√ov IS IN THE AIR: THE EXTREME MULTIFUNCTIONALITY OF THE SLOVENIAN AFFIX OV

√OV IS IN THE AIR: THE EXTREME MULTIFUNCTIONALITY OF THE SLOVENIAN AFFIX OV In this paper we consider several instances of the Slovenian affix ov, which surfaces in many, apparently unrelated contexts. Here we focus on (i) ov in verbs, where it can act as an imperfectivizer or a verbalizer, (ii) ov found in possessive adjectives and kind adjectives derived from nouns, (iii) ov which precedes the adjectiviser (e)n in denominal adjectives, and (iv) ov in nominal declension (acting as a genitive case ending in dual and plural or as a dual/plural augment). Building on the observation that certain affixes function either as inflectional or as derivational (see Simonović and Arsenijević 2020), and working within a Distributed Morphology approach which postulates that derivational affixes should be analyzed as roots (e.g. Lowenstamm 2014), we argue for a single multifunctional ov. This ov is a potentially meaningless root that can take as a complement other roots (thus forming a “radical core”) or phrases, resulting in different structures and consequently different stress patterns and meanings, but can also act as an Elsewhere allomorph, whose insertion is guided by an interplay of phonological and morphological constraints.

by Lowenstamm (2014), by answering the question what entry or entries for ov the Encyclopedia needs to contain.
In what follows, we first give an overview of the account which treats affixes as roots in section 2. Then section 3 focuses on the morpheme ov in contexts which would be traditionally classified as derivational: derived verbs, possessive and kind adjectives and adjectives with the affix ov(e)n. Section 4 focuses on ov in nominal declensions. Section 5 is the conclusion.

DERIVATIONAL AFFIXES AS ROOTS
One of the postulates of Distributed Morphology is that roots do not carry information about the category of the word (Halle/Marantz 1993, 1994Marantz 1996). Rather, the category of the word is determined by a categorial head. These heads can either have no phonological content (e.g. in travel, where the root √travel is combined with a mute categorizer v or n) or they can be phonologically realized, as is the case in most derivational affixes (e.g. in traveler). Recently, however, Lowenstamm (2014) showed that the assumption that derivational affixes are exponents of categorial heads runs into problems and instead proposed that derivational affixes are a subset of roots. Put differently, Lowenstamm (2014) proposes that derivational affixes are like roots in that they do not carry information about their category, but receive a category by being merged into the complement position of a categorial head. This move entails a separation between phonological/semantic content, on the one hand, and categorial heads, on the other, for both 'traditional' roots and affixes. Roots (including derivational affixes) have phonological content and/or meaning, while categorial heads are mute and have no semantic contribution. In this revised picture, affixes such as the English ic (which can be found in adjectives such as atomic below) are not analyzed as in (1a), but rather as in (1b), both of which taken form Lowenstamm (2014: 232, (6)).

IC ATOM
As (1b) already indicates, a crucial distinction between 'traditional' roots (e.g. √travel) and affixes comes from the affixes' requirement to take complements. Crucially, affixes can select roots, as ic in (1b), or categories, as is the case for ness, shown in (2) below.
(2) a. nP One argument for the 'affixes are roots' treatment comes from the observation that certain derivational affixes, such as the English ic, surface under different categorial embeddings, as shown in (3). Note that in this respect affixes are no different from regular roots such as √work or √travel.
(3) a. n: comic, academic, basics, sceptic b. a: comic, academic, basic, atomic One consequence of treating derivational affixes as roots is the necessity of storing them as such, an issue not explicitly tackled by Lowenstamm. Simonović (2020) argues that the categorial embeddings of affixal roots need to be stored in the Encyclopedia. Storing the categorial embedding of roots is classically assumed in DM for 'traditional' roots (such as √cat), as summarized in the title of Marantz (1996) 'Cat is a phrasal idiom'. Extending this to affixal roots, Simonović suggests that the categorial embedding of affixal roots should not be stored for each derived word, but once in an abstract schema, which also specifies the selectional behavior of the root. The Encyclopedia entries for the nominal and adjectival ic, and for ness would look as represented in (4).
(4) a. nP These Encyclopedia entries raise the further question of their stored meaning. Simonović suggests that, unlike 'traditional' roots, affixal roots are potentially stored without any meaning. This is compatible with the observation that the meaning of affixal roots is often extremely vague or unpredictable (especially of those that appear under different categorial embeddings). For example, as Creemers et al. (2017: 75) note, "[affixes such as ic] have, among other interpretations, meanings such as 'of,' 'relating to,' 'engaged in,' and 'connected with." Assuming (nearly) meaningless roots then leads to one further question, i.e. why such roots do not appear in all environments without any limitations. The tentative answer is that their insertion generally leads to well-formed items, but some items are dispreferred if a less complex structure is available. However, as will be shown, certain (nearly) meaningless roots do indeed appear in a variety of contexts, including inflection. One example of such an affixal root is ov, which we will consider in section 3.
The second argument for the root analysis of affixes comes from English stress. While we will not go into the details of Lowenstamm's account here, suffice it to say that under his approach the combination of affixes-as-roots analysis and phrasal spellout can account for the prosodic contrast between stress-affecting affixes, (e.g. ity in atomicity) and stress-neutral ones (e.g. ness in atomicness). The relevant trees are shown in (5) below. The stress-affecting behavior is not essentially a property of the affix ity, but of the structure it appears in. The structures with multiple roots heading each other, termed "radical cores" by Lowenstamm, always constitute a single phase and therefore a single stress-assignment domain. Here Lowenstamm assumes that phonological rules re-apply with each new root phrase, but the same result can be obtained by having the rules apply to the whole radical core at once, on the first spell-out.
Crucially for what follows, Simonović (2020) extends this approach to Slovenian and shows how observations about word stress first made in Marvin (2003) can be accounted for if we treat affixes as roots. Marvin observes that there exists a prosodic contrast in (apparently) deadjectival ost-nominalizations, illustrated by minimal pairs such as mládost 'youngness' (prosodically faithful to the adjective mlád 'young') and mladóst 'youth, young years'. Marvin's analysis, which Simonović follows, is that mládost is a deadjectival nominalization (i.e. the more productive type, compositionally interpreted and prosodically faithful to the base adjective), while mladóst is a root nominalization (i.e. the more rare, idiomatic type, characterized by stress-shifting). For Marvin, the stress-shifting behavior of ost is a consequence of a prosodic specification on the suffix, which is only realized phase-internally. Simonović dispenses with prosodic specifications on affixes altogether and proposes that the 'idiomatic' nominalization mladóst contains a radical core (a combination of two roots) and that radical cores always receive the default stress pattern.
In Slovenian, the default stress is stem-final. Stem-final stress is also the most common prosodic pattern in the language, as established based on the stress pattern of the 3,000 most frequent nouns, verbs and adjectives. For each of these 9,000 words, we marked the stress pattern and annotated whether stress is stem-final. Items which can either have stem-final stress or another stress pattern were excluded from the count. In each of the three categories the stem-final stress pattern is by far the most common one, and a majority of words have this pattern, specifically, 63% of verbs, 70% of nouns and 73% of adjectives receive the stem-final stress.
While mladóst has a default stress pattern, mládost is a deadjectival nominalization in which the root ost takes an adjective as its complement, which naturally leads to faithful prosody. The relevant trees are shown in (6).
To sum up the theoretical background presented in this section, we assume an approach under which derivational affixes can be treated as roots with potentially little or no meaning and according to which word stress is dependent on the structure of the word. In section 3, we show how this approach successfully captures the behavior of ov in the verbal and adjectival domain in Slovenian.

SLOVENIAN AFFIX OV IN THE DERIVATIONAL MORPHOLOGY
Before proceeding to the contexts in which ov can be found, we first need to note that ov surfaces either as ov or as ev, depending on the preceding consonant. This is a purely phonologically driven allomorphy, as illustrated in the possessive adjectives in (7).
In traditional grammars, ov was in some instances taken to be a morpheme, for example in possessive adjectives as in examples such as (7) above, but it was primarily treated as a part of larger morphemes. For example, Toporišič (2000: 184) lists several affixes that are used to derive collective nouns. Among these we can find je (as in cvet-je 'flowers', related to cvet 'flower') but also evje and ovje (as in borovničevje 'blueberry plants', related to borovnica 'blueberry', and cvetovje 'flowers', related to cvet 'flower'), stvo (as in članstvo 'members', related to član 'member') and ovstvo (as in judovstvo 'Judaism', related to jud 'Jew'), while ov is not listed as a morpheme in this context. Notably, such treatment suggests that stvo as a morpheme has nothing in common with ovstvo. Marvin (2003) makes a similar point regarding morphemes ec and je in deverbal nominalizations.
In order to avoid the undesirable reduplication of affixes, we assume as a null hypothesis that all instances of ov are instances of the same derivational affix, which can also combine with other derivational affixes. Especially illustrative of this behavior are the pairs with and without ov which have slightly different meanings and can be found among denominal adjectives (8), collective nouns (9) and denominal nouns (10). In each of these contexts, it is unclear what the semantic contribution of ov is or what conditions its presence.
(8) a. jezik-ov-en b. jezič-en language-ov-en language-en 'related to language' 'related to tongue' (9) a. grm-ov-je b. sad-je bush-ov-je fruit-je 'shrubbery' 'fruit' (10) a. bank-ov-ec b. obraz-ec bank-ov-ec face-ec 'banknote' 'form' In the above contexts, the affix ov is sandwiched between other categorized elements, therefore revealing no category with which it is associated. There are, however, instances of ov which would be analyzed as categorizers in classical Distributed Morphology. Such is the ov in possessive adjectives illustrated in (7). A similar analysis is plausible for denominal verbs in (11).
(11) a. pot-ov-a-ti (cf. pot) b. glas-ov-a-ti (cf. glas) travel-ov-theme-inf vote-ov-theme-inf 'to travel' 'path, travel' 'to vote' 'voice, vote' There appear to be no clear cases in which ov functions as a nominalizer (but we will argue that such cases actually exist in the inflectional domain in Section 4).
In sum, the derivational uses of ov point towards ov being an extremely multifunctional affix, comparable to the English ic, i.e. a root that has little to no semantic contribution, which can appear in various categorial contexts.
As is clear from the examples above, the contexts in which ov shows up are extremely numerous and analyzing all of them would go beyond the scope of this article. We therefore made a representative selection of the derivational uses of ov. In 3.1, we turn to the verbal ov, which functions as a verbalizer and as an imperfectivizer. In 3.2 we offer an analysis of the adjectival ov in denominal adjectives. Finally, in 3.3 we turn to a case where ov itself reveals no category, as it is followed by the adjectivizing affix -n.
In each case our discussion will be guided by the question what the prosodic and semantic effects of ov are and what needs to be stored in the Encyclopedia in order to obtain these effects.

The affix ov in verbs
Before turning to the combinatorial possibilities of the verbal ov, it should be pointed out that ov in verbal contexts is characterized by allomorphy. The version ov shows up in the non-finite forms, where it is accompanied by the theme vowel a, (12a), whereas the version u shows up in the non-finite forms, where it is accompanied by the theme vowel je (12b).
(12) a. pot-ov-á-ti b. pot-ú-je-mo travel-ov-theme-inf travel-u-theme-1sg.pres 'to travel' 'we travel' As the examples in (12) show, the two allomorphs display different prosodic patterns: while in the non-finite forms form the theme vowel is stressed, the stem-final vowel receives the stress in the present tense. As argued by Simonović (2020), these prosodic patterns are not a feature of the morpheme ov, but rather imposed by the theme vowels, since they are also attested with other roots (e.g. or-á-ti 'to plough', ór-je-mo 'we plough').
When it comes to the categories which the verbalizing ov selects and the effects it has, there seem to be two large classes. In most cases, ov shows up as an imperfectivizer, (13), which derives imperfective verbs (13a, c) from perfective ones (13b, d).
At first sight the imperfectivizing function seems to require a separate entry in the Encyclopedia. However, as proposed by Arsenijević (2018) for Serbo-Croatian, impefectivizers can be analyzed as re-verbalizers, which reverse the value of the verbal aspect to its default, which is imperfective. If this is the case, then the verbalized ov is a good example of a universal selector, which can have any kind of element (phrase or root) in its complement. The Encyclopedia entry of the verbal ov is shown in (15), where x represents any element.
Before turning to the adjectivized ov, a remark is in order concerning the stress of ov-verbs. The stress pattern described above and imposed by the theme vowels is preferred for all ov-verbs by all Slovenian speakers, and a majority of the consulted speakers report not ever using any other stress pattern on these verbs. However, some speakers (from Upper Carniola, Lower Carniola and Ljubljana) allow the exceptional pattern with the stress on the syllable preceding ov. These speakers have realizations such as pót-ov-a-ti 'to travel' and vér-ov-a-ti 'to believe', next to more common potov-á-ti and ver-ov-á-ti. We take this exceptional stress as evidence of the incorporation of an nP into the verb (in these cases of pót 'travel' and vér-a 'faith', respectively), which enables the preservation of the nominal stress. This is not unexpected given the cross-linguistic evidence that nominal lexical stress tends to be more strongly protected than that of verbs (Smith 2011). Assuming that the incorporation of nPs is the general way of capturing exceptional (i.e. not theme-controlled) stress in Slovenian verbs leads to the postulation of some nPs which do not surface independently. In our data set var and napred in the verbs vár-ov-a-ti 'to guard' and napréd-ov-a-ti 'to make progress' are instances of such non-attested nPs. However, while var and napred are not attested nouns, they are attested (with the expected stress pattern) inside adjectives vár-en 'safe' and napréd-en 'progressive', respectively.

The affix ov in possessive and kind adjectives
As already shown in the examples in (7), ov can be found in possessive adjectives, which are derived from either masculine or neuter animate nouns (but not feminine, which take the suffix -in). In (16)  Given the two groups of adjectives, which behave differently with respect to the gender of the noun they are derived from, the first option to be considered is that (i) possessive adjectives are derived from nPs (18a), and (ii) kind adjectives are derived from roots, (18b).
However, as we have seen in section 2 and as proposed in Simonović (2020), radical cores (i.e. roots directly adjacent to roots) trigger default stress, which in Slovenian is stem-final stress. This means that we would expect adjectives such as fižolov, (17b'), to be pronounced as *fižolóv (to be more precise, *fižol[ɔẃ]), which is not the case. In fact, the stress pattern of the two kind and possessive ov-adjectivizations is the same (and faithful to the stress of the nominal base).  Furthermore, kind adjectives always take nominal bases (and not bound roots or other categories), which again indicates that the structure in (18b) is not accurate. There is only a very small class of ov-adjectives which does have all the expected features of root adjectivizations (such as stem-final stress): (implying the unattested adjective kak-óv) This means that we need three distinct structures in order to capture the three groups of adjectives and two of them need to include an nP. We suggest that nPs can be selected either with the inflectional class/gender specified or without such a specification. Then, the structure for possessive adjectives (králj-ev) incorporates an nP with a declension class specified on the n, see (22a). On the other hand, the structure for kind adjectives (fižól-ov) incorporates an nP without a declension class specified on the n, (22b), which explains the fact that kind adjectives can also be derived from feminine bases. Finally, adjectives like kralj-év are genuine root adjectivizations and are all stored with a specific meaning.
As for the meaning of these adjectives, we argue that the morpheme ov has no concrete semantic contribution, but that the possessive and kind meanings are a consequence of the structure. That is, we suggest that the possessive meaning might be the default meaning for an adjective derived from an animate noun. This seems especially plausible given possessive constructions in other languages where no overt possessive morphology is used, but rather simple adjacency. Such a language is Egyptian Arabic in which possessive phrases have the structure as in (23) Similarly, we suggest that the mass meaning of kind adjectives is essentially a default. This meaning is achieved by simple adjacency in a variety of languages, including English, as shown in (24).
(24) a. bean soup b. lemon juice As for the third group, listed in (21), their meaning is simply stored.

The affix ov with n in adjectives
While staying in the adjectival domain, we now focus on a context in which ov does not appear to be immediately headed by any category, as it is selected by another root. The selecting root in this case is √n, which is part of the most general adjectivizer in Slovenian (surfacing as en in the citation form). As mentioned in Section 2, en-adjectives are one of the contexts in which minimal pairs with and without √ov are attested, as illustrated in (25).
Our analysis only assumes roots √n and √ov, which can be in the complement of an adjectival head, but can also combine with each other in more than one configuration, as shown by the prosody. Adjectives in oven display prosodic variation with a major and a minor pattern, comparable to the situation in the verbs in ovati in 3.1 and adjectives in ov in 3.2. The major pattern is stem-final prosody, while a few items have either optional or obligatory preservation of the prosodic pattern of the base noun. This once again points in the direction of the exceptional preservation of nominal prosody under other categorial embeddings. The three possibilities are illustrated in (26), where all the adjectives are in the definite form because the citation form (e.g. posloven), which has no ending, contains an epenthetic shwa vowel, which obscures the stem-final stress position.
In order to establish the quantitative relations between the three patterns, we extracted the adjectives in -oven attested more than ten times in the Slovenian national corpus Gigafida. The search yielded 210 adjectives which were plausibly analyzable as containing ov+n. The relevant figures are shown in the Table 1 below. Given the stem-final stress on the majority of the oven-adjectives, we propose that the structure in (27) is the stored structure. Note that we remain agnostic as to which category ov selects (a root or an nP), as we do not have enough data to tease apart these two options. More generally, the well-formedness of structures in which a category appears below a radical core and their predicted stress pattern need to be addressed by further research.
The instances of ov in what is traditionally considered to be derivation show that ov can be analyzed as a single root which has no specific meaning, can select different categories or roots and can in turn appear embedded under a root such as √n or different categories. In the next section, we turn to ov in inflection.

THE AFFIX OV IN THE NOMINAL DECLENSION
In inflection we can observe ov acting as the genitive case ending in dual and plural of the main masculine declension: Additionally, we can also observe ov in about 40 monosyllabic nouns, which take an ov augment in dual and plural (see Mirtič 2016 for a recent list of nouns taking the augment).
The genitive and the augment ov display an interaction. In the plural dual/paradigms where there is an augment, the genitive form only contains one ov, as shown by the genitive dual/plural form rob- [ɔ] show. The natural question is then which of the two ovs is pronounced in rob- [ɔ]v 'edge.gen.du/pl'. The stress pattern provides a clear clue. The augment ov is always stressed, as can be seen from the augmented forms of rob in the table above. On the other hand, the genitive ov is never stressed, as can be seen from the forms hríb-ov 'hill.gen.du/pl' and nasl[ɔ]v-ov 'address.gen.du/pl' (the only exceptions being the few nouns in which the ending contains the only stressable syllable nucleus in the word, e.g. ps-a 'dog.nom.du ', ps-i 'dog.nom.pl', ps-[ɔ]v 'dog.gen.du/pl'). The stress pattern in rob- [ɔ]v 'edge.gen.du/pl' is the same as in all the forms with the augment, but different assume that Slovenian has no specialized genitive dual/plural ending, and that what shows up in gen.du/pl is the Elsewhere allomorph.
Importantly, the insertion of the Elsewhere allomorph is restricted to very few cases, while being generally blocked. The competition between forms including the Elsewhere allomorph and those without it can be modeled in an Optimality Theory model. The issue of the relation between Lexical Insertion and phonology proper is a complex one (see Wolf 2013 for an overview). For the simplicity, we assume here that phonology (construed as an OT grammar) evaluates the candidates which are results of Lexical Insertion. Since Lexical Insertion is guided by the Subset principle, there is, as far as the Subset principle is concerned, no limit on inserting Elsewhere allomorphs. What filters those Elsewhere allomorphs out in most forms is a phonological constraint. One of the constraints that can be used to model this is the Optimality Theory's classic *structure (Zoll 1992;Prince/Smolensky 1993). This constraint militates against structure in general, always picking the candidate that has less structure. Given the Vocabulary Item in (32), any number of additions of ov are lexically sponsored, so they will never incur a violation of Faithfulness. However, each of them will incur a violation of *structure, which, for the purpose of this example we assume to assign a violation mark for each morpheme. In (34) this is illustrated using the tableau for the Instrumental Singular form of hrib.
The only situations in which the elsewhere allomorph can survive in Slovenian are cases in which its addition prevents a violation of another constraint. In this case, the relevant constraint is one militating against syncretism with the citation form. Pertsova (2015) proposed this constraint based on Russian Genitive plural. Interestingly, Pertsova shows that, diachronically, Slavic genitive plural ov became generalized as a consequence of homonymy avoidance.
In Slovenian, the problem of syncretism of genitive dual/plural with nominative singular does not arise in declensions of the type žensk-a and del-o (as they have overt nominative singular endings). The situation in the type hrib is different, since leaving the genitive dual/plural without an overt ending would lead to a violation of conc. This violation is avoided by allowing the Elsewhere allomorph in genitive dual/plural. The full picture obtained this way is that no big declension class in Slovenian violates conc, as can be verified from (36) The affix ov in the genitive dual/plural is therefore the Elsewhere morpheme used to avoid a conc violation.
As for the always-stressed augment ov, we propose that nouns like rob 'edge' have two stored allomorphs and the phonological constraints choose between them (as in various phonological models of root allomorphy, e.g. Kager 2008). One allomorph is the root rob with its category, the second is the root rob in a root complex with the root √ov. As we have seen above, the augment ov is always stressed (rob-ɔv-i), which provides evidence for the root complex structure:

OV ROB
A crucial point for the evaluation below is that phonology does not have access to the internal structure of complex roots so it will consider the two stem allomorphs (rób ~ robɔv) as introducing the same amount of structure. As a result, the two allomorphs will be deployed to cover the two different positions in which no overt ending is inserted.
(39) rób ~ robɔv CF: rob faithfulness contrast-citation *structure a. rob *! * b. ☛ robɔv * c. róbov **! c. robóvov **! Note that in the forms with an overt case ending, no preference can be established by the constraints. As a consequence, we assume that the forms which do have a preference (nom.sg and gen.du/pl) impose their selected allomorphs upon the remainder of the number sub-paradigm through a uniformity constraint, which leads to the entire singular sub-paradigm taking the root allomorph rob and the entire dual and plural subparadigms taking the root allomorph rob[ɔ]v. In sum, in this section we have shown that √ov figures as a root with no specific meaning and as the Elsewhere allomorph in the nominal declension.

CONCLUSION
In this paper we have focused on the Slovenian affix ov, which can be found in nouns, adjectives and verbs. While these occurrences of ov appear prima facie to be several different affixes which only coincidentally have the same phonological form, we argue for a single multifunctional ov, which can appear both in inflection and derivation. In line with proposals within Distributed Morphology which claim that derivational affixes should be analyzed as roots, we have treated ov as a potentially meaningless root which can take as a complement other roots (thus forming a "radical core") or phrases, resulting in different structures and consequently different stress patterns and meanings. In the nominal declension, ov acts as an Elsewhere allomorph, whose insertion is guided by an interplay of phonological and morphological constraints. Such a treatment of ov enables us both to give further support for the unification approach of affixes proposed in the literature and to further elaborate the affixes-as-roots model by addressing the question what the Encyclopedia entry or entries for ov need to contain.
Our consideration of the multifunctional morpheme ov in Slovenian is by no means intended as a definitive account (of this morpheme or of multifunctional morphemes in general). We therefore hope that further research will address our claims and predictions and broaden the data set, but also provide formal modeling of aspects we have not explored here. One aspect of the account which has not been entirely formalized is the ability of the (semantically light) root to surface as the Elsewhere allomorph in the nominal paradigm. While semantically light roots are more plausibly expected to appear as Elsewhere allomorphs than roots with a fully specified meaning, we leave it to future research to account for the relevant mechanism. Slovenian seems a good starting point for such an account, as the same kind of parallelism is attested in the verbal domain (see Simonović, this volume).