Avoiding stress on non-lexical material in nouns and verbs: predictable verb prosody in Serbo-Croatian stress standard varieties

3.1 Serbo-Croatian verb morphology

Serbo-Croatian verb forms minimally have the structure stem + theme vowel + inflectional ending(s). In (3) we give examples of verb forms represented by the infinitive and the first-person plural present-tense form.

As can be seen from the examples above, some verbs take the same theme vowel in all forms (pit-a-ti and plov-i-ti), while others take different theme vowels in different forms (all the other verbs in (3)). The theme vowels are themselves not an expression of an inflectional head but their selection is nonetheless influenced by the adjacent inflection. For most varieties, the distribution of the two theme vowels in the verbal paradigm can be analyzed as related to finiteness. If the adjacent inflection is non-finite, the theme vowel illustrated in the infinitive forms above will show up (e.g., the theme vowel a in zasp-a-ti, and nu in kre-nu-ti above), otherwise the alternate theme vowel will be inserted (e.g., the theme vowel i in zasp-i-mo, and ne in kre-ne-mo). Because the two exponents of the theme vowel are unpredictable based on each other, we illustrate Serbo-Croatian verbs using a non-finite and a finite form: the infinitive (ending: -ti) and the first-person-plural present-tense form (ending: -mo). We will also include the negated version of the present-tense form in cases where there is a prosodic interaction with the negative particle ne (e.g., in né d-a-mo ‘we don’t give’).

Theme vowel combinations are the most common way of describing Serbo-Croatian conjugation classes (see, e.g., Jelaska and Bošnjak-Botica [2019] for a recent overview). However, as we will also show in the following section, providing a complete list of theme-vowel classes is not at all straightforward. While an overwhelming majority of verbs can easily be classified into theme-vowel classes, there are many smaller classes that lend themselves to different possible analyses.

For easier presentation, until we refine the structure, we will reserve the term ‘stem’ for whatever material precedes the theme vowel, including prefix material (but not the negation particle), while everything added after the theme vowel will be termed ‘inflection’. Throughout the article, we will use the term theme vowel, although the theme can also contain a consonant (e.g., in kre-nu-ti) or be empty (e.g., in pas-∅-ti).

3.2 Dataset

Stress standard Serbo-Croatian is the de facto implementation of standard Serbo-Croatian in the urban centers outside of the traditional Neo-Štokavian area in both Serbia and Croatia. Despite its being extremely widespread, there are no detailed descriptions of this system. The modest literature consists of remarks on Zagreb SC in Kapović (2018) and general remarks in Simonović and Kager (2017, 2020. While the differences between the Croatian and the Serbian stress standard varieties are relatively few, in this article, we focus on the Serbian varieties.

In order to obtain a good overview of the stress standard SC situation, we assembled a database of the 3,000 most frequent verbs in the Serbian variety of standard Serbo-Croatian. For this purpose, we used the web corpus srWac (Ljubešić and Klubička 2014). We asked four stress standard SC speakers (two from Niš, one from Bor and one from Zaječar) to pronounce one finite and one non-finite form of each verb. The speakers were asked to use the variety they would speak when talking to a stranger from their own hometown in a relatively formal setting (e.g., at a doctor’s or a government office).^[2]

The 3,000 verbs were classified based on the theme-vowel combination. As foreshadowed in the previous section, there is a dichotomy, which will be crucial for our analysis in Section 3.3, between verbs in which theme vowels can be unproblematically isolated and those in which different analyses are a priori possible.

In order to illustrate approximate patterns of distribution, we provide a quantitative overview of the encountered types for the transparent group (Table 1) and the nontransparent group (Table 2). The transparent types are named after their respective theme-vowel combinations. The non-transparent types are illustrated using an example verb.

The transparent types constitute 80 % of the verbs in our data set. Within this group, the two largest classes (a/a and i/i) are far more frequent than all others and account for more than 63 % of all verbs in the language. The only class that always involves phonological alternation is a/je, where the theme je palatalizes the preceding consonant, so e.g., /pis-je-mo/ comes out as [piʃemo]. The existence of both a/je and a/e as separate theme vowel classes means that some verbs will be ambiguous between these two classes. For instance, the verb opirati∼opiremo ‘resist’, can be analyzed as belonging to either the a/je or the a/e class, because underlying /rj/ yields [r]. In such cases, we non-crucially assigned the verbs in question to the a/e class.

Nontransparent classes account for 20 % of all verbs and allow multiple possible parsings into stems and theme vowels, as also reflected in the different treatments in traditional descriptions, which resulted in different classifications (see, e.g., Jelaska and Bošnjak-Botica 2019 for a recent overview). This ambiguity will be elucidated using the two most common types from Table 2, illustrated by ispitivati∼ispitujemo ‘explore’ and školovati∼školujemo ‘school’, respectively.

These two types have alternating sequences iva∼uje and ova∼uje. One possible analysis is subsuming the two types under the transparent type a/je and assuming that affixes iv∼u and ov∼u display theme-vowel-triggered allomorphy. This would make it plausible to approach other, smaller types in a similar way, e.g., assuming parsings such as prodav-a-ti∼proda-je-mo and pljuv-a-ti∼plju-je-mo. While this allows grouping a multitude of apparently stray types under a transparent type, it also means assuming a large amount of stored allomorphy, possibly missing some predictable information. For instance, in the four types just discussed, we have to assume that it is purely accidental that the stem allomorphs that appear before the theme vowel a end in the consonant v in three out of the four cases. The same goes for the vowel u preceding the theme je.

There is a second a priori available approach, however. In order to avoid massive allomorphy, it is possible to assume that the entire alternating sequences in the nontransparent classes are thematic, so for ispitivati∼ispitujemo ‘explore’ and školovati∼školujemo ‘school’ the thematic sequences would be iva/uje, ova/uje. While this resolves the issue of allomorphy, it increases the list of theme-vowel classes in Serbo-Croatian and assumes many theme vowels which only exist in one, potentially very small class. For instance, the class va/je would unite 14 verbs from the types proda-va-ti∼proda-je-mo and plju-va-ti∼plju-je-mo. Moreover, the segmental content of the alternating thematic sequence va/je would, on the one hand, suspiciously contain that of the class a/je and be contained by those of two more numerous thematic classes: iva/uje and ova/uje. As we saw above in the discussion of the transparent classes a/je and a/e, such containment relations make verbs analyzable as belonging to various classes, without a clear criterion to distinguish between different analyses.

Third, an analysis is possible that allows multiple theme vowels per verb form. In this case, the last discussed four classes would be possibly analyzed in terms of i-va∼u-je and o-va∼u-je containing va∼je. A potential problem here would be that the themes that precede va and je are not elsewhere attested theme vowel combinations (i∼u and o∼u), which in turn means that we may end up with more theme vowel types than in the previous analysis.

The upshot of this discussion is that the nontransparent verbs present speakers with analytical choices, but mere consideration of the segmental content does not give a clear clue how these choices are made and if they are made by all speakers in the same way. As we will show in Section 3.3.2, our prosodic data testify that the analytical ambiguity has been resolved by the native speakers consistently for each particular type, as indicated by the position of stress. Thereby prosody becomes a litmus test for morphological structure.

One important aspect of the data to keep in mind is that the quantitative relation between the transparent and nontransparent types is 8:2. In 80 % of the verbs there is no doubt as to where the theme vowel begins. Their analysis provides a “blueprint” for resolving the morphological ambiguities of nontransparent verbs. We will start our data analysis in the following subsection by turning to the transparent verbs.

3.3 Data and OT analysis

3.4 Nontransparent verbs

As argued above, nontransparent verbs are analytically ambiguous and just based on segmental content, it is not clear which analysis is chosen by the speakers, or even if a specific analysis is chosen. However, a system in which prosody is guided by morphological boundaries necessitates analytical choices. The stress placement in the nontransparent verbs indicates that such choices have been made by native speakers, resolving the ambiguities concerning theme vowels. As our occasional references to Croatian stress standard varieties will show, this is the domain where Serbian and Croatian varieties occasionally made different choices, confirming that both options were indeed available and that picking one was a matter of conventionalization.

While for transparent verbs the segmentation into morphological domains was unambiguous and prosody depended on these boundaries, in considering nontransparent verbs, we will use prosody as a clue to establish which of the several possible segmentations was chosen.

In Table 3 we repeat the nontransparent types with stress markings in Serbian stress standard SC.

The only generalization that can be immediately read from the data in Table 3 is that, just as in transparent verbs, stress never ends up on inflectional endings. For the rest, in the seemingly chaotic picture, there are several strong generalizations.

1. In all infinitive forms which end in -ti, the vowel preceding this ending is unstressed.

2. In the small non-transparent group in which the infinitive form ends in -ći, the vowel preceding -ći is stressed.

3. In all present-tense forms (which end in -mo in the examples) the vowel preceding the inflectional ending is unstressed.

Focusing on the infinitive forms for a moment, we can say that the only non-transparent infinitives which allow stress on the syllable directly preceding the ending are the exceptional ći-infinitives (3 % of all verbs in our dataset), whereas the common ti-infinitives always have the vowel preceding the ending unstressed.

Ći-verbs are generally assumed to belong to the type ø/e (which has only 16 transparent members in our dataset, e.g., istrés-ø-ti ∼ istrés-e-mo ‘shake off’). If this is the case, the correct parsing is povu-ø-ći (or povuć-ø-i, see below)∼povuč-e-mo. In most traditional approaches, the ending -ći [t͡ɕi] is assumed to be the result of a historical process that fused one velar consonant of the stem and the consonant of the infinitive ending -ti, e.g., povući then originates from povukti (see the discussion in Stevanović 1970, 341–342). In the present-tense form, the stem-final velar got palatalized so that povuk-e-mo yielded povuč-e-mo [poʋut͡ʃemo]. Our prosodic data seem to testify to this zero-theme analysis of -ći forms (while, as we will see below, all ti-infinitives have an overt theme). Assuming the parsing povu-ø-ći/povuć-ø-i∼povuč-e-mo and our current phonological grammar, we indeed get the correct (stem-final) stress placement: poˈvu-ø-ći∼poˈvuč-e-mo.

In all other cases, the infinitive affix -ti is segmentally unaffected by the stem and the more canonical parsing is implemented: the vowel preceding the ending -ti is viewed as a theme vowel. This even happens in the type ˈdovesti∼doˈvedemo, which would traditionally be analyzed as featuring a zero theme, but the prosodic facts indicate that the parsing is ˈdov-e-sti∼doˈved-e-mo. The same is true of ˈdon-e-ti∼doˈnes-e-mo ‘bring’.

This parsing of all vowels preceding -ti as theme vowels reveals some theme vowels that segmentally correspond to already encountered theme vowels (e.g., i in ˈpop-i-ti) but it also reveals some themes which have not been assumed before (e.g., u in ˈob-u-ti).

The analysis of the infinitive forms appears to have influenced the analysis of the present-tense forms in some types. If a vowel marked as a theme vowel in the infinitive form reappears in the present-tense forms (e.g., in the type obuti∼obujemo), the same theme analysis is in some cases extended to the present-tense forms. This leads to certain present-tense forms being analyzed as having two theme vowels. In other words, once obuti is analyzed as ob-u-ti, it is harder to analyze u in obujemo as part of the stem. The attested stress pattern ˈobuti∼ˈobujemo points in the direction of an analysis of u as thematic in both forms: ˈob-u-ti∼ˈob-u-je-mo. This parsing straightforwardly accounts for the stress pattern in the types ˈpop-i-ti∼ˈpop-i-je-mo and ˈobuti∼ˈobujemo.

However, the double-theme analysis does not extend to all the types in which it is a priori possible. For instance, the type ustati∼ustanemo seems available for the same analysis: since a is a theme in ustati (as attested by the stress pattern ˈust-a-ti), this analysis of a as a theme could extend to the present-tense form ustanemo. However, judging by the attested prosodic pattern, the present-tense form has a single theme uˈsta-ne-mo, since the double theme analysis would predict *ˈust-a-nemo. This is an important point where a comparison with the Croatian stress standard varieties is instrumental. At least in Zagreb and Rijeka,ˈustanemo is the only possible form, perhaps the most prominent systematic difference between the Serbian and Croatian stress standard varieties.

The generalization for the Serbian varieties is that only verb forms with the theme vowel je allow a vowel directly preceding it to be segmented as a theme. This segmentation applies in all cases where je is directly preceded by a vowel. This generalization suffices to describe the assignment of double theme vowels in the Serbian stress standard and the algorithm does not need to make reference to the infinitive form. This is further confirmed by the types proˈdav-a-ti∼ˈprod-a-je-mo and ispiˈtiv-a-ti∼iˈspit-u-je-mo, where the theme is not carried over from the infinitive.

The parsing algorithm for Serbian stress standard SC verbs is summarized in (20).

In Table 4, we show all the nontransparent types parsed using the algorithm in (20).

Having the parsing in place, we can now move on to the OT analysis. For all types where a single theme vowel is isolated in all forms (poˈvu-ø-ći∼poˈvuč-e-mo, ˈdov-e-sti∼doˈved-e-mo, ˈust-a-ti∼uˈsta-ne-mo and ˈdon-e-ti∼doˈnes-e-mo), our current ranking predicts the full paradigm as is.

Our ranking also unproblematically deals with short verbs with multiple themes, e.g., the prefixless version of ˈpop-i-ti∼ˈpop-i-je-mo: ˈp-i-ti∼ˈp-i-je-mo ‘drink’. The paradigm is shown in (21) and the relevant tableaux in (22)–(24).

Note that candidate a. in tableau (23) with stress on the leftmost theme does not incur any violations of the *Infl-Stress constraints which are currently present in the ranking because the leftmost theme is not in the locality domain of the inflection.

Note that candidates a. and b. are not distinguished in terms of violations of the *Infl-Stress constraints. Similarly to Tableau (19), the correct result is due to the activity of Rightmost . We will return to this issue below and argue that ultimately, the rejection of stress by the negation particle has a different explanation, based on it being non-lexical material.

Next, we turn to the prefixed version of piti∼pijemo: popiti∼popijemo. The paradigm is shown in (25).

The infinitive form, which only has one theme vowel, can be handled using our current ranking, as shown in the tableau in (26).

The present-tense form ˈpop-i-je-mo is the first one where our current ranking predicts the wrong output: *poˈp-i-je-mo. Note that, unlike in our prose above, in our OT analysis so far, there has been no distinction between theme vowels outside the locality domain of the inflection and genuine lexical material. In order to make this distinction we add the final constraint to the *Infl-Stress family:

Including this last *Infl-Stress constraint in the ranking captures the present-tense forms of longer verbs with multiple theme vowels, as shown in the tableaux in (27) and (28).

Note that the inclusion of *Infl-Stress-3 in the ranking does not change anything for the already analyzed forms listed in (21), since all the vowels in these forms are part of non-lexical material and therefore all the candidates in (22), (23) and (24) violate *Infl-Stress-3. However, the inclusion of *Infl-Stress-3 slightly changes the analysis of negatives in tableaux (19) and (24). Previously, the non-stressing of the negation particle ne was accounted for by Rightmost ; now these forms are brought under the scope of *Infl-Stress-3.

Since *Infl-Stress-3 was the final constraint in the *Infl-Stress family, we can present the final overview of stressable morphological domains and *Infl-Stress violations that stressing them incurs. This overview is presented in Table 5.

In even longer forms with multiple theme vowels, the stress will always end up rightmost on the lexical material, due to Rightmost , as illustrated by the present tense form ispit-u-je-mo ‘we test’ in (29).

This concludes our analysis of nontransparent verbs. We have established the parsing algorithm for Serbian stress standard SC and completed the unpacking of the StemStress constraint into a family of constraints militating against stress on non-lexical material on three different levels: inflection ( *infl-Stress-1), non-lexical material in the locality domain of inflection ( *infl-Stress-2) and non-lexical material altogether ( *infl-Stress-3).

The careful reader will have noticed that there still remains one form in Table 4 which is not captured by our ranking: the infinitive form ˈškolov-a-ti. Our ranking predicts *škoˈlov-a-ti. Once again, the predicted form is exactly what we encounter in Croatian stress standard varieties, indicating that this is indeed a plausible stress placement for this word in stress standard SC. However, the Serbian varieties allow much more prosodic contrast in this type. We turn to this last piece of the puzzle in the next section.