The interpretation of BNs vs. that of other nominals

Spanish and Catalan object BNs are semantically distinct not only from proper names and definites but also from three other classes of indefinite nominal expressions: singular indefinites (see (2) above), bare plurals (henceforth, BPs; see (7a)), and bare mass terms (see (7b)).

1. (7)

   1. (a) Busco pisos. / Busco pisos. look.for.1sg apartments look.for.1sg apartments ‘I'm looking for apartments.’

   2. (b) Busco cerveza. / Busco cervesa. look.for.1sg beer look.for.1sg beer‘I'm looking for beer.’

At least two tests show that BNs are semantically more similar to BPs than they are to singular indefinites with a determiner, in spite of the English glosses for the examples in (1) and (3).Footnote ³ First, the data in (8) show that the scopal behavior of BNs resembles that of BPs rather than that of singular indefinites: whereas the bare nominals in (8a) and (8b) only take narrow scope with respect to negation, the singular indefinite in (8c) can take either narrow scope (i.e. the speaker is not looking for any apartment) or wide scope (i.e. there is an apartment the speaker is not looking for).

1. (8)

   1. (a) No busco piso. / No busco pis. not look.for.1sg apartment not look.for.1sg apartment ‘I'm not looking for an(y) apartment.’

   2. (b) No busco pisos. / No busco pisos.not look. for.1sg apartments not look.for.1sg apartments ‘I'm not looking for (any) apartments.’

   3. (c) No busco un piso. / No busco un pis. not look.for.1sg an apartment not look.for.1sg an apartment ‘I'm not looking for any apartment./There is an apartment I am not looking for.’

Second, the data in (9) illustrate – using Catalan – a difference involving the effect of the nominal on the telicity of a sentence. Neither BNs nor BPs can induce a telic reading on a verb otherwise unspecified for telicity. (9a) and (9b) exclude the temporal adjunct en ‘in’ admitted by telic predicates, allowing instead only a durative adjunct. In contrast, a singular indefinite can induce telicity, as the acceptable adjunct en in (9c) shows (In the examples here and afterwards ‘#’ indicates anomaly.)Footnote ⁴

1. (9)

   1. (a) Ha buscat pis #en una setmana. / durant una setmana.has looked.for apartment in a week during a week‘(S)he has looked for an apartment for a week.’

   2. (b) Ha buscat pisos #en una setmana. / durant una setmana.has looked.for apartments in a week during a week‘(S)he has looked for apartments for a week.’

   3. (c) Ha buscat un pis en una setmana / durant una setmana.has looked.for an apartment in one week during a week‘(S)he has looked for and found an apartment in a week./(S)he has looked for an apartment for a week.’

This similarity between BNs and BPs cannot be explained by positing that BNs are coerced into a mass interpretation. BNs are distinct from mass nouns,Footnote ⁵ as is most clearly evident in the fact that, unlike mass nouns, they cannot be modified by a measure phrase, as the contrast in (10), from Catalan, shows.

1. (10)

   1. (a) L'ampolla té doslitresd'aigua. (tenir aigua ‘have water’) the.bottle has two liters of.water ‘The bottle contains two liters of water.’

   2. (b) *L'edifici té unatoneladad'ascensor. (tenir ascensor ‘have a lift’)the.building has one ton of.lift

Despite these similarities, in other ways BNs contrast not only with singular indefinites but also with BPs. First, BNs, but not singular indefinites or BPs, are number neutral. Farkas & de Swart (Reference Farkas and Swart2003) argue for number neutrality on the basis of Hungarian examples like those in (11), which show that BNs in Hungarian can serve not only to describe what is likely to be an atomic individual (see (11a)), but also to describe what must be a group of objects (see (11b)).

1. (11)

   1. (a) Feri feleséget keres. Feri wife.acc seek ‘Feri is looking for a wife.’ (i.e. Feri is ‘wife-seeking’.)

   2. (b) Mari bélyeget güjt. Mari stamp.acc collect ‘Mari is collecting stamps.’ (i.e. Mari ‘stamp-collects’.)

Since there are no predicates like the English collect in Spanish and Catalan that take BN complements (see below), we use a slightly different diagnostic to illustrate number neutrality. When a singular indefinite or a BP is introduced in declarative discourse, the possibilities for felicitous continuations show sensitivity to number: if a singular indefinite is used, the default implication is that it describes an atomic individual; if a BP is used, the default implication is that a non-atomic individual is being described. The contrast in the acceptable vs. unacceptable continuations in the Catalan examples in (12a, b) illustrates. The first sentence in (12a) strongly implies that the speaker is looking for exactly one apartment; the first sentence in (12b) implies the opposite.

1. (12)

   1. (a) Busco un pis. {Un a Barcelona. / #Un a Barcelona i un a Girona} look.for.1sg an apartment one in Barcelona one in Barcelona and one in Girona ‘I'm looking for an apartment. One in Barcelona.’

   2. (b) Busco pisos. {#Un a Barcelona. / Un a Barcelona i un a Girona} look.for.1sg apartments one in Barcelona one in Barcelona and one in Girona ‘I'm looking for apartments. One in Barcelona and one in Girona.’

However, the behavior of BNs is strikingly different: the first sentence in (13) is equally compatible with subsequent reference to one apartment or to more than one, indicating that the BN carries no implication concerning the plurality or singularity of the individual(s) it describes.

1. (13) Busco pis. {Un a Barcelona. / Un a Barcelona i un a Girona}. look.for.1sg apartment one in Barcelona one in Barcelona and one in Girona ‘I'm looking for an apartment. {One in Barcelona./One in Barcelona and one in Girona.}’

In this respect Spanish and Catalan BNs closely resemble those in Hungarian.

A second contrast between BNs vs. BPs and singular indefinites involves the general failure of the former to take on any discourse referential properties. There are various ways in which this contrast manifests itself. Consider first discourse anaphora. To make this point we use Catalan, which has a richer system of pronominal anaphora than does Spanish. In addition to personal pronouns, Catalan, like French and Italian, has a partitive pronoun en. This pronoun picks up a common noun denotation only (leaving aside the contribution of numerals or determiners), as seen in the following contrast.

1. (14)

   1. (a) Pensava que veuria dos amics, però n'he vist tres. thought.1sg that see.cond.1sg two friends but part.have.1sg seen three ‘I thought I would see two friends, but I have seen three.’

   2. (b) *Pensava que veuria elmeuamic i n'he vist. thought.1sg that see.cond.1sg the my friend and part.have.1sg seen

In (14a) en is anaphoric to amics; (14b) is unacceptable because the context suggests that anaphora to el meu amic is intended, which is impossible with en. If anaphora to el meu amic is desired, a personal pronoun must be used.

1. (15) Pensava que veuria elmeuamic i l'he vist. thought.1sg that see.cond.1sg the my friend and him.acc.have.1sg seen ‘I thought I would see my friend, and I have seen him.’

With this background in mind, observe that while BPs and singular indefinites freely serve as antecedents for personal pronouns (assuming no scope-taking element intervenes to block discourse anaphora), BNs are severely limited in their ability to do so.Footnote ⁶ The contrast in (16), from Catalan, illustrates. Note that the sentence containing the BN is in an episodic tense and that there is no scope-taking element that could plausibly block discourse anaphora.

1. (16)

   1. (a) Avui porta faldilla. #La hi vam regalar l'any passat. today wear.3sg skirt it.acc her.datpast.1pl give.present the.year last ‘Today she is wearing a skirt. We gave it to her as a present last year.’

   2. (b) Avui porta unafaldilla. La hi vam regalar l'any passat. today wear.3sg a skirt. it.acc her.datpast.1pl give.present the.year last ‘Today she is wearing a skirt. We gave it to her as a present last year.’

   3. (c) Avui porta bracelets. Els hi vam regalar l'any passat. today wear.3sg bracelets them.acc her.datpast.1pl give.present the.year last ‘Today she is wearing bracelets. We gave them to her as a present last year.’

The only pronoun which can be systematically used to refer back to a BN is en:

1. (17) Avui porta faldilla. Li'n vam regalar una l'any passat. today wear.3sg skirt her.dat.part past.1pl give.present one the.year last ‘Today she is wearing a skirt. We gave her one as a present last year.’

These anaphora facts strongly suggest that BNs have the same denotation as common nouns – i.e. that they denote properties – and that they are not used to introduce discourse referents to token individuals of the sort that a personal pronoun identifies.Footnote ⁷

A second reference-related contrast between BNs, on the one hand, and singular indefinites and BPs, on the other, involves their ability to license non-restrictive relative clauses. Unlike singular indefinites and BPs, BNs are unable to host non-restrictive relative clauses which describe a token individual, as (18), from Catalan, shows:

1. (18)

   1. (a) *Per fi hem trobat pis, que començarem a reformar molt aviat. for final have.1sg found apartment that begin.fut.1pl to renovate very soon

   2. (b) Per fi hem trobat unpis, que començarem a reformar molt aviat. for final have.1sg found an apartment that begin.fut.1pl to renovate very soon ‘At last we have found an apartment, which we'll begin to renovate very soon.’

   3. (c) Per fi hem trobat pisos, que començarem a reformar molt aviat. for final have.1sg found apartments that begin.fut.1pl to renovate very soon ‘At last we have found apartments, which we'll begin to renovate very soon.’

As observed in Potts (Reference Potts2005) and elsewhere, non-restrictive relative clauses are only licensed if the (full) expression they modify denotes an entity of the sort the relative clause can apply to. The facts in (18), like those involving discourse anaphora, thus strongly indicate that BNs neither denote token individuals nor in any other way introduce discourse referents corresponding to such individuals.

A third contrast between BNs on the one hand, and singular indefinites and BPs, on the other, is that a BN cannot be the controller of an implicit subject, whereas singular indefinites and BPs can. See the Catalan data in (19).

1. (19)

   1. (a) *Busco pis per pro ser compartit. look.for.1sg apartment to be shared

   2. (b) Busco unpis per pro ser compartit. look.for.1sg an apartment to be shared‘I'm looking for an apartment to be shared.’

   3. (c) Busco pisos per pro ser compartits. look.for.1sg apartments to be shared‘I'm looking for apartments to be shared.’

Finally, the examples in (20)–(22), again from Catalan, show that locative modifiers (a la mà ‘in his hand’, al garatge ‘in the garage’) and secondary predicates (a punt ‘ready’) are incompatible with BNs, but not with singular indefinites or BPs.

1. (20)

   1. (a) En Joan porta motxilla (*a la mà). det Joan carries backpack at the hand ‘Joan carries a backpack in his hand.’

   2. (b) En Joan porta la motxilla a la mà. det Joan carries the backpack at the hand ‘Joan carries the backpack in his hand.’

1. (21)

   1. (a) Tinc cotxe (*al garatge ).have car at.the garage ‘I have a car in the garage.’

   2. (b) Tinc un cotxe/cotxes al garatge. have a car/cars at.the garage ‘I have a car/some cars in the garage.’

1. (22)

   1. (a) Tinc cotxe (*a punt). have car at point

   2. (b) Tinc el cotxe a punt. have the car at point ‘I have the car ready.’

We will argue below that for both control and small clause predication, the syntax requires the BN to appear in a subject position, and that such positions require some sort of discourse referent associated with them which neither the BN nor any other element in the sentence can provide.

Despite the robustness of the contrasts in (16)–(22), discourses such as those in (23) are acceptable, in which a BN does appear to license discourse anaphora to the third person accusative pronouns la and el.

1. (23)

   1. (a) Per la festa es va posar faldilla. Se l'havia comprat la tarda anterior. to the party clpast.3sg put.on skirt cl it.acc.had bought the afternoon before ‘She put on a skirt for the party. She had bought it the day before in the afternoon.’

   2. (b) Ja tinc pis. El vaig comprar ahir. already have apartment it.accpast.1sg buy yesterday ‘I already have an apartment. I bought it yesterday.’

However, we maintain that in such examples the pronoun in question is not directly anaphoric to the bare singular but rather has an antecedent that is accommodated by the hearer into the common ground (see e.g. Lewis Reference Lewis1979, Beaver & Zeevat Reference Beaver, Zeevat, Ramchand and Reiss2007 on accommodation). Though a full account of what permits accommodation in (23) but not (16a) must await future research, we offer the following suggestion. An important difference between the discourses in (16) and those in (23) lies in the rhetorical relations they manifest and the corresponding intuition about what the discourse topics are in each case (see Jasinskaja Reference Jasinskaja2008 for a recent discussion of rhetorical relations and topicality). The second sentence in each of the acceptable examples in (16) is connected to its predecessor only insofar as it provides a description of an object which is mentioned in the predecessor – a skirt or bracelets. Given that the verb in the first sentence (portar ‘wear’) has a low degree of informativity in relation to clothing/accessories and appears in the simple present tense, this sentence is most naturally understood as simply introducing the clothing/accessories as a discourse topic, and the second sentence in each case elaborates on the description of the item in question. (16a) thus shows that a BN cannot be used to introduce a new discourse topic – exactly what we expect if it fails to introduce a discourse referent corresponding to an individual.

In contrast, in (23), various elements in each sentence conspire to make it easier to interpret the discourse topic as something other than just an individual putatively corresponding to the BN. In (23a), two factors contribute to the intuition that the discourse topic could be something like ‘What did the subject wear on which occasion (and why)?’ First, the PP per la festa ‘to the party’ appears in a position associated with contrastive topics, facilitating the intuition that the party could be contrasted with some other occasion. Second, the continuation in (23a) places the skirt buying close in time to the skirt wearing, suggesting that (in contrast to what happens in (16a)), the continuation serves not so much to further describe the skirt as to explain why the skirt was worn. The implication of contrast (here, between what was worn to the party vs. on other occasions) has been shown by Ward, Sproat & McKoon (Reference Ward, Sproat and McKoon1991) to facilitate discourse anaphora in cases similar to those we are considering here.

In the case of (23b), the adverb ja ‘already’ implies the information that the speaker has an apartment sooner than might have been expected (see e.g. Mittwoch Reference Mittwoch1993 on already), indicating that this discourse must be added to a common ground in which some discussion of an apartment has already taken place, including possibly the specific apartment that the speaker bought – in other words, the discourse topic is more the apartment search and purchase than the apartment itself. However, Ward et al. (Reference Ward, Sproat and McKoon1991) also show that discourse anaphora can be facilitated when the pronoun refers to an object that is sufficiently topical. The apartment certainly qualifies in this case.

From the preceding discussion we conclude that if BNs introduced discourse referents to individuals (as singular indefinites and BPs sometimes do), anaphora to the BN with an accusative pronoun would always be possible in the same cases that such anaphora is possible for singular (in)definites and BPs. The fact that it is not clearly calls for a semantics that distinguishes BNs from other kinds of (in)definites.

The predicates licensing object BNs

Let us now consider the verbs that license these BNs. BNs in Spanish and Catalan are largely restricted to a particular class of verbs that we will call ‘have’-predicates, following Borthen (Reference Borthen2003), Footnote ⁸ who argues that a similar class is relevant for the distribution of BNs in Norwegian. According to Borthen (Reference Borthen2003: 190), a ‘have’-predicate is a word that introduces either explicitly or implicitly what she calls a profiledhave-relation, where ‘a have-relation is an asymmetrical coexistence relation between two arguments, called the possessor and the possessed, where the possessor is superior to the possessed rather than the other way round. … An argument can be superior to some other argument in terms of control, part–whole dependency, animacy, or point of view’. Though there are some differences between the distribution of BNs in Norwegian vs. Spanish and Catalan, to be discussed below, this definition offers a good first approximation of the class of verbs involved.

As the examples we have already used indicate, the class of predicates that license BNs in Spanish and Catalan is not limited to verbs of having, strictly speaking, such as tener/tenir ‘have’, or poseer/posseir ‘possess’, the latter exemplified in (24), from Spanish.Footnote ⁹

1. (24) Este proyecto posee licencia municipal. this project possesses permit municipal ‘This project has a permit from the city.’

Rather, the class also includes a few intensional transitive verbs that entail a relation that could be expressed via a verb of having in the relevant possible world, such as necesitar/necessitar ‘need’, buscar/buscar ‘look for’, as well as a small set of extensional verbs that entail a possessive or locative relation, essentially limited to ponerse/posar-se ‘put on (e.g. clothing)’, llevar/portar ‘carry’, usar/fer servir ‘use’, comprar/comprar ‘buy’, encontrar/trobar ‘find’, obtener/obtenir ‘obtain’, recibir/rebre ‘receive.’ (25) and (26) offer additional examples from Spanish and Catalan, respectively.

1. (25)

   1. (a) Encontramos taxi. found.1pl taxi ‘We found a taxi.’

   2. (b) Ha obtenido permiso de trabajo. has obtained permit of work ‘(S)he has obtained a work permit.’

1. (26)

   1. (a) M'acabo de comprar cotxe. refl.finish.1sg of buy car ‘I just bought myself a car.’

   2. (b) Ahir finalment vaig rebre carta. yesterday finally past.1sg receive letter ‘Yesterday I finally received a letter.’

The class also includes the existential predicate haber/haver-hi ‘there be’, which, while perhaps not an obvious candidate for a ‘have’-predicate, is composed literally of the verb haber/haver ‘have’ plus, in the case of Catalan, a locative clitic. We will show how this verb fits naturally into the class of ‘have’-predicates in Section 5.Footnote ¹⁰ On the other hand, a few predicates which might be expected to fall into this category, notably voler ‘want’ and desitjar ‘desire’, are severely restricted in their ability to combine with BNs (see Espinal & McNally Reference Espinal, McNally, Espinal, Leonetti and McNally2009). Although this fact requires further investigation, on our analysis the verb must undergo a lexical rule in order to be able to occur in the syntactic structure which licenses BNs and to combine with them semantically (see Sections 3 and 4 below), and thus it is not surprising that there might be some arbitrariness in the specific verbs involved.

Finally, note that although this class of verbs is similar to the class that combines with BNs in Norwegian (Borthen Reference Borthen2003), it contrasts sharply with the verbs that take BNs in Hindi. Dayal (Reference Dayal2003) observes that BNs in Hindi are licensed when the resulting predicate describes an institutionalized activity; that is, the BN has to describe an object which is typically a participant in the event described by the verb. This is not the case in Spanish or Catalan. Thus, in spite of the fact that reading books or selling newspapers might be considered institutionalized activities, both (27a) and (27b) are ill-formed in Spanish, even though their equivalents are acceptable in Hindi.

1. (27)

   1. (a) #Juan leía libro. Juan read.past.3sg book

   2. (b) #María vende diario. María sells newspaper

Also note that the distribution of BNs does not correlate with unaccusativity; BNs are excluded with most unaccusative verbs, as the Catalan examples in (28) show.

1. (28)

   1. (a) *Vam aconseguir que arribés metge. past.1pl manage that arrive doctor

   2. (b) *Passa tren. pass.3sg train

   3. (c) *Creix flor. grows flower

There are two reasons for this restriction. First and most obviously, unaccusative verbs do not qualify as ‘have’-predicates (the only prima facie exception being the existential predicate, on which see Section 5). Second, most unaccusatives have only one argument, and BNs are never licensed as subjects, not even of the passives of predicates that normally permit them – see Catalan (29a, b).Footnote ¹¹ They are also excluded as subjects of middles – see (29c).

1. (29)

   1. (a) *Pis ha estat comprat. (periphrastic passive) apartment has been bought

   2. (b) *Pis s'ha comprat. (pronominal passive) apartment cl.has bought

   3. (c) *Pis es compra fàcilment. (middle) apartment cl buys easily

In this respect, Spanish and Catalan BNs contrast with those in Norwegian, which can be subjects of passivized ‘have’-predicates (Borthen Reference Borthen2003).

In addition to these restrictions on the verbs that combine with BNs, there is a constraint on the interpretation of the resulting predicate: the BN is licensed only if, in the specific context of use, the verb phrase denotes a characterizing property of the external argument.Footnote ¹² As mentioned above (see (27)), this property is not necessarily a prototypical, stereotypical or institutionalized property: an examination of the general cultural context is not sufficient to determine which exact combinations of V+N are acceptable, as would be expected if the property had to be stereotypical or institutionalized. It also need not be a temporally stable property of the individual in question, as shown by examples such as those in (25) and (26). Rather, what we mean for a property to be characterizing is that it is relevant in the context to distinguish between whether or not an individual has the property in question. This entails that, while there are lexical restrictions on the verbs that allow BNs in object position in Romance, there is no such restriction on the set of nouns, the latter being subject only to these contextual restrictions. This is best shown by example. For instance, uttered out of the blue, the Catalan sentences in (30) sound odd.

1. (30)

   1. (a) (#)En Joan té joguina.Footnote ¹³det Joan has toy ‘Joan has a toy.’

   2. (b) (#)Aquest ordinador té virus.this computer has virus ‘This computer has a virus.’

If a hearer judges a sentence like (30a) as odd, this is due to the fact that, while it might be informative to hear that Joan has a toy, whether or not he has one does not strike the hearer as particularly relevant for any other purpose in the context. However, we can easily construct a context in which the sentence is felicitous. Imagine that a nursery school teacher is organizing an activity for which each child in the group must have a toy. Just as the activity is about to start, the teacher checks to see which children have toys and which do not. In such a situation, (30a) could be uttered as a confirmation that the child in question is ready to participate in the activity. Similarly, (30b) could be uttered by someone who is telling a system administrator which computers in a computer room need attention and which do not.

Of course, some properties that verb phrases with BNs describe are easy enough to imagine as relevant in this way, or are frequently enough treated in this way, so that they do not need any particular context at all to be felicitous or we can readily construct a context in which they would be. We would maintain that this is the case for all of the acceptable examples that have been used above.

With the Catalan existential predicate haver-hi ‘there be’, the contrast in (31) also shows that BNs are more or less acceptable depending on how easy it is to imagine the property in question as potentially characterizing what we will call a situational argument (represented by the clitic hi ‘there’, which is frequently bound to a left-dislocated locative or temporal phrase, as well).

1. (31)

   1. (a) #Al carrer hi ha cotxe. at.the street there has car

   2. (b) A aquest hotel hi ha piscina. at this hotel there has pool ‘There is a pool in this hotel.’

That is, we attribute the contrast that speakers find between the sentences in (31) to the fact that one can readily imagine situations in which it might be significant that a hotel has a pool or not, whereas streets are arguably less likely to be characterized according to whether they have cars on them at some particular moment or not. The requirement that the verb phrase be characterizing plays an important role in explaining the fact that not all types of nouns sound equally felicitous as BN complements with all types of ‘have’-predicates, and it points to an important contrast between BNs and both singular indefinite and BP complements.

The phrase ‘characterizing property’ might lead one to think that V+N combinations have something in common with so-called individual-level predicates (Carlson Reference Carlson1977). However, this is not the case, as we can easily find examples of predicates which clearly do not possess two of the three properties that Jäger (Reference Jäger2001) argues contribute to making a predicate behave as individual-level. First, unlike prototypical individual-level predicates, V+N combinations can denote transitory properties – indeed, they need not even be stative, as shown by the acceptability of the construction in the progressive, which resists stative predicates (see (32), from Spanish):

1. (32) Pedro está buscando apartamento. Pedro is looking.for apartment ‘Pedro is looking for an apartment.’

Second, again unlike individual-level predicates, V+N combinations can appear as complements to perceptions verbs:Footnote ¹⁴

1. (33) Hemos visto a Pedro usar bastón. have.ipl seen acc Pedro use cane ‘We have seen Pedro use a cane.’

Additional evidence for the characterizing nature of the resulting verb phrase is seen in the kinds of modification it does or does not allow. When the modifier clearly indicates that the property is not intended to characterize the individual, even for a short time (as implied with a les tres ‘at 3 o'clock’ in Catalan (34a)), the sentence is unacceptable. Conversely, the appearance of a DP structure together with a modifier such as a les tres induces a strong preference for a non-characterizing reading (see e.g. Bleam Reference Bleam2006).

1. (34)

   1. (a) En Joan ha portat motxilla toteldia /*alestres.det Joan has carried backpack whole the day at the three‘Joan has carried a backpack all day.’

   2. (b) En Joan ha portat una motxilla toteldia / alestres.det Joan has carried a backpack whole the day at the three‘Joan (has) carried a backpack all day/at 3 o'clock.’

To summarize this part of the discussion, we have shown that the set of verbal predicates that allow BNs in object position is lexically restricted to a class that we call the ‘have’-predicates. In contrast, there is no lexical restriction on the nominals that can combine with these verbs, as long as the resulting verb phrase denotes a characterizing property, in the sense we have introduced here, of the subject or a situational argument. Thus, the V+N construction is a kind of hybrid: it has a lexical dimension which places certain limits on the verbs that can participate in it, but it also has a productive dimension insofar as what can count as a characterizing predicate given the restriction on the verb is quite open. Thus, a lexical analysis of ‘have’-predicates must be combined with a general compositional rule that combines the verb and BN. But before we propose the semantics for this construction, we first turn to its syntax.

The starting point: Dayal's (Reference Dayal2003) semantics for pseudo-incorporation

What will become clear in this section is that – in the monadic structure in (35) – a nominal expression can appear in object position despite the fact that it does not semantically saturate the verbal predicate. This is the approach we will take: to treat Spanish and Catalan BNs semantically as true verb modifiers, and the participant they provide information about as an implicit semantic argument.

Perhaps the earliest attempt to formalize a semantics for object nominals as verb modifiers appears in de Hoop (Reference de Hoop1992). De Hoop assigns what she terms weak object nominals to type 〈〈e,t〉,〈e,t〉〉; the nominal then takes the verb it combines with as its argument. However, we want to avoid such an abstract semantic type for BNs, treating them instead as denoting (first-order, type 〈e,t〉) properties without any specification for number or (thus) the semantic and pragmatics effects number might bring along with it (see Farkas & de Swart Reference Farkas and Swart2003 for relevant discussion in relation to the contrasts between BNs and BPs). Very briefly, our reasons for assigning BNs such a denotation are the following.

First, a property-type denotation is the null hypothesis given the usual treatment of common nouns in formal semantics and the fact that Spanish and Catalan have standard determiner systems in which the determiner combines with a property-type nominal to form either an entity- or quantifier-denoting expression. Second, it is in line with specific proposals in recent work on bare nominals (see e.g. Borthen Reference Borthen2003 for Norwegian, Dayal Reference Dayal2003 for Hindi, Farkas & de Swart Reference Farkas and Swart2003 for Hungarian, and Dobrovie-Sorin et al. Reference Dobrovie-Sorin, Bleam, Espinal, Tasmowski and Vogeleer2006 for Romance). Thirdly, there are various arguments against assigning an entity- or quantifier-type denotation to the BNs. For example, such denotations fail to account for the obligatory narrow scope of BNs (see (8a)); other examples of entity- and quantifier-denoting expressions in Spanish and Catalan – notably, singular indefinite nominals – can in principle take wide scope, as observed above. In addition, an entity- or quantifier-type denotation fails to explain the very limited distribution of BNs: if BNs denoted entities or quantifiers, we would expect to find them in all kinds of positions, contrary to fact. Finally, a property denotation, but no other type of denotation, accounts directly for the discourse anaphoric behavior of BNs which, as shown in (17), only license anaphora to the pronoun en.

Our goal, therefore, is a semantic analysis of BNs on which they denote properties and function as verb modifiers. Such an analysis is Dayal's (Reference Dayal2003) semantics for pseudo-incorporation in Hindi, which, as mentioned above, addresses a combination of verb with object BN similar to the one under discussion here. Dayal's proposal suppresses the theme in the ‘deep’ lexical semantics of the verb and treats the direct object as forming a kind of complex predicate with the verb. However, the technical means by which this is achieved does not involve reducing the superficial valency of the verb. Rather, pseudo-incorporating verbs are lexically specified to select for two expressions: one that denotes a property, and one that denotes an entity, as in (39). (In this latter respect, Dayal's analysis resembles Semantic Incorporation; see Section 4.4). In addition, these verbs introduce the condition that the event they describe is appropriately classificatory, as defined in (40) (see Zimmer Reference Zimmer1972, Downing Reference Downing1977, Dowty Reference Dowty1979).

1. (39) λP _〈e,t〉λxλe[P–V(e)∧Ag(e)=x∧appropriately-classificatory(e)]

1. (40) An event denoted by a predicate δ that incorporates a property γ is appropriately classificatory iff

   ◊_probable(∃e[δ(e)∧∃y[Ag(e)=y]∧∃x[γ(x)∧Th(e)=x]]) (extensional verbs)

   ◊_probable(∃e[δ(e)∧∃y[Ag(e)=y]∧Th(e)=γ]) (intensional verbs)

Intuitively, (40) says that an event of type δ is appropriately classificatory if and only if its theme is typically an individual of type γ. As we observed earlier, however, this notion, though apt for Hindi, explains neither the unacceptability of Spanish examples such as those in (27) nor the acceptability of Catalan sentences like (41): it cannot be plausibly argued that a (drinking) glass is an object that one is highly likely to have in the way that a book is something that one is highly likely to read.

1. (41) No tinc got. not have glass ‘I don't have a glass.’

Rather, we would maintain that (41) is perfectly natural because in a situation like preparations for a meal, for instance, it is relevant to know whether an individual has a complete place setting or not, so that any missing items can be supplied.

Let us now consider the way in which the BN combines with the verb on this analysis. The event argument plays a key role, as it constitutes a locus for property ascription, crucially allowing the BN to contribute descriptive content without forcing the presence of an individual variable associated with it. Suppressing the theme is what blocks discourse anaphora to the incorporated nominal: if there is no existentially bound variable corresponding to the theme argument in the logical representation, Dayal's reasoning goes, there will be no variable for a subsequent pronoun to be coindexed with. Note that, in contrast, nothing need block the possibility of anaphora with a property-denoting pronoun – the pronoun simply needs to be able to refer back to the property itself, which is explicitly represented.

In addition to accounting for the anaphora facts, Dayal's proposal stands out in its attempt to capture the intuition that the BN, despite functioning syntactically as an object, is semantically a modifier. What remains unclear is how exactly P– V in (39) is to be understood: the relationship between the verbal predicate V and the bare singular predicate P is left to intuition. Dobrovie-Sorin et al. (Reference Dobrovie-Sorin, Bleam, Espinal, Tasmowski and Vogeleer2006) attempt to make this more explicit with the representation in (42).

1. (42) λPλxλe[V(e)∧Ag(e)=x∧Th(e) has P∧appropriately-classificatory(e)]

Specifically, (42) encodes the fact that the property denoted by the BN is attributed to the understood theme of the verb, which is not represented directly in the combinatorial semantics but is nonetheless manifest in the lexical semantics of the verb as Th(e).Footnote ²⁰

On the other hand, both the proposal in (39) and that in (42) have the inelegant characteristic of purportedly suppressing the verb's internal argument while maintaining a bivalent combinatorial semantics for the verb. There could be two reasons for doing this: (i) maintaining a bivalent verb is one way to guarantee that the verb will actually combine with two nominals (as it must); and (ii) it keeps semantic composition maximally simple, as it allows composition to proceed via function application. But, if we can guarantee in the syntax that the verb will have the complement it needs, and if we account for the suppression of the internal semantic argument by means of a lexical rule while keeping the idea that some nominals are modifiers of the verbs they occur with by a compositional rule, we will avoid the inelegance just mentioned.

The lexical rule for ‘have’-predicates

We begin by developing the lexical rule we need. As mentioned in Section 2.2, a promising starting point for a lexical analysis of the verbs we are considering is found in Borthen (Reference Borthen2003). Borthen proposes that BNs in Norwegian are licensed (inter alia) as the possessed argument of a predicate that introduces an asymmetrical have-relation, either explicitly or implicitly. However, though Borthen's characterization is closer to what is needed for Spanish and Catalan, there are two facts it does not directly account for. First, it does not have anything to say about the fact that the resulting verb phrase must denote a characterizing property of an external argument, because this condition simply does not hold in Norwegian. Second, Borthen's analysis does not block BNs from subject position of passivized ‘have’-predicates or from subject position of secondary predicates. Although we have already provided a syntactic account that blocks Spanish and Catalan BNs from these positions, our analysis would benefit from semantic motivation for that account, as well as an explanation for the difference between Spanish and Catalan, on the one hand, and Norwegian, which allows BN subjects, on the other. We will return to these differences at the end of Section 4.4, once our entire semantic analysis is in place.

Our proposal consists of a lexical rule which applies only to those predicates we mentioned in Section 2.2, and which supresses the theme argument of the predicate and adds a condition on use to take into account its potentially characterizing nature.Footnote ²¹ We define this rule in a generalized fashion in (43), although we could also simply directly posit alternative lexical entries fitting the general pattern of the output in (43) for the individual verbs that take BN complements.

1. (43) Input: λyλe[V(e)∧θ(e)=y∧∃w[C(w)][∃e′[depend(e,e′,w)∧have(e′)∧ havee(e′)=y]]]

   Output: λe[V(e)∧∃w[C(w)][∃e′[depend(e,e′,w)∧have(e′)∧havee(e′)= θ(e)]]]

   Condition on use of output: The issue of whether the referent introduced by the external argument participates or does not participate in e must be crucial for characterizing that referent in some way that is immediately relevant in the context.

The input to this rule specifies that whatever situation it denotes must be one which depends in some way on the existence of a have-relation involving the eventual subject referent and some other individual in some (not necessarily actual) world w (which might be subject to contextual restrictions as represented by the variable C). We represent this condition with the predicate depend(e,e′,w) and the accompanying requirement that the subjacent event e′ be a have-relation (have(e′)), with y as the havee. Our representation follows Parsons (Reference Parsons1990) insofar as it treats the verb as a predicate of events and separates out reference to the verb's participants, and follows Kratzer (Reference Kratzer, Rooryck and Zaring1996) in positing that the external argument is introduced via a functional projection; we also follow Dobrovie-Sorin et al. (Reference Dobrovie-Sorin, Bleam, Espinal, Tasmowski and Vogeleer2006) in treating thematic roles as functions from events to the individuals bearing those roles. The lexical rule in (43) can apply to Spanish and Catalan tener/tenir ‘have’, which depends on what we might call a prototypical have-relation holding in the very same world of evaluation, but it can also apply to necesitar/necessitar ‘need’, which requires the have-relation to hold only in those (not necessarily actual) worlds in which the subject's needs are met, and to the other ‘have’-predicates described in Section 2.Footnote ²²

Now consider the output of this rule. As mentioned above, the object argument disappears, which, as in the case of Dayal's analysis, accounts for the anaphora facts described above. This, however, leaves us for the moment with the problem of how to combine the BN with the verb. The key to solving this problem, as we will show in the next section, is the fact that we crucially do not assume the disappearance of the entailment that the verb describes a situation involving two participants. We refer to the participant corresponding to the suppressed internal argument as θ(e) and treat it as part of the lexical semantics of the verb, subject to description by a modifier, rather than associating it with an interpretively active variable that corresponds to a Theme argument. Note, furthermore, that any entailment of existence in the actual world for this implicit participant will depend on the lexical semantics of the particular verb involved: if the have-relation that the situation described by the resulting predicate depends on is one that must hold in the actual world, its satisfaction conditions will guarantee that the havee exists in the actual world; if not, it won't.

Finally, the output of the lexical rule introduces a condition on the felicitous use of the V+N sequence in order to capture the requirement that the predicate be potentially characterizing, as described in Section 2.2.

The condition on use in (43) replaces Dayal's notion of ‘appropriately classificatory’; however, unlike Dayal, we do not include the condition as a regular entailment. If it were a simple entailment, it would be possible to assert a sentence like (41) on the grounds that having or not having a glass is irrelevant for characterizing the speaker in the context. But this is not the case: (41) can only be false if the speaker lacks a glass, and even if it is false, the condition remains that having (or not having) a glass could serve to characterize the speaker in the context.

A compositional semantics for V+N combinations

With the output of the lexical rule in hand, we can now return to the question of how to combine the verb with a BN. We want to preserve from Dayal's (Reference Dayal2003) and Dobrovie-Sorin et al.'s (Reference Dobrovie-Sorin, Bleam, Espinal, Tasmowski and Vogeleer2006) the treatment of the BN as a verb modifier rather than a contributor of information about an explicit argument; we also want to maintain the maximally simple hypothesis that the BN denotes a first order property. However, we want to avoid having the verb semantically select for the BN. In order to do this, we propose that the verb and the BN combine not via function application, but rather via the intersective composition rule in (44).

1. (44) If ⟦V⟧=λe[V(e)] and θ is an implicit role function defined for V, and if ⟦N⟧=N, a property, then ⟦[_VV N]⟧=λe[V(e)∧N(θ(e))].

The rule in (44) applies to verbs which are missing only an external argument (setting aside the event argument) but for which a participant role beyond that ultimately contributed by the external argument is entailed as part of the lexical semantics of the verb. A nominal in the complement position of such a verb will, under this rule, act as a modifier, placing a restriction on the object that satisfies the unexpressed participant role of the event. Note, crucially, that the requirement of having a role function defined will correctly prevent the rule in (44) from applying to other [_VV N] structures where it would not be desirable, such as in analytic verbal expressions (e.g. do work), because the verb do and other light verbs that appear in such expressions do not have such role functions defined for them.

We now show how the output of the lexical rule in (43) fits the requirements for the compositional rule in (44) to apply. The havee role in (43) can serve as the relevant role function for the verb; (44) will thus guarantee that the property denoted by the BN describes the value of this role function. Consequently, the output of (44), applied to a Catalan example such as portar motxilla ‘carry a backpack’ (lit. ‘carry backpack’), will be as in (45).

1. (45) ⟦portar motxilla⟧=λe[portar(e)∧∃w[C(w)] [∃e′[depend(e,e′,w)∧have(e′)∧ havee(e′)=θ(e)]]∧motxilla(θ(e))]

Although (44) is a construction-specific rule associated with the [_VV N] structure, it is not ad hoc insofar as it bears a striking similarity to a modification rule which has already been argued for in a different domain, namely Selective Binding (Pustejovsky Reference Pustejovsky1995).Footnote ²³ Selective Binding was proposed to account for the modifying effect of adjectives such as fast in phrases such as fast car: a fast car is a car that can move quickly or be driven quickly. In this sense, fast describes not the car directly but rather an event that a car can participate in. In Pustejovsky's semantics, this event is represented in the noun car's lexical semantics, and Selective Binding accesses it for the application of the adjectival predicate (see Pustejovsky Reference Pustejovsky1995 for details).

Though there are important differences in the formal implementation, what Selective Binding and the rule in (44) share is the fact that the modifier – the adjective in the former case, the BN in the latter – provides information about an aspect of the modified expression's lexical semantics that is not manifest as an argument. In V+N combinations the BN provides information about an implicit participant, and thus describes the situation the verb denotes indirectly insofar as it restricts the identity of that participant. In the case of Selective Binding, the adjective describes an entity indirectly (such as a car) by providing a description of a type of eventuality in which that entity can or does occur.

With this semantic analysis in hand, we can account for the various properties of the construction which were described in Section 2 and for which we have not yet provided an explicit explanation. Some of the facts follow from assigning BNs a property-type denotation, as we pointed out in Section 4.1. As a property-type expression, the BN is a constant and thus inherently scopeless. It ends up being interpreted as if it had obligatory narrow scope because when it combines with the verb, any existential force that ends up being associated with it is due to the lexical entailments of the verb, which are always inside the scope of any other operators in the sentence; this is of course essentially the same account of obligatory narrow scope as the one given in Carlson (Reference Carlson1977) for bare plurals. The fact that the BN denotes a property also accounts for the fact that it freely antecedes only the pronoun en in Catalan which, as noted above, itself denotes a property. Moreover, since we have proposed that this property not be specified for number, we account for the fact that the BN is number neutral, unlike BPs or singular indefinites, which are specified for number. Finally, the number neutral, property-type denotation accounts for the failure of the BN to induce telicity (recall (9)). In order for a verb that is underspecified for telicity to get a telic interpretation, it must combine with some expression that is able to turn the verb into the description of a situation with a natural endpoint; and for a predicate to describe a situation s with a natural endpoint, it must apply to s and, in addition, it must be guaranteed not to apply to arbitrary incremental continuations of s.Footnote ²⁴ A V+N sequence based on such an underspecified verb will never count as telic on this definition because, thanks to the number neutrality of the BN, it will always be able to apply in principle to any arbitrary incremental continuation of any situation it describes.

Other facts discussed in Section 2 follow directly or indirectly from the lexical rule in (43). Its input condition limits the class of the verbs that can take BN complements to ‘have’-predicates. No direct restrictions are placed on the BN, though the condition on use of the V+N structure which is added by the lexical rule will ensure that only those BNs that serve in the context to create potentially characterizing properties will be licensed. Finally, in suppressing the internal argument, the rule leaves the verb no option but to combine with the bare nominal via the compositional rule in (44), guaranteeing that the BN will serve as a modifier.

Finally, the elimination of the internal argument, the assignment of a number neutral, property-type denotation to the BN, and the fact that the verb and BN combine via the rule in (44), all together lead to a natural account both of the failure of BNs to license discourse anaphora to token individual-denoting expressions such as personal pronouns, and of the inability of BNs to take non-restrictive relative clauses and to be controllers of implicit arguments. However, as a preliminary to presenting this account it will be useful to briefly compare our semantics to three alternative proposals for combining verbs with property-type nominals, namely Semantic Incorporation (Van Geenhoven Reference Van Geenhoven1996), Restrict (Chung & Ladusaw Reference Chung and Ladusaw2004), and Unification of Thematic Arguments (Farkas & de Swart Reference Farkas and Swart2003).

A comparison with other approaches to the semantics of V+N

Though Semantic Incorporation was originally proposed by Van Geenhoven (Reference Van Geenhoven1996) for noun incorporation in West Greenlandic, the semantics can in principle be extended to other cases where a verb which normally takes an entity-type argument is forced to combine with a property-type expression. On Van Geenhoven's analysis, incorporating verbs are assigned a lexical entry (possibly alongside other lexical entries) according to which the verb formally denotes a relation between an individual and a property. The property is ascribed to the entity that corresponds to the verb's internal argument, which is introduced via existential quantification. Semantic Incorporation, unlike our analysis, thus does not alter the argument structure of the verb in question in any deep way. Rather, its principal effect is to shift the introduction of the existential force and the introduction of the corresponding discourse referent, both of which are typically contributed by the nominal that the verb combines with, to the verb itself. To exemplify, a Semantic Incorporation analysis for a predicate like Catalan portar motxilla ‘carry backpack’ would be as in (46).

1. (46) λPλx∃y[portar(x,y)∧P(y)](motxilla)=λx∃y[portar(x,y)∧motxilla(y)]

In the dynamic version of Semantic Incorporation that Van Geenhoven develops, one effect of introducing an existentially quantified internal argument is to license the introduction of a discourse referent that can support discourse anaphora to token individuals, which is generally possible in West Greenlandic.Footnote ²⁵ This is the principal problem that such an analysis has for Spanish and Catalan. Since BNs in the latter languages do not license discourse anaphora to token individuals, a Semantic Incorporation analysis would have to be supplemented by a separate account of why discourse anaphora is blocked. We therefore will not adopt such an analysis, though by no means do we exclude the possibility that Semantic Incorporation might be the right analysis for other kinds of data in other languages.

An analysis in terms of Chung & Ladusaw's (Reference Chung and Ladusaw2004) Restrict rule faces a similar problem. Restrict accomplishes via semantic composition rules what Semantic Incorporation accomplishes in the lexicon: it allows a predicate which denotes a relation between individuals to combine with a property-type expression, the property simply adding descriptive content to one of the arguments without saturating that argument. Saturation results either via a separate rule of Existential Closure or when the verb phrase combines (by function application) with an extra object that denotes either an entity or a quantifier. The output of Restrict for portar motxilla ‘carry backpack’ is presented in (47a); the output following Existential Closure (EC) appears in (47b) (see Chung & Ladusaw Reference Chung and Ladusaw2004 for further details).

1. (47)

   1. (a) Restrict(λyλx[portar(x,y)],motxilla)=λyλx[portar(x,y)∧motxilla(y)]

   2. (b) EC(λyλx[portar(x,y)∧motxilla(y)])=λx∃y[portar(x,y)∧motxilla(y)]

Saturation by an extra object is illustrated by the Chamorro example in (48):

1. (48) Gäi-ga' un ga'lagu ennao na patgun. agr.have.pet a dog that linker child ‘That child has a pet dog.' (Chung & Ladusaw Reference Chung and Ladusaw2004: 89, ex. (29a))

An initial, if perhaps not insurmountable, obstacle to an analysis of Spanish and Catalan in terms of Restrict is the fact that, as mentioned in Section 1, doubling is not attested in either of these languages:

1. (49)

   1. (a) *Tinc pis un duplex. (see (5a)) have apartment a duplex

   2. (b) *El va buscar cambrer. him past.3sg look.for waiter

More seriously, as was the case with Semantic Incorporation, and for essentially the same reasons, a Restrict analysis predicts the possibility of having discourse anaphora to token individuals, which, as noted, must then be blocked for the cases which interest us here. Chung & Ladusaw (Reference Chung and Ladusaw2004: 121, 127) suggest that such blocking should be the job of morphosyntax, rather than something that follows from the semantic analysis. However, they do not offer any specific clues as to what such morphosyntactic mechanisms would be. We therefore do not adopt an analysis of BNs in terms of Restrict, either, although, as with Semantic Incorporation, we do not dispute its viability as a composition rule for other sorts of data.

A superficially similar, but ultimately rather different analysis of incorporation appears in Farkas & de Swart (Reference Farkas and Swart2003). These authors propose an enrichment of Discourse Representation Theory (DRT, Kamp Reference Kamp, Groenendijk, Janssen and Stokhof1984) in which they distinguish between discourse referents – those variables that instantiate the arguments of a predicate – and what they call Thematic Arguments. They then propose two different kinds of semantic composition rules: (i) Instantiation, which allows a variable representing a discourse referent to replace a thematic argument in the discourse representation of a predicate; and (ii) Unification of Thematic Arguments, which, under the right syntactic conditions, has the ultimate effect of allowing information contributed by a verbal predicate and a nominal to become associated with a single thematic argument. Instantiation is used to combine in DPs; Unification is used for BNs and BPs. However, the authors posit that the number feature on BPs leads to the accommodation of a discourse referent which eventually instantiates a thematic argument.

An illustration of the result of applying these rules in the analysis of the sentence in (50) is given in (51), where u is an instantiating discourse referent, and y is a variable representing a thematic argument.

1. (50) En Joan porta motxilla. det Joan carries backpack ‘Joan carries a backpack.’

1. (51)

   

The Discourse Representation Structure (DRS) in (51) is verified if there is a sequence of entities 〈e ₁, e ₂〉 in the universe of discourse such that the following conditions are met (where f assigns a value to the variable u):

1. (52)

   1. (a) 〈e ₁, e ₂〉 ∊ ⟦portar⟧

   2. (b) f(u)=e ₁

   3. (c) f(u)=⟦Joan⟧

   4. (d) e ₂ ∊ ⟦motxilla⟧

A crucial aspect of this theory is that it allows for final DRSs in which only Unification, but not Instantiation, has applied to some argument. In this respect the proposal differs from the Restrict analysis. While the latter requires all arguments to end up saturated (whether via Existential Closure or in combination with an appropriate external object via function application), the Unification analysis does not impose this requirement on thematic arguments. Farkas & de Swart allow Unification to apply without Instantiation in order to account for cases of blocked discourse anaphora: since discourse anaphora to a nominal is possible only when there is an instantiated discourse referent corresponding to that nominal, any nominal that combines via Unification with no accompanying application of Instantiation via doubling or the accommodation of a discourse referent (in the case of BPs) will fail to license discourse anaphora.

Although the differences at the level of formal detail make it difficult to compare the Unification analysis exactly with the other analyses discussed here, it seems clear that thematic arguments as proposed by Farkas & de Swart are not arguments in the usual sense of the term, or in the sense that the y variables are in the Semantic Incorporation and Restrict representations in (46) and (47), respectively. Rather, thematic arguments are more like the implicit role functions used in Dayal's and our analyses. In this respect, Unification could in principle be a viable candidate for analyzing Spanish and Catalan BNs, in substitution for the composition rule in (44). However, in order to be adapted to Spanish and Catalan, Unification would still have to be supplemented by a lexical rule like the one in (43), to restrict the verbs to which it can apply and to introduce the condition that the resulting verb phrase be potentially characterizing, and it would also have to be restricted to apply only to the syntactic configuration identified in (44), under the conditions identified in that rule.

While the empirical differences between using Unification vs. the rule in (44) may turn out not be substantial, developing (44) as an alternative to Unification and to the other analyses discussed here has had the positive effect of leading to more general reflection on two points which have received little attention in the previous literature. First, as mentioned in Section 4.3, the analysis establishes a clear parallel between the modification of verbs by BNs and the modification of nouns by adjectives. Since Unification applies specifically to thematic arguments, and since the sorts of elements within a noun's semantics that are modified by adjectives under rules like Selective Binding do not correspond to thematic arguments, this parallel was not previously visible. We believe it is worth exploring further in an effort to better understand modification as a cross-categorial phenomenon.

Second, Unification is designed to interact with other aspects of syntax and semantics in such a way that not only languages like Hungarian or Spanish and Catalan, but also languages like West Greenlandic and Chamorro can be accounted for (and something similar can arguably be said for Restrict). While this broad coverage is prima facie an advantage, it also deemphasizes the relevance of cross-linguistic variation in incorporation and incorporation-like phenomena. But we suspect that this variation could well be a sign of significant variation in the semantics for combinations of verbs with property-type object nominals, and suggests that this possibility should be given careful scrutiny.

The comparison of the different analyses presented in this section can help shed light on the remaining facts to be accounted for: the discourse anaphoric behavior of BNs, on the one hand, and the differences in the licensing of BNs in Spanish and Catalan vs. Norwegian, on the other. A detailed analysis of discourse anaphora would require converting our semantics to a dynamic framework such as DRT; though it is beyond the scope of this paper to do so here, we can sketch how the account would work. In the standard construction rules for DRSs, certain kinds of expressions (e.g. DPs) introduce discourse referents to token individuals; others do not. BNs, as property-denoting expressions, fall into the latter category. That means that if a BN is to be associated with a discourse referent, that referent must be supplied in some other way, for instance via the verb, as on Van Geenhoven's analysis, or via the accommodation licensed by number, as Farkas & de Swart propose for BPs. On our analysis, neither of these alternative means of introducing discourse referents is available. Thus, the BN will not be associated with any discourse referent, and as a result it will not directly license discourse anaphora to token individuals. The failure of the BN to be associated with a discourse referent also accounts for its failure to license modification by a non-restrictive relative clause, assuming that the latter are integrated into the semantics as independent propositions ascribed to the referent introduced by the expression they modify, as in e.g. Potts (Reference Potts2005).

Finally, recall that Norwegian BNs differ from those in Spanish and Catalan in that the former can appear as subjects of passivized ‘have’-predicates and secondary predications, while the latter cannot. According to the analysis we advocated in Section 3, the failure of Spanish and Catalan BNs to appear in subject position was due to their failure to project Number or D; this lack of morphosyntax correlates with their failure to instantiate a discourse referent. The underlying assumption is that the contents of subject positions need to be associated with such a referent. The composition rule in (44) also fails to provide a referent, but we have seen that both Semantic Incorporation and Restrict are able to contribute one. We therefore suggest that Norwegian BNs combine with their verbs not by a rule such as Unification or (44), but rather by Semantic Incorporation or a lexically-governed Restrict. This correctly predicts another contrast between Norwegian BNs and Spanish and Catalan BNs, which is that the former, unlike the latter, freely licenses token discourse anaphora (Borthen Reference Borthen2003).

We now turn to the final issue to be addressed in this paper, namely the extension of the analysis to existential sentences.