The syntax of number marking: The view from bare nouns in Wolof

Several languages allow for their nominals to occur without any functional morphology; they are dubbed ‘bare nominals’. BNs are often number-neutral, i.e., there is no commitment to a singular or plural interpretation. In Wolof, however, BNs are singular when unmodified. A plural interpretation becomes available only when a nominal-internal plural feature is exponed in the form of complementizer or possessum agreement. I propose an extension of Béjar & Rezac’s (2009) Person Licensing Condition to number: a marked number feature (i.e. plural) must be licensed by Agree. BNs in Wolof can in principle be singular or plural. In the absence of a nominal-internal probe that Agrees with the plural feature of the BN, the Number Licensing Condition is violated, causing the derivation to crash. Unmarked number, i.e., singular, does not obey the NLC, so the derivation converges, yielding a singular BN. However, if there is a nominal-internal number probe, which is realized as complementizer or possessum agreement, the NLC is satisfied, allowing a derivation to converge where the BN is plural. If correct, this analysis accounts for the unusual behavior of BNs in Wolof and provides further empirical support for the view that valued features are responsible for nominal licensing (Kalin, 2017, 2019).

(2) Gis-na-a see-NA-1SG nonggo darra student senegalee. Senegalese 'I saw a Senegalese student/*some Senegalese students.' I assume that BNs are nominals that lack the functional morphology displayed by their overt counterparts (1). Thus, BNs in Wolof lack a(n overt determiner) and the class marker attached to it. Because of the absence of latter, there is also no overt number morphology.
Mareem has a very specific preference and she has married several, different dancers.
Several, unrelated languages have BNs too. One of them is Mandarin (Rullmann & You, 2006;Jenks, 2018), which is illustrated below. As can be gleaned from the translation in (5), BNs may have a number neutral interpretation, that is, they lack a commitment to a singular or plural interpretation (Corbett, 2000). Number neutrality can be demonstrated by, among other things, the possibility of the BN to saturate a collective predicate and to be referred back to with either a singular or plural pronoun. Not every nominal modifier, however, has the same effect in the number interpretation of a Wolof BN. In particular, if a BN is merged with a modifier (underlined) that does not have the syntax of a relative clause, as in (8a), it still behaves as if it were singular as regards the saturation of collective predicates and discourse anaphora.  One of the differences between relative clauses and nationality modifiers lies in whether there is plural morphology in the modifier or not. The same difference will be shown to arise in two types of possessive constructions, one that has number morphology and one which does not; the presence or absence of number morphology will also correlate with the number interpretation of the BN. In view of this distinction, this paper aims at addressing the following questions: (10) a. How can we account for the exclusively singular interpretation (and not number neutral) interpretation of unmodified BNs in Wolof? b. Why does a BN without any plural morphology behave as if it were singular, while a BN merged that does contain plural morphology behaves as if it were plural?
I will propose an extension of Béjar & Rezac's (2009) Person Licensing Condition to number, according to which an interpretable instance of the feature [PLURAL] must be licensed by the operation Agree. A generalized need for certain features to be licensed has been proposed by Kalin (2017Kalin ( , 2019. The gist of the analysis is that BNs in Wolof can bear either a [SINGULAR] or a [PLURAL] feature. However, because [PLURAL] requires licensing by Agree, the equivalent construal only arises in the BN when there is a number probe inside the nominal that Agrees with the matching feature in the BN. This would be the case, for instance, of relative clauses (8a), where number morphology appears in the class marker (CM) prefixed to the relative complementizer u. Conversely, in the absence of such a probe, as in (9a), only a BN with a [SINGULAR] feature leads to a convergent derivation, as this unmarked feature is stipulated not to require licensing by Agree.
2. Unmodified BNs in Wolof are singular. In this section, we will examine data that suggest that BNs in Wolof are singular.
First, a BN in Wolof cannot saturate a collective predicate. The verb dajeele 'gather' requires a plural object. The teacher gathered *a child/some children in the park.' As we have already seen, a BN in Wolof cannot be the object of that verb (cf. Mandarin (6b)). In sum, the data investigated above suggest that BNs in Wolof are singular, and not number neutral, even though this is claimed to a cross-linguistically stable property of BNs. We may then ask the following question: how can we account for the exclusively singular interpretation (and not number neutral) interpretation of BNs in Wolof?
However, the generalization that BNs in Wolof are singular only holds if they are unmodified. If there is plural morphology in the nominal, either in the form of complementizer agreement or possessum agreement, the BN has a plural interpretation. In the next section, we will see that the addition of modifiers may have a pluralizing effect in the BN. If the modifier does not expone a plural feature, the BN thus modified remains exclusively singular. In §4, we will see a similar contrast in the behavior of two possessive nominals. If the possessive construction contains plural morphology, a BN is interpreted in the plural. If the possessive nominal does contain any number morphology, no pluralizing effect is witnessed.
3. Adding a modifier: relative clauses vs. plain modifiers. Nominal modifiers in Wolof predominantly have the syntax of a relative clause. The exception are expressions for nationalities. Either one of these modifiers can combine with BNs, but each option has consequences for the number interpretation of the BN. This contrast will serve as the basis for the proposal to be put forth in this paper, where the source of number interpretation in Wolof BNs is nominal-internal. We discuss each modifier in turn.
In Wolof, a relative clause contains a class marker attached to the relative complementizer u (Torrence, 2013). As mentioned above, the same morpheme is found attached to determiners. The class marker of the relative clause and that of the head of the relative clause must match. ]. ] 'Samba closed some window that is dirty/some windows that are dirty.' We saw in §2 that BNs in Wolof are singular and that they behave like a singular full DP. We may ask then how they can be able to be modified by a relative clause with a plural class marker (y), while their singular full DP counterpart cannot (19b). In fact, the behavior of BNs now resembles that of plural DPs (19b). We may further ask if BNs modified by a plural relative clause may behave like full plural DPs in other aspects as well. In this section, we will go back to the properties investigated above and see that the answer to this question is positive.
A BN modified by a plural RC can be the object of the collective predicate dajeele 'gather' (21). It can, in fact, must, be referred back to with a plural pronoun (22). It can also be the antecedent of a reciprocal (23) and of a plural reflexive (24). 3 Plain modifiers like those that express nationalities do not have a pluralizing effect, like relative clauses do. Why does an unmodified BN behave as if it were singular, while a BN modified by a plural relative clause behaves as if it were plural? Why does adding a plain (i.e. numberless) nominal modifier not have the same effect? Before we move on to an analysis, we will examine two possessive constructions from which we can draw the same questions.
4. Two types of possessive nominals. In Wolof, there are at least two types of possessive nominals. In (29), the possessive determiner sama 'my' is used. It precedes the possessum xaj 'dog'. A definite determiner bi 'the' can be part of the same nominal. In (30), the genitive suffix u is used. It is affixed to the possessum muus 'cat', which precedes the possesor Mareem. 'The cook saw some cats of Mareem's.' These constructions differ in whether or not they contain some number morphology. When a BN is used in these possessive constructions, their behavior resemble that of plural relative clauses and plain modifiers. We discuss each possessive construction in turn. Starting with (29), the possessive determiner may be suffixed with a morpheme that is sensitive to the number of the possessum. If the possessum is singular, -y may not occur (31a); the opposite holds when the possessum is plural (31b). 5 The emergent generalization is that BNs in Wolof are singular, unless there is nominalinternal morphology that expones a plural feature. The latter can be realized as relative complementizer agreement or possessum agreement. In the absence of such an exponent, which happens when the BN is unmodified or when the modifier is itself number-less, the BN is construed in the singular, exclusively. In the next section, I propose an analysis that is based on the extension of an independent analysis of Person Case Constraint phenomena.

Analysis.
In order to account for the PCC (Person Case Constraint), Béjar & Rezac (2009) propose the following condition: Person-Licensing Condition (Béjar & Rezac, 2009, (13)) A π-feature [F] must be licensed by Agree of some segment in a feature structure of which [F] is a subset. Kalin (2017Kalin ( , 2019 argues for a generalized version of (44) where interpretable, valued features are responsible for nominal licensing. The empirical support that Kalin furnishes is a series of similarities between the PCC and DOM (Differential Object Marking). The author contends that a unified analysis can be provided that is based on the need of certain interpretable nominal features to be licensed via Agree. In a similar vein, a I propose a further extension of the PLC (44) to the marked number feature, i.e. plural: 8

(45)
Number-Licensing Condition A marked number feature (i.e. plural) must be licensed by Agree.
I will argue that (45) can account for the number interpretation restrictions observed in the distribution of BNs in Wolof. The first step in the analysis is to provide a structure for full nominals, as this will allow us to better grasp the missing elements in a BN. To recall, BNs in Wolof lack both a(n overt) determiner and the class marker affixed to it. I hypothesize that a BN has a truncated structure. Under this view, we must then ask which nominal layers are missing in a BN. Torrence (2013) analyzes the class marker that is prefixed to the relative complementizer u as an instance of complementizer agreement. I extend this analysis and propose that the class marker that appears affixed to determiners is also the result of Agree. A suggestion that this proposal is on the right track is supplied by the fact that a class marker can appear more than once in the same nominal (46a). In this case, the class markers must match (46b).  The Wolof class markers can be found in (47).

(47)
Class markers in Wolof (Tamba et al. , 2012, tab. 17 It seems clear from (47) that there are more class markers for singular nouns than for plural ones. We could assume that there are as many vocabulary items as there are class markers. While this potential analysis is consistent with the facts, it misses the asymmetry in the amount of singular and plural class markers. In order to capture these facts, I propose that a class marker is a feature which is a specification of n. I follow Acquaviva (2009) in assuming that gender and other rootspecific morphology is encoded in the categorizer that merges with the root. Furthermore, I postulate a single head (AgrP) that probes for a class marker and a number feature. It is this single head (Agr), I contend, that is exponed as the class marker morpheme in (47). We now have the pieces to derive full nominals like a-b xale 'a child' (INDEF-CM.SG child) and a-y xale 'some children' (INDEF-CM.SG child). In both (50a) and (50b) the root and hence the class marker in n are the same. What differs is the number: singular in (50a) and plural in (50b). The exponent of Agr in (50a) is thus b (49a) and in (50a) In the plural nominal (50b), the stipulated NLC (45) can be satisfied because Agr has a Number feature to be valued, thereby licensing the marked plural feature via Agree. (50a) satisfies the NLC vacuously due to the absence of a plural feature. As mentioned, I assume that BNs have a truncated structure; this has already been proposed for BNs in other languages (cf. Massam 2001, a.o.) Specifically, I propose that BNs in Wolof lack an AgrP layer, since they lack a class marker, here, to reiterate, analyzed as the exponent of Agree. Conversely, NumP is retained under the assumption that this is the only locus of number interpretation (Ritter, 1991(Ritter, , 1992Harbour, 2011). Hence, the BN structure is as in (52) below. (I am so far agnostic regarding the projection of a silent DP layer.) All things equal, since Wolof has both singular and plural full nominals, I hypothesize that the same range of possibilities should be available to BNs as well. A potential suggestion that BNs can indeed have a number feature is provided by the fact that they can (in fact, must) be cross-referenced by plural morphology when conjoined in the subject position.  (45). With these tools in hand, we can move on to the derivation of relative clauses that modify BNs. Following Torrence (2013), assume a raising analysis for relative clauses in Wolof. Also following this author, I assume that the class marker that appears affixed to the relative complementizer is the result of complementizer agreement. For uniformity, I assume that this class marker is also the exponent of an Agr projection now at the relative CP level and which is probing for both a number and a class marker feature. (53)  Licensing of [NUMBER] by Agree is also possible in the possessive construction, if the possessum-sensitive y is the exponent of Agree. The derivation of (33b) would be as in (54) In brief, the analysis proposed to account for the exclusively singular (as opposed to the more commonly attested number neutrality) interpretation of BNs in Wolof by proposing that it obeys the NLC (45). This is a condition that imposes that [PLURAL] be licensed via Agree, an extension of Béjar & Rezac's condition on [PARTICIPANT] features and, more generally, of Kalin's (2017; 2019) nominal licensing system. 6. Concluding remarks. In this paper, we investigated BNs in Wolof. Specifically, the goal was to provide answers to the questions in (10). Unlike what happens in other languages, BNs in this language are not number neutral, but rather exclusively singular. However, this generalization only holds if the BN is unmodified and there is no morphology exponing a plural feature. This is the case of relative clauses, where there is complementizer agreement and possessive nominals where there is agreement with the possessum. In order to account for this generalization, I proposed an extension to Béjar & Rezac's (2009) PLC to number (the NLC in (45)). If correct, this analysis accounts for the typologically unusual behavior of BNs in Wolof and provides support for the view that valued features may be responsible for nominal licensing (Kalin, 2017(Kalin, , 2019.