Weak crossover in Spanish relative clauses

2.1 Spanish relative clauses: general observations

Just like many other languages, Spanish shows an asymmetry between appositive relative clauses, which add information about an already-identified referent, and restrictive relative clauses, whose function is to specify a referent (see RAE-ASALE 2009: §44.1l). Consider the following contrast between an appositive (5) and restrictive relative clause (6), illustrated by subject relativization:^[2]

Appositive relative clauses are known to lack crossover effects (Bianchi 2002a, 2002b; Safir 1986, 2017), and we will not consider them further until the discussion section.^[3] Relatedly, we will not consider (semi)free relatives (RAE-ASALE 2009: §44.1g), as they do not create an environment for WCO. In the rest of this section, we will focus exclusively on externally headed restrictive relative clauses.

We start with the relativization of subjects, with a special emphasis on active transitive clauses, the configuration where we will be testing WCO effects. Such subject relatives are introduced by the element que, irrespective of the human/nonhuman status of the relativized noun, which we will refer to as the target. Note that we keep consistent all of our glosses in English and use “that” as the element introducing relative clauses, without making any commitment that the example under assessment is structurally parallel to the English gloss; in other words, no conclusions should be drawn about our structural analysis from the English translations.

No other option is possible. Attempting to introduce the relative clause via quien (9) or D(ef) + que (10) is ill-formed, regardless of the human/nonhuman status of the target.

In sum, restrictive relative clauses targeting a subject can only be introduced by que, regardless of the human/nonhuman status of the head noun.

Let us now consider the relativization of a direct object, the structural configuration in which we can assess WCO effects. Crucial for our purposes is that, in contrast to subject relativization, several encoding options are available for the relativization of objects. The availability of these multiple strategies depends on the human/nonhuman status of the target. Consider first the relativization of a human target such as la doctora “the (female) doctor” in the following examples. It is possible for the relative clause to be introduced by que (11a), a quien (11b), or a + D(ef) + que (11c) (see Zagona 2002: 56–60). The marker a observed in (11) is the so-called “a personal,” that is, the differential object marker (DOM) whose appearance is conditioned by animacy, definiteness, and specificity. Spanish DOM has been widely discussed in the literature (Torrego 1998; Zagona 2002; Leonetti 2008, 2012; López 2012, a.o.):

Turning now to inanimate targets, only the que strategy in (12a) is available (the same holds for nonhuman animate nouns such as those denoting animals, but we omit the data).^[4]

An important characteristic of Spanish object relative clauses has to do with the flexible order of the verb and subject inside the relative clause. Consider two ordering possibilities for each of the relativization strategies illustrated in (11), one involving the order subject-verb (13), and the other verb-subject (14):

Information-structural factors determine the most felicitous order given a specific context (see Brucart 1999: 400). Gutiérrez-Bravo (2003) reports that the verb-subject order is felicitous as an answer to a broad question such as What happened?, whereas the subject-verb order is not. In other words, the postverbal position inside the relative clause is reserved for new elements, whereas the preverbal position is used for elements that have been introduced previously. Consider the following examples that illustrate this generalization. Given a prior discourse where a referent is introduced (here, “the teacher” in the preamble (15)), a preverbal subject inside the relative clause is felicitous (15a), whereas a postverbal subject is not felicitous (15b).^[5]

The placement of the subject inside the relative clause in relation to the verb is important for the purposes of our experiment, which, as discussed below, includes a discourse context for each experimental item. Given the discourse leading to the experimental stimuli, all of our examples exhibited subject-verb order so that they would be felicitous.

There is one final observation regarding the availability of two orderings for the subject and verb inside the relative clause. Recall that que relativization is the only strategy that is available for both subject (7) and object relatives (11) – in fact, it is required for the former. In relation to the real-time processing of a relative clause, this fact entails that, upon encountering the relativizer que, a Spanish speaker will begin to consider two competing parses: namely, either a subject is being relativized or an object is. The word order between subject and verb within the relative clause immediately resolves any ambiguity when we encounter the DP following que, since this DP lacks the DOM exponent a. The points of temporary ambiguity at which two readings are being considered is notated in red font below, whereas blue indicates the disambiguation point at which the object relativization reading becomes the only one that is available.

In contrast, when the ordering within the relative clause is verb-subject, the two readings are still in play when the verb is encountered. It is only once the post-verbal DP (which lacks the DOM exponent) is encountered that the object relativization reading wins out.

In a nutshell, there is a longer period of temporary syntactic ambiguity with que relative clauses if the ordering is verb-subject, as in (17), relative to the subject-verb ordering in (16). This longer temporary ambiguity might lead to a (more) pronounced garden-path effect. In contrast, no garden path arises for relative clauses introduced by a quien, since these relative clauses unambiguously target an object. As we will see in Section 3, all of our test items involved subject-verb order as a means to minimize temporary syntactic ambiguity.

It is worthwhile at this juncture to briefly illustrate the major analyses of the syntax of restrictive relative clauses through the lens of the Spanish data that we will use to assess WCO. We will address here the head-external and head-raising (head-internal) analyses (see Bianchi 2002a, 2002b; Hulsey and Sauerland 2006; Salzmann 2017; De Vries 2018).

Let us start with the head-external analysis (Chomsky 1977; Jackendoff 1977). Observe an object relative clause “the doctor that you met” with que (18) and a quien (19), the two relativization strategies we will compare:

(18)

Head-external analysis; que

(19)

Head-external analysis; a quien

The analysis illustrated above is usually contrasted with the head-raising approach (Carlson 1977; Kayne 1994; Schachter 1973; Vergnaud 1974). Under this alternative, the head noun originates within the relative clause:

(20)

Head-raising analysis; que

(21)

Head-raising analysis; a quien

In addition to these two options, researchers have also proposed the matching analysis, which posits that the head noun is represented twice in the structure: once in its surface position (i.e., the matrix clause) and once within the relative clause (Hulsey and Sauerland 2006; Salzmann 2006; Sauerland 2004). Finally, Cinque (2020) builds on previous work to pursue the idea that matching and raising approaches are not mutually exclusive but, rather, structural variants that emerge from one and the same “double-headed” underlying structure (which we do not represent here).

Regardless of the syntactic analysis, we appear to have a configuration that ought to give rise to WCO, namely those cases where the movement dependency internal to the relative clause crosses a non–c-commanding pronoun. We use the head-external analysis in (22) to illustrate this configuration.

(22)

a.

la princesa que su reina amaba

“the princess that her queen loved”

b.

la princesa a quien su reina amaba

“the princess that her queen loved”

Equipped with this background regarding relativization in the language, which leads to the expectation of WCO violations, we now turn to the literature on WCO in Spanish restrictive relative clauses.

2.2 Do Spanish relative clauses exhibit WCO effects?

Systematic and controlled discussion of WCO effects in Spanish relative clauses is scant. For example, in relation to a fragment example that involves indirect object relativization (23), Bosque and Gutiérrez-Rexach (2009: 584) note: “The effects of weak crossover are less robust than those of strong crossover. In fact, weak crossover gives rise to grammatical strings in relative structures…” (translation ours).^[6]

Given the authors of the cited book, we assume that the dialect represented by the relevant example is Peninsular Spanish; see Bosque and Gutiérrez-Rexach (2009: 46) for their discussion on the kind of data used (e.g., acceptability judgments). Other dialects (such as Guatemalan Spanish, spoken by author Ranero) would require that the relativized nominal be doubled by the indirect object clitic le as in (24), independently of the presence of a manipulation that is meant to induce WCO.^[7]

Note, however, that the fragment above would only be felicitous if the postverbal subject inside the relative clause constitutes a new discourse element. Thus, we would want to ensure that the judgment was given in the appropriate context (recall our discussion of word order in Section 2.1). Regardless of this detail, if we took (23) and its accompanying passage as representative of the entire empirical landscape in Spanish, one could classify the language as one that does not display WCO in relative clauses.

A subset of the data discussed in Agüero Bautista (2014: 8) is also used to demonstrate that Spanish relative clauses lack WCO effects. Consider the example in (25), which displays the a quien strategy for object relativization.

However, Agüero Bautista (2014) goes on to argue that WCO arises in the same structural context if the possession strategy inside the relative clause is manipulated. WCO is thus taken to explain why (26) is ill-formed – presumably, the issue arises because, instead of using the possessive form su as in (25), the construction uses a DP of pronoun structure (akin to English “of him/her”).

We disagree with this interpretation of the data and the concomitant explanation of the lack of WCO effects in (25) as (presumably) contrasted with (26). WCO aside, there are independent reasons why the possessive structure DP of pronoun is degraded in this context. Choices in the expression of possession vary with respect to focus and contrast. When a pronoun is explicitly used to express the possessor in a complex DP (in contrast to the possessive su), this pronoun receives the focused or contrastive reading. Such a focused/contrastive status of the overt pronoun makes it hard or impossible to interpret it as referring to the same entity as another element in the sentence; in other words, contrast drives the disjoint reading. This observation about how overt pronouns affect information structure has been noted by researchers like Zagona (2002).

Crucially, this effect occurs regardless of whether the configuration could induce WCO or not. Consider the following dialogue without WCO as an illustration.

Note that the question forces a coreferential reading, one which is incompatible with the answer offered by B’. What seems to be triggering the deviance, then, both in our example (27B’) and in (26), is that the relevant possession strategy induces contrast on the overt pronoun, rendering the bound reading impossible. Another example showing the same effect appears in (28b):

A final illustration involving subject relativization is observed below in (29b). Notice that the same contrast persists, even though this is an environment (again) where WCO could not arise.

We thus maintain that Agüero Bautista seems to provide evidence against the existence of WCO in Spanish relative clauses via examples like (25); however, his argument that manipulating the possession strategy gives rise to WCO does not go through. Moreover, the comparison between (26), which is degraded due to double confounds of possession and contrast, and (25), where we expect to find a WCO effect, might mask any degradedness in (25). To foreshadow our experimental findings below, examples like (25) received ratings on par with other WCO configurations, thus attesting to WCO in this context.

Another study that briefly discusses the WCO configurations under relativization is Fernández Soriano (1995), although in structural contexts that are distinct from those we assess in our experimental study. In (30), we see two examples of possessive relativization that are surface-similar to our data, given by Fernández Soriano, and that are judged well-formed. Only que-relativization is available in these configurations.

First, note that these examples are not object relatives, however, and (30b) involves long-distance relativization (the main type of data that Fernández Soriano focuses on). A plausible baseline comparison for (30a) would be a parallel example using the relative pronoun cuyo/a “whose” (31a). However, Picallo and Rigau (1999: 1002) remark that the use of this relativization strategy is limited in many dialects (e.g., it has “practically disappeared” in Mexican Spanish; see also Elvira 2023: 180). Picallo and Rigau provide (31b) (a similar example to (30a)) in order to illustrate possessive relativization in “colloquial” speech (i.e., quesuismo, see (32a) below), as well as a third possibility employing the verb tener “to have” inside the relative clause (31c).

Crucially, there is a fundamental structural distinction between Fernández Soriano’s examples in (30) and our target relative clauses: Fernández Soriano’s data pattern like correlatives, in which the apparent head noun has no structural representation within the embedded clause. Thus, they more closely resemble English such that-relative clauses (consider the boy such that his father is a doctor). Research has independently demonstrated that correlative constructions do not exhibit crossover effects (e.g., Postal 1993; Bhatt 2003), and the Spanish data support that idea.

We have mentioned several studies maintaining that WCO effects are not observed in Spanish relative clauses.^[8] We are not aware of any studies that have taken an opposing view. As a result, there does not seem to exist a debate in the literature regarding the status of WCO configurations in Spanish relative clauses, in contrast to the conflicting viewpoints that one can easily find regarding English relatives.

Significantly, WCO receives little or no attention in the relevant literature. For instance, WCO effects are not discussed at all in the extensive chapter focusing on relativization in the 1999 Gramática Descriptiva de la lengua Española (Brucart 1999). They are not mentioned in other works either: Eguren (2017) (which analyzes relative clauses and possessives), the collected essays on Spanish relativization in Rivero (1991) (which includes reprints of that author’s earlier work such as Rivero 1982), and the overview articles in Francom (2012), Brucart (2016), Elvira (2023).

Similarly, the discussion of relativization in RAE-ASALE (2009: §44.1-§44.10) does not explicitly mention strong/weak crossover effects (Spanish cruce fuerte/débil), even in the sections devoted to possession inside relative clauses (§44.8g-e). In the grammar’s section on relativization, a single corpus example is provided that might qualify as a configuration where the effect of WCO could be expected, (32a), but it displays two characteristics that make it unlike the data we are interested in here. First, it involves the relativization of an indirect object (recall (23)). Second, the configuration exhibits a clitic inside the relative clause under the scope of the possessed nominal, and this clitic doubles the target of relativization (see bolded elements). It is also of note that this kind of example is described as dispreferred on prescriptive grounds by RAE-ASALE (2009), which names the phenomenon quesuismo. The prescriptively preferred version would use the possessive cuyo “whose”, which we construct and provide as a comparison in (32b).

To summarize, some prior literature has offered the viewpoint that there is no WCO in Spanish restrictive relative clauses, but this generalization has been made through a limited dataset that does not systematically compare relativization strategies (Bosque and Gutiérrez-Rexach 2009). Others have claimed that baseline configurations do not give rise to WCO, but the effect surfaces under specific manipulations related to possession (Agüero Bautista 2014). However, we assessed the data used to argue for that asymmetry and showed that they were confounded and ruled out independently of any WCO configuration. Finally, detailed studies of relativization do not mention crossover effects at all (e.g., RAE-ASALE 2009).

With this background in mind, we now turn to our experimental investigation, which seeks to clarify the status of WCO in Spanish restrictive relative clauses.

3.1 Participants

We recruited 103 participants located in Mexico (48 female, 53 male, 2 did not say; mean age: 30) via the Prolific.com crowd-sourcing service; 99 identified as native speakers of Spanish.^[9] Of those, 39 passed the attention checks described below; we include their data in the analysis.

3.2 Materials and design

We created a total of 24 items, with eight conditions per item. Where possible, items were a direct translation of the materials from Howitt et al. (to appear). We manipulated three factors: WCO (presence vs. absence of a configuration that should yield WCO), determiner (definite vs. quantifier), and relativization strategy (a quien vs. que). All data involved direct object relativization (see Section 2.2 for discussion of other types of relativization). An example item appears in Table 1.^[10]

Recall that there are three main relativization strategies for direct objects (11). We chose to compare que (11a) and a quien clauses (11b), instead of incorporating a+D(ef)+que clauses (11c), for two reasons. First, the determiner adds an additional layer of complexity that would complicate the interpretation of any results. Second, the number of elements involved in that relativization strategy is conditioned by the grammatical gender of the (singular) head noun – i.e., two elements with masculine nouns (e.g., cada príncipe al que / *a el que …“each prince that…”) versus three elements with feminine nouns (e.g., cada princesa a la que …“each princess that…”). Given that the head nouns in our items were a mixture of grammatically masculine, feminine, and neutral nouns (i.e., nouns that are compatible with either determiner such as el/la asistente “the assistant”), we chose to avoid the inconsistency introduced by a+D(ef)+que clauses.

There are also several reasons why we used direct object relativization in our study instead of indirect object relativization. For starters, doing so allows for a straightforward comparison to Howitt et al.’s (to appear) results on English, which showed a clear WCO effect in restrictive relative clauses. Since we are interested in laying the groundwork for broader cross-linguistic studies on the status of WCO in relative clauses, we sought to minimize any differences in experimental design. Furthermore, the pair of relativization strategies that can be compared systematically via indirect object relativization is not ideal. We can assess the status of WCO with quien relativization (33b), but we cannot assess the status of bare que relativization (33a), which is ill-formed. Instead, we would need to assess the strategy that uses a determiner (33c), the additional structural component mentioned before.

Note how the data above illustrate that some Spanish dialects require that a relativized indirect object be doubled by a clitic in base position inside the relative clause. These doubling facts complicate the interpretation of any result related to WCO configurations, since several analytical possibilities exist to derive the dependency between the doubled clitic and its associate. Furthermore, it is conceivable that our participants could vary with respect to clitic-doubling of indirect objects, which would have introduced a serious confound (see our Section 5 for the possibility of the resumption or doubling of direct objects in certain dialects, an issue left for future work). Finally, assessing indirect object relativization necessarily requires that the role of the direct object be considered carefully in designing the test items. This introduces an additional variable that could obscure any results related to the presence or absence of WCO effects.

Let us return, then, to our target sentences. They were presented in a story context meant to deliver a bound/coreferential interpretation that would yield WCO violations. Participants evaluated the naturalness of the target sentences in the context of the story using a slider with endpoints labeled nada natural “not natural at all” (coded as 0) and natural “natural” (coded as 1). Participants were told that all of the sentences were true in an attempt to discourage truth-value judgments and highlight that the task was to evaluate naturalness. An example trial for the item in Table 1 appears in Figure 1.

Figure 1:

Example experimental trial with the item in Table 1; this trial features a WCO, definite determiner, a quien relativizer condition.

After reading the instructions and completing two practice trials, participants completed a total of 30 trials in random order: 24 test trials (one from each item; three from each of the eight conditions) together with six fillers/attention checks. The task took an average of 14 minutes to complete. The fillers were designed to elicit a clear response, either as natural or unnatural. For natural fillers, we expected a slider response of 0.5 or greater; for unnatural fillers, we expected a slider response of less than 0.5. Our fillers are available in the appendix linked in footnote 10. Only the 39 participants who provided the expected response on at least five of the six fillers passed our attention check.

3.3 Results

Figure 2 plots average naturalness ratings grouped by experimental condition. We fit a linear mixed-effects regression predicting rating by WCO, determiner, and relativizer, together with their interactions; the model included random intercepts by participant and by item. We find an effect of determiner (β = −0.07, t = −2.31, p < 0.05), such that quantifiers receive lower ratings than definites, as well as an interaction between WCO and relativizer (β = −0.12, t = −3.03, p < 0.01), such that the effect of WCO is significant only when the relativizer is que. No other effects reached significance.

Figure 2:

Average slider rating grouped by condition. Error bars represent bootstrapped 95 % confidence intervals.

4.1 Appositive parse

As mentioned in Section 2 above, appositive relative clauses are known to lack crossover effects (e.g., Safir 1986, 2017; Bianchi 2002a, 2002b). One could posit that we did not detect a WCO effect in a quien relative clauses because participants judged such clauses as appositive, rather than restrictive.

The logic of this explanation is as follows. It is independently known that que cannot introduce appositive object relative clauses, as evidenced by the contrast in (34) (Zagona 2002: 58).^[11]

In other words, Spanish speakers are aware that an appositive reading is impossible in que object relative clauses. Therefore, when an object relative clause is introduced by a quien, as in (34b) and as in half of our experimental trials, participants may take the presence of a quien as a sign that the appositive interpretation is warranted on the basis of paradigmatic contrast: two grammatical constructions serve similar but not identical functions, and so one expects to find some differences between them. If a quien object relative clauses are interpreted as appositive (again, in contrast to que object relative clauses), the nonoccurrence of WCO violations for a quien relative clauses is explained. In other words, since the appositive parse is unavailable in general for que object relative clauses, it should be the case that the WCO effect shines through in that condition, unencumbered.

However, there is more to the story when it comes to appositive parses of a quien relative clauses. In the condition involving a quantifier, an appositive relative clause is ill-formed – its degraded status can also be observed in the English translation (35c).^[12]

As a result of this restriction, an appositive parse is not possible for our a quien conditions with a quantificational determiner (i.e., cada in our experimental items). We, therefore, expect that any rescuing effect attributable to an appositive parse should not be possible in conditions with a quantifier. And indeed, ratings for WCO-violating a quien relative clauses with a quantifier are on a par with ratings for the WCO-violating que relative clauses (Figure 2; 0.62/1 for a quien vs. 0.58 and 0.61 for que). It is the WCO-violating a quien relative clauses with a definite that receive higher ratings than expected (0.69), and, crucially, it is this specific condition that allows for the appositive rescue strategy. We, therefore, consider it likely that the absence of clear WCO effects for a quien in our results arises as a result of the appositive parse that is available to a quien with a definite head noun.

4.2 Weakest crossover

Speakers show varying degrees of tolerance for A ¯ -binding violations, even though all such violations stem from the same fundamental issue: disruption of c-command relationships in A ¯ -binding contexts. Some constructions remain acceptable despite involving both (i) an operator that locally A ¯ -binds a pronoun and a trace and (ii) a pronoun contained within an NP that c-commands the trace – in other words, the classic WCO configuration exemplified in (22) for both que and a quien. WCO configurations that do not give rise to WCO violations are known as weakest crossover contexts; such contexts offer the possibility of broadening our understanding of binding configurations. Weakest crossover has been identified in tough-movement, topicalization, and parasitic gap constructions (Lasnik and Stowell 1991). If we interpret our results with a quien as the absence of WCO violations, then a quien restrictive relative clauses are candidates for weakest crossover as well.

According to Lasnik and Stowell (1991), weakest crossover arises as a result of the logical status of the operator in A ¯ position, which licenses a new kind of empty category, a null epithet, that gets treated as an R-expression not subject to the principles giving rise to WCO violations. Quantificational operators must bind variables, and variables give rise to WCO; nonquantificational operators are associated with null epithets immune from WCO. In an attempt to explain the absence of WCO with a quien, one could hypothesize that a quien itself serves as the relative-clause operator (or has a silent operator enclosed in it, as proposed in Arregi 1998), and that this operator is nonquantificational, thereby licensing a null epithet in place of the variable trace that is subject to WCO with que.

Setting aside the thorny issue of identifying a difference in the quantificational status of (the operators associated with) que and a quien on the basis of Lasnik and Stowell’s characterization, Safir (2017) points out a major drawback of the account, namely the absence of independent evidence for their new kind of empty category. Without such evidence, positing the new category offers little by way of explanation. There are alternatives to Lasnik and Stowell’s story that do not worry about quantificational status or positing a new type of empty category. For example, Ruys (2004) rejects characterizing WCO in terms of an A ¯ -binding violation and instead focuses on A-binding requirements (see also Sauerland 1998); in cases of weakest crossover, the would-be offending pronoun has an appropriate A-binder. For present purposes, the trouble lies in finding a difference between que and a quien structures such that the latter, but not the former, features an appropriate A-binder. However, as we discuss in the following subsection, perhaps this search is not necessary.

4.3 Taking stock

To summarize, we have considered two possibilities for explaining the contrast between que and a quien observed in our data. Under the first, both que and a quien give rise to WCO violations, but only a quien can take advantage of an appositive rescuing strategy. Under the second possibility, only que (but not a quien) gives rise to WCO violations. Should we pursue the second possibility and attempt to explain away WCO for a quien, we saw that we have options with which to do so. But these options carry their own costs, and they compel us to ask further questions about the structure of Spanish relative clauses. More importantly, it is not clear whether our results warrant explaining away WCO for a quien. As we mentioned above, ratings for the WCO-violating a quien relative clauses with a quantifier (i.e., the a quien condition not compatible with an appositive parse) are on a par with the WCO-violating que conditions. If anything needs explaining, it is the absence of low ratings for WCO-violating a quien relative clauses with a definite – precisely the configuration compatible with an appositive parse.

On balance, it appears that the appositive rescuing strategy provides a better account of our data.