The Morpho-Semantics of Number in Brazilian Portuguese Bare Singulars

In this paper, I provide evidence against the idea that bare singulars in Brazilian Portuguese (BP) are morphologically singular but semantically number neuter. I argue instead that they are sometimes specified for number (singular) and sometimes under-specified and that this has semantic consequences. I develop an account according to which whether or not a bare noun can enter a syntactic derivation under-specified for number depends on the functional heads to which it is syntactically related. The conclusion is that, as far as number is concerned, morpho-syntactic specification should NOT be kept apart from semantic interpretation.

(1) is an episodic sentence with a bare singular NP occupying the object position.
(2) is a generic sentence with bare singulars in both subject and object positions. In this article, our focus will be on the use of bare singulars in episodic sentences, in which they receive an existential interpretation, roughly equivalent to 'there is one or more' or 'there is at least one'. 1 A common intuition about BP bare singulars in episodic sentences is that they are morpho-syntactically singular but semantically neuter (cf. Schmitt and Munn 1999, Munn and Schmitt 2005, Müller 2002, Müller and Oliveira 2004 In (3), we notice the absence of plural morphology on the bare noun and its possessive modifier and also the presence of singular morphology on the verb of the subject relative clause attached to the bare noun. In (4), the bare singular in the first sentence can serve as the antecedent of either a singular or a plural pronoun appearing on the subsequent discourse. These facts have been taken as evidence that despite being morphologically singular, BP bare singulars are semantically neuter, being able to introduce both atomic and non-atomic discourse referents. 3,4 However, one should be careful when evaluating this evidence. Plural discourse anaphora is possible even in the absence of a plural antecedent, as the following example attests: 5 1 There are syntactic restrictions affecting the distribution of bare singulars in episodic sentences that will not be discussed in this paper. For instance, bare singulars cannot appear as the subject of finite clauses under an episodic reading. The restriction does not apply to generic contexts, as attested in (2). This is, per se, an interesting topic. Further research is necessary to tell what the impact of this and related matters is to the points to be made in this paper.
2 Due to number neutrality, it is hard to give appropriate and concise English translations to the BP sentences containing bare singulars. Although it may affect readability a little bit, I decided to translate the bare nouns and related agreement morphology in pairs, such as a friend/friends. 3 In this paper, the term number neutrality should be understood as follows: an NP is number neutral if its denotation contains both atomic and non-atomic entities. 4 As pointed out by an anonymous reviewer, a different logical possibility is that BP has a number system with three categories -neuter, singular and plural -with neuter and singular being homophonous. Although this is certainly compatible with the data presented so far, we will see throughout this paper some evidence that certain instances of bare singulars should be treated as genuine singular NPs, in the sense of having only atomic entities in their denotations. 5 Schmitt and Munn (1999) acknowledge this point. See, in this respect, the work of Evans on e-type pronouns (Evans 1980) and Kamp and Reyle's (1993)  Here, the plural pronoun 'elas' refer to the apples eaten by the boys. Of course, there are differences between (4) and (5). For instance, the singular indefinite is under the scope of a universal quantifier in the latter but not in the former, but for the moment the example suffices to raise a warning flag and point out that it is not obvious that the bare singular in (4) should be semantically neuter or plural to license the plural pronoun. We will return to this point in section 7. 6 Moreover, semantic neutrality is not enough to legitimate morphological mismatches in discourse anaphora. Consider the following case: The noun 'criança' is morphologically feminine in BP, but semantically neuter, being able to refer to both boys and girls. However, it can only serve as the antecedent for feminine pronouns. As shown above the use of a masculine pronoun is infelicitous in this case. 7 This prohibition against gender mismatches abstraction. More on this topic in section 7. 6 Discourse anaphora is a complex topic and many non-trivial examples can be constructed linking pronouns to noun phases that have been previously mentioned. For instance, Krifka et al. (1995) present cases in which kind referring pronouns are linked to existential indefinite NPs that introduce individual referents into the discourse. Even more indirect links exist between noun phases and pronouns, as the following example provided by an anonymous reviewer shows: Here the pronoun them is linked to the proper noun Natal (a city in Brazil) and seems to refer to its inhabitants. What examples like this reveal is that noun phrases can introduce/activate several kinds of discourse referents, some of them only indirectly related to the NP denotation. In section 7, I will briefly discuss what kind of connection might be responsible for the relation between bare NPs and pronouns in examples such as (4) and (5). 7 This constraint against gender mismatches seems to be a very strong one. As shown by the following example (due to an anonymous reviewer), even when the context makes it clear that the child under discussion ought to be a boy, only the feminine pronoun is allowed: does not seem to apply to number mismatches, as confirmed by (7) (6) and (7) are cases of feature mismatch across sentences, why then is (6) bad, but (7) fine?
The goal of this paper is to argue that BP bare singulars are sometimes morpho-syntactically unspecified for number and sometimes fully specified for number (sg). Crucially, I will also argue that whether or not bare singulars enter a syntactic derivation specified for number has semantic consequences. In other words, the morpho-syntactic facts should not be kept apart from the semantic facts: with respect to number, morpho-syntactic under-specification equals semantic under-specification. 8 The rest of the paper is organized as follows. In section 2, I present data showing that bare singulars in BP are not always number neutral, and that under certain circumstances they are not compatible with plural interpretations. In section 3, I lay out some general assumptions concerning the interpretation of number features on both NPs and functional heads. In section 4, I discuss the number specification of nominal and functional heads in BP. In section 5, I explore the consequences of these assumptions when applied to BP bare nouns and show how they can account for the data presented in section 2. In section 6, I extend the analysis to full nominal phrases, that is, to cases in which an NP is not bare, but covered by a determiner projection. In section 7, I revisit the cases of discourse anaphora, paying attention to the dynamic behavior of bare singulars. Section 8 is a brief conclusion. According to the authors, "the data above suggest that bare singulars have no specification for number". But one can also construct minimal variations in which the bare singular can co-occur with a in-adverbial. (ii) below seems just fine: The data above seem to suggest that bare singulars MAY have no specification for number, but as an anonymous reviewer pointed out, it is far from obvious how to account for the contrast between (i) and (ii). Moreover, as another anonymous reviewer noticed, the bare noun in (i) is interpreted as one or more letters, whereas in (ii) it is interpreted as a single letter. This interaction between plurality and aspect is an interesting but intricate topic, which I leave for future work

Data
If bare singulars are semantically neutral with respect to number and contain both atomic and non-atomic entities in their denotations, they should be able to combine with certain expressions (or appear in certain environments) that require the presence of non-atomic arguments to trigger group-oriented interpretations. In this section, we will discuss three such expressions (environments) and compare the behavior of bare singulars and bare plurals to check whether or not these plural interpretations arise.

Anaphoric se
The clitic form 'se' in BP, when used as an anaphoric pronoun, can give rise to both reflexive and reciprocal interpretations, as illustrated in (8) This sentence can be used to talk about a group of students each of which praised himself or about a group whose members praised one another. As one might expect, in order for a reciprocal reading to arise, SE should have an antecedent that denotes a plurality. If its potential antecedent denotes a single individual, only the reflexive reading will show up: Both (10a) and (10b) have reciprocal readings. In the case of (10a) this is as it should be since the antecedent of SE is a bare plural. As for the case of (10b), it gives further support to the view that bare singulars are semantically neuter with respect to number. Just like (10a), it can be used to talk about students who were praising one another.
Consider now the following pair, again contrasting a bare plural and a bare singular: cumprimentando. geeting 'I saw a student/students who was/were greeting himself/themselves.' Once again, only the version with a bare plural has a reciprocal reading, whereas the version with a bare singular has only a (weird) reflexive reading.
Notice already that the contrasts in (11) and (12) are unexpected if bare singulars are neuter for number. They should trigger reciprocal (as well as reflexive) readings in both cases and no salient difference between bare plurals and bare singulars should be attested. 9 I will offer an account for the contrasts above in the next sections. What I would like to point out now is a difference between the examples in (10)  the ones in (11)/(12) that will turn out to be crucial to the explanation of these contrasts.
In (10), bare nouns are the (subject) argument of a gerundive verbal form. Gerunds in BP do not inflect for number (or person). Therefore, the same verbal form is used with bare singulars and bare plurals. On the other hand, in (12) the bare nouns (or the corresponding relative pronouns) are related to the subject position of a finite verbal form. Finite forms do inflect for number in BP, and, as shown in the examples, different forms surface depending on whether we have singular or plural subjects. As I mentioned above, this difference will be crucial to my proposal about the morpho-semantics of BP bare nouns. For the moment, let us look at some more examples highlighting the same point: These examples show the same pattern: reciprocal interpretations become unavailable for bare singulars when they are (or are related to) subjects of finite verbal forms. No such restriction is observed with bare plurals.

Cumulative Interpretations
Consider the following sentences: Both sentences have what may be called a cumulative interpretation. According to this interpretation, no friend of mine has to be living in many countries. What is required for the sentence to be true is that there is one or more friends of mine and a set of many countries, and each one of theses countries is the residence of at least one such friend of mine. In other words, the sentence can be used to talk about a group of 15 friends of mine and a group of 15 countries, with each one of these friends living in a different country. Consider now another pair of sentences: países. coutries 'There is/are a friend/friends of mine who is/are living in many countries Notice that we have replaced the gerundive forms in (16) with subject relative clauses with finite verbs in (17). Interestingly, only (17a), in which a bare plural is related to the finite verb, has a cumulative interpretation. (17b), with a bare singular, requires (at least) one friend of mine to be living in many countries.
Let países. countries 'There is/are a friend/friends of mine exiled in many countries'.
Here, only (18a) has a cumulative interpretation. (18b) requires some friend(s) of mine to be in many countries. This contrast is similar to the contrast observed above in (17)  países. countries 'There is a friend/friends of mine who exiled in many countries'.
Once again, only (19a) has a cumulative interpretation. (19b) requires some friend(s) of mine to be in many countries.
Crucially, participial verbal forms in Portuguese inflect for number (exilado/exilados), patterning thus with finite verbs and not with gerunds. We are facing the same contrast we had seen above with respect to anaphoric 'se': an overt morpho-syntactic agreement seem to (indirectly) correlate with semantic contrasts concerning the number interpretation of bare nouns. (20) illustrates the anaphoric reading. The second sentence says that John works in a project that has already been mentioned in the discourse, namely, the project on which that Mary works. (21) illustrates the group-internal reading. Both John and Mary are currently working in a single project. It is also compatible with a anaphoric interpretation. For instance, if what is being discussed is Pedro´s current project, then (21) can mean that John and Mary are both working in that project too. Of course, when uttered out of the blue, the anaphoric interpretation of (21) vanishes due to the absence of an appropriate discourse antecedent.
It should be clear that in order for the group-internal reading to arise, 'mesmoḿ ust relate to a group-denoting argument, such as a coordinating DP (John and Mary) or a plural definite (the employees). Otherwise, only the anaphoric reading will be available. In the second sentence in (20), for instance, 'mesmo' only has an anaphoric reading. (21), on the other hand, has both a group-internal and an anaphoric interpretation, as I have just pointed out.
Bearing this in mind, consider the following sentences: projeto. project 'I saw a student/students working on the same project.' In both sentences, mesmo has an anaphoric and a group-internal reading. Substituting a bare singular for a bare plural makes no salient difference in interpretation.
The situation changes when we replace gerunds with finite subject relative clauses: projeto. project 'I saw a student/students who was/were working on the same project'.
In this case, only in (23a) mesmo has a group-internal reading. (23b) can only be interpreted as saying that the student(s) that I saw were working in a project that had been previously mentioned in the discourse.
Below is another example: quarto. room 'There is a guest/guests who is/are sleeping in the same room'.
In (24), both sentences allow for a group-internal reading, whereas in (25), only the version with a bare plural allows for such reading.
Another contrast that is revealing is one in which bare singulars are used as the subject of a gerundive active verbal form and also of a gerundive passive verbal form, as shown below: quarto. room 'I saw a geust/guests who was/were being put in the same room'.
The relevant fact is that only in (26a) can we have a group-internal interpretation. Notice in (26b) the presence of the passive participle 'posto', which is inflected for number. If instead of a bare singular we had a bare plural, reciprocal readings would be possible in both sentences:

Interim Summary
The pattern that emerges from the data presented in this section can be summarized as follows: • When argumental bare singulars are syntactically related to verbal forms carrying singular morphology, "plural" interpretations are blocked.
• When argumental bare singulars are syntactically related to verbal forms carrying no number morphology, "plural" interpretations are allowed.
These are the generalizations that I will try to explain in the next sections. At this point, it is important to check whether the absence of plural interpretations discussed above correlates not with the denotation of bare nouns, but rather with the presence of overt singular morphology, as the generalizations above might seem to suggest. If this is the case, plural interpretations should not be available when singular verbal forms combine with group-denoting, singular arguments, as in (28)  The word 'criançada' is a group denoting expression meaning 'bunch of children'. 11 In (28), the word is singular and triggers singular morphology on the verb. Nevertheless, a reciprocal interpretation is readily available to the sentence. Other singular expressions that are group denoting behave the same way: What the data above suggests is that singular verbal morphology per se does not block plural readings of their agreeing subjects. Therefore, the reason behind the asymmetries found in the previous section between bare singulars and bare plurals should lie elsewhere, most likely on the denotation of these bare nouns. This is the road I will take on the following sections. 12

On the Interpretability of Number Features
In this section, I lay out some assumptions regarding the interpretation of number features on noun phrases and also on functional heads. In the next section, these general assumptions will be applied to BP data focusing on the relation between the morphology and the semantics of number.

Number Features on Nouns
I assume that number features -singular (sg) and plural (pl) -are interpretable on nouns. I follow Krifka (1992) and Kratzer (2004) and assume that uninflected nominal denotations are cumulative and that the role of sg and pl is to restrict the denotation of a noun, eliminating sums and atoms, respectively.
(31) N-= {a, b, c, a⊕b, a⊕c, b⊕c, a⊕b⊕c} N-sg = {a, b, c} N-pl = {a⊕b, a⊕c, b⊕c, a⊕b⊕c} Thus, an uninflected bare noun has both atoms and sums in its extension, being therefore neutral with respect to number. Number-inflected nouns, on the other 12 Just to make the parallel with many of the examples from the previous section clearer, adding relative clauses do not change the judgements: In both cases, the subject DP refers to a group whose members hugged each other. Thus, (singular) relative clause formation should not be blamed for the lack of plural readings either.
hand, denote sets of atoms (singular nouns) or sets of mereological sums (plural nouns). Whether or not number features host their own projection under DP/NP will not be important for our purposes in this discussion and for the sake of simplicity, I will represent them as features attached directly to nouns.

Number Features on Functional Heads
As for number features on functional heads (I, v, D, ...), I take them to be uninterpretable at the syntax-semantics interface and follow Chomsky (2000) and much subsequent work in assuming that they must be deleted (via agreement) before the syntactic representation is sent to semantic interpretation. 13 More precisely, I assume that a number-specified functional head F requires another number-specified expression X(P) in the same structure so that they can enter into an agreement relation and F´s uninterpretable features be eliminated. Only in (33a) did F find an appropriate number-specified XP that can check and consequently eliminate its uninterpretable number features. In (33b), F´s number feature will persist throughout the syntactic derivation and the resulting structure will not be legible by the semantic component. Notice that this is the case only under the assumption that F is specified for number. Were it not, the resulting syntactic structure would be fine and compatible with a number neutral X(P).

Number Specification in BP
With this general background in mind, we now turn to the number specification of functional and nominal heads in BP, paying special attention to the relation between morphology and semantics.

Verbal Functional Heads
The idea I would like to pursue for the number specification of BP verbal heads is very simple and intuitive. Verbal heads that overtly display overt number morphology are specified for number and those that do not display any overt number morphology are not specified for number, i.e., do not bear any number feature.
As the following table shows, this intuition separates finite and participial heads (of passives and reduced relatives) on one side and gerunds and little v (which hosts object-verb agreement) on the other. As we demonstrate later, this partition has important repercussions in our account for the interpretive asymmetries between bare singulars and bare plurals discussed in section 2. But first, we will look at the number specification on BP nouns.

Noun Phrases
My core assumption concerning number marking on BP nouns is stated in (34): (34) BP nouns may enter a syntactic derivation without number features.
I will capitalize on the fact that there is no overt singular morpheme on BP nouns, whereas there is a plural one (-s). The following table shows the spellout rules for singular, plural and unspecified nouns: 14 (35) Number Spell-out on BP nouns Since there is no overt singular morpheme in BP (sg is a zero morpheme), a bare singular NP such as menino is ambiguous: it can be interpreted as singular or number-neutral. However, it is important to notice that although leaving a noun unspecified for number is an option in BP, whether or not this will result in a well-formed structure ultimately depends on the funcional heads with which it is related, as we show in detail in the following section.

BP Bare Nouns and Agreement
According to what we proposed in section 4, finite T and adjectival participles are specified for number. Since the number features on these heads are uninterpretable, they require that a related argument be specified for number. In the case of bare nouns, this excludes the possibility of having a number neutral bare singular as an agreeing argument for both finite T and participial heads, as illustrated below: In (37a), T and the subject NP enter into an agreement relation and the uninterpretable number feature of T is deleted. In (37b), the deletion does not occur due to the fact that the subject NP is not specified for number. Both little v and Gerunds behave differently since they are not specified for number and can be related with underspecified bare nouns. An immediate consequence of all this is that interpretations that require a non-atomic argument will not arise in the presence of a bare singular and a related finite T or a participle. Thus, although bare singulars can in principle be singular or number neutral, the surrounding structure may impose a number specification (singular) on them and induce a more restricted set of possible interpretations. In this sense, morphology and semantics go hand in hand as far as number specification on bare nouns is concerned.

Discourse Anaphora Revisited
As we pointed out in the introduction, the main source for the intuition that BP bare singulars are number neutral is the fact that they can serve as discourse antecedents of plural anaphoric pronouns, as in (45) We attributed the absence of a reciprocal interpretation for cases such as (46) to a singular number feature attached to the bare singular NP. The question that arises at this point is whether or not this bare singular can serve as the discourse antecedent of a plural anaphoric pronoun. One might expect a negative answer here, since being morphologically and semantically singular, the bare noun should behave on a pair with singular indefinites, which are not good antecedents for plural pronouns under these circumstances: However, even a bare singular that is interpreted as singular can serve as an antecedent for a plural discourse pronoun, as (48) below attests. This behavior contrasts thus with singular indefinites, as shown in (49) The situation might look paradoxical, but it is not. Let us see why. First, notice that singular indefinites and bare singulars are both interpreted as existential in-definites and make the same contribution to the truth-conditions of the sentences in which they appear. Thus, the first sentences of both (48) and (49) have the meaning roughly represented in (50): In (50), x is a variable ranging over atomic individuals. The sentence says that there was one guest who was greeting himself. Notice, however, that (50) is compatible with the existence of more than one such person. Although the static, truth-conditional contribution of the singular indefinite in (48) and the bare singular in (49) is the same, their dynamic, discourse-related contribution need not mimic this similarity. This is exactly what I want to suggest. In a nutshell, I assume that whereas a singular indefinite introduces a (atomic) discourse referent, along the lines of Kamp (1981) and Heim (1982), bare singulars do not. Their potential for serving as discourse antecedents comes from a different process, called abstraction in Kamp and Reyle (1993). 15 .
Roughly speaking, abstraction is a process of set formation corresponding to the intersection of what is usually called the restrictor and the nuclear scope of certain determiners. In the case of (48), we have the intersection between the set of guests and the set of individuals who were greeting themselves. It is this set that provides the referent for the plural pronoun in the subsequent discourse. A similar case can be made for determiners such as 'mais de um/more than one', as in the example below: 16 In (51), it is assumed that the non-referring expression mais de um aluno ('more than one student') combines with the predicate VP, resulting in truth-conditions stating that the intersection between the set of students and the set of individuals who answered the question has cardinality greater than one. It is to this set that the pronoun eles ('they') refers. What I am proposing is that the same process applies to bare singulars in BP (assuming the presence of a covert existential determiner for bare NPs). Thus, we need no new machinery to explain the dynamic differences between singular indefinites and bare singulars.
We are also in a position to address a difference pointed out in the introduction between gender and number mismatch. The relevant example is repeated below: crying 'There is a child in the room. She/he is crying'.
Recall that the noun 'criança' is morphologically feminine, but semantically neuter and can refer to both boys and girls. However, it cannot serve as the antecedent of a masculine anaphoric pronoun. Contrastingly, the morphologically singular NP 'criança' can serve as the antecedent of a plural anaphoric pronoun, as we have just emphasized. Here, we may credit the asymmetry between gender and number to the non-existence of anything akin to abstraction in the domain of gender.
Summing up, even when bare singulars are morphologically specified for number and has the same contribution to the static meaning of a sentence as singular indefinites do, their dynamic, discourse potential is different, and through a process of set formation they can still (although indirectly) serve as antecedents to plural discourse anaphora.

Conclusion
In this paper, I provided evidence against the idea that BP bare singulars are morphologically singular but semantically number neuter. I have argued instead that they are sometimes specified for number (singular) and sometimes underspecified and that this has semantic consequences. I developed an account according to which whether or not a bare noun can enter a syntactic derivation under-specified for number depends on the functional heads to which it is syntactically related. The main conclusion is that, as far as number features in BP noun phrases are concerned, morpho-syntactic specification should NOT be kept apart from semantic interpretation.