It is a well-established observation that there exists a substantial asymmetry between subject relative clauses (srcs) and object relative clauses (orcs). Indeed, srcs are typically easier for children to comprehend (de Villiers et al. 1979; Tavakolian 1981; Goodluck & Tavakolian 1982; Corrêa 1995; Friedmann et al. 2009; Lau & Tanaka 2021) and similarly, they pose fewer challenges for adults during processing (Frauenfelder et al. 1980; Ford 1983; Clifton & Frazier 1989; Schriefers et al. 1995; Mak et al. 2002). With regard to child populations in particular, a greater difficulty in the comprehension of orcs as compared to srcs has been consistently observed in several languages over the last two decades (Adani et al. 2010 for Italian; Bentea et al. 2016 for French; Morrill and Gavarró 2010 for Catalan; Sevcenco and Avram 2012 for Romanian a.o.). The first aim of this study is to investigate whether this asymmetry holds true for Spanish, a language which, to the best of our knowledge, has hardly been investigated in this particular respect before.1 To this end, we tested Spanish monolingual children’s comprehension of both srcs and orcs. The results show that, as expected, the former are overwhelmingly better comprehended.

Another point that has been frequently addressed in the literature concerns the role of language-specific factors in the comprehension of orcs in particular. Interestingly, languages have been shown to differ with respect to how idiosyncratic properties of their grammar impact on children’s comprehension of these structures (Belletti et al. 2012; Guasti et al. 2012; Bentea 2016). For this reason, in a second comprehension task, we take into account two particular properties of Spanish orcs, namely the position of the subject within the relative clause and the prepositional marking of (animate) direct objects, widely known as Differential Object Marking (dom). The manipulation of these two variables produces four types of orcs, whose variation depends on whether the subject is pre- or post-verbal and whether dom is overtly realised or not (cf. Section 2.1 for details). We found that these differences are non-trivial as the statistical analysis reveals an effect of both word order and dom. Specifically, pre-verbal subjects and overt dom bring about higher comprehension rates. These facts strike us as particularly revealing, for both factors can be used to pin down the thematic relations that inform the sentences at issue, as will be extensively discussed in Section 2. We integrate these original data with available cross-linguistic evidence and advance a first empirical generalisation of which properties can ameliorate children’s comprehension of orcs in which languages. In a nutshell, we characterise so-called ameliorating cues as grammatical properties that allow the parser to distinguish the subject from the object. Another crucial aspect of our findings is that (overt) dom does ameliorate children’s comprehension of orcs, but only when it is not concurrent with a cue that is based on word order (in our case pre-verbal subjects). Moreover, we show that the impact of dom strictly depends on its acquisition. Based on dom-mastery scores assessed through a sentence-repetition-task, we found that dom has an impact on orcs’ comprehension only with those children who have associated dom with its grammatical function, whereas no dom effect emerges with children who have not acquired it yet.

We interpret these results in the broader debate on the source of locality effects, taking into account the well-known divide that holds between “grammar-based approaches” and “processing-based approaches” (for a discussion, see Boeckx 2008b; 2012; Hofmeister & Sag 2010; Belletti & Rizzi 2013; Phillips 2013; Sprouse et al. 2013 a.o.). Whereas the former construe locality effects as the outcome of deep grammatical principles, the latter try to derive the same set of phenomena calling upon independent principles of cognition. Using Sprouse et al.’s (2013) wording, the central difference between the two is that processing accounts “seek to reduce [locality] effects to one or more components of the sentence-processing system that are motivated by language-independent perceptual or cognitive properties.” (Sprouse et al. 2013: 23). With regard to this, Phillips (2013) and Sprouse et al. (2012) suggested that processing-based approaches be rather called “reductionist approaches”, exactly because they aim at reducing locality to factors that are outside the grammar and fully independent from it. Note that the cited literature has typically addressed the problem from the perspective of (extraction from) syntactic islands. In this study, we broaden the scope of this debate to include another instance of locality constraints, namely intervention locality.2 Many scholars have defended accounts to intervention effects and their alleviations that rely on extra-grammatical factors. A paradigmatic example of these is represented by so-called “experience-based” theories (a.o. Bever 1970; MacDonald & Christiansen 2002; Wu et al. 2018), in which the general prediction is that “[…] comprehenders are sensitive to the statistical regularities in lexical and structural input, and can use their prior experience with linguistic signals to predict the upcoming structure” (Wu et al. 2018: 24).3 In a broader sense, however, all approaches that resort to domain-general constraints on language processing to explain the relative cost of orcs (see e.g. Gordon et al. 2001; 2004; Warren & Gibson 2002; 2005) can be legitimately categorised as “reductionist”. On the other hand, a theory like Relativised Minimality (Rizzi 1990; 2004; 2018; Starke 2001; Friedmann et al. 2009), which defines intervention effects as solely affected by syntactic rules, is a natural representative of aforementioned grammar-based approaches.4 Considering the results of our experiments, a processing-based approach is seemingly better equipped to interpret at least some aspects of the emergent pattern. For instance, the negative impact of simultaneous cues on orcs’ comprehension can be construed as a direct consequence of the child parser being overloaded. At the same time, though, a grammar-based approach to (intervention) locality proves superior in accounting for the fact that dom’s effectiveness additionally depends on its acquisition. Evidently, such an outcome is only predicted if the strategies at the base of amelioration effects hinge on grammatical representations and presuppose grammatical competence. Nonetheless, we will illustrate that even a tout-court grammatical approach like Relativised Minimality is challenged by our data, especially (but not only) because the latter cannot predict a negative interaction between two otherwise effective ameliorating cues (i.e. word order cues and dom cues). For these and other reasons that will be elaborated on in the next sections we advance the hypothesis that, at least as far as intervention locality is concerned, neither approach is entirely flawless, and that a more suitable option lies in a third, intermediate perspective. Specifically, we argue that intervention effects in Spanish orcs and ameliorations thereof are naturally predicted if their source resides at the interface between the grammar and performance systems. In such a scenario, both language-specific factors and more general constraints on sentence processing are expected to contribute to surfacing locality effects, and this is exactly what our findings suggest. Indeed, we illustrate that whereas the integration of ameliorating cues is subject to the cognitive limitations of the child parser (two cues are overly taxing), amelioration strategies must be necessarily anchored to deep grammatical representations (dom is only effective if acquired). In essence, our proposal conforms to the minimalist effort to pursue interface-based accounts and look at domain-general properties to free the grammar from dispensable stipulative mechanisms (Chomsky 2005; 2017; Berwick & Chomsky 2016; Hornstein 2018; Chomsky et al. 2019).5 An endeavour, this, that has been regarded as particularly desirable for locality-related phenomena (Boeckx 2012; Dillon & Hornstein 2013, and Presotto 2024a for a discussion).

The paper is organised as follows. Section 2 provides the theoretical background. First, we take into account cross-linguistic data on children’s comprehension of orcs and put forth an empirical generalisation of what qualifies as an ameliorating cue. Then, we illustrate the main properties of Differential Object Marking and relative clauses in Spanish (Section 2.1). Section 3 presents our study and deploys our research questions. Section 4 details our methods and the results of the experiments. In section 5 we discuss our findings. Section 6 concludes.

Robust evidence has been provided that, albeit overall taxing, orcs can be made easier under certain conditions. For instance, the seminal work by Friedmann et al. (2009) showed that Hebrew-speaking children find it easier to comprehend orcs in which the object, but not the subject, is lexically restricted. A comparable result has been found for Italian, where a mismatch in number between the subject and the relative head facilitates children’s comprehension of orcs (Adani et al. 2010). Similarly, German children get more accurate when orcs are disambiguated by morphological case (Arosio et al. 2012). These types of alleviations to the typical penalty associated with orcs are generally referred to as “amelioration effects”. And the particular property that triggers the amelioration is defined as an “ameliorating cue”.

What is especially interesting about these facts is that ameliorating cues are not universally valid. Rather, speakers of different languages seem to benefit from different properties of their grammar when it comes to comprehending arduous structures like orcs.6 In a most influential paper, Belletti et al. (2012) tested Hebrew- and Italian-speaking children with gender-matched and gender-mismatched orcs. They found that gender mismatches ease orcs’ comprehension in Hebrew, but crucially not in Italian. Hebrew stimuli were like (1a) in the gender-match condition and (1b) in the gender-mismatch condition.

Likewise, the subject and the object in the tested Italian orcs were specified for either the same (2a), or different (2b) gender features.

A central difference between the two languages is that, differently from Italian, Hebrew’s main verb inflects for gender in undergoing agreement with the subject (Shlonsky 1997). The authors interpreted the results as a corroboration of the featural Relativised Minimality approach (Starke 2001; Rizzi 2004), whereby the degree of disruption in the integration of orcs and, more generally, Object A-bar dependencies, is modulated by the featural specification of the subject and the object. In particular, the fact that gender is only relevant in Hebrew would support the claim that only features that are “syntactically active” are taken into account in the computation of intervention effects. Somewhat simplifying, active features are features that are capable of triggering movement operations (cf. Rizzi 2013; 2018 for a discussion). Since gender qualifies as a movement-triggering feature in Hebrew, but not in Italian, the intervention configurations of Hebrew and Italian orcs will differ with respect to the presence of this very feature. That is, gender is involved in the computation of locality effects in Hebrew, but not in Italian, as exemplified below:7

The fact that gender is represented in (3) defines two distinct set-theoretic relations between the relative head and the intervening subject. In (3a), the featural bundle of the object includes that of the subject, whereas in (3b) the two bundles intersect. In the case of Italian instead, gender is never involved in such a calculation, hence the unchanged inclusion relation in (4a) and (4b). This is crucial, for it is exactly this type of set-relation that determines the degree of disruption of orcs, with inclusion being significantly harder than intersection.8 Again, this would explain why only Hebrew-speaking children can benefit from gender-mismatched orcs.

This type of analysis has been further maintained (among others) for French, where number-mismatched orcs seem to be better comprehended by children than their number-matched counterparts (Bentea & Durrleman 2017; Durrleman & Bentea 2021). Such a result straightforwardly follows from Relativised Minimality since number features in French are attractors for movement and, thus, visible to the syntactic computation of intervention effects.

The Relativised Minimality approach has undoubtedly unveiled relevant aspects of children’s comprehension of orcs and their ameliorations, as well as provided meticulous tools for the investigation of the phenomenon in a cross-linguistic perspective. In spite of its thorough theoretical basis, however, Relativised Minimality is not always unquestionably supported by empirical data. For instance, Guasti et al. (2012) tested the comprehension of orcs by Greek and Italian children and found that the former perform better when overt case marking disambiguates argumental DPs (much as what Arosio et al. 2012 found for German). Even though the authors do call upon Relativised Minimality to construe their findings, it is not obvious why case should bring about amelioration effects, for it is unclear whether it can be analysed as a “syntactically active” feature. As a matter of fact, case is explicitly predicted to not count as an ameliorating feature in this framework because it is not a trigger for movement (Friedmann et al. 2017).

Another crucial assumption in this approach is that the lexical restriction on a DP, characterised as a [+NP] feature, modulates intervention effects qua licensor of syntactic positions (Rizzi 2018: 349–355). The prediction for orcs is that anytime the object and the (intervening) subject share a [+NP] feature specified with the same phi-features (cf. Belletti et al. 2012 et seq.), intervention effects arise resulting in poorer comprehension of the sentence by children. Conversely, when either the subject or the object is not lexically restricted (as is the case with pronouns), an amelioration effect is predicted. This is borne out in a large number of studies. Still, in a paper on Romanian children’s comprehension of relative clauses, Sevcenco & Avram (2012) present results that go in the opposite direction. In one task, the authors tested sentences like those displayed in (5). In (5a) an overt subject DP (piticul) occurs in postverbal position, whereas in (5b) and (5c) the subject is expressed by a null pronoun. As pointed out in the paper, sentences like (5c) are ambiguous between a src and orc reading.9 Sentences like (5b) are, instead, never ambiguous, for the authors ensured that the verb was always plural and the relative head singular.

Participants (monolingual children aged 4;0–7;5) understood sentences like (5a) in 80.29% of cases but were only 67.9% accurate with sentences like (5b). Moreover, they assigned an orc interpretation to sentences like (5c) only 51.2% of the time. Independently of the results with sentences like (5c), whose interpretation is complicated by their inherent ambiguity, children seem to perform better with sentences like (5a) rather than (5b). That is, the presence of a lexical restriction (i.e. [+NP] feature) on the subject eased children’s comprehension despite the relative head being lexically restricted as well.10 Conversely, the [+NP]-mismatch between the object and the (null) subject produced significantly lower accuracy.

Based on these observations, we look at these phenomena from a fairly different angle. That is, we adopt a broad perspective on the empirical evidence on amelioration effects in orcs and try to surmise their commonalities across languages.

What we believe can be extrapolated from such data is that children tend to perform better when orcs can be somehow disambiguated from srcs. More in detail, we contend that children’s comprehension likely improves when morphosyntactic cues are available that allow the parser to tease apart the object from the subject and, hence, construe the relations between the verb and its arguments.11 We will also clarify in the discussion that such morphosyntactic cues need to be acquired in order to trigger ameliorations, which crucially ties their effectiveness to their grammatical function. For now, however, we limit ourselves to build upon these considerations to put forth a first descriptive generalisation that demarcates the properties of a (potentially ameliorating) morphosyntactic cue:

Based on this, not only can we infer what can constitute an ameliorating cue in which language, but also make sense of amelioration effects that seemingly fall outside the reach of Relativised Minimality’s predictions. Before we do that, let us clarify some aspects of (6).

First, by no means does our generalisation imply that anytime a morphosyntactic cue is available amelioration effects arise. It simply states that when these emerge, there is at least one morphosyntactic cue that can disambiguate the structure. Second, the distinction between presence and absence of a given grammatical property is grounded on the observation that sometimes even just the lack of, for instance, some relevant morphological exponent can rule out certain interpretations and disambiguate the sentence, as is the case in the Spanish orcs discussed in Section 2.1. Third, and perhaps most importantly, we need to elucidate the ontological status of our generalisation. Taken alone, it is not much more than an empirical observation of the facts. It surely provides accurate predictions on amelioration effects, but it is not a theory of amelioration effects in its own right. For this reason, we pair the descriptive power of (6) with a broader speculation on the system underlying ameliorations to intervention locality. This is pursued in the discussion. We first try to corroborate our claim by taking into account the literature discussed above.

All the amelioration effects we reviewed conform to (6). For instance, the fact that a gender-mismatch between the subject and the object increases children’s comprehension of Hebrew orcs straightforwardly follows from it, as in such a case the gender morphology on the verb unambiguously points to the subject DP. Of course, though, verbal inflection is not the only expected ameliorating cue. We deliberately adopt a general term though “grammatical property” to make sure we embrace other relevant factors. Among these are case marking, word order, and, in fact, anything that reveals the subject/object status of a nominal. This way, ameliorations triggered by (unambiguous) case morphology in e.g. German (Arosio et al. 2012) or Greek (Guasti et al. 2012) naturally ensue. Word order is also expected to play a role insofar as it allows to discern the subject or the object. We will show that the results of our study go in this exact direction (cf. sections 4.3 and 5).

Another welcome outcome of our generalisation is that it avoids altogether the issues that some ameliorating cues pose to Relativised Minimality. In this regard, we have already mentioned how case is inherently predicted to contribute to amelioration effects. Moreover, the data from Romanian orcs like (5) are also fully in accord with (6). As discussed by Sevcenco & Avram (2012), indeed, the presence of an np subject alone guarantees the disambiguation between orc- and src-readings. So whereas (7a) is only unambiguous if the direct object clitic (in this case îl) and the relative head have distinct phi-features, (7b) does not need such a proviso since “[a]n overtly realised pre-verbal or post-verbal subject […] in the embedded clause bans the [src-] reading” (Sevcenco & Avram 2012: 221).12

To summarise, we suggest that available evidence on children’s comprehension of orcs allows us to draw the following conclusion: it is only when a disambiguation between the subject and the object is possible that orcs can be made easier. We formalised this observation in (6), which provides a descriptive generalisation of the properties constituting morphosyntactic cues. These, we claim, represent a necessary (but not sufficient) condition for amelioration effects to arise.

Below we introduce some properties of Spanish relative clauses that meet our definition of morphosyntactic cue, namely Differential Object marking and subject position. The design of our experiment, presented in Section 4, builds upon those very properties in order to test whether they correlate with facilitations in children’s comprehension of orcs.

In the previous sections we discussed how orcs are more difficult than srcs. We also pointed out that the burden associated with orcs can be alleviated by morphosyntactic properties that qualify as ameliorating cues. Based on the literature on amelioration effects, we put forth a descriptive generalisation of what constitutes an ameliorating cue, which we repeat here in (13):

We can now clarify the reason why we distinguish between the presence and the absence of a specific grammatical property. As illustrated in Section 2.1, in cases like (12a) it is the sheer absence of dom that manifests the subjecthood of the np within the relative clauses. Once more, because futbolista is [human; definite; specific], it needs to be marked with the dom particle a (in fact, the amalgam al) any time it occurs in object position. If dom is not realised, then, it follows that the np at issue is not the object but, rather, the subject of the clause. We distinguished cues of the latter type from cues that hinge on properties available in the surface structure of a sentence, be these overt morphological exponents, as in (12b), or related to the order of the linearised constituents (12c). We defined the former in-absentia cues, and the latter in-praesentia cues. Although these labels only rest on a descriptive distinction of what, other things being equal, involves the same implications, we do suspect in-absentia and in-praesentia cues be potentially distinct in their amelioration capacity. In particular, in-absentia cues are conceivably costlier than in-praesentia cues as the information they convey, we suggest, is not as easily accessible to the parser (cf. Section 5).

Morphosyntactic cues can be further classified based on the grammatical property that defines them. As shown in (12), Spanish orcs can be disambiguated via morphological cues (i.e. dom) and/or purely syntactic cues determined by the linear order of constituents (in our case, the position of the subject). In this study, we aim to find out precisely whether and to what extent different types of disambiguating cues can alleviate the arduousness typically associated with orcs. More in detail, we test how in-absentia dom, in-praesentia dom, and word order properties (henceforth wo) impact on children’s comprehension of Spanish orcs. Furthermore, we seek to understand whether different effects emerge based on the number of available cues and whether these interact in some way. First, though, we need to establish whether a src-orc asymmetry holds true in Spanish children’s comprehension. Albeit expected, this outcome is yet to be validated experimentally, which is why our first research question is the following:

Once the relative cost of orcs is determined, we will be able to infer whether they can be ameliorated and under which conditions. Concretely, our first experiment tests children’s comprehension of srcs as opposed to orcs, using orcs with post-verbal subjects and in-absentia dom. Our second experiment focuses instead on the comprehension of orcs disambiguated by, respectively, one in-absentia cue (i.e. in-absentia dom), one in-praesentia cue (i.e. either dom or wo), or two simultaneous in-praesentia cues (i.e. both dom and wo). These conditions were represented in the paradigm in (12). Testing children with such structures will allow us to address our second research question:

In a previous study, we tested the exact same sentences with an adult population and took into account their accuracy in the comprehension of such structures, as well as the reaction times (rts) associated with their responses (Presotto 2024b). Unsurprisingly, comprehension of srcs was at ceiling and, consistent with the results in Betancort et al. (2009), the correspondent rts were overall faster than those of orcs. Interestingly, however, not all orcs brought about the same outcome. Participants’ accuracy was virtually at ceiling with all orcs but the ones that were only disambiguated by a single in-absentia cue (like (12a)), which produced drastically lower rates (58.85%). As for the respective rts, we found that these were the highest when only in-absentia dom occurred, and significantly decreased as soon as dom was overt or there was a pre-verbal subject.19

Hindrances in adult on-line processing have been argued to be connected with children’s difficulties in comprehension (Belletti & Rizzi 2013; Phillips & Ehrenhofer 2015 and references therein). Thus a natural prediction for our current experiments would be that children, much as adults, exhibit worse performance with orcs as a whole. Among the latter, then, those featuring a single in-absentia cue should result in even poorer accuracy rates.

As mentioned in the introduction, though, we hypothesise that the acquisition of the grammar that underlies a specific cue might play a non-negligible role. In fact, this aspect could conceivably determine some discontinuity between child off-line data and adult on-line data. Whereas the adult parser likely relies upon a number of substantially automatised procedures (Baddeley & Hitch 1974; Ferreira et al. 2002; Givón 2002; Jefferies et al. 2004; Sturt et al. 2004; Ferreira & Patson 2007; Ferreira & Lowder 2016 and references therein), this is not always necessarily true for children, especially when dealing with material that is being acquired. In this regard, it has been maintained that children employ greater resources to analyse aspects of the target grammar that are in the process of being acquired or have been recently acquired. According to Biberauer’s Maximise Minimal Means model (Biberauer 2019a; b), children are indeed predicted to maximise the resources at their disposal and “consistently focus on input that is ‘just right’, […] minimising attention to any [input] which has become ‘too simple’ (as this has already been accounted for and is therefore ‘uninteresting’ and not worthy of cognitive resources)” (Biberauer 2019a: 216), eventually suggesting that the parser be particularly committed when attending to new aspects of the target grammar.20 In a way then, cues that turn on different aspects of the target grammar are likely to be processed differently. Those that characterise as material to acquire (hence “worthy of attention”) will draw more resources, whereas those that build on consolidated grammar will rather be more automatised. We could say that the computational load that is associated with a cue varies as a function of the acquisition of the piece of grammar that is to analyse: recently/not yet acquired aspects of the target grammar will be more burdensome than long consolidated ones.

If the above hypothesis is on the right track, we expect that wo cues will be processed more easily and in a more automatised, adult-like fashion, for word order properties are acquired particularly early in Spanish (Grinstead 1998; 2000; 2004; Bel 2001; 2003 a.o.).21 On the other hand, dom has been shown to be acquired only later (Rodríguez-Mondoñedo 2008),22 so that the (missed) mastery of its grammar can have an interesting reflection on how dom cues affect children’s comprehension.

Concretely, we predict that dom cues are more costly than wo cues for those children who have acquired dom, but nonetheless bring about an amelioration effect due to the fact that dom allows for structural disambiguation. For participants who, instead, have not acquired dom yet, we foresee two possibilities. The first possibility is that they simply neglect dom cues, resulting in lack of amelioration effects. Alternatively, they could benefit from dom cues based on probabilistic biases that associate the sheer presence of the sound a/al between the relative head and the relativiser with an orc reading, as predicted by the aforementioned experience-based accounts to relative clause processing (cf. Section 1 and in particular fn. 3).23 Note that the results will not only tell which alternative is better supported, but also shed new light on some critical aspects of the theory of locality in general. Let us suppose that dom cues trigger amelioration effects regardless of whether dom as a grammatical property has been acquired. In such a scenario, the effectiveness of a given cue would be hardly connected with its underlying grammatical representation, and a reductionist approach to intervention effects would naturally account for the facts. If, instead, ameliorations only arise when dom is associated with its function in the grammar, then the grammar itself must be acknowledged a crucial role in accounting for amelioration effects. In sum, the acquisition of dom-grammar is key to the understanding of intervention locality and thus constitutes the core of our third research question:

In the next section, we describe how we gauged participants’ acquisition of dom and illustrate our overall methodology. Then, we will discuss the results and, in developing the above considerations, clarify their import on the broader debate regarding the source of locality.

This study is divided into two main parts. The first part was designed to test the comprehension of srcs and orcs by Spanish children. The second part was designed to test whether and to what extent different types of disambiguating cues and their combinations impact on the comprehension of orcs specifically. In addition, we measured children’s mastery of dom grammar to understand how this impacts on their ability to comprehend orcs disambiguated by dom cues alone. Overall, the study comprised a total of three experiments: two sentence-picture-matching tasks and one sentence repetition task.

The results of our first experiment and their statistical interpretation unequivocally reveal that srcs are much easier than orcs for Spanish children to comprehend. This appears to hold true independently of the type of orc considered, as confirmed by the fact that even the condition that reached the highest proportion of accurate answers in Experiment 2 was far below the performance obtained with srcs.30 Thus, as expected, a subject-object asymmetry in children’s comprehension of relative clauses holds true for Spanish, much as in previously tested languages (cf. Section 2). This answers our first research question, which we repeat here:

We additionally found that older children, as well as children with higher syntactic abilities performed better with srcs.31 Somewhat more puzzling is the observed interaction between age and type of relative clause, which was also replicated in Experiment 2 (cf. Figure 3 and Table 3). In particular, it is unexpected that older children perform worse than younger ones in the comprehension of orcs with post-verbal subjects and in-absentia dom cues. One could argue that younger children show a lesser degree of commitment and resort to guessing more frequently than older children, who rather seem to favour the (wrong) src-reading more consistently. If so, the accuracy rates of the latter would be lower as they experience garden path-related errors in a more adult-like fashion as discussed in Section 3. Still, this interpretation is at odds with the fact that the great majority of the children (n = 45) only gave 0 or 1 accurate answers out of the 4 items corresponding to this particular condition (cf. Table 6 in the Supplementary Materials for a visualisation of children’s response patterns across conditions). Another possibility is that the emerging pattern stems from a classic instance of U-shape, whereby the mastery of a specific category is subject to a temporary regression before progressing again. Assessing this hypothesis would, of course, require testing older children. However, we leave this aspect of our results open to further interpretation.

Back to the main finding, the fact that srcs are understood better than orcs is surely not unexpected. In fact, a theory of intervention locality like Relativised Minimality explicitly predicts such an outcome: a disruption arises in the integration of configurations where two structural positions that are linked by a syntactic dependency are separated by an intervening constituent. Within such a framework, intervention is defined in terms of c-command, and the intervener needs to be syntactically similar to the target of the dependency. Differently from srcs, orcs represent a case of critical intervention configuration: the subject structurally intervenes between the relative head and the internal argument of the relative clause, which are linked by a syntactic dependency. Hence the asymmetry between srcs and orcs.

Relativised Minimality not only offers a neat account for intervention effects, but also provides an elaborated set of predictions of when these are weakened. To iterate, the degree of disruption is defined as a function of the featural distinctness between the target (here the relative head) and the intervener (here the subject), where: a) the distinctness is computed based on the set-theoretic relations between the features of the relevant syntactic objects, and b) only syntactically active features enter such computation. In other words, a mismatch in relevant features between the subject and the object should trigger an amelioration effect and ease children’s comprehension of orcs. If we look at our results, however, it is unclear whether Relativised Minimality’s predictions are entirely met. To see why, let us recapitulate some crucial aspects that emerged from the results of Experiment 2.

We found that dom ameliorates Spanish orcs, but only when it is overtly realised, for in-absentia dom does not aid children’s comprehension. Indeed, although children’s accuracy is fairly low even when dom is overt, the latter clearly ameliorates participants’ performance as compared with orcs featuring in-absentia dom and a post-verbal subject. Another determining factor that emerges from the data is wo: when the subject within the relative clause occurs pre-verbally children’s comprehension strongly increases. Thus, both in-praesentia dom cues and wo cues in isolation bring about an amelioration, with the latter being by far the most effective ones as they generate much higher performances. Crucially, though, when in-praesentia dom and wo cues are concurrently present in the same sentence, children’s accuracy significantly decreases.

Now, back to Relativised Minimality, is dom actually predicted to constitute an ameliorating factor? If dom is treated as a case marker (as suggested in e.g. Bossong 1985; Brugè & Brugger 1996; Aissen 2003),32 it should not fall within the definition of syntactically active features and thus should be excluded from the set of ameliorating cues/features (cf. section 2 and the sources cited therein). This is, for that matter, what Bentea (2012; 2016) and Sevcenco & Avram (2012) conclude for Romanian, where dom seems to not trigger a clear amelioration effect in orcs.33 The fact that Spanish dom does ease orcs’ comprehension is then problematic for such a framework. Another potentially puzzling aspect of our results is that wo is characterised as an ameliorating cue. Why would pre-verbal subjects help children’s comprehension under a Relativised Minimality perspective? A possible interpretation is that such subjects are specified with a [focus] feature (or, more in general, an information-structure related one) that is responsible for their raising to the pre-verbal position. This way, the target and the intervener would bear distinct sets of relevant features, accounting for the observed amelioration.34 Still, a particularly critical problem arises if we consider the third main finding of our experiment, namely the fact that when dom and wo are simultaneously present children’s comprehension decreases. Why would two cues that trigger ameliorations when occurring in isolation constitute an impediment as soon as they are realised together? If anything, their concurrent presence should strengthen the distinctness of the target and the intervener and actually boost even better performances. Or, in the worst case, these sentences should be at least as well-comprehended as those with only one wo cue. Again though, what we see is that the simultaneous presence of two cues hinders children’s integration of orcs, which cannot be straightforwardly made sense of calling upon (just) Relativised Minimality.

As a matter of fact, a closer look at the data opens an alternative and seemingly plainer interpretation. Indeed, our results are quite naturally accounted for if we take into account general principles of sentence processing and language acquisition. The fact that in-absentia dom does not trigger an amelioration effect, for instance, is substantially expected given the nature of this type of cue. Independent evidence shows that covert elements are challenging for children (Biberauer & Roberts 2012; 2015; Crisma et al. 2020 and references therein). Therefore, it is likely that in-absentia cues are not robust enough, especially as compared to overt ones. Arguably, then, children cannot build upon the information conveyed by the latter and, thus, are unable to disambiguate the structure. This would explain why orcs like (16a) are consistently assigned a src-reading, which is, all else being equal, the favoured/default interpretation.35 In addition, these orcs likely constitute a case of garden-path sentence, where the disambiguation can only be carried out as the last np is analysed. Considering that this causes a significant slow-down in adults’ processing (cf. section 2), it is understandable that children also struggle with this condition. In fact, a different result would have been very surprising.

Another aspect of our results that can be reasonably ascribed to processing principles is the decreased comprehension with orcs disambiguated by two cues at once. In sentences like (16d), the src-reading is banned by both subject position and overt dom, yet children’s accuracy gets lower as compared to that of orcs disambiguated by just wo. We contend that this outcome is the reflection of working memory (wm) limitations affecting the child parser. Under this view, each grammatical atom is associated with a certain computational cost that weighs on the parser. Cues like dom and wo are of course no exception and can thus conceivably lead to a wm-overload when simultaneously present. This effect might be even more evident when children are faced with demanding structures like orcs, which assuredly constitute an already burdensome base for the parser to operate on. If this is on the right track, the fact that cues only produce ameliorations when occurring alone immediately ensues. The reasoning is the following: the integration of orcs is a demanding task, but can be alleviated via morphosyntactic cues like those in question. These, however, besides favouring the disambiguation of the structure, load the parser with additional material to analyse. Therefore, their effect is beneficial inasmuch as their computational cost counterweights that of, say, intervention effects. If too many cues occur, the parser can suffer from an exceeding load, resulting in poorer comprehension.

Note that, within this perspective, the fact that wo and dom (in isolation) bring about remarkably different accuracy rates (cf. Figure 4) is actually a welcome result, one that confirms the hypothesis we explored in Section 3. As discussed therein, consolidated grammatical properties are taken to be processed more easily and in a mostly automatised fashion. Conversely, recently acquired aspects of the target grammar require a greater amount of resources (as the parser is more receptive towards newer information). Because wo properties are acquired early (see above), it makes perfect sense that wo cues represent the most effective means of amelioration: they disambiguate the structure without weighing too much on the parser. dom, instead, is acquired later and arguably weighs more on the sentence-analyser. In other words, the facilitative effect of a cue reflects its timing of acquisition (Tsimpli 2014; Schulz & Grimm 2018; Listanti & Torregrossa 2023). Further support for this interpretation is provided by the data we observed with adults (cf. Supplementary Materials). There, the reaction times associated with orcs featuring dom and wo simultaneously were even slightly lower than those of orcs featuring dom or wo alone. This naturally follows from the fact that adults have long acquired dom-grammar, whose integration is then as easy and automatised as different properties like, for instance, constituents’ ordering.

At this point, we can answer our second research question:

Our third research question was about the impact of the acquisition of a cue’s underlying grammar on its effectiveness. In other words, we wondered whether the ameliorating capacities of a cue are related to the knowledge of its particular grammar. To answer this last point we can look at the results of the srt presented in section 4.3, where we reported a three-way interaction between overt dom, wo, and dom-score. In Figure 5 we showed that the effect of (in-praesentia) dom cues is a function of dom’s acquisition: as children’s mastery of dom increases, their accuracy improves when dealing with isolated dom cues, and declines when these cues are concurrent with wo cues. In plain words, the impact of dom cues (be it positive or negative) only emerges with children who know that an a/al particle in front of a [human; definite; specific] np indicates that the latter is a direct object. If such knowledge is still missing, the child parser simply neglects the dom morphemes. We can then provide an answer to our last research question:

That being the case, a rather inescapable consequence is that the grammar must be attributed a quite determinant role: without the relevant grammatical knowledge, morphosyntactic cues are virtually irrelevant. This conclusion is far from being trivial. As a matter of fact, the identification of such a key role of grammar is downright essential to the overall interpretation of our results. To see why, let us briefly summarise what we have said so far. We maintained that, as far as Experiment 2 is concerned, an account based on Relativised Minimality, which fully modulates intervention locality within the grammar, is not unproblematic. Without some ad-hoc revisions, indeed, the calculation of the strength of intervention effects in Relativised Minimality terms cannot predict the observed results. We also contended that an approach that looks at non-strictly grammatical aspects of our cognitive system (e.g. computational resources, wm-limitations, parser’s receptiveness) seems to be more compatible with our findings. At this point, the question arises as to what import these considerations might have on the broader enquiry on locality. As already mentioned in the introduction, a long-standing debate persists about the source of locality effects in the human language. So-called grammar-based approaches argue that the syntax or, more generally, the grammar is responsible for emergence of these effects. On the other hand, processing-based or “reductionist” theories account for the same facts by assigning a crucial role to the performance system. So how can the data of our study contribute to this debate? Well, the question of the knowledge of the grammar underlying a morphosyntactic cue is, in this respect, a central one. In a way, to understand the role of the acquisition of a cue’s grammar is to understand the role of grammar itself. Because, in our case, dom only triggers an amelioration effect when its particular grammar has been acquired, we can deduce that cues’ effectiveness is dependent on, and tightly linked to their function within the grammatical system. If, on the other hand, dom cues were found to be relevant irrespective of such an acquisitional factor, we would conclude that the grammar does not play a pivotal role in the alleviation of intervention locality. Now, we said that the amelioration effects in Experiment 2 could be naturally derived from independent principles of our cognitive system. A fact that taken as such lends support to reductionist theories. At the same time, though, we just illustrated that since morphosyntactic cues need to be connected with an acquired grammatical function (being otherwise ineffective), evidence is there in favour of a grammar-based approach to locality, namely one that appoints the grammar as the origin and regulator of locality constraints.

In sum, the results of Experiment 2 and those of the srt, put together, seem to point to opposite directions if interpreted through the lens of the processing vs. grammar dichotomy that informs the debate on the source of locality. In spite of this apparent contrast, we argue that our findings need not be taken as conflicting evidence. Rather, it is our contention that they naturally open to a third theoretical stance, one where the mechanisms driving ameliorations to intervention-locality are neither entirely regulated by the grammar nor completely independent from it. That is to say, these mechanics reside at the interface between the grammar and the performance system and are thus predicted to be somehow affected by an interaction of the two. To iterate the point, we could say that both language-specific and language-independent factors contribute to the surface ameliorations observed in Experiment 2, and that the strategies responsible for amelioration effects have a twofold nature: on the one hand, they are driven by processing mechanisms, on the other, they hinge on purely grammatical representations. The need to abide by the constraints on the sentence-processor accounts for the impact of, for instance, wm-limitation. The centrality of grammar instead explains why these mechanisms operate exclusively on acquired linguistic material. In this light, we believe we need to orient ourselves towards an account that naturally comprise this “double nature”. Note that this need not be done through the rejection of current theories in toto. On the contrary, we uphold and strengthen specific aspects of them in moulding a third, intermediate analysis/hypothesis.36 We hold on to Relativised Minimality, for instance, in pursuing the idea that “the ingredients of the computation, and of the calculation of locality in particular, remain inextricably linked to language-specific properties” (Rizzi 2013: 184). At the same time, however, we eschew the whole set-theoretic apparatus of grammatically-encoded principles that Relativised Minimality builds upon in favour of, we believe, “lighter” assumptions.37 In this particular regard, we rather follow a typical reductionist perspective and contend that the machinery that drives the computation of (ameliorations to) locality can be straightforwardly derived from non-task-specific principles, namely general constraints on human cognition.

It remains an open question whether the hypothesis of an interaction between the grammar and performance systems can be expanded to other phenomena in the realm of locality theory. And the issue of the source of locality effects would undoubtedly be at the centre of such endeavour. For sure, further enquiry is needed to elaborate this line of reasoning, but we can already say that it would be somehow desirable to confirm that what originates locality effects belongs to the same interface system that drives their ameliorations.38 If true, not only amelioration strategies, but also locality proper could be construed as an interface phenomenon. We hope to further develop these hypotheses in future work.

In this study we tested the comprehension of srcs and orcs by Spanish monolingual children. Our results provide novel evidence that the long-observed subject-object asymmetry in child comprehension of relative clauses holds true for Spanish. Moreover, by testing different types of orcs we were able to determine that both overt dom-marking and wo properties (i.e. the subject position within the relative clause) constitute ameliorating factors in the comprehension of orcs. These results, paired with available evidence from the vast literature on children’s comprehension of relative clauses, corroborated our descriptive generalisation regarding amelioration effects: it is only when a disambiguation between src- and orc-readings is available that ameliorations arise, and the morphosyntactic properties that are responsible for the disambiguation qualify as ameliorating cues.

Spanish dom and wo cues are not equally effective, however. Children proved to be more sensitive to wo properties than dom-marking, with pre-verbal subjects bringing about better accuracy rates than overt dom. Even more strikingly, we showed that these cues interact when occurring together: children’s comprehension decreases when orcs with a pre-verbal subject also feature an overt dom-marker. We additionally measured children’s knowledge of the grammar underlying dom through an ad-hoc srt and found that dom only proves to be an ameliorating cue with children who have acquired it. We interpreted our results as evidence for both processing-oriented theories and grammar-based accounts to intervention locality. While the interaction between dom and wo is immediately explained as an effect of wm-overload, the fact that cues need to be associated with an acquired grammatical representation inevitably ties amelioration effects to language-specific factors. We proposed that this apparent contrast is actually predicted if the strategies at the base of amelioration effects are regulated at the interface between the grammar and the performance system. In this scenario, both domain-specific and domain-general factors are expected to contribute to the resolution of so-called intervention effects, which is, as discussed in Section 5, exactly what our findings suggest.

Although we are aware that an interface-based approach as the one sketched here is not (yet) completely immune to criticism, it is our firm conviction that this type of explanations represents a promising way to capture the multi-faceted nature of locality effects. A position, this, that finds support in the very premises of the minimalist programme, which openly favours the effort to achieve the “easiest possible grammar” through the substitution, when possible, of ad-hoc syntactic rules with domain-general principles. In this light, the long-standing dichotomy between grammatical and reductionist approaches to locality can be abandoned in favour of an intermediate position that integrates the power and predictions of both theories, as many scholars have recently intimated (cf. Presotto 2024a for a discussion). We leave the detailed description of such an interface system to future research.