed until now. Thus it is doubtful whether this phenomenon can be analyzed as a strict phonological process, without any reference to morphological information.

To sum up this sub-section, the Reduplication Hypothesis is based on the observation that -iva unexpectedly triggers a morphological truncation of -(v)a. Coats (1974) proposes to avoid this truncation issue by assuming that -iva results from a succession of -(v)a suffixes: i. the first belongs to the perfective stem; ii. the second is an imperfectivizing marker. Then, in order to account for the presence of -iva in verbs derived from stems with no -(v)a, Coats suggests a reduplication of the imperfectivizing marker. This configuration is supported by the a-mutation co-occurring with -iva (Enguehard 2015a). Finally, Feinberg (1980) extends this reduplication mechanism to all verbs ending with -iva. Such an assumption supposes that the emergence of -i- is morphologically conditioned by a specific reduplicated structure.

Of course, it is possible to assume that the suffix responsible for the a-mutation is distinct from the suffix realized in -iva. For instance, Gribanova (2015) suggests that the a-mutation is due to a set of melodic features present in the Asp head of the verb. I do not aim to exclude this possibility. However, one can legitimately wonder how to validate or invalidate such a lexically specified set of features which is not encountered elsewhere in Russian morphology. By assuming a reduplication of the imperfectivizing marker, the present analysis takes risks purposefully. The factors conditioning -iva and the a-mutation being structural, not lexical, they can be validated or invalidated by external data.

To conclude this section, I addressed analyses proposing that -i- and -j- can be deduced from their context. Accordingly, they should not be part of the underlying structure. However, I pointed out that their derivation cannot be deduced from phonology only. They both involve a certain degree of morphological conditioning. Now, I move on to my proposal. I will show that the Reduplication Hypothesis can motivate the emergence of -i- and -j-.

4.1 Identity Avoidance

Hoekstra (1984), Grimshaw (1997) and Van Riemsdijk (1998) (among others) shed light on an identity avoidance constraint banning the succession of two identical heads in morphosyntactic trees. This constraint is sometimes referred to as an OCP effect applying to the syntactic structure (e.g. Van Riemsdijk 2008). Just like Obligatory Contour Principle in phonology, OCP in morphosyntax triggers repair mechanisms like deletion and/or addition of material (see Grimshaw 1997 about clitics in Italian and Spanish).²²

Returning to the representation of -iva in (18) above, one can notice that it involves a head v selecting another head v. Following the identity avoidance constraint, such a configuration should normally be ill-formed. This theoretical ill-formedness of the representation in (18) motivates the occurrence of -i-. The emergence of such a phonologically unpredictable segment in a repetition context is very similar to reduplication phenomena with a fixed segmentism (e.g. table-shmable) addressed in Yip (1992, 1998) and Alderete et al. (1999), among others. These authors argue that the fixed segmentism at issue is a repair mechanism motivated by a constraint banning repetition. This constraint is what makes the representation in (18) ill-formed. Thus, my hypothesis is that -i- is, on the surface, a sort of fixed segmentism, just like -shm- in table-shmable.

But what is the underlying status of this fixed segmentism in Russian? In the present case, -i- has three properties: i. it is phonologically unpredictable (see Section 3); ii. it is semantically null; and iii. it is hypothetically conditioned by a structural requirement. An item with the same three properties is discussed in Faust (2011: 222–223; 2013: 23–29). Focusing on the morphonology of Modern Hebrew, Faust identifies an unexpected segment /i/ in the feminine form of loanwords (e.g. astronawt-it ‘astronaut-F’). The occurrence of this segment /i/ constrats with the usual form of the feminine marker -a(t).²³ Thus, what does motivate its insertion? The segment /i/ seems not to be directly related to gender. Indeed, the feminine form of derived adjectives – which always show a feminine maker -it – also have a masculine counterpart ending with -i (merxav ‘space’ → merxav-i (M), merxav-i-t (F) ‘spatial’). In this case, it is possible to parse -it into -i-t. Faust argues that the segment /i/ found in the feminine form of loanwords is the same morphological item as the segment /i/ observed in derived adjectives. Accordingly, he proposes an analysis where the function of this /i/ is to license certain derivational processes. More specifically, Faust assumes that: i. loanwords are borrowed with their category; and ii. categories and the feminine marker select only roots. Following these two assumptions, the expected underlying structures of denominal adjectives (i.e. [_a[_nmerxav]-it]) and the feminine form of loanwords (i.e. [[_nastronawt]-it]) should be ill-formed. As a consequence, Faust proposes that /i/ is the realization of an expletive root node inserted in order to license a merge of heads which would normally be incompatible.²⁴

I argue that Russian -i- is similar to the expletive root of Faust (2011, 2013). It is inserted in order to license a merge of heads that would normally be incompatible. The only difference with Faust’s proposal is that the selectional incompatibility in (18) specifically results from Identity Avoidance. I assume -i- to be the realization of a root √I (the front/palatal element of Element Theory, see Kaye et al. 1985) with no specific meaning. This root is inserted in order to repair the ill-formed representation in (18) (compare with (19)). From this perspective, -i- turns out to be the morphosyntactic equivalent of an epenthesis: i.e. a sort of buffer surfacing in order to satisfy a structural requirement (here: Identity Avoidance).

At this stage of the analysis, the occurrence of -i- is motivated in morphosyntax. To validate the analysis of -i- as an epenthetic root inserted between two identical heads, I now show that the same root is inserted in another suffix we have examined: -Vj(V). This case will confirm that the insertion of -i- is conditioned by reduplication in general, not by the lexical properties of the Secondary Imperfective suffix in particular.

4.2 Another case of reduplication: Long Adjectives

I now go back to LA suffixes. I argue that these are another case of reduplication involving an expletive root √I.

We already saw in Section 2.2 that most LA suffixes show two identical gender markers with an intervening consonant -j-. The masculine suffix -ój/-ij is the only exception. Given that we defined the masculine marker of short adjectives as -Ø, we should not expect the occurrence of a vowel preceding -j-. In order to account for this issue, it is necessary to refer to the history of Russian. In Old Slavic, the masculine marker was a vowel called back yer (traditionally represented by ъ). In Modern Russian, back yers disappeared in most cases, except when followed by another yer (see Lightner 1965). However, Lightner (1965), Melvold (1989) and Scheer (2001) agree that these back yers are maintained as floating features in the underlying representation of Russian words. More specifically, Hamilton (1980) and Melvold (1989: 35, 50) even argue that the masculine marker never ceased to be an underlying /-ъ/. Based on this assumption, I propose that all LA suffixes involve a reduplication of their SA counterpart (see Table 5). Regarding masculine /-ъ-ъ/ → -ój/-ij, only the first yer is realized (because it is followed by another yer).

In the following sub-section, I show how the reduplication can derive the intervening segment -j- in parallel with -i-.

4.3 Phonological derivation

In this sub-section, I aim to show how the expletive root √I is phonologically derived as -i- in -iva and as -j- in LA suffixes. My proposal follows the CVCV framework introduced in Lowenstamm (1996).

5.1 Semantics of reduplication

As we saw in Section 2, the semantic value of -iva is assumed to be imperfectivity. But the semantic value of LA suffixes is more obscure. I argue that it is to some extent similar to imperfectivity.

First, it must be noticed that long and short adjectives contrast only in predicate position. This contrast can take various forms (see Comtet 1997: 136 for details). However, short adjectives generally express a restriction of time or quality. As an example, compare (26a) and (26b). In (26a), the short form bólen expresses a state that is limited in time. In (26b), the long form bol’n-ój denotes a chronic state.

This contrast between stage-level (26a) and individual-level predicates (26b) is comparable to the contrast expressed by Russian grammatical aspects. On the one hand, both stage-level and perfective express a state or an action over a delimited period of time. On the other hand, both individual-level and imperfective express a state or an action over an undelimited period of time. Such a correlation between individual/stage-level predicates and aspect was already explored in Luján (1981) and Camacho (2012) for Spanish ser/estar (see also Arche 2012 for criticism). The contribution of the present analysis to this issue is to connect two properties of individual-level adjectives and imperfective aspect in Russian: i. they both express an undelimited period of time; and ii. they can be both realized by reduplication.

The accurate semantics of this reduplication requires further investigation (the present paper focused on a morphonological reasoning only). An iconic relation between imperfectivity and reduplication was already pointed out in several studies, including Naylor (1986), Zack (1994) and Mattes and Schwaiger (2014). But it is important to mention that reduplication in -iva seems not to show any aspectual contrast with its non reduplicated variant -va (both express imperfective).

Nevertheless, this observation is not absolutely true. In unprefixed verbs, -iva can have a specific iterative value (e.g. xáž-yva-t’ ‘to visit several times’) (Garde 1980: §605), which is not shared by -(v)a. Moreover, Karcevski (1927: 93) analyzes the historical spread of forms with -iva as a the result of a mechanism repairing an ambiguity between -(v)a and a (phonologically similar) theme vowel -a. It is thus possible that reduplication does not provide any semantic value in verbs, but reafirms the aspectual value of the suffix -va.²⁹

5.2 Quality of the expletive root

Given that √I is a root, it is supposed to be stored in the lexicon as such. Accordingly, its quality cannot be predicted phonologically. But a question arises: do we have independent arguments to confirm that the intervening element |I| found in -iva and -Vj(V) is a lexical item?

Interestingly, the phonological form of the expletive root √I corresponds to the phonological form of the Russian coordinating conjunction i ‘and’. Moreover, the function of the coordinating conjunction is similar to the function of the expletive root: both coordinate morphosyntactic nodes. More specifically, i (i.e. ‘and’) coordinates items exclusively belonging to the same category (e.g. compare Máša i Pë́tr ‘MaryN and PeterN’ and *Máša i xorošó ‘MaryN and goodA’). This is exactly the function of √I in reduplication processes. Thus it could be proposed that the Russian coordinating conjunction is a root √I which can intervene between phrases, words or bound morphemes. Given that √I has no semantic value in -iva and -Vj(V), the present supposition opens perspectives for the general analysis of coordinating conjunctions. These could be items inserted only under structural requirements, and not for lexical purposes.

A similar hypothesis was proposed in Fábregas (2005: 238–242) to account for the linking element /i/ occurring in Spanish compounds. However, unlike Spanish, √I is not a usual linking element in Russian compounds. These generally involve a vowel -o- (e.g. sam-o-vár : sam+var ‘self+boil’).³⁰ In the meantime, it could be supposed that √I and -o- have different functions: e.g. √I coordinates identical categorial nodes, when -o- coordinates roots. The relation between reduplication with an expletive root and compounding with a linking element deserves to be addressed in a further study.