What the Schwartzes Told Me about Allomorph Priority

Abstract

In Standard Yiddish, -s and -ən are used as default allomorphs for plural word formation. It is argued here that the choice is left to the phonology, with -s acting as a default within a default. This status is used to explain the exclusive use of -s in the pluralization of proper names, which are claimed to be formed with no sensitivity to the phonological form of the base.

צום אָנדענק פֿון זשאַן ספּעקטאָרן. זשאַנס אַזעלכע געפֿינט מען שוין נישט.

1. Introduction

This short article is about two regularities in the formation of plural nouns in Standard Yiddish and the way these two regularities are related. The first regularity concerns the selection of either -s or -ən, which occurs only in items that are not associated lexically with any other plural suffix. The second regularity is the use of -s only in the plurals of surnames or proper names.

The claim in the article is the following. In the relevant regular nouns, the selection between -s and -ən is made by the phonology, with -s crucially acting as default. In contrast, when proper names are pluralized in Yiddish, their phonological form is not considered—and so the default -s is used.

There are several pluralization strategies in Yiddish. Some nouns, mostly of Germanic origin, can be pluralized by umlaut alone (1a), or by both umlaut and a suffix -ər (1b); these pluralization strategies are by and large unproductive. Other nouns, mostly of Semitic origin, are pluralized with -əm or -əs (1c,d)—these correspond to the Yiddish pronunciation of the original Hebrew suffixes [im] and [oθ]. Yet another group of nouns, mostly of Slavic origin, are pluralized with -əs, too (1e); many of these nouns carry one of a class of suffixes, such as those in (1f). Finally, a special plural -əχ is attached to bases with a final diminutive marker -l̩; bases ending with a nondiminutive syllabic l̩ generally do not take this ending (e.g., fojgl̩ ‘bird’, pl. fejgl̩). Stress is marked with an acute accent.

All of these suffixes are specific to their lexical entry or lexical stratum, even though it may be possible to find some exceptions to the generalizations above. In this sense, their distribution is limited. When none of the conditions above holds, one of two other suffixes -s or -ən∼-n̩ (henceforth -ən) is used. As can be seen in (2), this is the case in several “native” Germanic nouns (2a, b, f–h), but also in loans (2c–e, i, j). Importantly for the present purpose, both suffixes are applicable to nouns of Semitic origins, when the Semitic plural of the noun is not used (2e, j).

The first goal of this article is to describe and motivate the distribution of these default allomorphs.

The second goal of the article concerns the pluralization of proper names, as in the English phrase “the two Jacks that I know.” Given the fact that proper name plurals are relatively rare, and that either -ən or -s are used in nouns that do not have a lexicalized plural, one expects the same to hold for the plurals of proper names. Instead, one finds only -s in such scenarios (3). This is the case even if the proper name is identical to a real noun that is pluralized with -ən (3a). Note, in addition, that an epenthetic ə occurs between the stem and the suffix -s if the latter ends in a sibilant (3b,c); but the same sibilant-final sequences, when they are not proper names, are not pluralized with -s. The question, given these facts, is why the suffix -ən is ruled out in the pluralization of proper names.

Section 2 presents the results of a dictionary search, from which it emerges that -ən is used when the last syllable of the word is stressed or if the word ends in a nonsonorant consonant. The suffix -s is used elsewhere. A short comparison follows between the Yiddish facts and those of German and Dutch, highlighting the differences between the languages. Section 3 provides a simple optimality theoretic formalization of the Yiddish facts. Two allomorphs are served to the phonology with a crucial priority relation: -s has priority over -ən. The section also includes a brief discussion of the analyses of the Dutch and German facts, and why they cannot be immediately extended to Yiddish. In section 4, the priority relation between allomorphs is claimed to underlie the single pluralization strategy of proper names. Assuming that the phonological form of the proper name is not available at the stage of the allomorphic choice, -s is expected. Parallel facts from German are discussed. A morphosyntactic rationale is then provided for this unavailability. Section 5 provides a conclusion to the article.

2. The Distribution of -s and -ən

2.1 Methodology

Previous work on the topic of pluralization in Standard Yiddish (for example, Mark 1954–1956, Jacobs 2005) was based, like this study, on dictionary searches (complemented by native-speaker intuitions, when necessary). Three recent studies on the same topic in different varieties of Hassidic Yiddish (Abugov & Ravid 2014a,b, Abugov & Gillis 2016) took a more experimental approach based solely on elicitation from native speakers. None of the previous studies explicitly identifies the default status of -s/-ən, as opposed to the varying degrees of lexical specificity of the rest of the suffixes. Because they regard the system as a whole, without making the distinction between default and lexically specific strategies, these studies, while they do point to the phonological generalizations relevant for -s/-ən, make it seem like these are a tendency rather than a rule.

An effective way of identifying the productive, default strategy of pluralization in a language is a “wug test” (Berko 1958). However, this strategy was not possible for the present study, due to the scarcity of speakers of a dialect close to Standard Yiddish in my community. Instead, I scanned Niborski’s (2012) dictionary of the Hebrew elements in Yiddish, focusing specifically on nouns. While a majority of Hebrew nouns in Yiddish are borrowed along with their plurals (1c,d above), for quite a few others, the only borrowed form is the singular. In those cases, the plural form given by the dictionary belongs to the Germanic stratum. Therefore, I made a list of all of the many nouns whose plural was not borrowed from Hebrew.

The results, which were very uniform, are illustrated by the plurals of orthographic symbols (letters, diacritics) and some other nouns given in (4).

The generalizations are straightforward. If the final syllable of the word is unstressed and the word ends either in a sonorant l,m,n,r or a vowel, one finds -s. In the rest of the cases—if the final syllable of the word is stressed, or if it isn’t and the word ends in a non-sonorant consonant—one finds -ən (recall this form stands for -n̩, too). If the word ends in a syllabic l̩, then both -ən and -s are possible.

Since the generalizations about the distribution of allomorphs are purely phonological, the next section will put forth an analysis in which the choice between the allomorphs is left to the phonology. But before proceeding to the formal analysis, the next subsection compares the Yiddish facts with those of two closely related languages, namely, German and Dutch.

2.2 Pluralization in German and Dutch

Even though German seems to be the closest language to Yiddish among the Germanic languages, pluralization in the two languages is quite distinct. German involves many plurals with a suffix -ə, for example, ʃuː, ʃuː-ə ‘shoe’ or ʃaf, ʃaf-ə ‘sheep’ (Wiese 1996, 2009), whereas Yiddish lacks this strategy altogether. Secondly, the distribution of -ən is sensitive to the gender of the noun (Wiese 1996, 2009), in that sometimes it is used for feminine nouns when other suffixes are expected on phonological grounds; no such sensitivity is found in Yiddish. More importantly, while German does employ a plural suffix -s, its distribution is not at all parallel to that of -s in Yiddish. For instance, Wiese (2009:138) provides examples of the use of -s which do not follow the generalization above, such as baʁ, baʁs ‘bar’, cf. (4u); and this suffix is altogether excluded from the otherwise comprehensive analysis in Trommer (2021:611), who refers to Wiese (2009) and references therein for “the special status of the plural -s.” To summarize, formal accounts of German pluralization agree that -ən and -s in German are not allomorphs whose selection is relegated to exclusively phonological considerations, as is clearly the case in Yiddish.

Yiddish does share this logic with another Germanic language, namely Dutch. As reported for instance in Booij (1998) and van der Hulst & Kooij (1998), Dutch employs both -ən and -s, and phonological considerations are at play in the selection of these suffixes. According to Booij (1998:145), the rule is “-s after an unstressed syllable, -ən after a stressed syllable,” for example, káːnɔn, káːnɔns ‘canon’, kaːnɔ́n, kaːnɔ́nən ‘gun’. This generalization is very similar to the ones given for Yiddish above. Booij’s analysis suggests that the choice between these allomorphs is entirely phonological, as I have proposed for Yiddish.

There are nevertheless several differences between Dutch and Yiddish. First, as reported by van der Hulst & Kooij (1998:188), monosyllabic loanwords are exempt from competition: only -s is available in this case, for example, the loan bal, bals ‘dance party’ versus the native noun [bal, balən] ‘ball’. In Yiddish, as we saw, it is precisely in loans that one finds an overwhelming regularity. Second, Booij reports that ə-final nouns may have either plural, for example, [kadə, kadən∼kadəs] ‘quai’; in Yiddish, only -s is possible in this case. Other complications involve the quality of the final vowel in V-final words and the length of the last vowel of C-final stems. I have only found passing mention of the effect of the quality of the final consonant, a finding that was central in the Yiddish facts. Importantly for what follows, van Wijk (2002:219) concludes, based on a corpus study, that “it is unlikely that Dutch has a unique default.”

To summarize, the facts from both German and Dutch are different from what I found in my dictionary search in Yiddish. That said, Dutch seems to be much closer to Yiddish, in that mostly phonological considerations are relevant for the choice between -ən and -s. In the next section, I provide the motivation for the distribution of these allomorphs in Yiddish and discuss the differences with respect to both German and Dutch.

3. Motivating the Distribution of the Plural Allomorphs

It is well known that Germanic languages prefer nonfinal stress. ¹ Yiddish is no exception to this rule: the great majority of polysyllabic items are penultimately or antepenultimately stressed, even if one can occasionally find monomorphemic words with final stress (e.g., jaríd ‘market’).

The original analyses of Dutch plurals in both Booij (1998) and van der Hulst & Kooij (1998) capitalize on this fact: the allomorph chosen is the one that will allow for nonfinal stress. The same choice of plural allomorph in Yiddish is also clearly affected by this consideration. Vowel-final stems with nonfinal stress such as dómino or rófe in (4) above take -s, which does not change their stress profile. But a form like ʃvo takes -ən, specifically because this choice allows the output ʃvóən to lose its final stress; a goal that would not be obtained by *ʃvo-s. The same is true for zémər versus gvar: the former takes -s because it is not stress-final, whereas the latter takes -ən in order to avoid being stress-final. In other words, the suffix -ən adds a syllable to the base and is therefore capable of improving the stress profile of stress-final bases by distancing the stress from the final syllable of the plural word. For this reason, it is used whenever the base is stress-final.

The other major consideration mentioned above is the identity of the final consonant; unlike nonfinality of stress, this consideration does not play a clear role in the analyses of Dutch plurals. As shown in (4), Yiddish bases like áləf, even though they are not stress-final, still take -ən, not -s (in Dutch, the plural would be áləfs). This, I submit, is due to sonority considerations. ² For the present discussion, only the difference between sonorants on the one hand and obstruents on the other is important. The sonority hierarchy (see, for instance, Selkirk 1984) places sonorants closer to vowels than obstruents. Thus, an output like *áləfs creates a syllable ending with a sonority plateau fs; and this scenario is avoided by the selection of -ən in aləfn̩. I will return to the variation in giml̩s ∼ gimlən below.

From this perspective, items like zémər and rófə become important: for these, both -s and -ən would produce forms with nonfinal stress, and neither would be problematic in terms of sonority. I suggest that this indicates a relation of priority between the two default allomorphs: -ən is used only if -s poses a problem, by leaving the stem stress-final or creating a plateau (for the notion of allomorph priority, see, for instance, Bonet et al. 2007). In other words, there is a default within the default.

This state-of-affairs, whereby one generally dispreferred output is selected in specific environments despite its dispreferred status, can be captured elegantly in Optimality Theory (Prince & Smolensky 1993/2004), with its violable constraints. In the ensuing analysis, I will assume the following constraints:

The first three constraints are clearly markedness constraints motivated by typological considerations. Languages with sonority plateaus also have sonority slopes, but the opposite is not necessarily true; languages with syllabic consonants also have non-syllabic ones, but the opposite is not necessarily true; and nonfinal stress is the universal tendency (Allen & Hawkins 1978, Bat-El 2018), at least in weight-insensitive languages like Yiddish (Hayes 1995). ³

I will now show that, given a certain constraint hierarchy, the distribution of allomorphs can be formalized. The hierarchy I propose crucially ranks the first two markedness constraints, Nonfin(ality) and *Plateau, above the priority requirement. Example (6) presents the case of vowel-final singulars. The input to the phonology is the two allomorphs, with a priority relation signaled by “>”. With a stress-final input like ʃvó, the selection of -s violates Nonfin(ality), and therefore -ən is selected, even though it violates the lower-ranked Priority. With an input like géto, no relevant markedness constraint is violated by either -s or -ən, and the priority relation is the decisive factor. ⁴

The sonorant-final bases in (7) present a similar case. For a stress-final input like mem, selecting -s again violates Nonfin(ality), and therefore -ən is selected even though it violates the lower-ranked Priority. With an input like gójlem, in contrast, no markedness constraint is violated by either -s or -ən, and the priority relation is the decisive factor.

Obstruent-final bases like áləf violate *Plateau if they take -s; therefore, -ən is selected.

The facts above do not provide an argument for the ranking of Nonfin and *Plateau. Either ranking will yield the correct result. A different type of unranking holds between *C̩ and Priority. In this case, the two constraints are “crucially unranked” (Anttila 1997): both rankings are considered possible, such that the variation between gimlən and giml̩s is yielded: ⁵

Interestingly, all three markedness constraints, while being typologically motivated, are repeatedly violated in Yiddish. The language has many syllabic consonants, quite a few words ending in sonority plateau, and many cases of syllabic consonants. Thus, the selection between allomorphs is an instance of the “Emergence of the Unmarked” (McCarthy & Prince 1994): under specific circumstances, a universal tendency is revealed to be active in a language which otherwise does not obey it. Here, the universal tendencies are revealed because the phonology has a choice between two plural exponents.

The present analysis reveals interesting differences between Yiddish on the one hand and German and Dutch on the other. Two aspects of the Yiddish account stand out in this respect. First, the selection between -ən and -s is left to the phonology; this suggests that no analysis of German can be extended into Yiddish (or vice versa). As discussed above, all other analyses are in agreement that the two do not compete in German. ⁶ Second, the -s allomorph was regarded as default in Yiddish, a characterization that does not figure in analyses of the Dutch facts and is explicitly argued against by van Wijk (2002). In addition, the nature of the final consonant is central in Yiddish but does not play a central role in analyses of Dutch. Thus, while the rationales in Dutch and Yiddish are certainly similar in that both leave the choice to the phonology, they are also distinct from each other in more than one way.

Taking a step back to more theoretical considerations, there are two arguably controversial aspects of the analysis above. First, phonology is served two underlying representations rather than one, after which it selects the better one. The mere existence of disjoint underlying representations, and the fact that phonology can be sensitive to this disjoint nature, is not accepted by all phonologists. Second, in choosing the correct allomorph, phonology must be sensitive to the priority relation: in order to do that, phonology must “understand” the symbol “>”, which is definitely not a piece of phonological information. In other words, upon reception of the disjoint exponent, phonology must somehow know not only that there are two options, but also that one is preferable. Such sensitivity may be regarded as a drawback by phonologists who adhere to a strict view of modularity, according to which phonology can only understand phonological “vocabulary” such as consonants, vowels, syllables, etc., to the exclusion of lexical information such as category or relations between exponent (Scheer 2012).

While I acknowledge these difficulties, I have nothing to add regarding them. Instead, I wish to show that the priority given to -s is a possible reason for its exclusive status in the pluralization of proper names.

4. Plurals of Proper Names

As discussed in the introduction, proper names (PNs) exhibit only -s in plurals regardless of their phonological form. This is the case even when they end in obstruents and/or are stress final, for example, frojd-s ‘Freuds’, ʃpinóza-s ‘Shpinozas’, ugó-s ‘Hugos’. ⁷ In addition, only in this configuration can the -s allomorph be accompanied by epenthesis -əs (e.g., maʁánʦ-əs). In this section, I link this fact to the prioritized status of this allomorph within a default way of marking plurals.

4.1 Plurals of Proper Names and Phonological Form

Plurals of PNs behave in a nonuniform fashion cross-linguistically (Schlücker & Ackermann 2017). Some languages do not display inflection on PNs at all. In French, for instance, many singular nouns ending in the sequence al have corresponding plurals where al is replaced by o, for example, ʃəval, ʃəvo ‘horse’. But the surname ʃəval cannot be pluralized ʃəvo; instead, plural number is only marked on the article, for example, l-e ʃəval ‘def-pl Cheval’ (Nübling 2017). Other languages, such as English, apply the regular plural inflection: bɪlz can refer to checks or to people called Bill. In some cases, there is even phonological sensitivity to the form of the base. In Modern Hebrew morphology, for example, feminine plurals are marked by replacing the feminine singular suffix with -ot, for example, ʃiʁ-á ‘poetry’, ʃiʁ-ót ‘poetries’, χav-í-t ‘barrel’, χav-ij-ót ‘barrels’; and the same is true for plurals of feminine PNs which carry the feminine singular ending -a, for example, ʃíʁa ‘Shira (PN)’, ʃíʁ-ot ‘Shiras’. Interestingly, the form of PN plurals like ʃíʁ-ot is nevertheless somewhat exceptional: in the native vocabulary, -ot attracts stress (see ‘barrels’ above, as well as the otherwise homophonous ‘poetries’), but in the PN plural, stress remains on the stem.

Although plural formation in PNs varies cross-linguistically, there is one commonality across different languages: the tendency to tamper relatively little—less than in the case of regular nouns—with the phonological form of the base PN.

In this context, the Yiddish pattern is of interest. As argued in the previous section, -s and -ən are, in a very specific sense, the same morpheme: they are served as one to the phonology, and the phonological computation decides on which of the two to apply. Thus, the analysis above dictates that -ən should be available as an exponent with PNs. The question of why it is never used with PNs is therefore especially troublesome.

I submit that the answer is precisely the priority status given to -s within the complex exponent {/s/>/ən/}. As mentioned above, the phonological form of PNs is relatively inalterable. Suppose that, for a reason to be speculated upon below, the phonological form of the PN is unavailable to the computation upon the selection of the plural allomorph. In the absence of that information, the lexical priority correctly predicts that -s will always be selected:

Subsequently, the selected allomorph is concatenated with the base PN. Since selection has already occurred, -ən is not an option. Similarly, if a sequence of sibilants emerges from the concatenation, for example, maranʦ+s, it cannot be resolved through the selection of -ən; instead, epenthesis is applied.

The assumption that phonological form is unavailable for the pluralization of PNs, together with the prioritized status of -s, motivates the absence of -ən from this domain. If adequate, the analysis provides a novel, original argument in favor of phonological inputs with multiple, ordered allomorphs.

There are, of course, alternative analyses for the exclusive use of -s with PNs. It can be claimed that such items are marked with the feature [+proper], and a realization rule states that only -s is compatible with this feature. However, this view only states the facts formally—it does not explain them. Alternatively, one can argue more vaguely that /s/ is used when one wants to change as little as possible in the base. PNs and loans are such a case, as suggested by Nübling (2017), for instance. Formally, such a view amounts to a proposal that there are different phonologies for PNs, on the one hand, and loans and regular nouns, on the other, with faithfulness constraints ranked higher for the former. In other words, it also assumes a feature [+proper] which triggers a specific phonological grammar. ⁸ While there is probably independent evidence for a feature [proper] (for instance, Longobardi 1994, Matushansky 2006 or Roehrs 2020 for German phrasal PNs), not all grammars must be sensitive to it. I have proposed that Yiddish is one possible case where PN pluralization can be regarded as following the general grammar, specifically because -s is the default marker.

4.2 The Logic behind Phonological Insensitivity

One may wish to raise the question of why there is no phonological “communication” between the plural exponent and the base in the case of PNs. This is a relatively independent question, on which I can only speculate. Consider the morphosyntactic structure attributed to nouns in Distributed Morphology (Halle & Marantz 1993) in (11). In this structure, the root is situated at the basis of the derivation, and the head num introduces the number feature. In Yiddish, the complex allomorphic exponent is inserted for the plural feature, as shown below. The base nP moves to spec,numP in order to derive a count noun and the right morpheme order. Realization is triggered by the head D; when this head is merged, all of the exponents of its complement are considered (that is, everything below the dotted line). Thus, the underlying, phonemic representations of both the head noun and the plural are considered together. Because the base is stress-final and obstruent-final, the plural allomorph chosen is the otherwise disfavored one -ən.

It is well known at least since Longobardi (1994) that a PN like maranʦ cannot be of the same structure as the singular noun. One clue as to what structure it may correspond to comes from the interaction of PNs with definite and indefinite articles. In many languages, a PN which refers to a single person cannot ordinarily be combined with articles. For instance, in Modern Hebrew, the definite article ha cannot be combined with PNs: ha-tapuz ‘def-orange’, but not *ha-maʁanʦ ‘def-Marantz’ when referring to a person named Marantz. The accusative marker et, which must precede definite DPs when they are direct objects, is added to PN objects even though they do not carry a definite marker overtly: et=ha-tapuz ‘acc=def-orange’ versus et=maʁanʦ ‘acc=Marantz’. In other languages, such as most varieties of Austrian German (Markus Pöchtrager, personal communication), PNs must carry definite articles in most uses. Such phenomena suggest that PNs are inherently definite, either obviating a definite article or requiring it.

Longobardi (1994) proposed that in Italian, PNs, but not common nouns, are a case of N to D movement. Matushansky (2006) updates this account by assuming that PNs, unlike regular nouns, occupy entire definite DPs, in the sense that they involve what Matushansky calls m(orphological)-merger of D and N. A PN therefore spells out the entire N to D stretch, as depicted in (12). Languages in which a definite article is obligatory with PNs, like Austrian German, are the default for Matushansky (2006:285); in languages like Standard Yiddish, where this is not possible, this impossibility can be conditioned by the semantics of PNs.

Matushansky proposes that in many languages, some cases of overt morphological marking on PNs, such as plural marking, disrupt the morphological merger in (12), such that the regular realization of D can be used. A case in point is English PN plurals like the Marantzes: once plural morphology occurs, the realization of the definite article is possible and even obligatory. In this case, the structure of the PN plural becomes identical to that of a regular plural in (11). ⁹ Yet if the present analysis is on the right track, Yiddish cannot be like English. This would make the wrong prediction that proper names should be able to pluralize with -ən.

Other languages may resort to PN-specific alternatives strategies in order to break the N-to-D sequence more explicitly. For instance, Dutch often forms diminutives in order to pluralize surnames (Marijke De Belder, personal communication), as seen in the title of the reality TV series De Verhulst-je-s about the Verhulst family. I propose that Yiddish employs another strategy, namely, to use a dummy structure based on an empty nominal predicate √PRO, as in (13). This √PRO carries the approximate meaning ‘individual’, such that the nP refers to an individual of the type Marantz. ¹⁰ The PN DP is hosted in spec,nP in order for the stretch realized by the PN to remain undisrupted. When the allomorphic plural exponent is considered, the noun to be pluralized is phonologically null. In the absence of any phonological information, the prioritized -s is selected, and -ən is unavailable upon the subsequent concatenation with the base. If the base ends in a sibilant, as is the case in (13), epenthesis ensues. Since this structure is not headed by the PN DP, the realization of a [+def] feature on the top D is not neutralized by the semantics of the PN; it can therefore be overt [di].

To summarize, the language-wide phonologically conditioned allomorphy in Yiddish does not apply to PNs; instead, one finds only the prioritized allomorph in all cases. Under a priority-based analysis such as that of section 4, this distributional fact follows if there is no phonological communication between the stem and the suffix at the moment of allomorph selection in the pluralization of PNs. In this subsection, a proposal was presented to explain this state of affairs in which PNs are DPs which might resist regular pluralization. One solution is to pluralize an nP based on a null pronominal structure, which will be modified by the PN DP in its specifier. ¹¹

5. Conclusion

This short article aimed to explain two related regularities in Yiddish plural word-formation: (i) the distribution of the non-lexical plurals -ən and -s, and (ii) the exclusivity of the latter in plurals of proper names.

The first regularity, I proposed, is a case of phonologically conditioned, optimizing allomorphy: two options are lexically available to the phonology, and therefore the choice is dictated by phonological considerations. Importantly, the two allomorphs are lexically ordered, such that -s is prioritized when phonological considerations are not decisive. The second regularity follows from the view of the two allomorphs as prioritized, if it is accepted that the choice cannot be affected by the phonological form of the base in PNs. Thus, the analysis works as a whole: because the choice is made in the phonology, and because -s is prioritized, it is the only allomorph available in PN pluralization.

PN pluralization involves associating a kind interpretation to an item for which it is not canonically available. Different languages may go about this task in different ways. The proposal made for Yiddish would fit well for other languages in which, for any inflectional nominal feature, there is phonologically conditioned, optimizing allomorphy with a prioritized exponent, and that exponent is used in the inflection of PNs. Future exploration will reveal if, indeed, other languages follow this path.