1 Introduction

The occurrence of null objects varies across languages. Some languages, such as English, generally do not license object drop, while languages such as Japanese impose fewer constraints on object drop (see Cole 1987 for an overview). Spanish and Portuguese license null objects under specific conditions. While European Portuguese (henceforth, EP) allows for both definite null objects and indefinite null objects (henceforth, INO) (Raposo 1986; Cyrino 2001; 2016; Barbosa 2019; 2024), European Spanish (henceforth, ES) typically permits INOs, but not definite null objects (Campos 1986; 1999; Brucart 1999; Clements 2006), as illustrated in (1).¹

Most of the research on ES focuses on factors related to the definiteness of the antecedent of the INO, its interpretation and the properties of the corresponding pronominal system. However, as pointed out in Martínez-García (to appear), the definiteness restriction on the antecedent cannot explain examples like the ones in (2), where even though the antecedents of each INO are indefinite, non-specific (concretely, bare nouns), the sentences are ungrammatical.

Considering the contrast between (1b) and (2), it seems that there should be an additional restriction affecting the occurrence of INOs in ES. Following intuitions in Martínez-García (to appear), in this paper we aim to show that cases like the ones in (2) can be accounted for if lexical constraints are taken into consideration. Additionally, we will show that the same lexical constraints govern the distribution of object bare nouns and INOs in ES.

Interestingly, a relation between the occurrence of bare nouns and null objects has been proposed for other languages, such as EP (Raposo 2004; Barbosa 2024), Hungarian and Polish (Ruda 2017), Japanese (Tomioka 2003) and ES (Martínez-García to appear). Concretely, Raposo (2004) proposes that the occurrence of null objects is less restricted in EP than in ES because the occurrence of object bare nouns is less restricted in EP. We will elaborate on this crosslinguistic difference and, based on corpus data, discuss the consequences for the analysis of null objects. Based on the correlation between the licensing of object bare nouns and INOs, we will propose an implicational hierarchy underlying the distribution of INOs crosslinguistically.

2 Indefinite object drop in Spanish

Let us first clarify what we refer to with the term INO. Crucially, INOs like the one in (3a) need to be distinguished from so-called cognate null objects in (3b), which also receive indefinite, non-specific readings. Unlike INOs, cognate null objects occur without discourse antecedents (i.e., they are non-anaphoric) and their interpretation is derived from the meaning of the root of the verb (Cummins & Roberge 2005). This paper focuses exclusively on INOs like the one in (3a).²

Based on the assumption that ES clitics are by default definite and receive specific readings (Leonetti 2006), it is assumed that INOs are used in place of clitics when referring to indefinite antecedents (Campos 1986; 1999), as in (4b), contrary to (4a). That is, INOs occur where clitics cannot appear (Dimitriadis 1994). This correctly predicts that INOs are not possible in languages with partitive clitics, like Catalan, Italian and French (Clements 2006) because these languages (in contrast to ES) show a corresponding clitic for indefinite, non-specific contexts.

The generalization that clitics are restricted to definite antecedents and INOs to indefinite antecedents has been refined based on examples like the ones in (5), showing that an indefinite, non-specific antecedent is compatible with an INO (see 5a), while an indefinite, specific antecedent is not (see 5b).

The difference is that the INO in (5a) is a property anaphor, as it refers back to the property denoted by any sum of individuals referred to by means of the antecedent, while the one in (5b) denotes an object anaphor that takes a particular object as its referent (Keller & Lapata 1998). The definition of property anaphor amounts to the claim in Laca (2013) that the INO denotes a property. When the INO refers to a bare plural, it denotes a property of pluralities. In Laca (2013), the property is interpreted as a restrictive modifier of the verbal predicate and the relevant variable is bound under Existential Closure. This is what gives us the indefinite interpretation. Thus, we assume that type anaphor (Keller & Lapata 1998), property anaphor (Tomioka 2003; Laca 2013) and identity of sense anaphor (Bresnan 1971) are synonymous.

Note that in languages like English there are indefinite individual anaphors, as shown in (6a), which is an English variation of (5b). Here, the indefinite ‘one’ is also a property anaphor, but it cannot be replaced by an INO. Similar examples can be also found in Spanish, like that in (6b), where the INO has been substituted by the indefinite unos ‘some’.

Examples like the one in (6b) also display indefinite readings and can be genuinely considered to behave as property anaphors, like INOs in Spanish. However, the difference between (6b) and (5a) is that the former can display a specific reading due to the presence of an explicit D (i.e., unos ‘some’), contrary to the latter, which can only display non-specific readings. Thus, INOs always refer to the property and never to the object level, contrary to the indefinite pronoun uno ‘one’, which refers to an object that has the property of the antecedent. In other words, the indefinite pronoun uno ‘one’ displays object reference, while INOs display property reference. To capture that semantic difference, Spanish grammar displays at least two kinds of property anaphors: INOs referring to bare nouns, like the example in (5a), and nominal ellipsis, where D must be stranded, like the example in (6b). Although both cases behave as property anaphors, they display different grammatical properties (i.e., they are different phenomena).

Hence, INOs in ES can only behave as property anaphors, while clitics and indefinite uno ‘one’ behave as object anaphors. In other words, correferentiality is involved in (5b), while identity of sense anaphora is involved in (5a) (Grinder & Postal 1971; Hankamer & Sag 1976; Büring 2005; see Bresnan 1971; Depiante 2000 for discussion). That is why a clitic is preferred in (5b).³

Since INOs behave as property anaphors, bare nouns emerge as the most suitable ‘candidate’ for a referent of INOs, given that they receive indefinite, non-specific, existential readings (Laca 1996; 1999; 2013; Dobrovie-Sorin & Laca 2003), as shown in (7). Crucially, in similar contexts, clitics are not adequate, as shown in (8).

However, a deeper examination of additional data raises further questions. As stated in Leonetti (2011), both clitics and INOs can be found in (9) to refer to bare nouns. Thus, the question arises: what distinguishes the interpretation of clitics from that of INOs in (9)?

Following Leonetti (2011), the occurrence of the clitic triggers a generic interpretation. As stated in Laca (1996), bare nouns provide this context, as they display a close association with generic contexts, like habits, attitudes, tendencies, or dispositions. The condition that the occurrence of clitics triggers generic interpretations predicts the ungrammaticality of the sequence with the clitic in (10), given that the context does not enable a generic interpretation (Leonetti 2011; see Espinal 2010: 992 for more data; see Laca 1996: 247 for further insights).⁴

Examples like the one in (11a) pose a further challenge. Despite the context allowing for a generic reading, the use of the clitic remains prohibited. Leonetti (2011) addresses this apparent problem by appealing to the selectional properties of the predicate selecting the clitic. The predicate proteger ‘protect’ only licenses definite DPs. This selectional property is incompatible with the indefinite, non-specific interpretation of the topic selected by the clitic los ‘them’ (i.e., tiburones ballena ‘whale sharks’). On the contrary, the example in (11b) is grammatical, since the verb ver ‘see’ can select object bare nouns.

To sum up, INOs relate to indefinite, non-specific antecedents and behave as property anaphors. Their antecedents are indefinite, non-specific phrases, typically bare nouns (henceforth, BN). BNs are exceptionally related to clitics if a generic interpretation is available.

Crucially, the availability of object BNs can be determined by the properties of the predicate. In the next sections, we will have a closer look at the interpretive properties and distribution of object BNs in ES and establish a strict relation between the availability of object BNs and INOs.

3 Object bare nouns in Spanish

In this section, we focus on the interpretation and distribution of object BNs in ES. While BNs are prohibited in subject position in ES (under ‘neutral’ intonation patterns), as opposed to English (see Carlson 1977; Suñer 1982 for an overview), they can occur in object position. It has been mentioned in the literature that BNs receive indefinite, non-specific readings in ES, like INOs. Contrary to English BNs, Spanish BNs cannot receive kind readings (see Carlson 1977; Heim 1982; Krifka et al. 1995 on English; Laca 1996; 1999; 2013 on Spanish; Dobrovie-Sorin & Laca 2003 on Romance). Additionally, BNs have been said to refer to properties but not to individuals, as they do not refer to specific tokens in the world (see Dobrovie-Sorin & Laca 2003; McNally 2004 on BNs in Romance). This distinguishes BNs from indefinites in ES.

Some approaches analyze object BNs as incorporated nouns, given that they share properties with object BNs in incorporation languages, like Frisian and Greenlandic (Masullo 1996; Espinal & McNally 2010; Martí 2011). Following these approaches, object BNs are interpreted as modifiers of the verb. Predicates with BNs do not semantically select the object BN but rather allow it to modify the interpretation of the verb by restricting the properties of the object. Thus, object BNs are syntactic arguments but not semantic arguments. This analysis has been extended to cognate objects and INOs (see Martí 2011; Armstrong 2014 for details). We assume with Espinal & McNally (2008; 2010) that, firstly, bare singulars are different from bare plurals and mass nouns, since they display different grammatical properties and, secondly, that bare singulars are semantically but not syntactically incorporated into the verb, given that dislocation is possible, contrary to expected (see section 3.1).

Although all object BNs share basic properties, they differ in the types of predicates that permit them (Dobrovie-Sorin & Laca 2003). Some predicates do not allow for BNs at all, while others only allow for a specific kind of object BN and still others allow for all types of object BNs. According to the literature (Laca 1996; 1999; 2013; Dobrovie-Sorin & Laca 2003), Spanish displays three types of BNs: bare singulars (henceforth, BS), mass nouns (henceforth, MN) and bare plurals (henceforth, BP), which will be discussed in this section.

4 Correlating indefinite object drop and bare nouns in Spanish

In the previous chapters, we reviewed the basic conditions governing the occurrence of INOs according to the literature and the interpretation and distribution of object BNs in ES. This section aims to put object BNs and INOs together, showing that they share basic interpretive and syntactic properties in ES, thereby indicating a correlation between them.

From a semantic point of view, INOs and BNs share default indefinite, non-specific, existential readings. That is why INOs must refer to BNs and be interpreted as BNs. We follow Laca (2013), who offers a precise account of how INOs, which denote contextually salient properties, receive an existential interpretation. Essentially, the property is interpreted as a restrictive modifier of the verbal predicate and the relevant variable is bound under Existential Closure. This results in the indefinite interpretation, with mandatory narrow scope.

The first piece of evidence for the similarities between INOs and object BNs comes from their scopal properties. INOs are considered to be ‘scopally inert’ in ES (Martí 2011; Armstrong 2014). Concretely, they must take narrow scope with regard to other sentential operators, like negation. This is shown in (24), where the INO, preceded by negation, only takes narrow scope. The proposition in (24) means ‘it is not the case that Juan took any part of the available fish’ (narrow scope), but not ‘there are some fish that Juan didn’t grab’ (wide scope).

Interestingly, the same restriction holds for object BNs (Laca 1996; 1999; Dobrovie-Sorin & Laca 2003). In (25), the BN only takes narrow scope with regard to negation. This suggests that there is a parallelism between INOs and object BNs regarding scope. Thus, in order to predict this property of INOs, they should be syntactically represented as BNs in ES.

There has been a large debate on the categorial nature of BNs. It has been claimed in the literature that BNs bear a null D (Longobardi 1994) or a null quantifier (Contreras 1996). Alternatively, they have been analyzed as nominal phrases without D (Dobrovie-Sorin & Laca 2003). We will pursue the latter account for the following reasons. The first piece of evidence comes from predication. With a full DP, as in (26b), the adjective can either modify the noun phrase or be interpreted as depictive. In (26a), with a BN, a depictive reading is not available.

This follows from the generalization that depictives are only possible with definite arguments (Demonte & Masullo 1999); i.e. when the argument they are combined with bears D (Landau 2010). Thus, BNs do not bear D in ES.⁶

Crucially, if INOs are not only interpreted but also syntactically represented as BNs, they should not be visible as subjects of predication, resembling cases like the one in (26a). Interestingly, the adjective nuevos ‘new’ in (27) cannot be interpreted as a depictive, but only as a modifier of the INO. Thus, the same grammatical restriction operates on ES BNs and INOs.

In light of (27), it seems that INOs are not only interpreted but also represented as true object BNs. It follows that INOs do not bear D, just like BNs (see section 7).

In sum, object BNs and INOs share the following properties: (i) they receive indefinite, non-specific, existential readings (related to property anaphors), (ii) they are not visible as subjects of predication (i.e., they lack D) and (iii) they take narrow scope with regard to operators. Thus, we conclude that the element filling the null object position is a BN, whose value is obtained from an antecedent via an identity condition. We assume that the identity condition is both syntactic and semantic in nature (see Merchant 2001; 2013 on this point).

Bearing the previous considerations in mind, it seems plausible to state that INOs essentially behave as object BNs in ES, the crucial difference being that INOs lack phonological content and must have an antecedent of the same type (see Laca 2013 on this issue). Hence, we suggest that the occurrence of INOs in ES must be subject to the same distributional restrictions reviewed for object BNs in previous chapters. Crucially, we propose that the same lexical restrictions of the predicate hold: for a given predicate to permit INOs in ES, it must be able to select object BNs. Given this proposal, we can now connect the occurrence of INOs and the occurrence of object BNs and formulate a first (tentative) hypothesis, illustrated in (28).

Additionally, we suggest that, depending on the type of object BN that is permitted, there are three basic types of predicates permitting INOs in ES: (a) those that allow for INOs referring to all kinds of BNs, (b) those that exclusively allow for INOs referring to BPs and MNs and (c) those that do not allow for INOs. Hence, we suggest that there are three types of predicates in ES. Thus, a more fine-grained version of the hypothesis in (28) is given in (29).

Note that the hypothesis in (29) assumes, in line with claims in previous sections, that the occurrence of object BSs is highly restricted, while the occurrence of object BPs and object MNs is almost free. This idea is illustrated in the scale in (30). The scale operates from left to right: if a predicate permits BSs, it will allow for BPs and MNs, since their occurrence is less restricted.

In the upcoming section, we explore the realization of INOs with the categories of predicates mentioned above and show a strict correlation between the occurrence of object BNs and INOs.

5 Towards a typology of predicates

In the previous sections, we showed that INOs and object BNs share basic interpretive and syntactic properties and suggested that there is a correlation between the occurrence of object BNs and INOs in ES.

In this section, we propose that there are three types of predicates regarding the licensing of object BNs. (i) Type A predicates are characterizing ‘have’-predicates that allow for all kinds of object BNs (e.g., encontrar ‘find’, llevar ‘carry’, necesitar ‘need’, etc.). (ii) Type B predicates are essentially consumption and activity predicates that only allow for object BPs and object MNs (e.g., comprar ‘buy’, comer ‘eat’, vender ‘sell’, etc.). (iii) Type C predicates do not allow for any kind of object BNs. These include psychological predicates (e.g., adorar ‘adore’, detestar ‘hate’, etc.), specific change-of-state verbs (e.g., quemar ‘burn’, etc.) and verbs displaying animacy restrictions (e.g., pegar ‘hit’, etc.). This section illustrates each predicate type and correlates the occurrence of object BNs with the occurrence of INOs.

6 Extending the proposal to indefinite object drop in Portuguese

As mentioned in section 1, EP differs from ES with regard to definite object drop, which is licensed in EP but not in ES. The question arises, whether EP is also more permissive than ES with INOs and whether these differences can be related to the licensing of BNs in EP.

According to previous literature, EP differs in two respects from ES when it comes to object BNs (Barbosa, Müller & Oliveira 2001). First, it is claimed that BNs in EP can be interpreted as kinds, in contrast to ES (Barbosa et al. 2001; Raposo 2004), as shown in (45) and (46).

Note that examples (45) and (46) include psychological predicates, which are incompatible with BNs in ES (see section 5.3). In section 6.3, we will argue that the crucial difference between EP and ES concerns the availability of BNs with this type of predicates in EP.

Second, some authors claim that EP does not allow for BSs, concluding that EP “would never license a singular count noun in a bare form in argument position” (Brito & Lopes 2016: 268). This is illustrated in (47). However, example (47b) includes a perception verb like ver ‘see’, which, according to our typology of predicates, belongs to Type B predicates and is therefore expected to not occur with BSs. Hence, the ungrammaticality of (47b) does not necessarily exclude that EP does allow for BSs with characterizing ‘have’-predicates (see section 6.1).

In the next sections, we will explore the distribution of BNs and INOs in EP, in order to ascertain whether the correlation established for ES is true for EP as well. We expect that INOs in EP are less constrained compared to those in ES, based on examples like (45).

7 Conclusions: Crosslinguistic implications and theoretical consequences

We have shown that the distributional properties of INOs can be essentially derived from the distributional properties of object BNs in ES (Laca 1996; 1999; 2013; Dobrovie-Sorin & Laca 2003; Espinal & McNally 2008; 2010). This is an elegant proposal that avoids introducing additional, unnecessary constraints on the occurrence of INOs. In the remainder of this section, we summarize our key findings concerning the interpretation, the syntactic analysis, the lexical restrictions determining their distribution and the typology of INOs.

First, concerning their interpretation, we assume that INOs can be described as type anaphora (Keller & Lapata 1998), property anaphora (Tomioka 2003; Laca 2013) or identity of sense anaphora (Bresnan 1971). Assuming that the three terms are basically synonymous in this sense, this view is based on Laca (2013), who argues that BNs and INOs denote properties and are interpreted as restrictive modifiers of the verbal predicate. BNs are the most suitable antecedent for INOs, who represent a copy of the antecedent expression, given that both receive indefinite, non-specific, existential readings (Laca 1996; 1999; 2013; Dobrovie-Sorin & Laca 2003).

INOs cannot be analysed as null pronominals because pronouns can only be definite in Spanish, as there are no partitive clitics available (contrary to Catalan, Italian and French). This also sheds light on the difference between null subjects and null objects: while the former are typically considered to behave as a null pronominal (pro) in consistent null-subject languages like Spanish, the latter behave as a full NP (but see Martínez-García 2023, to appear on null subjects in internal argument positions; Duguine 2013 for criticism on null subjects as pro).

Second, concerning their syntactic analysis, we assume that INOs are elided NPs, representing BNs without phonological content (see Laca 2013; Martínez-García 2023; 2025, to appear for further insights; Cyrino 2019 for an alternative perspective). Specifically, the operation deriving NP-ellipsis deletes the phonological features at PF. PF-deletion relies on the deletion of where the phonological features of the NP at PF is deleted and assumes that the ellipsis site remains syntactically projected and visible at the syntactic structure and LF. This idea is based on the fact that INOs show sensitivity to syntactic operations, like modification and scope. Additionally, although we do not delve into this topic in this paper, the fact that INOs can be antecedents of the same anaphoric expressions that refer to BNs suggests that INOs are elided BNs in Spanish.

Third, concerning the lexical restrictions determining the distribution of INOs and BNs, we showed that three types of predicates can be distinguished. Type A predicates are so called characterizing ‘have’-predicates (Espinal & McNally 2008; 2010). They are the most permissive predicates, allowing for BSs, BPs and MNs and INOs relating to these types of BNs. The reason they allow for BSs is that they denote a relation between two individuals, allowing for predicate+BN to express a characterizing property of a subject (e.g. ‘someone seeking an appartment’ as being an ‘apartment-seeker’). Type C predicates, like desear (‘desire’) or querer (‘want’), are psychological predicates. They cannot refer to BSs because they denote situations and not a relation between two individuals. BPs and MNs also cannot serve as objects of Type C predicates because these predicates select specific entities, due to the presupposition of the existence of the individual entities that they entail (Carlson 1977; Laca 1996; Seres & Espinal 2018). Similarly, change-of-state verbs, like quemar (‘burn’) and verbs selecting animate objects, like pegar (‘hit’) select definite entities, therefore they cannot combine with non-delimited entities introduced by BPs. Type B predicates like consumption and activity predicates are in-between type A and type C as they do not allow for BSs because they cannot behave as characterizing ‘have’-predicates, but as they do not impose a definiteness restriction on their object, they license BPs and MNs and INOs referring to this type of BNs.

Fourth, concerning the typology of INOs and building upon Raposo (2004) and Barbosa (2024) for EP, Ruda (2017) on Hungarian and Polish and Tomioka (2003) for Japanese, we suggested that the variation observed in the distribution of INOs and object BNs reflects a universal implicational hierarchy. Concretely, from a typological perspective, languages with fewer restrictions on the occurrence of object BNs are predicted to allow for INOs more freely.

Concerning the occurrence of BNs and INOs, ES and EP show similarities and differences. Crucially, EP is more permissive in two respects. First, Type B predicates do allow for BSs, BPs and MNs in EP (and INOs related to these types of BNs) and second, Type C predicates allow for BPs and MNs and INOs related to this kind of antecedents.

We claim that the asymmetry concerning the distribution of BNs can be partially related to the different ways of assigning kind readings in ES and EP. While such readings are impossible for object BNs in ES, they are possible for object BSs in EP. Specifically, kind readings in ES require the presence of a determiner (e.g., María detesta las zanahorias ‘María hates carrots’), contrary to EP. Hence, kind readings materialize with DPs in ES, while in EP they can appear with NPs. This underscores a significant crosslinguistic difference in the distribution of object BNs in EP and ES, aligning with predictions made by Raposo (2004) regarding the broader availability of object BNs in EP.

We have shown that, in accordance with our hypotheses in (29), EP displays less restrictions on indefinite object drop than ES and can be positioned on the right side of ES on a scale like (63). For languages utilizing partitive clitics, like Italian, Catalan and French, we assume that they impose the most restrictions on indefinite object drop, as they require the use of partitive clitics. Thus, our generalization applies primarily to Romance languages lacking partitive clitics.

In sum, our generalization correctly explains the distribution of INOs in ES and the crosslinguistic differences between ES and EP with regard to the availability of object BNs and INOs. However, further research should be conducted to better understand whether the predictions of our proposal apply to other languages as well. For example, it is well-known that Modern Greek and ES share core properties concerning indefinite object drop and their pronominal systems (see (64) from Dimitriadis 1994; Giannakidou & Merchant 1997 on Modern Greek). This suggests that our proposal could be extended to Modern Greek INOs as well. If Modern Greek and ES are truly similar regarding indefinite object drop, they should allow BNs and INOs with similar predicates and occupy similar positions on the scale in (63).

Furthermore, our generalization entails that languages without articles, such as Russian, Polish and Czech, would freely allow for indefinite object drop, since these languages only display BNs. This correlation between INOs and BNs seems to hold true for these languages (see McShane 2002 on Russian and Czech; Ruda 2017 on Polish).

At the other end of the scale are languages without BNs for which we predict that they are not able to allow for indefinite object drop. However, Basque as a language that lacks BNs (with the exception of French loanwords) challenges this expectation by permitting indefinite object drop even more freely than ES (Duguine 2013). A tentative way of solving this issue is to propose that the generalization made above applies primarily to languages not displaying object agreement. In languages like Basque with object agreement, the mechanism of agreement between the verb and the object may be responsible for a general licensing of (definite and indefinite) null objects. In this way, these languages circumvent the restrictions imposed by other languages like ES and EP without agreement. But further research is needed to shed light on these issues.

In conclusion, this paper suggested reconsidering the need for lexical restrictions for allowing indefinite object drop in ES and EP and offered a critical perspective against viewing indefinite object drop as only dependent on referential and definiteness factors. By emphasizing the fundamental grammatical properties shared by these phenomena across languages, we highlighted the correlation between the distribution of object BNs and INOs.

Abbreviations

EP: European Portuguese

BN: bare noun

BP: bare plural

BS: bare singular

ES: European Spanish

INO: indefinite null object

MN: mass noun

Funding information

This research was funded by the Spanish Ministerio de Universidades, under the auspices of FPU20/04298, EST23/00059 and EST24/00098, by the Spanish Agencia Estatal de Investigación, with funded project PID2021-123617NB-C3, and by Departamento de Lengua Española y Teoría de la Literatura (Universidad Complutense de Madrid).

Acknowledgements

We are indebted to the audience at Going Romance 2024 (Universidade do Minho, December 6, 2024), where we presented an earlier version of this paper, for their valuable comments. We are also grateful to the anonymous reviewers of Glossa. All mistakes and shortcomings in the work are no one’s fault but our own.

Competing interests

The authors have no competing interests to declare.