There is considerable psycholinguistic evidence that object relative clauses (ORCs) (1b) are more difficult to process than subject relative clauses (SRCs) (1a): they give rise to higher latencies in reading times of critical regions, as well as higher reaction times and lower accuracy rates in comprehension tasks.

This asymmetry between SRCs and ORCs has been observed across different populations: (a) comprehension of ORCs is impaired in individuals with agrammatic aphasia (Grodzinsky 1989; Friedmann & Shapiro 2003; Garraffa & Grillo 2008; Sánchez et al. 2017; a. o.), (b) these structures are acquired later during language development (de Villiers et al. 1979; Mann et al. 1984; Friedmann et al. 2009; Adani et al. 2014; Guasti et al. 2018; in Spanish, Presotto & Torregrossa 2024, Torrens 2024; Llompart et al. 2025), and (c) even for adults with no language pathologies, the processing of ORCs is more costly than that of SRCs, as manifested in both online and offline measures (King & Just 1991; Gordon et al. 2001; Mak et al. 2002; Gibson et al. 2005; Fedorenko et al. 2012; Guasti et al. 2018; a. o.). Moreover, this asymmetry has been found in studies focusing on several languages: English (e.g., King & Just 1991; Gibson et al. 1994; Pickering 1994; Caplan et al. 2001; Gordon et al. 2001; Traxler et al. 2002), Dutch (e.g., Frazier 1987; Mak et al. 2002), German (e.g., Mecklinger et al. 1995; Schriefers et al. 1995), French (e.g., Holmes & O’Regan 1981; Cohen & Mehler 1996) and, of particular relevance for this article, in Spanish (e.g., del RÍo & López-Higes 2006; Betancort et al. 2009; del RÍo et al. 2012; Arancibia Gutiérrez et al. 2014; Reali 2014; Sánchez et al. 2017; Manoiloff et al. 2018; Murujosa et al. 2021; Dąbrowska et al. 2022; and see Murujosa & Sevilla 2022 for a revision).

A wide array of explanations has been proposed to account for this processing asymmetry. Most explanations can be grouped as based on (a) memory resources, (b) underlying structure, or (c) distributional patterns. Memory-based accounts assume that the language processor needs to hold the information of one end of the dependency in working memory until encountering the other end. In this view, the subject-object asymmetry is caused by the working memory cost of retrieving and integrating the relative pronoun and its co-indexed antecedent upon encountering the embedded verb (Gibson 1998; Grodner & Gibson 2005; Lewis & Vasishth 2005). Structure-based accounts, on the other hand, emphasize the underlying morphosyntactic structure of the sentences studied and propose explanations based on either universal parsing strategies (Frazier & Clifton 1989) or specific properties of the computational component of language (O’Grady 1997; Friedmann et al. 2009). Other structure-based accounts highlight the simultaneous influence of syntactic and non-syntactic information, such as lexical and semantic properties (Mak et al. 2002; 2006; Traxler et al. 2002; Bornkessel & Schlesewsky 2006; Gennari & MacDonald 2008; Dunagan et al. 2023). Finally, distribution-based accounts operate on the premise that frequently occurring structures in natural speech are processed more easily. Numerous studies have investigated how different distributional properties of RCs influence processing and acquisition (Hale 2003; Kidd et al. 2007; Reali & Christiansen 2007; Levy 2008; Diessel 2009; Reali 2014; Ambridge et al. 2015). Recently, studies on this processing asymmetry have shifted from studying the sentences as monolithic entities (e.g., SRC vs ORC) to studying how the fine-grained aspects of the semantic and morphosyntactic configurations of the sentences influence their processing. Research has focused, for example, on how animacy information and prominence of nominal phrases play a role during the processing of RC (Mak et al. 2002; Traxler et al. 2002; 2005; Mak & Vonk & Schriefers 2006; 2008; Betancort et al. 2009; Lin 2018) or on how different configurations of morphosyntactic features, such as case, gender and number, can impact the parsing of the sentence (Adani 2011; Belletti et al. 2012; Adani et al. 2014; Adani & Stegenwallner-Schütz & Niesel 2017; Adelt et al. 2017; Friedmann & Rizzi & Belletti 2017; Terzi & Nanousi 2018; Biondo et al. 2023). Furthermore, more studies have been carried out regarding the processing of RC in different languages, with special emphasis on the study of languages that have not always been in the focus of attention of psycholinguistic research and that manifest diverse typological variations in RC constructions (Blasi et al. 2022; Berghoff & Bylund 2025, for a discussion on the importance of cross-linguistic research). Some typological variations that have been studied include word order and headedness (e.g., in Chamorro: Wagers et al. 2018; in Tagalog: Pizarro-Guevara 2020; in Korean: Kwon et al. 2013; in Turkish: Bulut et al. 2019) and morphosyntactic alignment (e.g., in Basque: Carreiras et al. 2010; in Avar: Polinsky et al. 2012; in Tongan: Ono et al. 2020).

These studies have shed new light on the processing asymmetries and the overall nature of sentence parsing. In some instances, the previously well-established subject advantage of RCs has been questioned, giving rise to examples of languages where ORCs were found to be easier to process than SRC (Kidd et al. 2007; Chen et al. 2008; Lin & Garnsey 2011; Wagers et al. 2018) or examples of sentences in a given language in which the previously reported ORC disadvantage is ameliorated by a mismatch of relevant features (Kidd et al. 2007; Chen et al. 2008; Adani et al. 2010; Lin & Garnsey 2011; Belletti et al. 2012; Bentea & Durrleman & Rizzi 2016; Wagers et al. 2018; Bentea & Durrleman 2022; Papadopoulou & Douka & Paspali 2025). Overall, these antecedents have helped to focus on the analysis of the grammatical properties of languages and the fine-grained morphosyntactic configurations of the sentences under study, and their impact during comprehension.

Several studies on RCs comprehension and the subject/object processing asymmetry have shown that underlying morphosyntactic structures play an important role in the processing of complex sentences across languages. While different explanations have been proposed to account for this asymmetry, approaches focusing on the role of morphosyntactic structure, such as Relativized Minimality, have explained the observed difficulty in ORCs comprehension as a syntactic intervention effect (e.g. Friedmann et al. 2009). However, this processing difficulty has mostly been examined with prototypical transitive nominative-accusative verbs.

A main aim of the present study was to extend the empirical domain of the psycholinguistic literature on RCs by exploring a new type of structure: RCs with dative-marked object experiencer psychological predicates in Spanish. To do so, we took advantage of some properties of this language and used a specific set of sentences that allow an almost one-to-one comparison: activity dative assigning verbs (ACT) vs object experiencer psychological dative assigning verbs (PSY). The contrast between PSY and ACT verbs in Spanish in both subject and object RC comprehension was useful for testing (a) how the underlying different syntactic structures of superficially identical sentences affect their comprehension; and (b) whether, and to what extent, syntactic intervention can account for processing evidence obtained from a larger sample of types of filler-gap dependencies. To this end, an auditory comprehension task was carried out in which the processing of SRCs and ORCs (the Type of RC variable) with dative-marked object experiencer psychological verbs, in contrast to dative assigning activity verbs (Verb Type variable), was tested in native speakers of Rioplatense Spanish.

Two novel results emerged: First, a main effect of Verb Type was found in both accuracy and RTs, showing that RC sentences with PSY verbs were overall more difficult than RC sentences with ACT verbs. Second, there was an interaction between Verb Type and Type of RC in both accuracy and RTs, and this interaction showed a mirrored pattern effect of the Type of RC variable in sentences with different verb types.

The main effect of Verb Type obtained shows that sentences with PSY predicates were generally more difficult to understand than sentences with ACT verbs in terms of lower accuracy and higher latencies. Previous studies have shown that object experiencer psych verbs in simple declarative sentences lead to greater processing difficulty according to various metrics, such as increased reading times (Cupples 2002; Brennan & Pylkkänen 2010), and decreased comprehension accuracy in patients with Alzheimer’s disease (Manouilidou et al. 2009) and aphasia (Beretta & Campbell 2001; Piñango 2009; Thompson & Lee 2009). Results from studies that used ERPs (Bornkessel et al. 2003) and neuroimaging techniques (Bornkessel et al. 2003) also found differential brain activity after the presentation of sentences containing each type of verb. This added difficulty has been attested as well for dative-marked object experiencer verbs in Spanish declarative sentences (Gattei et al. 2015a; Gattei et al. 2015b; Gattei et al. 2017; Gattei et al. 2021; Taboh et al. 2024).

Among the factors that have been proposed to account for this effect in declarative sentences, differences in the semantic structure of psychological predicates have been pointed to as a source of increased processing load. In particular, distinctions between stative and eventive readings, as well as variation in the internal semantic composition of experiencer events, have been argued to give rise to richer or less prototypical event representations, which in turn affect comprehension (e.g., Gennari & Poeppel 2003; Brennan & Pylkkänen 2010). Such semantic properties are closely related to well-known difficulties at the syntax–semantics interface, insofar as psychological verbs frequently involve non-canonical patterns of argument linking, including configurations in which the experiencer is semantically prominent but not realized in a canonical subject position. These mismatches between semantic prominence and syntactic realization have been shown to challenge standard thematic hierarchies and to incur additional processing costs (Gennari & MacDonald 2009; Gattei et al. 2015a).

In addition, evidence from corpus-based and experimental studies suggests that speakers exploit distributional regularities linking verb classes and word-order configurations to generate predictive expectations during sentence comprehension. Corpus analyses of Spanish psychological predicates indicate systematic associations between verb type and surface order, which can be taken as a source of probabilistic expectations (e.g., Mateu 2022). Experimental work further shows that readers use these distributional cues online, and that deviations from expected verb–structure pairings—such as psychological verbs occurring in less frequent configurations—can modulate both accuracy and response times (Gattei et al. 2015b). In our specific study, an additional factor may be contributing to this Verb Type effect: the greater difficulty in extracting relevant information (i.e., the psychological states of the depicted participants) from the images used in the comprehension task.

Interestingly, our results show that the difficulty observed in simple declarative sentences remains during the comprehension of RCs with psychological verbs, a structural context in which it has not been fully investigated. While consistent with prior evidence from declaratives, the exact nature of this effect is far from being clarified through our manipulation. In this sense, accounting for the exact cause of this phenomenon exceeds the aim of this work and should be examined separately in future research, as the adult comprehension and production of sentences with psychological predicates is still an understudied field, especially regarding Spanish and other Romance languages.5 It is important to note, however, that while RM provides a precise account of the subject/object asymmetry in RCs by linking processing costs to the reconfiguration of featural intervention induced by the verb’s underlying syntactic structure, the theory, as a model of locality in A-bar dependencies (Rizzi 2004; Friedmann et al. 2009), does not directly predict or explain the main effect of Verb Type itself. This main effect seems to reflect more general processing costs associated with object experiencer psychological predicates independent of filler-gap dependencies. This is not a shortcoming of RM, which is not designed as a comprehensive sentence processing model, but rather a strength of its targeted scope: it successfully captures how verb-induced syntactic differences modulate intervention in relative clauses, yielding the observed reversal of the classic asymmetry –which is discussed below.

The results obtained from RCs with activity verbs in this experiment replicate previous studies comparing subject and object RC with prototypical accusative-nominative verbs in Spanish. Thus, RCs with ACT verbs showed the already known processing asymmetry: sentences with ORCs, in comparison to SRCs, led to more mistakes in the comprehension task. SRCs were answered almost at ceiling level (95.8%, SE = 1.28%), which could also explain the marginally significant main effect of Type of RC, where SRCs were globally answered more accurately than ORCs. The observed asymmetry was also found in reaction times: ORCs yielded higher latencies than SRCs. This shows that, although our experimental sentences with ACT dative-assigning verbs are different from the previously studied prototypical nominative-accusative sentences, both structures are comparable in terms of relevant properties: both types of verbs (dative-assigning vs accusative-assigning ACTs) project a canonical word order (SVO) and a sentence structure with the object in a lower syntactic position than the subject.

Importantly, RCs with PSY verbs showed a mirrored pattern: as predicted by RM, ORCs led to fewer mistakes and lower latencies, in comparison to SRCs. Accuracy rates and RTs for each experimental condition showed that, in Spanish, ORCs are not always more difficult to process than SRCs. These results suggest that the difficulty associated with the processing of RCs is specific to underlying syntactic configurations rather than monolithic structure types.

Our data regarding RCs with PSY verbs and ACT verbs align with the hypothesis of syntactic intervention by inclusion of features proposed by RM, presented in §1. The establishment of the syntactic dependency between filler and gap becomes more difficult when there is an intervening element, and this translates into a decrease in the performance of the participants in experimental comprehension tasks and higher response times. In the case of the ACTs (8), effects of syntactic intervention are triggered during the derivation of the ORC, as shown in (8b), since the embedded subject DP intervenes in the establishment of the syntactic dependence. In the case of the PSYs (9), it is during the derivation of the SRC (9a) that syntactic intervention effects are triggered since the embedded object DP functions as an intervener.

Overall, these results are consistent with previous findings showing that not all RCs in all languages behave in the same way and that the behavior in each language depends, to a large extent, on its grammatical properties (e.g., Hsiao & Gibson 2003; Carreiras et al. 2010; Kwon et al. 2013). In Spanish, for example, the results obtained show structural differences between the syntactic configurations of PSY and ACT verbs with dative arguments and argue in favor of a syntactic description in which the experiencer argument of PSY is generated in a higher position than the theme argument.

Our results, and this syntactic description of PSY, are consistent with the research carried out by Mateu (2022). The author studied the comprehension of wh-questions with dative-assigning ACTs and PSYs in both Spanish speaking adults and children; and found that PSY dependencies were harder to comprehend when the wh-questions were subject-extracted. Also, within the RM framework, and assuming the high applicative proposal to account for PSYs, the author explains these results in terms of syntactic intervention. The subject-extracted wh-questions with PSYs involve moving the subject wh-phrase past an intervener, the experiencer DP, which hinders the establishment of the syntactic dependency. Mateu’s results, like ours, are in line with the predictions of RM.

As we have argued, under the proposed analysis for both types of dative assigning verbs, ACTs and PSYs (Cuervo 2003; 2020; Pujalte 2015), the structures used in this study can be considered parallel (and opposite) in configurational terms. However, even under this analysis, they differ in other fundamental features, including the thematic properties of their arguments. The extent to which the intervention effects found can be explained solely by the position of the +NP intervener or can be modulated by other features remains a valid question, as has been mentioned in the Introduction section. For example, it could be hypothesized that differences in the feature configurations involved in (12a) and (12b), driven by thematic features, may give rise to distinct intervention effects.

Within the RM framework, studies have investigated whether other features, other than the lexical restriction of the +NP feature, play a role during syntactic intervention (Adani et al. 2010; Adani 2011; Belletti et al. 2012; Bentea et al. 2016; Adani et al. 2017; Friedmann et al. 2017; Terzi & Nanousi 2018; Papadopoulou et al. 2025; Pettenon & Sanfelici & Mateu 2026). In this line, RM argues in favor of a hierarchical configuration of features relevant during the establishment of syntactic dependencies. Following Rizzi (2018), the first distinction is between criterial and non-criterial features: the intervention of criterial features, like the +Q feature, gives rise to grammatically degraded sentences, such as weak islands like (13); while the intervention of non-criterial features, like +NP, gives rise to derivational and processing difficulties (but not to agrammaticality).

Research has focused on the role that other non-criterial features, besides the +NP, may play. For example, in some languages a match of animacy, case, number or gender features of the DPs can magnify the intervention effect, while a mismatch ameliorates it. For example, in the above-mentioned study, Mateu (2022) also manipulated the number feature of the constituents, creating both number match and mismatch conditions between subject and object, and found that a mismatch in number improved children’s performance only for object wh-questions with ACTs and subject wh-questions with PSYs (but not for ACTs subject wh-questions nor PSYs object wh-questions).

Moreover, the size of this ameliorating effect may be different for each feature: Adani et al. (2010) found that number mismatch had a bigger facilitating effect than a gender mismatch in Italian RCs. However, more studies are needed to determine what features are relevant or not for the establishment of filler-gap dependencies in different languages (see Belletti et al. 2012, for a discussion) and whether different values of the features may behave differently as well. Evidence from research on the establishment of local dependencies, such as subject-verb agreement, has found that certain values of the number or gender features have a stronger hindering effect, for example, the plural value (Bock & Cutting 1992; Bock & Miller 1991; Fayol et al. 1994; Sánchez et al. 2018) and feminine value (Badecker & Kuminiak 2007; Slioussar & Malko 2016; Sánchez et al. 2016; 2018) act as stronger attractors in agreement production.

As it can be observed in Figures 2 and 4, the extent to which comprehension is more costly for participants according to the Verb Type and Type of RC variables does not seem to be different between ACT and PSY predicates and the intervention effect appears to be mirrored in ACTs and PSYs. These results are consistent with the featural RM claim that only morphosyntactic features that trigger movement enter into the computation of intervention (e.g., Belletti et al. 2012; Bentea et al. 2016; Rizzi 2018). While thematic features may have structural repercussions, they do not trigger movement and, according to the featural RM claim, should not be computed.

Therefore, our results are consistent with the predictions of RM regarding intervention effects. However, as noted in the Introduction, other theoretical accounts have been proposed that could explain the subject-object processing asymmetry in sentences with activity verbs. Some of these theories could also account for the data obtained in this study. According to memory-based accounts, the difficulties observed in the processing of ORCs have been explained as (a) a distance effect, generated by the additional cost of intervening material between the antecedent and its integration point (e.g., Gibson 1998; 2000) or by the decay in activation of the stored representation (Lewis & Vasishth 2005), or/and an interference effect, generated by the similarity between competing stored representations (e.g., Gordon et al. 2001; 2004, and also Lewis & Vasishth 2005). These proposals focus on possible processing effects of retrieval and integration at the embedded verb and the main verb of the sentence. While these theories could explain the general processing asymmetry effect, the RCs used in this study raise a new question, as the embedded subject appeared in a post-verbal position. Explanations based on integration/retrieval effects at the embedded verb as the main source of additional processing load and/or retrieval difficulty are challenged by the word order exhibited by Spanish (for a discussion of this aspect of RCs in Spanish, see del RÍo et al. 2012).

Given the type of verb manipulation performed, our results may be of interest for structure-based interactive models that consider both morphosyntactic and semantic features of noun phrases, such as the extended Argument Dependency Model (eADM). In this direction, Bornkessel & Schlesewsky (2006) propose that morphosyntactic (e.g. word order, case marking) and semantic (e.g. animacy, definiteness) features of NPs are extracted to compute the relative prominence of one argument over others and to assign a plausible thematic role. The model proposes that the sentence parser will assign the first NP the highest possible thematic role aligned with those features and predict a verb type with a lexical-semantic structure that matches such thematic role. This model might predict higher cognitive costs for SRCs with psych verbs compared to SRCs with activity verbs, and conversely, greater costs for ORCs with activity verbs compared to ORCs with psych verbs, which is not the comparison of interest in this paper. However, the eADM derives its predictions primarily from studies examining electrophysiological and neurobiological changes during online comprehension of simple sentences6 (but see Dunagan et al. 2023 for a recent work on ORCs processing in English and Chinese).

Finally, with regard to usage-based approaches (Reali & Christiansen 2007; Diessel 2009; Tomasello 2009; Reali 2014; Ambridge et al. 2015), the explanation for the difficulty associated with ORC is related to its low frequency in spoken and written language. These studies mostly explain the behavioral data obtained by taking into account the distribution patterns of the sentences under study, found in corpus analyses or sentence completion (cloze) tasks. However, no corpus studies have been carried out to investigate the frequency of RCs with dative assigning verbs in Spanish. There has been, nonetheless, some data reported regarding the frequency of word order with PSYs. For example, Mateu (2022) examined the frequency of sentences with the verb gustar ‘to like’ produced by both children and adults interacting with children in Spanish CHILDES corpora (MacWhinney 2000). The results showed that children begin to produce the Exp-V-Theme (OVS) order earlier than the Theme-V-Exp (SVO) order; and that both adults and children produce more instances of the Exp-V-Theme (OVS) order than Theme-V-Exp (SVO) utterances.7 Furthermore, in a study conducted by Gattei et al (2015b), Rioplatense Spanish adult speakers were asked to complete a sentence completion task in which critical items began with either a nominative or a dative phrase, and both phrases were followed by the dative clitic le. Participants therefore had to complete the sentences with a verb. The authors report that when the sentences began with a nominative phrase, participants responded more often with an activity verb; but when the sentence began with a dative phrase, participants responded with a dative-marked object experiencer psychological verb (with gustar ‘to like’ being the most common). If this distribution pattern were to be replicated in RCs, the data obtained in this study could also be explained.

While assessing the applicability of such models is beyond the scope of the present investigation, broadening the theoretical scope beyond RM raises interesting avenues for future research and merits further in-depth exploration.

In this study we set out to revisit the subject/object asymmetry in the comprehension of sentences with relative clauses in Spanish. The RCs contained either a dative-marked object experiencer psychological predicate or a dative assigning activity predicate. This contrast allowed us to manipulate the relative structural positions of the subject and object within the clauses. The results of the current experiment show that subject relatives are not always easier to comprehend than object relatives in Spanish. When the RC contained a dative-marked psychological predicate the traditional subject/object pattern was inversed: object-extracted RCs elicited a higher accuracy rate and lower reaction times.

Our results are in line with accounts of the processing asymmetry that take into consideration the fine-grained aspects of the morphosyntactic structural configurations that underlie sentence comprehension, such as the RM framework. Moreover, our work highlights the importance of carrying out experimental studies in a wide array of languages, with diverse grammatical properties, and exploring how different morphosyntactic aspects affect language processing.