We employed complex intransitive constructions in the present study, formed using an event nominal (physical: “pani” fever, or mental: “santhosham” happiness) that provides most of the meaning and activates its argument structure, followed by a semantically underspecified light verb (“vannu’ come). We employed a 2 × 2 factorial design, similar to that of Shalu et al. (2025), manipulating the subject case (nominative or dative) and predicate type (ME or PE), keeping the light verb identical across conditions. Since the sentence-initial subject was uniformly dative in all non-anomalous conditions and nominative in all violation conditions, any ERP differences at the verb for ME violations vs. PE violations would be directly attributable to a processing difference between the two verb types.

Our hypotheses were as follows. If ME and PE in Malayalam complex constructions are processed similarly, then qualitatively similar ERP correlates should be expected for both verb types.

However, since a large part of the predicative meaning originates from the event nominal within the complex construction (Butt, 2010), and the nominals in the study represent two very different types of experiences, namely, mental vs. physical (Jayaseelan, 2004), we should observe qualitative neurophysiological differences between the two verb types.

Twenty-eight first-language speakers of Malayalam (mean age = 27.8; 11 women and 17 men) participated in the experiment in exchange for payment. They had learned Malayalam before the age of six and studied the language as an academic subject in school until 10th grade. All participants were dominantly right-handed, as determined by an adapted version of the Edinburgh Handedness Inventory (Oldfield, 1971) in Malayalam. Data from seven participants were excluded from analysis due to excessive artifacts in their data epochs as determined by an amplitude threshold of ±100 μV (further details are provided in the Supplementary material).

We employed critical stimuli in 4 conditions (Box 1), with 36 sentences in each critical condition, resulting in a total of 144 critical sentences. These were interspersed with fillers and pseudorandomly presented. Further details are provided in the Supplementary material.

Participants were seated in a dimly lit, sound-attenuated room during the experiment, which included a brief practice session. Stimuli were presented using E-Prime 2.0 (Psychology Software Tools, Pittsburgh, PA, United States) in a rapid serial visual presentation. Participants performed an acceptability judgment and a probe task after each trial. Further details are provided in the Supplementary material.

Scalp activity was recorded using 32 Ag/AgCl electrodes fixed to the scalp using Hydrocel Geodesic Sensor Net 32, with Cz as the online reference. EEG data were pre-processed using EEGLAB (version 14; Delorme and Makeig, 2004, sccn.ucsd.edu) in MATLAB (version R2023b; The MathWorks, Inc.). Data epochs were extracted at the critical position (verb; −200 to 1,200 ms) from the continuous recordings for each participant and analyzed using the eeguana package (version 0.1.11.9001; Nicenboim, 2018) in R (version 4.4.3; R Core Team, 2024).

The mean acceptability ratings and probe detection accuracy for the critical conditions were calculated for the trials included in the ERP analysis. As Table 1 shows, the non-anomalous conditions were rated as highly acceptable (>98%; ceiling), whereas the acceptability for the violation conditions was very low across the board (<21%). The probe detection accuracy reached a ceiling for all the conditions. Figure 1 shows raincloud plots (Allen et al., 2021) of the behavioral acceptability judgments.

The behavioral acceptability and accuracy were analyzed by fitting generalized linear mixed models using the lme4 package in R. Categorical fixed factors used scaled sum contrasts (effect coding). In the analysis of acceptability data, the statistical model included the fixed factors Case (Nominative vs. Dative) and Verb type (Mental experiencer vs. Physical experiencer), with random intercepts for participants and items and by-participant random slopes for the effects of Case, Verb type, and their interaction term. Type II Wald chi-squared tests of the fitted model (AIC = 1571.012) of the acceptability data showed a main effect of Case [χ²(1) = 109.66, p < 0.001, s = 82.83] and the interaction of Verb × Case [χ²(1) = 7.49, p = 0.006, s = 7.33]. Estimated marginal means on the response scale were computed on the model using the emmeans package (Lenth Russell, 2021) to resolve this interaction, which showed that, for both verb types, the estimates for the conditions with a dative subject were higher compared to those for conditions with a nominative subject. Pairwise contrasts of these estimates within each verb type revealed a simple effect of Case for both mental experiencer light verbs [estimate = 0.878, standard error (SE) = 0.038, p < 0.001, s = 373.79] and physical experiencer light verbs [estimate = 0.873, SE = 0.040, p < 0.001, s = 346.06]. These findings reflect the substantial difference in acceptability between conditions with dative subjects, which engendered very high acceptability, vs. conditions with nominative subjects, which were rated as hardly acceptable.

In the analysis of probe detection accuracy, models with an interaction term of the fixed factors Verb and Case, as well as those with random slopes, were singular. The model with Verb and Case as fixed factors with by-participant and by-item random intercepts (AIC = 816.1534) did not detect any effect involving Verb or Case (s < 2.5), reflecting the highly similar (ceiling) accuracy of probe detection across conditions.

The ERPs at the verb are shown in Figure 2. Visual inspection of the ERP data indicated a negativity effect in the violation conditions relative to the non-anomalous ones. Based on both the ERP components identified in previous research (Shalu et al., 2025) and visual inspection, we selected the 400–600-ms time window for analysis. The single-trial ERP mean amplitudes extracted in the analysis time window from a total of 3,238 data epochs entered the analysis. The raw data collected during the experiment, the pre-processing pipeline used, and the pre-processed data are available in the online data repository.

We computed a linear mixed-effects model with the fixed factors Verb type, Case of the subject noun, ROI, the −200 to 0-ms pre-stimulus baseline mean amplitude as a covariate (scaled and centered), along with by-participant and by-item random intercepts. The analysis code and full model outputs are available as R notebooks in the online analysis repository. Type II Wald Chi-squared tests on this model (AIC = 434325.1) showed a main effect of Case [χ²(1) = 6.84, p = 0.008, s = 6.81] and interaction effects of ROI × Verb [χ²(1) = 14.79, p = 0.03, s = 4.69] and ROI × Case [χ²(1) = 14.82, p = 0.03, s = 4.70]. Estimated marginal means on the response scale were computed using the emmeans package (Lenth, 2021) in the model to resolve these interactions. The pairwise contrasts of estimates for Case within each level of ROI revealed simple effects of Case in the right-frontal (estimate = 1.245, SE = 0.364, p < 0.001, s = 10.60), right-central (estimate = 1.067, SE = 0.364, p = 0.003, s = 8.19), right-parietal (estimate = 1.012, SE = 0.364, p = 0.005, s = 7.50), mid-fronto-central (estimate = 0.947, SE = 0.365, p = 0.009, s = 6.71) and mid-parieto-occipital (estimate = 0.566, SE = 0.319, p = 0.07, s = 3.71) regions. The estimates for nominative subjects were more negative than those for dative subjects in these regions. The pairwise contrasts of estimates for Verb type within each level of ROI revealed a simple effect of Verb type in the left-frontal region (estimate = −0.839, SE = 0.364, p = 0.02, s = 5.54). The estimate for mental experiencer verbs was more negative than that for physical experiencer verbs in this region. A more complex model in which the by-participant random slopes for Case, Verb type, and their interaction term were included in the random effects specification¹ showed that this pattern of results remained largely intact despite numerical differences. In sum, both mental and physical experiencer verbs elicited a negativity effect when the subject was in the nominative case as opposed to the dative case. Further, a general verb-type effect ensued for mental experiencer verbs, which elicited a negativity effect in the left-anterior region compared to physical experiencer verbs, regardless of grammaticality.

The N400 effect found for both ME and PE is perhaps unsurprising because both verb types have an identical argument structure and realization (Belletti and Rizzi, 1988; Grimshaw, 1990; Levin and Hovav, 2005). This result is consistent with findings from Shalu et al. (2025). As in their study, we interpret the negative effects that we found in the present study as a consequence of a violation of interpretively relevant rules, which aligns with Choudhary et al. (2009) and Nieuwland et al. (2013).

In Malayalam, a clear morphosyntactic constraint governs the case marking of the subject argument in ME and PE in complex constructions. Regardless of whether the experience meant by the experiencer verb is mental or physical, it is expressed as though it is moving from an abstract space to the subject argument, which serves as the experiencer-goal of the abstract movement. In this respect, the dative subject experiencer is similar to other dative arguments that are the targets/goals of some movement (see sentences 5–7). Consequently, the subject is in the dative case, which is the general obligatory marker for targets of movement in Malayalam (Nizar, 2010; Mohanan, 1990; Krishnan, 2020; Jayaseelan, 2004). The violation of this interpretively relevant rule in the present study led to an N400 effect for both verb types, regardless of the experience conveyed by the verb.

The classic focal LAN effect is typically elicited by anomalous inflectional morphology (Gunter et al., 2000) and is generally understood as an indicator of the detection of morphosyntactic inconsistencies (Münte et al., 1998; Friederici, 2002; see Molinaro et al., 2011 for a review). It is observed in the same time window as the N400, with often the only difference between the two effects being the topography. Although the effect we found for ME is in the left-anterior region and within the same time window as the N400 effect, there are at least two points that speak against interpreting this LAN-like effect as an instance of the classic focal LAN. First, as already mentioned, our critical stimuli did not constitute a morphosyntactic violation of the type typically associated with the LAN. Second, and more importantly, the LAN-like effect is a general verb-type effect for ME, regardless of anomaly. That is, the same effect is elicited by both anomalous and non-anomalous sentences of the ME type.

Instead, a finding that would be more pertinent for interpreting this effect stems from previous imaging studies on abstract verbs, which have shown that left-lateralized brain regions are activated when processing these verbs (Rodríguez-Ferreiro et al., 2011; Binder et al., 2005; Sabsevitz et al., 2005; Noppeney and Price, 2004; Perani et al., 1999). In line with this, it is plausible that the LAN-like effect observed in our study for ME could be explained by the degree of abstractness and non-bodily involvement associated with these verbs compared to PE (Borghi, 2020). ME refers to internal states, often linked to emotions, thoughts, etc., and other intangible aspects that cannot be directly observed. By contrast, PE describes physical sensations or states. Compared to ME, which does not directly engage with the body, PE is closely linked to the physical body and the tangible sensations it can perceive and feel. That is, even though PE is abstract, they have some observable effects in the body that make it more perceivable compared to relatively more abstract mental experiences. For instance, an individual suffering from fever may have a flushed face and sweating, an individual who is experiencing hunger may have a growling sound from the stomach, etc. Thus, the LAN-like effect observed for ME can be plausibly attributed to their more abstract nature compared to PE. This interpretation is necessarily tentative, and further neurophysiological research on the finer distinctions of abstract verbs would be needed to confirm this.

Alternatively, the LAN-like effect for ME in comparison to PE could plausibly be explained in terms of the ± control/agency associated with the respective verb types (Schlesinger, 1992; Mohanan, 1990; Jayaseelan, 2004). ME refers to a state where the experiencer has volitional control/agency over their mental state, allowing them to change or influence it without taking physical action. For instance, someone can willfully suppress sadness to maintain composure. In contrast, PE describes a state where the experiencer has little control/agency over the situation and must take tangible actions to change it. For example, a person experiencing freezing cold must actively do something to alleviate the discomfort, such as wearing warm clothes or turning on the heater. In light of this, it is plausible to consider the LAN-like effect observed in our study as a marker of the control/agency difference between ME and PE. However, this interpretation also remains speculative, and more studies comparing agentive verbs with non-agentive verbs would be needed to gain further insights in this regard.

Verb types differ in terms of their internal semantic structure (Davidson, 1971; Dowty, 1979; Jackendoff, 1991; Levin and Hovav, 2005), and several processing studies have attributed the differences observed between verb types to this aspect. However, a limitation of many of these studies is that it is very difficult to isolate syntactic differences from semantic differences (Brennan, 2015). Our stimuli did not have this issue because all critical sentences in our study were syntactically identical and differed only with respect to the type of experiencer verb. Therefore, the LAN-like effect that we found for ME regardless of grammaticality in comparison to PE appears to show inherent semantic differences between these two verb types, including but not limited to their degree of abstractness and agency.

In conclusion, the present study explored the processing of mental and physical experiencer verbs in Malayalam complex experiencer constructions. Results revealed a negativity effect for subject case violations involving mental as well as physical experiencer verbs, suggesting that they are processed qualitatively similarly. However, we also found a LAN-like impact for mental experiencer verbs, regardless of grammaticality, compared to physical experiencer verbs. In sum, although both mental and physical experiencer verbs are processed qualitatively similarly, our data nevertheless provide converging evidence to conclude that effects of inherent differences between mental and physical experiencer verbs in Malayalam are still discernible even in syntactically identical constructions involving the two verb types.