I hates Mondays: ERP effects of emotion on person agreement

ABSTRACT Recent evidence indicates that emotion influences the computation of agreement dependencies based on number or gender. In this event-related potential study, we examined the role of emotion in the processing of person information. Participants made grammatical judgements to sentences with positive, negative and neutral verbs that either matched or mismatched person features of a preceding pronominal subject. Emotion did not modulate P600 amplitude enhancements to agreement violations. Importantly, whereas enhanced LAN effects to all ungrammatical sentences were observed in a cluster of left fronto-central electrodes, only neutral verbs that violated person agreement elicited enhanced LAN amplitudes in a sub-cluster of left frontal electrodes. The narrow distribution of LAN effects to emotion verbs suggests that feature-checking operations dealing with the early detection of person agreement errors are less efficient when words signal affective biologically salient content. Our results favour lexicalist views arguing that lexical and conceptual information influences agreement.


Introduction
Neurobiological and behavioural studies examining the interplay between language and emotion have shown that emotion has an impact on language at several levels (see Herbert, 2020;Hinojosa et al., 2020aHinojosa et al., , 2020b;;Kissler, 2020;van Berkum, 2020, for reviews), including vocabulary acquisition (Grosse et al., 2021;Nook et al., 2017), lexico-semantic processing in the native language (Kissler & Herbert, 2013;Méndez-Bértolo et al., 2011;Palazova et al., 2011) and in non-native languages (Chen et al., 2015;Opitz & Degner, 2012), or linguistic communicative contexts (Aguado et al., 2019;Schindler et al., 2019;Schindler & Kissler, 2016).Recently, emotional modulations during morpho-syntactic processes have been also observed.In this line, there is evidence pointing to a role of emotion in the processing of agreement dependencies between sentence constituents based on gender (Hinojosa, Albert, Fernández-Folgueiras, et al., 2014) or number (Martín-Loeches et al., 2012) features.In this work, we will extend prior research by examining the relationship between emotion and the processing of person agreement, a question that remains unexplored to the best of our knowledge.
Across languages, agreement is a pervasive phenomenon to codify structural links between two words in a sentence (e.g.determiners and nouns, nouns and verbs, …) by means of the variation of morphological categories that indicate person, number or gender features of verbs, nouns or adjectives (Acuña-Fariña, 2009;Corbett, 2006;Mancini et al., 2011a;Wechsler, 2009).Establishing agreement dependencies among sentence constituents involves displacing information about these features from the controller (e.g. a subject) to the target (e.g. a verb) (Mancini et al., 2013).Prior event-related potentials (ERPs) studies have observed enhanced amplitudes in two waves when agreement dependencies between sentence constituents are violated relative to correct sentences in grammaticality judgement tasks.An early left anterior negativity (LAN) emerges between 300 and 500 ms at left frontal electrodes, which is thought to index the early identification of agreement errors (Gunter et al., 2000;Molinaro et al., 2011Molinaro et al., , 2015)).Thereafter, reanalysis and repair operations that reflect efforts to integrate agreement mismatches into prior sentence contexts are indexed by a late posterior positivity (P600) that occurs approximately at 500 ms at centro-parietal scalp locations (Kuperberg, 2007;Osterhout & Holcomb, 1992).Of note, the amplitude of the P600 is sensitive not only to agreement errors but also to certain semantic-thematic aspects such as the violation of verb-argument relationships (Kim & Osterhout, 2005), or animacy constraints between nouns and verbs (Nieuwland & Van Berkum, 2005).Several ERP studies have reported processing dissociations between number, gender and/or person features.These differences mainly consisted of enhanced P600 amplitudes for person relative to gender agreement violations, or gender relative to number agreement mismatches, as well as different LAN and P600 topographies for person vs number agreement errors (Alemán Bañón & Rothman, 2016;Barber & Carreiras, 2005;Mancini et al., 2011a;Nevins et al., 2007;Zawiszewski et al., 2016;but see Alemán Bañón et al., 2012;and Silva-Pereyra & Carreiras, 2007).Dissociations are thought to reflect a hierarchical typology in which person is at the top, followed by number and gender (Person > Number > Gender; Carminati, 2005;Greenberg, 1963).This privileged status comes from the fact that person occurs independently from gender and number in most languages (Carminati, 2005), and has greater cognitive salience (Harley & Ritter, 2002;Nevins et al., 2007).Also, it has been suggested that dissociations between some of these features may arise from differences in the anchoring requirements that are needed for their interpretation.Under this view, interpretation of number requires that the parser checks the specification of this feature against the nominal argument, whereas the interpretation of person relies on the inspection of the association between arguments and speech participants' roles (Mancini et al., 2011a(Mancini et al., , 2014;;Sigurdsson, 2004).
Of note, following a two dimensional conception of emotions (Russell, 2003) in which the affective space is represented by the dimensions of valence (from negative to positive) and arousal (from calmed to activated), several studies in Spanish have examined the interplay between emotion and agreement.Despite of the lack of effects in the P600 component, interactions have been observed in the LAN wave.In this sense, Martín-Loeches et al. (2012) reported that emotional features elicited enhanced LAN amplitudes following number agreement violations between nouns and negative adjectives (e.g.La chica sg feas pl baila/The ugly pl girl sg dances), which were not observed for neutral or positive adjectives.The authors concluded that the processing of agreement relations had higher costs in negative words.Subsequent studies have relied in the analyses of gender features.LAN modulations by emotion have been found for gender agreement errors in neutral (e.g.El camarero m rubia f /The blonde f waiter m ) relative to negative target adjectives following controller nouns in noun phrases (Hinojosa, Albert, Fernández-Folgueiras, et al., 2014).Similarly, LAN effects were observed only for neutral masked adjectivesbut not for positive and negative masked adjectivesembedded within a sentence when preceding an unmasked neutral adjective containing a gender agreement mismatch with a noun (Jiménez-Ortega et al., 2017).This lack of LAN effects to gender agreement mismatches in emotional compared to neutral words have been related to a facilitated processing of agreement anomalies in sentence elements conveying biologically salient meaning (Hinojosa, Albert, Fernández-Folgueiras, et al., 2014).In contrast, Fraga et al. (2021) found enhanced LAN amplitudes to negative adjectives that violated gender agreement relations with preceding nouns, a result that was thought to reflect the difficulty of processing gender errors in negative words.Another set of studies failed to report emotional modulations on gender agreement.In this sense, Díaz-Lago et al. (2015) found similar LAN effects to positive and neutral adjectives that violated gender agreement relations with their preceding controller nouns.In the same vein, subsequent studies examining gender agreement dependencies between controller nouns and target adjectives did not observe differences in the LAN amplitudes elicited by neutral and negative adjectives (Fraga et al., 2017, Exp. 1), by positive, neutral and negative adjectives (Fraga et al., 2017, Exp. 2), or by negative and positive adjectives (Padrón et al., 2020).Overall, it seems that emotional effects on agreement processing are likely, although evidence is still inconclusive.Divergent findings might be attributed to differences in the structural distance (local or distal position) between the agreement controller and the target (Alemán Bañón et al., 2012;Biondo et al., 2018) or the morphological markedness of the sentence constituents (Alemán Bañón & Rothman, 2016), as well as to individual differences in morphosyntactic processing (Fraga et al., 2021) (see Fraga, 2020;and Mancini et al., 2021 for further discussion).
Based on prior findings showing (1) processing dissociations possibly arising from the different sources of information accessed during the computation of gender (either syntactic or lexical information), number (conceptual information concerning the number of participants) and/or person (semantic and pragmatic information about discourse roles) agreement, and (2) emotional effects on number and gender agreement, it seems important to extend this line of research by examining the influence of emotional features on the processing of agreement dependencies that rely on person features.To this aim, we used as stimuli short sentences with pronominal subjects (e.g.Yo 1sg /I 1sg) , and negative (e.g.Mentir/To lie), positive (e.g.Recompensar/To reward), and neutral (e.g.Coger/To take) verbs that either agreed or disagreed in person features, while participants judged their grammatical acceptability.Our focus was on those ERP components that have been typically associated with the computation of person agreement dependencies (LAN, P600).Higher amplitudes in the person disagreement relative to the person agreement condition were hypothesised in both LAN and P600 waves (Molinaro et al., 2015).Bearing in mind the results from studies on the interplay between emotion and gender or number agreement, predictions could be made as follow with respect to the interaction between emotion and person agreement.In line with those lexicalist approaches assuming interactions between different levels of linguistic representation during the computation of agreement relations (Adger & Smith, 2010;Wechsler, 2011), modulations in the LAN component should be expected if emotional features influence the processing of person agreement between sentence elements (Hinojosa, Albert, Fernández-Folgueiras, et al., 2014;Martín-Loeches et al., 2012).Alternatively, encapsulated views of agreement, such as the feature-copying model (Chomsky, 2001), assume that the computation of agreement dependencies is a purely syntactic-driven process that is unaffected by the semantics (e.g.emotional content) of controllers (e.g. a subject) and targets (e.g. a verb).Therefore, a lack of interaction between Emotion and Grammaticality in the LAN wave should be expected (Díaz-Lago et al., 2015;Fraga et al., 2017).Regarding the P600, we hypothesise a main effect of Grammaticality with no further modulations by Emotion (Fraga et al., 2017;Hinojosa, Albert, Fernández-Folgueiras, et al., 2014;Padrón et al., 2020).Encapsulated and lexicalist views would not make different predictions about this component since its amplitude is modulated by both lexico-semantic and syntactic information (Kuperberg, 2007).

Participants
Our sample size was determined based on an a priori power analysis using G*Power (Faul et al., 2007).Assuming a α = 0.05 significance level, we estimated that a total sample size of 28 participants would provide 95% power to detect effects.(medium size effect d = 0.5).Considering potential drop-outs, we recruited 32 participants to exceed the criterion.Of the 32 recruited participants two were excluded due to excessive artefacts in the recording.The remaining sample consisted of 30 native speakers of Spanish (20 females, 10 males), who participated in the study after providing informed consent.The age range of the participants was 18-29 (mean 21.7).All participants were right handed according to the Edinburgh Handedness Inventory (Oldfield, 1971).They had normal or corrected to normal vision, and reported no history of neurological or psychiatric impairment.Participants were students from the Complutense University of Madrid that received credit course.Ethical approval has been obtained from a local ethics committee.
The verbs were used to create a total of 240 regular present tense sentences (see Appendix A).The length of all experimental sentences was 4 or 5 words and had either a Pronominal subject -Verb -Direct object (e.g.Yo agredo al estudiante/I attack the student) or a Pronominal subject -Verb -Prepositional phrase (e.g.Tú oras en la iglesia/You pray in church) structure.There were 80 sentences per condition (80 with a positive verb, 80 with a negative verb, 80 with a neutral verb).Each sentence included a 1 st or 2 nd person pronominal subject with its corresponding grammatical or ungrammatical verb.Hence, this resulted in two grammatical and two ungrammatical versions of each sentence (e.g.Yo 1sg canto 1sg en la ducha/I 1sg sing 1sg in the shower and Tú 2nd cantas 2nd en la ducha/You 2nd sing 2nd in the shower are grammatical sentences, whose ungrammatical versions are Yo 1st cantas 2nd en la ducha/I 1st sing 2nd in the shower and Tú 2nd canto 1st en la ducha, You 2nd sing 1st in the shower).Therefore, four lists were created so that each participant read only one of these versions of each sentence.Each stimulus list also included a set of 72 filler sentences that consisted of half grammatical and half ungrammatical sentences displaying person agreement and disagreement between pronouns and verbs (3rd singular, 1st plural, 2nd plural, 3rd plural). 1 Sentence order within each list was randomised.In total, each participant read 312 sentences.The within-participants design included two factors were entered in a 3 (Emotion: negative, neutral positive) x 2 (Grammaticality: correct, incorrect).The person features of the pronoun (1 st or 2 nd person) were not considered as an experimental condition.

Procedure
Participants performed the experimental task seated comfortably in an electrically shielded and sound-attenuated room.They were seated in front of a computer monitor, on which sentences were visually presented word by word.Words were displayed in black lowercase letters (with the exception of the first letter of each sentence that was displayed in upper-case letter) on a soft-grey background.Participants performed a grammatical judgement task.They were asked to assess the acceptability of each sentence by pressing one of two buttons with the middle and the index fingers.The assignment of correct and incorrect responses was counterbalanced across participants.Each trial began with a fixation point for 1000 ms, followed by a 100 ms blank screen.Thereafter, each word was presented for 300 ms, followed by a 300-ms blank screen.One second after the offset of the last word, a question mark indicated that participants had to judge the sentences for grammaticality (Hinojosa, Albert, Fernández-Folgueiras, et al., 2014;Martín-Loeches et al., 2012;Taylor-Clarke et al., 2002).The intertrial interval was 500 ms.The whole set of 312 words were randomly presented to each participants in 5 blocks.A 10 trials practice block was allowed before the beginning of the first experimental block.

Electroencephalographic (EEG) recording and analysis
EEG activity was recorded trough a 64-channels Brai-nAmp system.Sixty electrodes mounted in an Electro-Cap recording cap based on the International 10-10 System.All electrodes were referenced to the average activity of the two mastoids.The electrooculographic (EOG) activity was recorded using vertical and horizontal bipolar electrodes.These electrodes were placed at supra-infraorbital level of the left eye and on the outer canthus of both eyes, respectively.Impedances were kept below 5 KΩ for scalp and mastoids electrodes, and below10 KΩ for EOG electrodes.Recordings were amplified using BrainAmp amplifiers (BrainProducts, Munich, Germany), continuously digitised at a sample rate of 1000 Hz, and filtered online with a frequency band-pass of 0.01-100 Hz.
EEG data were analyzed with the EEGlab toolbox (Delorme & Makeig, 2004), a toolbox implemented in the Matlab environment (The MathWorks, Natick, MA).The signal was down-sampled to 500 Hz and down pass filtered with a low cut-off at 20 Hz.The continuous EEG was segmented time-locked to the target verb presentation (−200 ms to 800 ms).Data were baseline corrected (−200 ms).The ERP analyses included only epochs with correct responses.Epochs with recording channels exceeding ± 150 μV were removed.Subsequently, an independent components analysis (Makeig et al., 1997) was performed to eliminate the blink artefacts (Jung et al., 2000).Finally, epochs with artefacts were individually rejected with a visual inspection criterion.Following this procedure, we retained on average 29.67 (SD = 4.91) trials in the positive agreement, 30 (SD = 4.47) trials in positive disagreement, 28.87 (SD = 5.82) trials in the negative agreement, 30.23 (SD = 6.31) trials in the negative disagreement, 29.1 (SD = 5.57) trials in the neutral agreement, and 28.6 (SD = 5.86) trials in the neutral disagreement conditions.Finally, grand-averages ERPs were calculated for each condition across participants.
The current study focused on the two ERP components that have been related to the processing of Table 1.Means (and SDs) of valence (1 = highly negative, 9 = highly positive), arousal (1 = highly relaxes, 9 = highly activated) No. of letters, No. of syllables, log frequency and concreteness (1 = highly abstract, 9 = highly concrete).person agreement features, the LAN and the P600 (Hinojosa et al., 2003;Mancini et al., 2011a;Silva-Pereyra & Carreiras, 2007).Statistical differences between conditions in these waves were assessed following a nonparametric cluster-based random permutation analysis approach (Maris & Oostenveld, 2007).This approach overrides the problem of a priori choosing a spatiotemporal window or electrodes by computing a mass univariate test in every channel-time pair.The mass univariate approach generates a large number of statistical comparisons, which increases the type 1 error.To counteract the increased probability of false positive results, the significant effects between conditions are not assessed in the spatiotemporal pairs individual level, but in the cluster level, formed as the sum of significant adjacent spatio-temporal statistical values.The significance of the cluster statistic is evaluated by comparing its value with the sampling distribution obtained by permutation tests that randomly shuffles the label of the data.The analytic steps were as follows.First, a mass univariate test was conducted at each time-electrode pair (a dependent-samples t-test for the Congruency contrast or a dependent F-test for the Emotion and interaction contrasts).Clusters were formed with the adjacent spatiotemporal points with p-values below 0.05.Cluster-level test statistic was calculated by taking the sum of all the individual statistics values within that cluster.Then, a null distribution was created by computing 1000 randomised cluster-level test statistics.Finally, the actually observed cluster-level test statistics were compared against the null distribution and only clusters falling above the 95th percentile were considered significant (p < 0.05).

Performance
Behavioural data from two participants were not recorded due to technical issues.Therefore, we report data from 28 participants.As can be seen in

LAN effects
Based on prior studies (Mancini et al., 2011a;Zawiszewski et al., 2016), differences between experimental conditions in the LAN component were assessed in the 300-500 ms time-window.As shown in Figure 1, the cluster-based analysis approach revealed a more negative component in the disagreement compared to the agreement condition in a cluster of left frontal electrodes (FP1, AF3, F7, F5, F3, FC5, FC3, TP7, T7, C5, C3, CP3) between 360 and 460 ms (p = 0.098).The interaction between Emotion and Grammaticality was also examined in the LAN spatiotemporal window obtained by the previous cluster analysis.These theoreticallybased post-hoc analyses showed different LAN effects in the three emotional conditions in a cluster formed by four frontal electrodes (FP1, AF3, F7, F5) (p < 0.05) (Figure 1).Post-hoc Bonferroni pairwise comparisons revealed that differences between agreement and disagreement conditions were only statistically significant in the neutral emotional condition (p < 0.05, Bonferroni corrected).Also, the results of post-hoc comparisons between negative, neutral and positive condition either in correct or incorrect grammatical conditions showed no significant effects.Finally, the statistical analysis of the main factor of Emotion did not reveal any cluster of differences between conditions.

Discussion
We sought to understand the relations between emotional and person features during the processing of local agreement mismatches between pronominal subjects (controllers) and negative, neutral and positive Figure 2. P600 component.Topographies and grand-average ERP waveforms of the main effect of Grammaticality in the P600 component (collapsed for all electrodes forming the significant cluster).Topography of the incorrect minus correct conditions averaged in the time window 560-800 ms.White points represent the significant electrodes that form part of the main effect of Grammaticality in the P600 cluster.Topographies and grand-averaged ERP waveforms (collapsed for all electrodes forming the significant cluster) of each of the Grammaticality conditions (p-value <0.05).
verbs (targets).In agreement with the findings of prior studies (Barber & Carreiras, 2005;O'Rourke & Van Petten, 2011;Silva-Pereyra & Carreiras, 2007), we observed enhanced LAN 2 responses for neutral verbs that violated person agreement dependencies with a preceding pronoun relative to the grammatically correct condition.Importantly, the central and novel finding of our study is the lack of LAN modulations in a subset of left frontal electrodes when person agreement errors occurred in verbs with negative and positive valence.Thus, our data suggests that LAN feature-checking operations dealing with the early detection of person agreement errors are less efficient when the target verb signals positive and negative biologically salient content.In a subsequent processing stage, the system becomes aware of this mismatch and enables repair and reanalysis operations in order to complete parsing, conceptual interpretation and referential processes needed for discourse understanding (Kempen, 1998).In particular, it has been suggested that repairing agreement violations entails the deletion of a feature specification and the insertion of a feature specification once that the parser realises that features on the agreeing target are not consistent with the feature specification of the controller (Ackema & Neeleman, 2019).These additional computations are reflected in similar enhanced P600 amplitudes to person agreement errors in positive, negative and neutral verbs relative to grammatical sentences.Overall, our data is in line with prior research that has shown emotional modulations in number (Martín-Loeches et al., 2012) and gender (Fraga et al., 2021;Hinojosa, Albert, Fernández-Folgueiras, et al., 2014;Jiménez-Ortega et al., 2017) consistency checking during the establishment of agreement relations between sentence elements.Prior eye-tracking and ERP evidence pointed to the existence of processing dissociations in the computation of person relative to number agreement (Biondo et al., 2018;Mancini et al., 2013;Zawiszewski et al., 2016), which suggest an independent representation of these features (Mancini et al., 2021;Rizzi & Cinque, 2016).These differences have been related to the higher cognitive salience of person (Nevins et al., 2007), to distinct interpretive properties related to feature-mapping options for person (e.g. the role of the subject in the discourse) and number (e.g. the cardinality of the subject, that is the number of entities that this sentence element entails) (Biondo et al., 2018;Mancini et al., 2017Mancini et al., , 2021)), or to different checking-for-agreement mechanisms (den Dikken, 2019).Although processing differences between person and gender features are less understood, it seems plausible to hypothesise that emotion could exert a different impact in the computation of person relative to both number or gender agreement relations.In this line, Martín-Loeches et al. ( 2012) observed higher processing costs for the detection of number agreement anomalies between nouns and negative adjectives, which is at odds with the less efficient detection of gender (Hinojosa, Albert, Fernández-Folgueiras, et al., 2014) or person (current data) agreement mismatches in emotional words.Nonetheless, the use of pronominal subjects to control subject-verb agreement dependencies in the current study may at least partially account for similar emotional effects found for the processing of gender and person information.Pronouns are specified for both gender and person, while noun phrases (i.e.R-expressions) are specified for gender only.According to some theoretical views, differences in the processing of person compared to number or gender agreement features would be only observed when person agreement mismatches occur between verbs that have person information, and noun phrases that do not carry person specification (Ackema & Neeleman, 2013, 2019).Indeed, prior ERP studies that failed to report processing dissociations between person and number used agreement violations between pronominal subjects and verbs (Silva-Pereyra & Carreiras, 2007).Thus, the detection and repair of an unexpected person agreement form when the parser checks verbal features against the pronominal subject may share similar mechanisms with computations involved in the identification of gender disagreement between verbs and noun phrase subjects (as in Hinojosa, Albert, Fernández-Folgueiras, et al., 2014).These operations seem to be similarly affected by emotional features.Future studies should deal with a direct comparison of subject-verb person agreement mismatches using noun phrases (e.g.*John 3sg shoot 2sg the gun) and pronominal subjects (e.g.*He 3sg shoot 2sg the gun).
Models of sentence parsing (e.g.Bornkessel-Schlesewsky & Schlesewsky, 2013; Friederici, 2011;Hagoort & Indefrey, 2014;Vosse & Kempen, 2000) appear to be largely underspecified with respect to the impact that emotion may have during the establishment of different syntactic relations between sentence constituents.However, the finding that the emotional content of words is accessed during the computation of agreement dependencies based on person features has some implications for theoretical models of agreement.Our results argue against a view in which agreement is purely syntactically-driven and encapsulated with respect to semantic, pragmatic or discourse features (Chomsky, 2014;Franck et al., 2006).Instead, our data favours lexicalist views, which assume that the syntactic parser is exposed to lexical, conceptual and referential external influences while computing agreement dependencies between sentence elements (Adger & Smith, 2010;Vigliocco & Hartsuiker, 2002;Wechsler, 2011).In this vein, some proposals with a particular focus on the processing of person features have emphasised the contribution of semantic-pragmatic and discourse informationrelated to the role of the participant in the speech event signalled by the verb argument's referentwhile establishing agreement relations between sentence elements based on person information (Mancini et al., 2011a(Mancini et al., , 2014)).Thus, current evidence leaves open the possibility of modulatory effects by emotional features when assigning discourse participant roles (i.e.speaker, addressee, or non-participant) to interpret person information.
An important aspect that deserves further attention relates to the issue of how emotional features of words are involved in agreement processing in a broad sense.Some lexicalist accounts have claimed that agreement dependencies are computed through a cue-based retrieval mechanism for accessing and comparing information stored in memory from previously processed constituents (Lewis & Vasishth, 2005;Wagers et al., 2009).According to this proposal, an agreement mismatch between a verb and its preceding subject engages a reanalysis process that makes use of several sources of information provided by the verb, such as agreement features (e.g.number or person) or structural cues (e.g.thematic roles).Under this logic, if emotional features behave as semantic cues that contribute to the processing of agreement violations, additional computation efforts to detect agreement errors could be expected when verbs and their subjects disagree not only in person but also in emotional features.The observation of LAN effects at fronto-central and central electrodes to agreement mismatches in both positive and negative verbs preceded by rather neutral pronominal subjects is in partial agreement with this view.However, some cautious is needed since the narrower distribution of the LAN for emotion words relative to neutral words suggests that the detection of errors in neutral verbs which were preceded by emotionally congruent neutral pronominal subjects is more efficient and activates feature-checking mechanisms to a greater extend.Alternatively, the Unification model (Hagoort, 2003a;Hagoort & Indefrey, 2014), assumes a parallel processing in which different sources of linguistic information, such as morphosyntactic or semantic, interact as soon as they become available.Of note, the results of prior studies suggest that the temporal course of the detection of agreement errors between 300 and 500 ms overlaps with the automatic engagement of attentional resources devoted to the processing of the emotional features of words (Hinojosa, Albert, López-Martín, et al., 2014;Kissler & Herbert, 2013).Thus, in line with the proposal by Hagoort (2003a), we suggest that the computation of agreement relationships would benefit from the privileged access to the emotional information of words, as reflected in the interaction between Emotion and Grammaticality found in the current study.Nonetheless, additional work is needed to further examine the mechanisms underlying emotional effects on agreement.In this sense, it would be relevant to directly compare agreement errors in verbs with emotional and nonemotional subjects.Also, the lack of main effects of Emotion in the pattern of ERPs suggests that task demands might be a critical aspect in the processing of emotional words embedded in sentences (see Hinojosa, Albert, Fernández-Folgueiras, et al., 2014 for further discussion).Therefore, the comparison of the results elicited by grammaticality judgement and emotional categorisation tasks might shed some light about how emotional and morphosyntactic information interact during the processing of agreement.Finally, another productive approach to test parallel and interactive morphosyntactic and emotional effects might be the use of experimental paradigms that include double agreement and semantic violations (e.g.Hagoort, 2003b) in target emotional words.
In conclusion, the results of the present study indicate that prior interactions between word emotional content and the processing of agreement dependencies based on gender and number are also observed for person agreement relations between pronominal subjects and verbs.Our data show that the mechanism dealing with the early detection of agreement mismatches during structure-building operations is less efficient in emotional compared to neutral words.This result suggests an advantage in processing emotional utterances containing agreement errors, which points to the need of a rapid communication of information that might be particularly relevant for individuals during communicative interactions.Our findings also highlight the importance of considering the contribution of emotional features when outlining neurobiological and psycholinguistic models of language.Future research within this domain should deal with open questions such as the direct comparison of emotional effects in the processing of number, gender and person agreement by means of double mismatches, the study of person agreement errors in feature distant (non-local) structural relations, or the impact of individual differences in morphosyntactic processing during the computation of person agreement.
Notes 1.As one reviewer noted, agreement errors always occurred following the sentence initial pronouns in both fillers and experimental sentences, which might have biased participants towards more word-pair based and less syntactically based processing.2. Tanner and Van Hell (2014) have suggested that biphasic LAN-P600 effects to verb agreement mismatches could be a result of grand mean responses from N400 (related to lexico-semantic processes) and P600 components since some individuals depicted centrally scalp distributed waveforms in the LAN time-window.However, this possibility seems unlikely in the current study since our data-driven analyses showed that the interaction between Grammaticality and Emotion reached significance in a cluster of left-frontal electrodes.Also, Tanner and Van Hell examined individual differences in LAN/N400-P600 effects and followed a data-analysis approach based on a visual inspection of Regions of Interest that included a-priori selected electrodes.Of note, the size of our sample does not allow investigating individual differences as in Tanner and Van Hell's study, an issue that deserves further examination in future research.

Figure 1 .
Figure1.LAN component.Topographies and grand-average ERP waveforms of the LAN component (collapsed for all electrodes forming the significant cluster).Topography of the subtraction of the incorrect minus the correct condition in the time window 360-460 ms.White points represent the significant electrodes that form part of the main effect of Grammaticality in the LAN cluster.Circles illustrate the electrodes that integrate the significant cluster of the Grammaticality x Emotion interaction.Topographies and grand-averaged ERP waveforms (collapsed for all electrodes forming the significant cluster) of each of the Grammaticality conditions (cluster p-value <0.1).Topographies and grand-averaged ERP waveforms of the agreement and disagreement conditions in each of the emotion conditions (p-value <0.05).