Subject-verb number agreement in Spanish, as in many other languages, conforms to the rule of having the number morphological features of the verb agreeing with those of the subject noun-phrase (NP) (see Acuña-Fariña, 2009 for a review). However, findings from the psycholinguistic literature have shown that agreement is not susceptible to influences coming neither only from the head noun nor only from syntax. For example, number agreement can be affected by factors, such as the difference between the morphological singular number of the head nouns and the morphological plural number of the closest-to-the-verb (local) noun (Bock and Miller, 1991; Haskell and MacDonald, 2003; Bock and Middleton, 2011) or the distance between a mismatching feature embedded in a prepositional phrase (PP) and the subject head noun (e.g., Franck et al., 2002). Evidence showing that agreement does not come entirely from a syntactic source and that it is sensitive to the role of semantics comes from studies showing that plural local nouns elicit more plural agreement attraction errors when the noun phrases have a notional distributive reading (e.g., The label on the bottles) than when they do not (e.g., The house of my cousins; Vigliocco et al., 1995, 1996). These findings suggest that grammatical information of a subject can be overridden by the number specification of the conceptual representation of the subject phrase and that conceptual factors may impact grammatical encoding. In all the above cases, the observed result is that the verb seems to be “attracted” to the plural number of the preceding noun when it mismatches the number of the head noun. This may happen mainly for three reasons: (i) the plural feature of the local noun may override the default assignment by being mistakenly detected by the verb-agreement mechanism (feature percolation account; Bock and Eberhard, 1993); (ii) singular number (being morphologically unmarked) is more vulnerable to the influence of plural number (Eberhard, 1997); or (iii) due to unsuccessful number reconciliation of the number features of the subject noun when agreement features are transmitted to the target verb (marking and morphing account; Bock et al., 2001).

In comprehension studies, attraction effects have been demonstrated mainly in the form of “grammatical illusions.” Thus, sentences, such as “The key to the cabinets *are …,” seem to be processed faster, as shown in reaction time (e.g., Wagers et al., 2009) and as if they are grammatically correct, as demonstrated in the reduced negative left anterior negativity (LAN/N400) or positive (P600) components of agreement violation detection as compared to other ungrammatical sentences, such as “The key to the cabinet *are…” (e.g., Tanner and Van Hell, 2014). Such illusions have also been explained by a feature percolation account that assumes erroneous percolation into a subject noun position when syntactic constituents are hierarchically integrated into a processing structure (e.g., Bock and Cutting, 1992). In addition, they have been accounted for by a cue-based retrieval failure during syntactic integration caused by feature slippage or misidentification of the correct controller of agreement (e.g., Wagers et al., 2009).

However, there are findings to suggest that even in grammatical sentences, effects of attraction can be shown in the form of “ungrammatical illusions.” Nicol et al. (1997) found slower reading times and less accurate responses when a singular head noun was followed by a plural than a singular attractor, as in “The author of the speeches is…” vs. “The author of the speech is…” in a Maze task where participants had to decide which of the two words was a better continuation for a sentence. Laurinavichyute and von-der-Malsburg (2019) found that a plural noun that mismatched the number of a subject noun in grammatical sentences, such as “The admirer of the singers supposedly thinks that…” slowed down processing on the verb. Processing difficulty in correct sentences of NP-mismatch conditions with a singular head (Experiment 3: The key to the cabinets was…), as displayed in slower reading times, was also reported by Pearlmutter et al. (1999). Similarly, Franck et al. (2015) showed that plural object interveners slowed down the grammaticality judgment of subject-verb dependencies. In an Event Related Potential (ERP) study, Kaan (2002) found that the grammatical condition of a singular subject and a plural object in German yielded an enhanced early positivity at the critical verb. In another ERP study, Martin et al. (2012) found evidence of processing costs during comprehension of elliptical sentences, such as “Marta se compró la camiseta que estaba al lado del vestido y Miren cogió otra…” (Marta bought the t-shirt_[FEM] that was next to the dress_[MASC] and Miren took another_[FEM]…): the gender of a mismatching attractor emitted larger negativity and larger late positivity in the condition where the attractor had a different gender from the antecedent. In other words, when the attractor did not match the retrieval cue, this had an impact on the processing of grammatical sentences as well. Finally, in a sentence completion study in Basque, the only study where the number of attraction effects have been examined in grammatical sentences at an electrophysiological level, Santesteban et al. (2020) found slower subject-verb production when the subject and object mismatched in number than when they matched. Mismatching objects elicited an early production P2 followed by a negative component, showing the difficulty of number feature retrieval and monitoring during correct subject-verb agreement production.

Recent studies that have looked at the impact of semantic factors on agreement processing have considered the case of emotional meaning. This is not surprising as emotional content due to its salience, with affective information being prioritized over non-affective information and capturing attention resources (e.g., Zajonc, 1980, 2001; Delaney-Busch and Kuperberg, 2013), has been found to affect lexical processing and shows its signature in neural implementation as well. More enhanced effects are reported for emotional than non-emotional words (e.g., knife vs. sink) whether they are processed in isolation or embedded in sentences and interact with other semantic or morphosyntactic information (see Kissler et al., 2006; Citron, 2012 and Hinojosa et al., 2020 for reviews). The majority of ERP studies at the sentential level that have examined the influence of emotional valence on the agreement have considered the case of gender agreement in Spanish (e.g., Hinojosa et al., 2014; Díaz-Lago et al., 2015; Fraga et al., 2017; Jiménez-Ortega et al., 2017). Some have found that the detection of gender agreement violations between adjectives and nouns can be affected at the early stages of morphosyntactic processing by whether the content of the agreeing element is emotional, as reflected by the interaction between grammaticality and valence in the LAN/N400 time window (Hinojosa et al., 2014; Jiménez-Ortega et al., 2017; Fraga et al., 2021). With regard to number agreement and sentence comprehension with ERPs, which is the focus of the present study, to our knowledge, there are only two studies that have investigated the role of emotional words (Martín-Loeches et al., 2012; Jiménez-Ortega et al., 2017 where the manipulation of emotional information involved a subliminal presentation of adjectives as the question of interest was centered on the automaticity of syntactic processing). Here, we focus on the study by Martín-Loeches et al. (2012) because in their experimental procedure the variables of valence and number were manipulated supraliminally, as in our study. The researchers used Spanish sentences of determiner-noun-adjective-verb structure for syntactic processing (Experiment 1) and manipulated emotional valence (positive, negative, neutral adjectives) and grammaticality (syntactically correct vs. incorrect). Thus, the adjective, which was the critical word, either matched or mismatched the number of the noun it modified and in the latter case resulted in the creation of syntactic violations: e.g., La hermana querida acude (The loved_[SG] sister arrives) vs. *La hermana queridas acude (The loved_[PL] sister arrives); La chica fea baila (The ugly_[SG] girl dances) vs. *La chica feas baila (The ugly_[PL] girl dances); and El espejo ovalado refleja (The oval_[SG] mirror reflects) vs. *El espejo ovalados_[PL] refleja.

Event-Related Potentials elicited during the performance of a grammaticality judgment task revealed a larger LAN (350–450 ms) component for incorrect than correct sentences only when the adjective was negative rather than neutral. Thus, morphosyntactic processes were found to be sensitive to the emotional information carried by the syntactically anomalous emotional words and affect the detection of number agreement of violations between the adjective and its noun. A late positive component (P600; 600–700 ms) was elicited by ungrammatical sentences and was not modulated by valence, suggesting that emotional information modulated grammatical processing only at the early stages of agreement computation (reflected in the LAN component).

Despite the merits of the study of Martín-Loeches et al. (2012), some aspects of the materials and the design may not have offered the best condition for clear evidence of valence and morphosyntactic effects on number agreement (for a review of methodological and procedural issues that may have contributed to discrepancies in studies of emotional impact on gender and number agreement see Fraga, 2020). Apart from emotionality, animacy was different between the conditions of interest, as the subject noun in the emotional conditions (positive/negative) was animate, whereas in the neutral condition it was inanimate. Importantly, the critical word (the adjective) was the element that bore both the valence and agreement manipulations making rather difficult the attribution of effects. Finally, language use is predominantly based on computation and processing of well-formed utterances both in comprehension and in production, and given that the existing relevant literature has mainly considered syntactic violations, more research is needed to address number agreement processing in grammatically correct contexts.

Thus, in the present study, we sought to investigate the interplay between morphosyntactic and semantic information in number agreement not examined so far. Specifically, we investigated the extent to which the emotional content of a local noun, i.e., of an element not directly relevant to the computation of subject-verb number agreement but of reported salience, might be considered by the parser along with its morphosyntactic features and affect the application of syntactic rules for the comprehension of grammatical sentences. That is, we tested the effect of emotion under the most stringent agreement conditions. Our focus was on two ERP components because they are the most relevant ones for current purposes: a negative component/N400 between 300 and 500 ms that when yielded is associated with initial emotional analysis (e.g., Delaney-Busch and Kuperberg, 2013) and a late posterior positive component/late posterior positivity (LPP)/late positive component (LPC)/P600 (after 500 ms) that is typically associated with sustained attention to emotional input and elaborate processing (e.g., Bayer et al., 2010; see Hinojosa et al., 2020 for a review). Regarding syntactic computations of ERP components in the same time windows, a (left anterior) negativity has been suggested to index the processing of dependency relations and is emitted when morphosyntactic violations or mismatches are detected. A late (centroparietal) positivity is associated with effects of reanalysis of syntactic violations or of expectations (present study) of agreement relations, which are not consistent with the syntactic analysis taking place (e.g., Osterhout and Mobley, 1995; Kutas and Federmeier, 2011; see Kuperberg, 2007 and Molinaro et al., 2011 for reviews.) Importantly, to obtain a clear picture of the effect of emotional attractors as emotionally loaded words (on a purely semantic level) and as syntactic interfering elements (on a syntactic level), we performed separate analyses in the position of the attractor (local noun) and of the target (verb), respectively.

Considering the crosstalk between brain networks that contain linguistic and emotional information (Pulvermüller, 1999; Palazova, 2014; see Hinojosa et al., 2020 for a recent review) and the attention-grabbing effects that affective meaning appears to have on lexical processing (e.g., Delaney-Busch and Kuperberg, 2013), in the present ERP study, we explored the extent to which semantic information of affective nature may impact access to syntactic representations during subject-verb number agreement. Thus, we investigated the attraction effect of emotional local nouns, i.e., interference of structurally irrelevant items, on grammatical sentences during sentence comprehension.

Unlike previous studies on agreement dependencies, where the influence of emotional content was considered in tandem with morphosyntactic violations (Martín-Loeches et al., 2012; Hinojosa et al., 2014; Fraga et al., 2017), the design of the present study allowed us to tease apart semantic from syntactic effects and provide a clearer picture of the impact valence that may have on number agreement processing. The manipulation of valence (negative vs. neutral) and number (singular vs. plural) of the attractor showed different effects on the processing of the attractor as a lexical item (local noun) and on the way, the retrieval of the features of the attractor affected subject-verb agreement computation.

With regard to the former, as reflected in the negative ERP component that was yielded (300–500 ms), we found a strong valence effect only for singular attractors. The fact that the valence effect was absent for plural attractors may be due to the costlier processing of morphologically and semantically richer plural nouns, which might have canceled out the larger saliency of nouns with negative vs. neutral emotional valence. Thus, if plurality and valence effects were of similar size, they might have canceled each other out. Additionally, the fact that attractor nouns that matched in both number and valence with the (singular and neutral) subject noun tended to elicit a larger late positive ERP component (780–880 ms) suggests that the parser had more difficulty identifying the subject and discarding the attractor noun as a possible candidate to be assigned the subject role.

The effects shown on the agreeing verb are particularly interesting, as in verb position, one would expect to find the attraction effect in the form of an ungrammatical illusion and the impact of valence (if there was). The negative ERP component showed that subjecthood feature checking was costlier when the attractor had the same valence as the subject noun that was neutral. This suggests that head and attractor noun features were retrieved for agreement computation. Importantly, the late ERP positive component showed sensitivity to the emotional content of the local noun during feature integration and reanalysis. The fact that “ungrammatical illusions” caused by attraction on grammatical sentences (e.g., Franck et al., 2015) only showed with neutral attractors suggests that the saliency of emotionally negative attractors facilitated discarding the features of the attractor as the agreement source.

Despite being an underinvestigated topic, studying emotional effects in agreement, especially attraction effects in grammatical sentences, can have important implications. It can provide information about what is considered by the parser during online the processing of agreement relations and offer explanations about whether “ungrammatical illusions” can be attributed to the faulty mental encoding of linguistic representations or difficulty in accessing the right morphosyntactic information due to factors, such as the emotional value of an attractor. It can also shed light on the debate between strongly modular models that assume distinct sequential processes between syntactic and semantic representations (e.g., Ferreira and Clifton, 1986; Friederici and Weissenborn, 2007) and fully interactive models that assume that syntactic and semantic constraints interact simultaneously at the message-level representation (e.g., Hagoort, 2003) or intermediate accounts (e.g., Kim and Osterhout, 2005). Our findings suggest that number agreement is not insensitive to affective meaning but that emotional information of an attractor is considered in different operations and at different stages during grammatical sentence processing: for the retrieval of lexical and syntactic representations of the subject-NP and during subject-verb number agreement. Regarding number agreement processing, valence seems to be considered at an early stage of feature checking, where it acts as a cue for the selection of agreeing elements (a local noun with the same valence as the subject noun triggered similarity-based interference). At a late stage of reanalysis, both valence and number features of the attractor are checked to confirm grammaticality.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The study was reviewed and approved by the Ethics Committee of the UPV/EHU (Comité de Ética para las Investigaciones relacionadas con Seres Humanos, CEISH; Project code: M10_2019_167). Participants provided their written informed consent to participate in this study.

AH and MS conceived and designed the study, collected the data, performed the statistical analyses, and drafted the manuscript. Both authors agreed on the final content prior to submission.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.