Glossa Psycholinguistics

Introduction

Linking is the process by which the human language processing system maps syntactic function of constituents to semantic information provided by the thematic roles established in the lexico-semantic structure of the verb. The result of this process allows speakers to understand ‘who is doing what to whom’ in a given utterance. Thus, accurate argument interpretation requires that linking is achieved correctly. To this end, during incremental processing speakers have to determine which participant of the sentence bears the most active thematic role, and which one bears the least active one. This process is known as argument hierarchization (Bornkessel et al., 2005). While different types of linguistic information may serve as cues for establishing the correct linking between syntax and semantics (e.g. word order, morphological case marking, animacy and definiteness of the constituents being the most important ones), the weight each cue imposes over argument interpretation is language-specific (MacWhinney et al., 1984). For instance, while English is a language that heavily relies on word order for argument interpretation, speakers of languages with a less rigid word order (e.g. German, Finnish, Russian) may use morphological case marking as a more informative linguistic cue (see Lamers and De Swart, 2012: for a collection of works that focus on this issue).

Several studies have inquired about how language-specific characteristics may affect the way linking is achieved during online processing, introducing evidence for psycholinguistic models that attempt to explain how this process is carried out (see Bornkessel and Schlesewsky, 2006; Bader and Bayer, 2006: for different models that focus on linking). Of particular interest for the current study, in a group of comprehension studies conducted in Spanish, Gattei et al. (2015a; 2015b; 2017) have shown that in this language, morphosyntactic information provided by case marking and word order, along with arguments’ span (the amount of available arguments previous to the verb) trigger predictions about the type of event that will be described, and thus syntax-to-semantics linking of the type of verb predicted. Results of these studies show that when reading sentences with different word orders (SVO as in (1)a and (1)c and OVS as in (1)b and (1)d), the appearance of a verb that is incongruent with the type of event predicted leads to higher reading times and higher error rates (Gattei et al., 2015a; 2017), differential neural correlates (Gattei et al., 2015b) and higher amount of incoming regressions to the verb and the arguments regions (Gattei et al., 2017). The authors interpreted the outcome as a reflection of higher processing cost when the sentence word order did not reflect the order of arguments in the lexico-semantic structure of the verb, which positions the most prominent argument of the structure in first position.

Furthermore, the authors found higher reading times for sentences that required a non-canonical linking between syntax and semantics (Gattei et al., 2015a: i.e. when the nominative-marked constituent is linked to the lowest-ranked thematic role) than for sentences with direct linking (i.e. the highest constituent in the syntactic hierarchy, is linked to the highest thematic role). This effect persisted even for the response times and accuracy rates of comprehension questions in both Gattei et al. (2015b) and Gattei et al. (2017). A question that remains unanswered relates to the role of the verb classes used in these experiments. The authors of the above mentioned studies aimed at maintaining the syntactic structure of sentences constant and used two-argument sentences with a nominative and a dative-marked object. They evaluated the interpretation of events that belong to different semantic classes that corresponded either to the class of activity verbs (e.g. ‘gritar’, ‘to yell’, ‘mentir’, ‘to lie’), or to the class of Object Experiencer psychological verbs (hereafter ObjExp psych verbs, e.g. ‘gustar’, ‘to like’, ‘encantar’, ‘to love’, and see Dröge et al., 2014; Bornkessel et al., 2003; 2005: for similar examples in German and Italian respectively).

Prior studies have shown that verb processing may differ according to several factors, including verb type (Pulvermüller et al., 2001), type of subcategorization and possible thematic roles (Shetreet et al., 2007), and semantic complexity (Brennan and Pylkkänen, 2010). Thus, it is unclear whether the verb type effects found in Gattei et al. (2015a) and Gattei et al. (2017) reflect greater difficulties related to linking complexity or if other factors may play a role.

The current study aims at responding to this question in order to shed light on how linking is achieved during real-time comprehension when the same class of verbs is used. It also goes a step further from other studies that investigated linking by using sentences with a special class of Spanish ObjExp psych verbs: Psych verbs which can bear both accusative or dative case marking.

The use of this subclass of psych verbs may provide further evidence on how a grammatical feature such as case marking serves as a valid cue for sentence overall interpretation in this language (see MacWhinney et al., 1984: for an experimental example of how different syntactic and semantic features modulate interpretation according to language-specific characteristics).

The study is organized as follows: we first introduce a few theoretical considerations about this class of verbs and discuss how they differentiate from psych verbs used in previous psycholinguistic studies. Then, we elaborate on the hypotheses and predictions for online comprehension of Spanish sentences with different word order. We finally test these hypotheses by means of a sentence comprehension eye-tracking study and an acceptability judgement task, and discuss the results under the light of the previous evidence on this topic and different psycholinguistic models that aim at explaining how linking between syntax and semantics is achieved during on-going processing.

Theoretical considerations about Spanish psych verbs with double case marking

The theoretical categorization and distinction of the different classes of psychological verbs (hereafter psych verbs) have been of great interest for the formal studies about the syntax-to-semantics linking, in particular, since the appearance of the influential work by Belletti and Rizzi (1988) about psych verbs and Θ Theory in Italian. In general terms, the authors classify Italian psych verbs in three classes. In the first class of verbs (the so-called ‘temere’ class), the Experiencer takes nominative case, while in the second class (the ‘preoccupare’ class), the Experiencer bares accusative case. Finally, the third class of psych verbs (the ‘piacere’-like class) assigns dative case to their Experiencer.

Spanish exhibits the same case-marking pattern for its first and third psych verbs classes, with verbs like ‘amar’ (‘to love’) and ‘gustar’ (‘to like’) respectively. However, it is still part of a controversial discussion what the most probable case marking is for the different exemplars that belong to the second class of psych verbs. For instance, the verb ‘molestar’ (to bother) allows accusative, dative and nominative case marking for the constituent related to its Experiencer, as it can be read in (2).

In these sentences, the Experiencer is the event role performed by a masculine NP (“him” /“he”). This same role is realized by an accusative pronominal clitic in (2a), a dative pronominal clitic in (2b) and as nominative subject in (2c), entailing different types of linking between syntactic constituents and semantic arguments. Furthermore, the Experiencer in each example presents a different status regarding its hierarchization against the other arguments of the sentence, thus affecting the way linking is achieved. (2a) represents direct linking since the highest ranked participant (e.g. the Agent) is linked to the subject while the lowest ranked one (e.g. the Experiencer) is linked to the direct object. Conversely, (2b) represents inverse linking because the highest-ranked participant of the event (e.g. the Experiencer) is not linked to subject (nor direct object) and, in consequence, it is marked dative morphology. The alternating nature of linking allows for these realizations but the fundamental question is whether these different couplings convey different meanings.

The literature regarding object case-marking alternation -with particular emphasis on causative constructions- has proposed that transitivity (Perlmutter, 1978; Comrie, 1976; Gibson and Raposo, 1986) and semantic / pragmatic factors (Hyman and Zimmer, 1976; Dorel, 1980; Reed, 1992) play an important role on the selection of arguments’ case codification. As for the ‘molestar-class’ psych verbs, several authors have tried to theoretically explain which aspects of syntax and semantics motivate different linking patterns (Treviño, 1990; Franco, 1992; Ackerman and Moore, 1999; Honrubia, 2015; Mendivil, 2002; Fernández-Soriano, 1999; Cuervo, 1999; Masullo, 1992). Although there may be important differences among the assumptions that each approach entails, a number of commonalities may be established. Most importantly, in all accounts linking between syntax and semantics plays an essential role in capturing the semantic differences in the event’s interpretation, either if linking is explained in terms of mapping the syntactic constituents with their specific Θ roles, proto-roles or generalized-semantic roles. Second, most accounts emphasize the dichotomic interpretation of the ‘molestar-class’ of event as being [+ causative ] / [– stative ] when accompanied with an accusative-marked object or [– causative ] / [+ stative] when the object bears dative case. Following Van Valin and LaPolla (1997), events described by verbs similar to ‘molestar’ can be interpreted either as a causative state or as a state, entailing two different types of linking between syntax and semantics, as it may be observed in Figure 1. Linking for the accusative-marked verb is direct, while verbs with dative case-marking show inverse linking.

Figure 1

An illustration of the relation between syntactic templates and arguments’ linking to generalized semantic roles for sentences with the same verb but different case-marking encoding (accusative vs dative) according to Role and Reference Grammar. Both sentences show the same syntactic template, but differ on the type of linking between constituents and the arguments of the lexico-semantic structure of the verb. While in sentence A the nominative-marked constituent is linked to the Actor (Agent) of a Causative State, in sentence B, it is linked to the Undergoer (Experiencer) of a State. Notice that the dative argument is directly associated to the lexico-semantic structure of the verb. CORE: the minimal phrase structure domain of the clause within which the nucleus and its arguments are realized; ARG: obligatory argument (noun phrase [NP] or prepositional phrase [PP]); NUC: nucleus (essentially equivalent to head, i.e., the element subcategorizing for the arguments); V: verb; AGX: agreement index (i.e. dependent of the NUCLEUS, it receives the agreement specifications of all core argument positions present in the Argument Structure); ACC: Accusative; NOM: Nominative; DAT: Dative.

In the following section, we attempt to explain the consequences of such characterization of the ‘molestar’-type of verbs for incremental sentence comprehension

Results

Total reading time

Mean total reading time was 6565 ms (SE = 123 ms). Figure 2C shows the average total reading time for the critical sentences used in the current eye-tracking experiment. The statistical analysis revealed an interaction between case and word order; β = –0.045, SE = 0.011, t = –4.096. Resolving this interaction showed that participants spent significantly longer time reading SVO sentences when the verb included a dative clitic than when they included an accusative clitic; β = 0.126, SE = 0.034, z = 4.025, p < 0.001. Besides, participants spent significantly longer time reading sentences with a dative-marked verb when they followed the SVO word order than when they followed the OVS order; β = 0.112, SE = 0.031, z = 3.592, p < 0.01. No significant differences between OVS conditions were found, p = .29.

Figure 2

Percentage of accurate answers (panel A), mean response times for the comprehension question (panel B, and mean total reading time for the critical sentences (panel C) in the current eye-tracking study according to case marking (Accusative vs. Dative) and word order (SVO vs. OVS). Error bars correspond to 95 % Confidence Intervals. ACC = Accusative-marked verb; DAT = Dative-marked verb; SVO = Subject-Verb-Object; OVS = Object-Verb-Subject.

Eye-tracking measures

Following Kliegl et al. (2006), we excluded the first word of all sentences for all reading duration analysis since the first word was positioned at the location of the fixation spot that triggered the appearance of the sentence. We also refrained from inspecting most reading measures for Region 2 (case marking preposition ‘a’) since the model could not be fit with any of the fixed variables. On the one hand, Region 2 only comprises sentences with OVS word order. On the other hand, case marking is only relevant for participants once they have read the clitic on Region 6. Thus, including these variables or the interaction between them did not make sense when fitting the model, and as it was explained at the data analysis section, control variables (i.e. frequency, and length) were kept fixed both at this region and Region 7. We thus inspected the reading measures FFD, FPRT, RBRT, RBRC and RPD from Region 3 onwards. However, at Region 1 and 2, we did analyse the probability of regressing to the first and second words (TIR) from other regions and the total fixation time.

Figure 3 summarizes the contrast between sentences with accusative case (causative psych states) and dative case (psych states) according to both word orders (SVO and OVS) for late eye-movement measures. Positive values mean that reading time and regression counts are higher for sentences with dative-marked verbs than for sentences with accusative-marked verbs. A positive black line (SVO conditions) and a negative grey dashed line (OVS conditions) correspond to an interaction between Case and Word Order. This representation makes the interaction visually clear. The asterisks show the regions where the interaction was significant. Table 5 shows mean values and standard errors for each condition at all regions and measures in which case marking yielded a significant difference.

Figure 3

The Figure shows the difference (Δ) in mean fixation times (ms) and the amount of regressive saccades (counts) between conditions with dative and conditions with accusative case marking according to the sentence word order. Error bars correspond to Standard Error of the Mean. Eye-tracking measures: TFT = Total Fixation Time; RPD = Regression Path Duration; RBRC = Right-Bounded Regression Count; TIR = Total Incoming Regressions. Word Order: SVO = Subject-Verb-Object; OVS = Object-Verb-Subject. The asterisk shows that the interaction between Word Order and Verb Type was significant.

Table 5

Mean values and standard error for each condition for all regions and measures in which case marking yielded a significant difference. TIR = Total Incoming Regressions; TFT = Total Fixation Time.

	TIR				TFT
	Region 6		Region 8		Region 6		Region 8
Condition	M	SE	M	SE	M	SE	M	SE
SVO.ACC	0.665	0.057	0.15	0.025	957	40	462	22
OVS.ACC	0.792	0.063	0.181	0.028	1080	46	539	27
SVO.DAT	1.069	0.068	0.227	0.03	1227	52	623	30
OVS.DAT	0.842	0.668	0.265	0.031	1099	54	597	31

We now provide the analysis of regions of interest for late eye-movement measures mentioned in the ‘Data analysis’ section. For each region, we first present the analysis of the interaction between Case and Word Order, since it is the contrast of interest of the current study. We then provide the relevant results of the multiple comparisons test whenever was needed. Finally, we report main effects of Case and Word Order. Appendix B shows the final converging models for each measure at each region. A table with all statistical coefficients including control factors (Length and Frequency), standard errors and t and p values for all measures, and the figures that depict the interaction between variables for early eye-movement measures may be found at Appendix C.

Regions 1 (Adverb) and 2 (Case marking preposition ‘a’)

Analysis of the probability of regressions into these regions (‘TIR’) and total fixation time (‘TFT’) revealed no interaction between case and word order.

Region 3 (First proper name)

Analysis of this region showed that TIR was affected by word order, showing a significantly higher probability of incoming regressions for object-marked constituents than for subject-marked constituents (OVS: M = 0.95, SE = 0.05; SVO: M = 0.65, SE = 0.04, β = 0.198, SE = 0.041, z = 4.791, p < 0.0001) .

Word order also modulated Total Fixation Time (‘TFT’). Participants fixated for significantly longer time at this region when the sentence began with the object-marked constituent than when it began with the subject constituent (OVS: M = 594 ms, SE = 21 ms; SVO: M = 482 ms, SE = 17 ms; β = 0.086, SE = 0.020, t = 4.361, p < 0.0001).

Region 4 (Auxiliary verb ‘ir’)

A main effect of word order was found for TFT measure, as well as for Regression Path Duration (‘RPD’). Participants took significantly longer to read this region and regressed for significantly longer time before continuing reading when they were reading an object-initial sentence (TFT: M = 424 ms, SE = 17 ms ; RPD: M = 248, ms, SE = 12 ms) than when they were reading a subject-initial sentence, TFT: M = 355 ms, SE = 14; β = 0.074, SE = 0.023, t = 3.212, p < 0.005; RPD: M = 200 ms, SE = 6; β = 0.091, SE = 0.030, t = 3.016, p = 0.005.

Region 5 (Auxiliary particle ‘a’)

Analysis of this region revealed no interaction between case and word order nor main effects of any of these two factors.

Region 6 (Disambiguating verb)

Analysis of this region showed a significant interaction between case and word order for TFT and TIR (TFT: β = –0.064, SE = 0.017, t = –3.866, p = 0.0001; TIR: β = –0.109, SE = 0.034, z = –3.192, p = 0.001). Post-hoc analyses of these measures revealed that, participants fixated for significantly longer time on this region and regressed significantly more times to this region in subject-initial sentences with a dative-marked clitic than in the other three conditions (TFT: SVO.DAT-SVO.ACC: β = 0.185, SE = 0.068, z = 2.707, p < 0.05; SVO.DAT-OVS.DAT: β = 0.165, SE = 0.047, z = 3.486, p < 0.005; SVO.DAT-OVS.ACC: β = 0.185, SE = 0.068, z = 2.707, p < 0.05; TIR: SVO.DAT-SVO.ACC: β = 0.500, SE = 0.123, z = 4.059, p < 0.001; SVO.DAT-OVS.DAT: β = 0.259, SE = 0.090, z = 2.881, p < 0.05; SVO.DAT-OVS.ACC: β = 0.324, SE = 0.120, z = 2.713, p < 0.05). The probability of regressing to this region was also higher for sentences with a dative-marked verb (M = 0.96, SE = 0.05) than for sentences with an accusative-marked verb (M = 0.73, SE = 0.04; β = 0.141, SE = 0.051, z = 2.749, p = 0.005).

Region 7 (Case marking preposition ‘a’)

Analysis of this region showed no significant interaction between case and word order nor main effects of any of these factors.

Region 8 (Second proper name)

Analysis of this region showed an interaction between case and word order for RPD and TFT (RPD: β = –0.067, SE = 0.024, t = –2.766 p = 0.005; TFT: β = –0.055; SE = 0.018; t = –3.034, p < 0.005). Resolving these interactions revealed that the regression path duration was significantly longer for SVO sentences with dative-marked verbs than for the other three conditions (SVO.DAT-SVO.ACC: β = 0.300, SE = 0.068, z = 4.360, p < 0.001; SVO.DAT-OVS.DAT: β = 0.205, SE = 0.069, z = 2.975, p < 0.05; SVO.DAT-OVS.ACC: β = 0.237, SE = 0.069, z = 3.458, p < 0.05). Planned comparisons of TFT showed that, in SVO sentences, participants fixated on this region for significantly longer time when the main verb included a dative clitic than when it included an accusative clitic, β = 0.285, SE = 0.052, z = 5.521, p < 0.001. Although no significant differences were found between OVS conditions, in sentences with an accusative-marked verb participants fixated for significantly longer time on this region when the sentence was object-initial than when it was subject initial, β = 0.160, SE = 0.051, z = 3.126, p < 0.05. Finally, the analysis also showed a main effect of case for RPD, TFT and TIR. On average, participants spent significantly longer time fixating on this region when the main verb included a dative clitic (M =618 ms, SE = 22 ms) than when it included an accusative clitic (M = 503 ms, SE = 18 ms β = 0.087, SE = 0.018, t = 4.819, p < 0.005). The probability of regressing to this region (TIR) was also significantly higher for sentences with dative-marked verbs (M = 0.24, SE = 0.02) than for sentences with accusative-marked verbs (M = 0.16, SE = 0.02; β = 0.199, SE = 0.069, z = 2.895, p < 0.005). Likewise, the regression path duration at this region was significantly higher for sentences with dative-marked verbs (M =626 ms, SE = 34 ms) than for sentences with accusative-marked verbs (M =504 ms, SE = 26 ms; β = 0.083, SE = 0.024, t = 3.428, p < 0.005).

Regions 9 (First word of the Spill-over region) and 10 (Second word of the Spill-over Region)

Analysis of these regions showed no interaction between case and word order nor main effects.

Discussion

The general purpose of the current study was to shed light on the role of linking for online sentence processing. The experiment conducted involved the use of Spanish sentences that included the same verb but which accepted alternative case marking. The rationale behind the use of these verbs was to test whether keeping semantic content of the verb constant while manipulating case marking would lead to differences in overall sentence interpretation due to differences in linking type. While accusative case marking entailed direct linking, the use of dative case marking conveyed an interpretation where the highest-ranked thematic role (the ‘Experiencer’) was linked to the lowest-ranked syntactic constituent, the Object. Sentences were also manipulated regarding their word order (SVO vs OVS) in order to evaluate whether argument incremental interpretation played any role during online processing of these sentences as evidenced by previous works on this language. Results of the eyetracking reading task showed that participants regressed significantly more times to regions 6 (dissambiguating verb) and 8 (second NP) when the verb included a dative clitic than when it included an accusative clitic. They also took longer time to read region 8 and spent longer time regressing to previous regions before continuing reading when the sentences entailed an inverse linking type. As shown at Table 5, case marking affected reading at both SVO and OVS sentences and, consistently, it modulated sentences’ total reading time as well. As for incremental argument interpretation, the eye-tracking measures of the current study showed an interaction between Case and Word Order for some of the late eye-movement measures. In particular, once participants found the main verb, they regressed significantly more to previous regions and fixated for significantly longer time at the verb and second NP when the sentence was in SVO word order and included a dative clitic (sentence (a) in Table 1 than in the other three conditions. Recall that in previous studies in this language, Gattei et al. (2017; 2021) found that this difficulty took place every time the participants read a sentence where the first argument was the least prominent one, for both SVO and OVS sentences. The implications of the current results will be discussed at the General Discussion section under the light of different psycholinguistic models that address how incremental linking takes place.

Results of the comprehension task showed an effect of word order and an interaction between word order and case marking for accuracy rates, showing higher accuracy for questions about SVO sentences in general and, in the case of OVS sentences, showing better comprehension when the sentence included a dative clitic than when it included an accusative clitic. For accusative-marked sentences, better performance appeared when participants had to respond to questions about SVO sentences than for questions about OVS sentences.

Following (Gattei et al., 2017), we wondered whether results of the comprehension task could be driven by sentences’ individual acceptability. Previous studies have shown an inverse correlation between the rate of acceptability associated with a specific sentence and the cognitive cost related to its comprehension, reflected in higher reading times, higher response times and higher error rates in comprehension tasks (Ditman et al., 2007; Wagers et al., 2009). A way to disentangle this possible confound was by running an acceptability judgement task on the sentences used in Experiment 1 and re-analysing the data collected at the comprehension task with acceptability scores as a control factor in the linear models used. We present this procedure in Experiment 2.

Experiment 2: Acceptability judgement task

All stimuli used in the eye-tracking experiment were used in an acceptability judgement task in order to disentangle whether difficulty that arose from the use of accusative or dative clitics in this group of verbs in sentences with different word order was related to overall sentence acceptability.

General discussion

The general purpose of the current study was to shed light on how linking is achieved during online processing. Substantial evidence from studies run in German (Bornkessel et al., 2005; Bornkessel-Schlesewsky and Schlesewsky, 2009; Bader and Bayer, 2006; Bornkessel et al., 2003; 2002; Frisch and Schlesewsky, 2001), English (Weckerly and Kutas, 1999), Chinese (Wang et al., 2009; 2012), and Italian (Droge et al., 2014) has shown that the parser does not wait until verbal information is available in order to make predictions about the argument structure of the sentence. These predictions are based on morphosyntactic (i.e. case morphology, word order) and semantic cues (i.e. animacy, definiteness) provided by the arguments of the sentence. Cross-linguistic evidence also suggests that the grade of relevance that each cue has for triggering these predictions may vary from language to language. Spanish is a language in which linking mainly depends on word order and case morphology. Previous studies in this language have shown that participants take longer time to read the second argument of an SVO or OVS sentence (Gattei et al., 2015a) and make longer and more regressions to previous regions (Gattei et al., 2017; 2021) when the order of semantic arguments does not reflect the order established by the argument structure of the verb. Evidence in this language has also shown that arguments’ misinterpretation modulates neurophysiological correlates (Gattei et al., 2015b). Interestingly, these studies revealed a main effect of verb type when participants had to integrate the second argument of the sentence (Gattei et al., 2015a; 2017) and when they had to respond questions about ‘who did / felt what to / for whom’ (Gattei et al., 2015b; 2017), with sentences with ObjExp psych verbs being more difficult than sentences with activity verbs. The authors suggested that such difficulty was related to differences in linking, since activity verbs entailed a direct linking, and sentences with ObjExp psych verbs involved an inverse linking between syntax and semantics. Although this suggestion goes in line with evidence found in other languages (Thompson and Lee, 2009), it remained unclear from these studies whether such differences are indeed related to linking or whether they could respond to differences in verb class (Pulvermüller et al., 2001), type of subcategorization and possible thematic roles (Shetreet et al., 2007), or semantic complexity (Brennan and Pylkkanen, 2010) of the two classes of verbs used (Activity verbs vs ObjExp psych verbs). Thus, we studied the comprehension of sentences that included a class of verbs that allowed different case marking for identical lexemes (see Ferreira, 1994: and subsequent works for studies addressing the choice of active or passive voice in sentence production through the comparison of different verb types). This case manipulation leads to different syntax-to-semantics linking, as shown in Table 1. On the one hand, accusative-marked sentences like (a) and (b) result in a direct linking between syntactic constituents and their semantic roles. On the other hand, dative-marked verbs like those in sentences (c) and (d) lead to sentences with inverse linking. The hypothesis of the current experiment was that if linking type played a role on the cognitive cost of sentence comprehension, differences according to case marking should have taken place.

On this regard, results of the current study show that during incremental reading, participants did regress significantly more times to regions 6 (dissambiguating verb) and 8 (second NP) when the verb included a dative clitic than when it included an accusative clitic. They also took longer time to read region 8 and spent longer time regressing to previous regions before continuing reading when the sentences entailed an inverse linking type. These results suggest that integrating verb and arguments information gives rise to a higher processing difficulty when linking between syntax and semantics is inverse than when it is direct, replicating similar findings on the self-paced reading and eye-tracking studies with different verb classes in Spanish (Gattei et al., 2015a; 2017; 2021). In other words, the current study managed to delink the verb type effect found in previous studies from the use of different verb classes and highlights the role of linking per se.

As for the results of the comprehension task, the current experiment did not find evidence showing that case (and hence linking type) alone modulated accuracy rates or reaction times as Verb Type did in Gattei et al. (2015a; 2015b; 2017). The fact that the linking effect found during incremental reading did not persist during the offline task could be explained by one main factor: differently to the questions formulated on the three previous experiments conducted in Spanish, comprehension questions on this experiment only focused on one of both participants of the event and used a reflexive version of the verb being tested so that participants had to respond to the same question for sentences with both case markings. For instance, for sentences like those in Table 1, the question was ‘¿María se va a molestar?’ (Is María going to feel bothered?). In the previous experiments, questions included a relative clause and both participants were somehow mentioned. For example, for sentences with a verb like ‘gustar’ (to like) or ‘gritar’ (to yell) questions were formulated as ‘¿Es María quien le gusta / grita a alguien?’ (Is María who likes / yells at someone?). Thus, we suggest that linking effects in response times and accuracy rates in questions of previous experiments could have been present mainly because participants had to process linking information from both the sentences and the questions arguments, which in turn included an embedded relative clause, leading to increasing parsing difficulty (see Murujosa et al., ress: for details on processing of relative clauses with psych predicates). In the current experiment, the questions were feasible to answer by only understanding if the participant mentioned was the Experiencer or not, both in accusative and dative marked sentences, and questions did not include a relative clause. Hence, although the current results do not invalidate previous findings, they call for further investigation of task-related effects when testing both comprehension and linking effects simultaneously.

Another result that requires analysis is the effect of word order found at region 4 (Auxiliary verb ‘va’). This effect was present for Regression Path Duration, Total Fixation Time and First Pass Reading Time (although not discussed at the Results section), showing longer reading times for OVS sentences than for SVO sentences. The appearance of this effect at the auxiliary verb region could be related to the fact that the OVS word order is not compatible with the lexical intransitive verb ‘ir’ (‘to go’). It is possible that participants initially tried to integrate the lexical form of this verb without success and soon realized that it was part of a periphrastic future construction, leading to longer reading times and regression path duration.

At the same time, the use of sentences with verbs that accept alternative case marking (accusative or dative), allowed us to assess whether the effects related to incremental argument interpretation found in previous studies comparing sentences with different verb types could also be found when sentences only differed in one character denoting case assignment.

As it was briefly mentioned in the Introduction section, different psycholinguistic models have attempted to explain how case marking is processed and thus, how linking takes place. While it is not the aim of the current study to fully commit to one of them, in the following section we introduce their general assumptions and how they would account for the results of the current study.

On the one hand, the Linking and Checking Model (LCM) is a behavioral model that bases its assumptions and predictions on the Minimalist Program of the Generative Grammar (Bader and Bayer, 2006). In order to explain how the human sentence processing mechanism deals with morphological case ambiguous sentences, the authors of this model thoroughly describe a group of studies that involved word order alternation in German. The model proposes that comprehension and linking between syntactic to semantic information is mostly syntactically-driven, with morphological case marking as the most important cue for predicting what type of event the sentence will describe. The Linking and Checking algorithm that the model assumes proposes that in case of ambiguity, accusative case marking should be preferred over dative case marking. Crucially for the sentences used in the current experiment, the LCM predicts that in the case of OVS sentences, in which the first constituent of the sentence is an ambiguous noun phrase, the appearance of a dative clitic at the verb region should entail a higher processing difficulty than an accusative clitic.

On the other hand, the extended Argument Dependency Model proposes that during sentence processing the parser attempts to assign the most active semantic role possible to the first constituent. In order to do so, the model assumes a prominence computation mechanism, that uses the available syntactic and semantic features of the language and the dependencies between them (Bornkessel and Schlesewsky, 2006). While this is a neurocognitive model that bases most of its assumptions and predictions on evidence from neurophysiological and brain imaging studies, it is still relevant for the current study as “one would expect to find at least a partial correspondence between eyetracking measures and different processing steps within the eADM.” (Bornkessel and Schlesewsky, 2006: p. 811). According to this model, Spanish prominence computation would be sensitive to case marking, word order and animacy alternations. In the case of sentences like the ones analysed in the current study, the use of different clitics yields different verb argument structures and different configurations of their arguments’ prominence status. For instance, accusative-marked sentences present a more-to-less prominent order of arguments in the case of SVO sentences and a less-to-more prominent structure in OVS sentences. Sentences with dative-marked clitics present the opposite pattern: a more-to-less prominent order of arguments in OVS sentences and a less-to-more prominent order in SVO sentences. Hence, the model predicts an interaction between word order and case marking, with SVO sentences with Dative case marking being more difficult than those with accusative case marking, and with OVS sentences with accusative case marking being more difficult than those with dative case marking.

As it may be seen in Table 2, the data collected in the current experiment partially supports the predictions of both models regarding case marking. In particular, results show that there was a significant interaction between case and word order at regions 6 and 8 (second NP) in late eye-movement measures TIR, TFT and RPD. However, this effect was not symmetric for both word orders. Subject-initial sentences with a dative clitic yielded more parsing difficulty than SVO sentences with accusative case marking but no differences were found between OVS conditions.

A third psycholinguistic model that is relevant for the current study is the Competition Model, proposed by MacWhinney et al. (1984); MacWhinney and Bates (1993); Bates et al. (1982). This model grounds most of its assumptions and predictions regarding linking on evidence from behavioral comprehension studies. It assumes that sentence processing is based on a direct form-to-function mapping that results from the interaction of several linguistic cues (e.g., word order, animacy, agreement, stress). The strength of each cue is language-specific, meaning that a cue may be a strong indicator for semantic interpretation in one language but a weak one in another. Cue validity is calculated according to the combination of cue applicability (how available a cue is) and cue reliability (how unambiguous a cue is) and, when considering what the most active argument of a sentence is, the different cues will interactively compete against each other to guide comprehension. This model shares certain similarities with the eADM in that they both try to explain cross-linguistic findings by resorting to a higher-level operation (i.e. prominence computation and cue validity respectively) that uses interacting information from different linguistic levels, and the outcome of overall sentence interpretation may be modeled according to the relevance of each cue for a specific language. While this was not the focus of the current study, accuracy results of the comprehension task showed that both position of the noun and case marking are relevant cues for Spanish sentence interpretation, but that participants do not rely on them similarly for all sentences. When trying to identify if the participant mentioned in the question was the most active argument of the sentence previously read (i.e. the Agent for accusative-marked sentences or the Experiencer in dative-marked ones), the results showed an interaction between word order and case which depended on case marking for OVS sentences and on word order for accusative-marked ones. This pattern suggests that when facing OVS sentences, participants relied more on a cue like the position of the noun, prioritizing preverbal nouns than case-marking of the object and yielding worse results for accusative than for dative-marked sentences. No differences were found between SVO conditions, showing that for these conditions, participants also relied on case marking of the object for interpretation. These hypotheses are compatible with the results of the eyetracking measures. Recall that during incremental reading the opposite pattern was found: higher difficulties to parse SVO sentences when they included a dative clitic than when they included an accusative one, and no differences between OVS sentences. It is possible that in subject-initial sentences, a higher effort was put on correcting mispredictions about case assignment -as well as processing word order- showing similar difficulty for both SVO sentences, and thus no differences between conditions. Conversely, it is possible that in the case of OVS sentences participants relied more on word order and engaged less in the clitic’s case marking, showing a higher rate of misinterpretation for those sentences in which OVS sentences is not the most frequent one (accusative-marked ones). On the other hand, interpretation was different for accusative than for dative-marked sentences. Participants showed worse performance at interpreting sentences with distinct word order in accusative-marked sentences but not in dative-marked ones, indicating that a noun being at preverbal position may have been a stronger cue for accusative-marked sentences than for their dative-counterparts. This pattern also complements the results of the eyetracking measures: during reading, participants regressed to the preverbal noun and main verb significantly more and spent more time reading these regions when the sentence included a dative than an accusative clitic. Once again, the longer time spent reading dative-marked sentences could have resulted in more accurate syntax-to-semantic linking and better performance, and differences regarding word order for accusative-marked sentences could have raised due to less reliance on noun position for these sentences. It is matter of future research to characterize whether this pattern of results is consistent with Spanish corpora analysis of cue reliability and applicability.

Conclusions

The current work aimed at providing further evidence on how linking is achieved during real-time comprehension when the same verbs class is used. It went a step further from other studies on the topic by using a special class of psych verbs in Spanish, which can bear both accusative or dative case marking, leading to verbs’ argument structures that require either direct or inverse syntax-to-semantics linking respectively. Together with the manipulation of word order, the sentences used in this experiment could disentangle whether processing inverse linking was more costly than direct linking. All in all, results showed that during incremental comprehension, inverse linking is more difficult than direct linking, irrespective of word order. Although the linking effect was not present at the comprehension task, it did take place at the sentence acceptability judgement task, with participants penalizing those sentences with indirect linking more than sentences with direct one. This dissociation of results, as well as the differences between the types of questions formulated for the current experiment and those from previous studies in the same language suggest that the type of questions used for the comprehension task (probably easier in terms of linking) could have led to a dissipation of the linking effect. Furthermore, results were briefly discussed under the light of different psycholinguistic models that address case processing and linking, providing a broader view on how the processing of this subclass of psych verbs could be informative for their own theoretical assumptions and experimental predictions.

Abbreviations

ACC = Accusative

Act = Activity

AGX = Agreement Index

DAT = Dative

eADM = extended Argument Dependency Model

fem = feminine

FFD = First Fixation Duration

FPRT = First Pass Reading Time

Infin = Infinitive

LCM = Linking and Checking Model

M = Mean

NOM = Nominative

ObjExp = Object Experiencer Psych Verbs

OVS = Object – Verb – Subject

pl = plural

REFL = Reflexive

RBRC = Right Bounded Regression Count

RPD = Regression Path Duration

SE = Standard Error

sg = Singular

SVO = Subject – Verb – Object

TFT = Total Fixation Time

TIR = Total Incoming Regressions

Data availability

The raw data, analyses’ scripts and supplementary files supporting the conclusions of this article are available at the OSF repository.

Ethics and consent

The experiments described in this work were reviewed and approved by the Ethics Board of Facultad de Filosofía y letras, Universidad de Buenos Aires, protocol # 16. All participants provided their written informed consent to take part in this study.

Funding information

This work was supported by the National Scientific and Technical Research Council (CONICET – Argentina) as part of the first author postdoctoral research project and by the James S. McDonnell Foundation Grant Number 220020334.

Authors contributions

CAG, LP, AJW and DS conceptualized the experiment and decided on the methodology, contributed on writing and on data curation and analysis. FA took the lead on participants’ recruitment and data collection. YS contributed with the revision of materials, writing and editing. Analyses’ code was written by CAG and DS.

Competing interests

The authors have no competing interests to declare.