Agents Strongly Preferred: ERP Evidence from Natives and Non-Natives Processing Intransitive Sentences in Spanish

Are non-native speakers able to process their second language in a native-like way? The present study used the Event-Related Potentials’ (ERPs) method to address this issue by focusing (1) on agent vs. agentless intransitive sentences and (2) on person vs. number agreement morphology. For that purpose, native and high proficiency and early non-native speakers of Spanish were tested while processing intransitive sentences containing grammatical and ungrammatical subject–verb agreement. Results reveal greater accuracy in the agent (unergative) condition as compared with the agentless (unaccusative) condition and different ERP patterns for both types of verbs in all participants, suggesting a larger processing cost for the agentless sentences than for the agentive ones. These effects were more pronounced in the native group as compared with the non-native one in the early time window (300–500 ms). Differences between person and number agreement processing were also found at both behavioral and electrophysiological levels, indicating that those morphological features are distinctively processed. Importantly, this pattern of results held for both native and non-native speakers, thus suggesting that native-like competence is attainable given early Age of Acquisition (AoA), frequent use and high proficiency.


Introduction
The growing scientific interest in bilingualism during the last decades has raised many questions (though not always answers) about language learning and processing; in particular, it has generated new evidence about how the two languages of the bilingual interact and are processed and controlled. This body of evidence strongly suggests that learning a second language is "harder" than learning a first language even when exposure to the second language is early and frequent [1][2][3][4][5] so that native competence might be difficult to achieve in a second language. Thus, it looks as if some aspects of the second nonnative grammar can be learned and processed as natives do, while others cannot (see [6], for an overview). Here, we focus on the Language Distance Hypothesis [7]. This hypothesis claims that early and proficient bilinguals can achieve native competence for grammatical properties shared by their two languages, whereas unshared grammatical properties pose a challenge for native-like syntactic processing. We present a novel behavioral and Event-Related Potential (ERP) study where early and proficient Basque-Spanish bilinguals behave native-like in their second language, Spanish, when processing (a) argument structure alternations in intransitive sentences involving agent vs. patient subjects and (b) subjectverb agreement, both of which are grammatical properties shared by the two languages of these bilinguals.
ERPs are a very reliable tool to determine the time course of language-related brain processes occurring at the electrophysiological level (see [8], for an overview). In general, three main types of language-related ERP components have been reported in the literature: (see also [18,19], for an overview of the literature on agent-first preference). This predilection for interpreting the initial argument as an agent has also been found among young children. More precisely, Abbot-Smith and collaborators [20] investigated, in an eye-tracking study, whether two-year-old (group 1) and three-year-old (group 2) children use a first-NP-asagent bias to process active transitive and passive sentences. The results showed that both age groups showed a tendency to map the first noun in a sentence onto an agent role. The assumption that agent-first arguments are preferred over patient-first arguments has been explicitly formulated within the extended argument dependency model (eADM) [17]. This proposal assumes two main grammatical roles, namely the actor and the undergoer, which correspond to the agent and patient prototypes, respectively. According to the authors, the assignment of a specific role depends on the particular language being processed: while in some languages (e.g., English) argument position is a reliable cue, in others the relevant information is provided by morphology (e.g., German, Japanese). The predictions made by the eADM have been largely corroborated by experimental data. For instance, ERP evidence from German [11] has revealed that processing inanimate agents is costlier than processing animate ones, that is, in the sentence "Paul asks himself [which teacher] ACC [the twig] NOM brushed" processing the twig is costlier in comparison with an animate NP in the same position, and this cost is reflected by an N400 component. This effect is presumably due to the increased processing cost associated with having to assign an actor role to an inanimate argument. In another ERP study carried out in Turkish, Demiral, Schlesewsky and Bornkessel-Schlesewsky [21] showed an increased processing difficulty (P600) when an initial ambiguous argument was disambiguated towards an object (=patient/theme) reading, as compared with agent subjects. In a similar vein, Bickel et al. [22] used ERPs to investigate agent vs. patient preference in Hindi and showed a clear agent-first preference when processing the initial arguments, that is, disambiguation of an initial NP towards a patient reading engendered a biphasic N400-P600 pattern. In sum, experimental evidence presented so far suggests that subjects that bear a patient relation to the verb (as in the child fell) will incur larger processing costs than subjects that bear an agent relation to the verb (as in the child laughed).
Regarding the second aspect to be examined in the present study, namely, the phifeatures person and number, only a few experimental works have investigated and compared their processing by means of ERPs. Silva-Pereyra and Carreiras [23] compared person, number and person + number violations in Spanish, and a similar Anterior Negativity followed by a P600 emerged for person and number. On the other hand, Mancini et al. [24] reported an N400 + P600 pattern for person violations, and an LAN + P600 pattern for number violations in Spanish. According to the authors an N400 effect is expected for violations that have an impact at the interface with the semantic-discourse representation of the sentence (person violations); while a LAN effect is expected for violations limited within the boundary of the morphosyntactic representation (number violations). Person and number features have also been tested in other languages, and this LAN/N400 difference has not been reported elsewhere. Zawiszewski et al. [25] compared person, number and person + number violations in Basque, and found an N400 followed by a P600 for all types of violations; however, person violations and combined person + number violations yielded a larger P600 than number violations, suggesting that person is more salient than number in terms of phi-features (see [26], for a more detailed description of person and number feature processing).

The Present Study
We examine these two dimensions of subjecthood in non-native language processing: (a) agent preference and (b) person/number agreement by means of ERPs in order to compare how native speakers of Spanish and native speakers of Basque who were highly proficient speakers of Spanish with an early AoA process (a) agent vs. patient intransitive subjects and (b) person vs. number agreement features in Spanish. The purpose of this comparison is to determine whether these early and proficient non-natives are native-like in their processing profiles. Spanish is a Subject Verb Object (SVO), nominative-accusative language where subjects carry nominative case, regardless of their semantic role; thus, besides prototypical agent subjects such as la niña agarra el juguete (the child grabs the toy), there are also patient subjects like la niña nació ayer (the child was born yesterday). On the other hand, Basque is a Subject Object Verb (SOV), ergative language where agent subjects bear ergative case (e.g., umeak jostailua hartzen du 'the child grabs the toy') and patient subjects are morphologically unmarked (e.g., umea atzo jaio zen 'the child was born yesterday') [27]. Since Spanish and Basque mark agent and patient subjects differently, our aim was to test the potential impact of these typological differences on SV agreement processing in native and non-native speakers. In other words, in Spanish all intransitive subjects are morphologically unmarked (nominative case) while in Basque intransitive agents carry an ergative marker (-k) and intransitive patients are unmarked (absolutive case). Previous studies on ergative morphology and non-native language processing (e.g., [5,7]) have revealed that typological differences between the L1 and L2 are an aspect that non-native speakers are particularly sensitive to; therefore, investigating SV agreement processing in native and non-native speakers of Spanish will allow us to look into this issue in more detail.

Hypotheses
In this study we use grammatical and ungrammatical subject-verb agreement dependencies in order to test the agent preference and the Feature Distinctness Hypothesis (FDH) ( [28]; see also [24]) in native and non-native speakers of Spanish. Given that in both Spanish and Basque (the native language of the non-native group) the verb agrees overtly with the subject, and, given the high proficiency and low AoA of the L2 speakers, according to the Language Distance Hypothesis [7], similar behavioral and electrophysiological responses to agreement violations are expected for all participants. Since the agent-first hypothesis has not been previously tested in intransitive sentences and nominative-accusative languages by means of ERPs, and based on previous evidence and literature regarding SV agreement (see [29], for an overview), we tentatively predict that SV violations will elicit a negative component (LAN or N400) followed by a P600. According to the agent-first hypothesis, a larger processing cost should be observed for sentences containing patient subject predicates than for those containing agent subject verbs. Similarly, according to previous work on phi-feature processing [24,25] differences between person and number feature processing should emerge as well, namely, person violations should elicit a larger P600 in comparison to the number violations. In addition, we might also find a negative component as a response to these violations (LAN/N400).

Participants
Twenty-seven native (10 men; mean age 20.5 years, SD = 2.9) and 25 non-native (5 men; mean age 21.8 years, SD = 2.9; AoA = 5.7 years, SD = 1.9) speakers of Spanish took part in the experiment. According to the Edinburgh Handedness inventory [30], all participants were right-handed and they were all paid for their participation (see Table 1 for details). Data from 3 native and 1 non-native participants were removed due to the insufficient number of segments available for statistical analyses. The study was approved by the Ethics Committee of the University of the Basque Country (UPV/EHU)) (M10_2020_182). Table 1. The following seven-point scale was applied for measuring the relative use of language: 1 = I speak only Spanish, 2 = I speak mostly Spanish, 3 = I speak Spanish 75% of the time, 4 = I speak Basque and Spanish with similar frequency, 5 = I speak Basque 75% of the time, 6 = I speak mostly Basque, 7 = only Basque. Proficiency level was determined by using the following seven-point scale: 7 = native-like proficiency, 6 = high proficiency, 5 = full proficiency, 4 = working proficiency, 3 = limited proficiency, 2 = low proficiency, 1 = very low proficiency. SDs values are in parentheses.

Materials
Two hundred fifty-six experimental sentences and 160 fillers (416 in total) were created and distributed over 4 counterbalanced lists. The materials were organized according to a 2 × 2 × 2 design: subject type (agent vs. patient), feature (person and number), and grammaticality (grammatical and ungrammatical) (see Table 2). The design of the materials was motivated by the assumption that 1st and 2nd person singular forms have a specification for the person feature, but not for number, while 3rd person is specified for number, but not for person [31,32]. Consequently, we followed the design used in Mancini et al. [24] and Martinez de la Hidalga et al. [28,33]. More precisely, for person conditions 2nd person was used in the grammatical condition and 1st person was used in the ungrammatical manipulation. For number conditions, 3rd singular vs. 3rd plural manipulations were used. A distance between subjects and verbs was created by adding three words and an average of 2.6 words (1, 2, or 3 words) were added after the critical word, controlled per condition. The materials were controlled for length and frequency.

Procedure
Personal computers (Windows 7 operating system) and Presentation software (version 16.3) were used to present the stimuli on the screen. Before the actual experiment began, participants were instructed about the EEG procedure and seated comfortably in a quiet room in front of a 24 in. monitor. The experiment was conducted in a silent room in the Experimental Linguistics Laboratory at the University of the Basque Country (UPV/EHU) in Vitoria-Gasteiz. Sentences were displayed in the middle of the screen word by word for 350 ms (ISI = 250). A fixation cross (+) indicated the beginning of each sentence trial. After each trial the words correcto? "correct?" or incorrecto? "incorrect?" appeared in the screen, and participants were asked to judge the acceptability of the previously displayed sentence as either correct (left Ctrl) or incorrect (right Intro). Half of the participants used the left hand for correct responses and the other half the right hand. All 416 sentences were distributed randomly in four blocks that lasted approximately 10 min each. Participants had a short break between each block which lasted as long as they needed. Before the actual experiment, participants ran a short training session of three trials. They were asked to avoid blinking or moving when the sentences were being displayed and to make the acceptability judgment as fast and accurate as possible. The whole experiment, including electrode-cap application and removal, lasted about 1 h 15 m.

EEG Recording
The EEG was recorded from 32 active electrodes secured in an elastic cap (Acticap System, Brain Products). Electrodes were placed on standard positions according to the extended Internationals 10-20 system in the following sites: Fp1/Fp2, Fz, F3/F4, F7/F8, FC5/FC6, FC1/FC2, T7/T8, C3/C4, Cz, CP5/CP6, CP1/CP2, P7/P8, P3/P4, Pz, O1/02, Oz, LM, VEOG and HEOG. All recordings were referenced to right mastoid position and re-referenced off-line to the linked mastoids. Vertical and horizontal eye movements and blinks were monitored by means of two electrodes positioned beneath and to the right of the right eye. Electrode impedance was kept below 5 kOhm for all scalp electrodes and below 10 kOhm for the eye electrodes. The electrical signals were digitized online at a rate of 500 Hz by a Brain Vision amplifier system and filtered offline within a band pass of 0.1-35 Hz. After the EEG data were recorded, the ocular correction procedure [34] as well as the artifact rejection procedure were applied (offline). Trials with other artifacts were removed when they indicated any voltage exceeding 150 µV and voltage steps between two sampling points exceeding 35 µV.

Data Analysis
For the data analysis, the following types of subject agreement structures were compared: agent subject with the grammatical and ungrammatical person (actuarás 'play.2SG.FUT.' vs. *actuaré 'play.1SG.FUT.'; conditions 1 vs. 2 in Table 2); agent subject with the grammatical and ungrammatical number (actuará 'play.3SG.FUT.' vs. *actuarán 'play.3PL.FUT'; conditions 3 vs. 4 in Table 2, respectively); patient subject with the grammatical and ungrammatical person (vendrás 'visit.2SG.FUT.' vs. *vendré 'visit.1SG.FUT.'; conditions 5 vs. 6 in Table 2); patient subject with the grammatical and ungrammatical number (vendrá 'visit.3SG.FUT.' vs. *vendrán 'visit.3PL.FUT.'; conditions 7 vs. 8 in Table 2, respectively). For the ERP measures, segments were created starting at 200 ms before and ending 1000 ms after the onset of the critical words (the verb) in the sentences. The trials associated with each subject type were averaged for each participant. The EEG 200 ms prior to the onset was also used as a baseline for all sentence type comparisons. After visualizing the data and based on the literature, 300-500 ms and 600-900 ms temporal windows were considered during statistical analysis in all conditions. After the stimuli were recorded and averaged, analyses of variance (ANOVA) were carried out in nine regions of interest that were computed out of 27 electrodes: lateral electrodes: left frontal (F7, F3, FC5), left central (T7, FP5, C3), left parietal (P7, P3, O1), right frontal (F4, F8, FC6), right central (C4, FP6, T8), and right parietal (P8, P4, O2); midline electrodes: frontal (Fp1, Fz, Fp2), central (FC1, Cz, FC2), and parietal (CP1, Pz, CP2). Repeated-measures ANOVAs were performed in all experimental manipulations and trials (correctly and incorrectly judged trials) for each window of time using five within-subjects factors: grammaticality (2 levels: grammatical, ungrammatical), subject type (2 levels: agent, patient), agreement feature (2 levels: person, number), hemisphere (2 levels: left, right), and region (3 levels: frontal, central and parietal). Midline (frontal, central, and parietal) electrodes were analyzed independently. Whenever the sphericity of variance was violated, a correction [35] was applied to all the data with greater than one degree of freedom in the numerator. Finally, further statistical comparisons were conducted (split by the grammaticality condition) whenever an interaction turned out to be statistically significant. Effects for subject type, agreement feature, hemisphere or region factors are only reported here when they interacted with the experimental manipulation of grammaticality.
For the behavioral results, error rates and response latencies of all the trials were submitted to repeated measures ANOVAs with grammaticality (two levels: grammatical, ungrammatical), subject type (two levels: agent, patient) and feature (two levels: person, number) conditions as within-subject factors. Subsequent comparisons (by subject and by item) were carried out whenever a grammatical interaction was significant.

Electrophysiological Results
Regarding the early time window (300-500 ms) (see Figure 1), the analysis of lateral electrodes revealed a marginally significant main GROUP effect (F(1,46) = 3.8, p = 0.057) indicating a larger negativity for natives than non-natives (−0.67 µV vs. 0.27 µV) (see Table 5 for details). Besides, a main FEATURE effect emerged as well (F(1,46) = 6.53, p = 0.014), indicating a larger negativity for number feature than for person feature manipulations   A marginally significant FEATURE*GRAMMATICALITY interaction emerged as well (F(1,46) = 3.28, p = 0.077). The subsequent analysis by grammaticality factor showed no differences between person feature violations and the corresponding grammatical sentences, and similarly, no differences between number feature violations and the corresponding grammatical sentences were found. The analysis by feature factor revealed no differences between grammatical verbs containing person and number manipulations, but number violations elicited a larger negativity than person violations (−0.53 µV vs. 0.15 µV) (F(1,46) = 6.9, p = 0.012). A GRAM*REGION*GROUP interaction (F(2,92) = 9.11, p < 0.001) revealed no differences between natives and non-natives driven by grammaticality or group factors.

Summary of the Results
Regarding accuracy, non-native speakers were less accurate than natives, and overall, participants were more accurate with sentences containing agents than with sentences containing patients as subjects. They were also more accurate with sentences containing number feature manipulations than with sentences containing person feature manipulations. As regards reaction times, non-natives were generally slower than natives, and participants reacted faster to number feature violations than to person feature violations. With respect to the differences between the agent subject and patient subject predicates, electrophysiological results in the early time-window (300-500 ms) revealed a frontal negative component as a response to patient subject violations in the L1 group and no effect in the L2 speakers. No negativity was elicited by agent subject predicate violations in either group. In addition, natives displayed larger negativity for grammatical agent subject predicates than for grammatical patient subject predicates, while no such effect was observed in the non-native group. Regarding feature processing, number feature violations yielded a left-lateralized negativity among all speakers and this negativity was significantly larger than that elicited by person violations. With regard to the late time-window (600-900 ms), in comparison with the grammatical sentences, all ungrammatical sentences yielded a positive component (P600). Additionally, grammatical patient subject verbs generated a slightly larger positivity than grammatical agent subject verbs, and agent subject verb violations elicited a larger positivity than patient subject verb violations. Regarding phi-features, both person and number violations produced a larger positivity than their grammatical versions. Furthermore, grammatical sentences involving person manipulations elicited more positive responses than grammatical sentences containing the number feature over central and posterior electrodes. Table 5. Statistical results (EEG data) Notes: GRAM: grammaticality (two levels); TYPE: predicate type (two levels); FEAT: feature type (two levels); HEM: Hemisphere (two levels); REG: Anterior-Posterior factor (3 levels); df: degrees of freedom. a p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001.

Discussion
In the present study we investigated how native and early and highly proficient non-native speakers of Spanish process intransitive predicates and phi-features in order to determine to what extent non-native speakers can do this in a native-like way. In the subsequent sections we will first discuss the similarities and differences found between native and non-native speakers and the implications for the hypotheses considered previously. Next, we will interpret our data in light of the agent-first hypothesis and Feature Distinctness Hypothesis.
Regarding the first aspect, in general, similar effects and interactions were found for non-native speakers as compared with native speakers of Spanish, that is, differences between the processing of agent subject and patient subject predicates were observed, and differences between person and number phi-features emerged for both groups. More specifically, all participants were more accurate when judging grammatical phi-features than when reading sentences containing ungrammatical phi-features. Similarly, they reacted faster to number violations than to their grammatical counterparts. L2 speakers were overall slower and less accurate than L1 speakers when judging the grammaticality of the sentences. Since both natives and non-natives stated native-like or very high proficiency level of Spanish, the differences can be hardly accounted for by a lower competence of the non-native speakers as compared with natives. In any case, despite some behavioral differences, similar effects and interactions between predicate type and phi-features were observed in all participants at the electrophysiological level. More precisely, all speakers displayed left-lateralized negativity as a response to number violations, while no differences between ungrammatical and grammatical person features were observed at this stage of processing in either group. This indicates that for all speakers it was easier to detect number violations than person violations. In a similar vein, agent subject verb and patient subject verb agreement violations yielded larger positivity than their grammatical agreement versions in the late time window (600-900 ms). Likewise, grammatical patient subject verbs elicited a slightly larger positivity than grammatical agent subject verbs, whereas a larger positivity was obtained for agent subject verb violations than for patient subject verb violations in all participants. This suggests that patient subject predicates were costlier to process than agent subject predicates for both native and non-native speakers.
Regarding the electrophysiological differences between L1 and L2 speakers, in the 300-500 ms time window natives showed higher sensitivity towards patient subject vs. agent subject predicate distinction than non-natives, that is, they displayed larger negativity for patient subject agreement violations than for grammatical patient subject sentences, while no such effect was observed for agent subject agreement conditions. Conversely, non-native speakers showed no negativity as a response to patient subject or agent subject predicate violations at this stage of processing. In addition, natives also showed larger negativity for grammatical agent subject agreement than for grammatical patient subject agreement. Conversely, non-natives processed both types of predicates alike. The fact that they showed smaller effects than natives (or no negativity at all) when presented with SV agreement violations is not new. This type of response has also been observed for L2 speakers of Basque [33], with the authors suggesting (after [36]) that smaller or absent effects may be due to a reduced degree of automaticity in the activation of processing resources. A potential (although speculative) explanation for this pattern of results may be the influence of case morphology on the way SV agreement is built and processed. More precisely, Spanish is a nominative-accusative language where agent and theme/patient subjects are morphologically identical (unmarked), while Basque is an ergative language where agent subjects are marked and patient subjects are not. Previous findings [5,7] indicate that case alignment is an aspect that non-native speakers are particularly sensitive to and that, even at low AoA and high proficiency, they do not process it similarly to natives if it diverges from the L1. Given that the L1 of the non-native speakers tested in the present study is Basque, those participants may rely to a larger extent on case morphology than native speakers of Spanish do when processing verb agreement, and consequently, they show no negativity to either agent subject or patient subject agreement violations because they can no longer rely on the morphological cues, present in their L1 (Basque) but absent in the L2 (Spanish). This explanation is in line with the Language Distance Hypothesis (LDH) [7], which claims that differences between native and non-native speakers will emerge if non-shared grammatical phenomena are tested [1,2,5,7].
In relation to the agent-first hypothesis, our results show that both native and nonnative speakers of Spanish process agent subject sentences with greater ease than patient subject sentences, as indicated by higher accuracy scores for sentences containing agents as subjects than for sentences containing patients as subjects. These findings are also supported by electrophysiological data: the positivity (P600) elicited by grammatical patient subject verbs was larger than that caused by grammatical agent subject, suggesting that the former are costlier to process than the latter. On the contrary, agent subject verb violations yielded larger positivity than patient subject verb violations suggesting that those structures are costlier to repair than those involving ungrammatical patient subject verb agreement. Our findings are also validated by previous experimental studies carried out in other (nominative-accusative) languages, where differences between agent subject and patient subject predicates were reported and larger processing costs for patient subjects than for agent subjects were found as well. Meltzer-Asscher et al. [37], for instance, examined native speakers of English by means of the fMRI method while processing transitive and intransitive agent subject verbs and intransitive patient subject verbs. The results revealed longer response times and larger activation of the left Inferior Frontal Gyrus (IFG) for patient subject verbs as compared with both transitive and intransitive agent subject verbs. Similarly, in a self-paced noncumulative moving-window experiment Dekydtspotter and Seo [38] tested L1 and L2 speakers of English while reading sentences containing agent subject and patient subject predicates and reported significantly greater loads after patient subject verbs than after agent subject verbs, indicating larger processing cost for the former than for the latter.
Despite different methodologies used in all these studies, our results are consistent with the main findings: intransitive agent subject verbs and patient subject verbs are processed differently, and in nominative-accusative languages patient subject verbs are costlier to process than agent subject ones. Although these differences between both types of predicates have been interpreted in terms of a syntactic movement [37], we believe that they can be also attributed to the general preference for interpreting the first unmarked argument of the sentence as an agent rather than a theme [18,19] (see also [8,17]). In other words, in nominative-accusative languages, in which all subjects are morphologically indistinguishable, a larger processing cost for patient subject verbs (unaccusatives) than for agent subject verbs (unergatives) would stem from a misinterpretation of the first (animate, human) argument as an agent instead of a theme and this would force the parser to reanalyze and revise the (wrongly) predicted structure at the verb position, leading to longer response times and an overall larger processing cost for patient subject sentences than for agent subject sentences.
In regard to the second aspect tested in the present study, namely, phi-feature processing, both L1 and L2 speakers showed faster reaction times for sentences containing number feature manipulations than for those containing person feature manipulations, and larger negativity for the number violations than for the grammatical number. Larger negativity was also observed for number violations than for person violations. Previous studies comparing person and number features in Spanish found some kind of negative component for person and number violations [23,24]. Person violations examined in the present experiment yielded no negativity at all, and though this outcome should be interpreted with caution (null effect), we tentatively attribute it to the type of materials used in our study. More precisely, sentences where the person feature was manipulated contained the 2nd person subjects, which participants may have initially considered vocatives, not an option for sentences containing number feature manipulations, which had 3rd person subjects. In that case, sentences containing person feature manipulations could initially leave more room for interpretation and even lead to a potential garden path, that is, sentences such as "Tú, ( . . . )" 'You, ( . . . )' could be also understood as "Hey, you, ( . . . )" and disambiguated at the verb position. As a consequence, person manipulations would be more difficult to process than number manipulations. Indeed, in the later stage of processing (600-900 ms) a larger positivity was observed for sentences containing a grammatical person than for those containing a grammatical number, thus suggesting that our interpretation may be on the right track. On the other hand, the negativities obtained for number violations are similar to the those found in previous studies in Spanish [23,24]. In sum, these results support the first claim of the Feature Distinctness Hypothesis, indicating that person and number features are processed differently, but do not support the idea that the person is more salient than the number, since the negativity was only observed for number violations, and similar P600 was found for both person and number violations.
All in all, our data show that non-native speakers of Spanish process agent subject and patients subject predicates as well as person and number features differently. In that sense, non-native speakers are similar to native speakers, who also revealed agent subject vs. patient subject and person vs. number processing differences, indicating that native-like competence can be attained whenever non-divergent grammatical properties are at play.

Conclusions
In this study, we examined the processing of intransitive predicates and phi-features by native and non-native speakers of Spanish. Our results lend support for the agentfirst hypotheses, as indicated by a larger processing cost for the patient subject sentences than for agent subject costs in both L1 and L2 speakers. In other words, our data can be accounted for by an initial assumption that all sentence-initial non-marked animate arguments are agents. Consequently, whenever that expectation is not met, as in the case of patient subject sentences in comparison to agent subject ones, a larger processing cost will be observed. Likewise, we provided additional evidence on phi-feature processing and showed that person and number features are not processed similarly. Lastly, this study further supports the idea that at high levels of proficiency and an early AoA, native-like processing is attainable as long as the properties are shared between the native and the non-native language.
Author Contributions: A.Z.: conceptualization, experimental design, assistance with data analysis, and writing. G.M.d.l.H.: data acquisition, analysis and writing (original draft). I.L.: conceptualization, experimental design and writing. All authors contributed to the article. All authors have read and agreed to the published version of the manuscript.

Institutional Review Board Statement:
The studies involving human participants were reviewed and approved by the Ethics Committee for Research involving human beings at the University of the Basque Country (UPV/EHU) (M10_2020_182). The participants provided their written informed consent to participate in this study.