The power of words depends on context: action congruence differentially shapes activation by words and sounds

Linguistic labels have been shown to facilitate visual recognition and categorization. Labels have also been shown to be better facilitators than sounds. However, less is known about how nouns in contrast with verbs and non-verbal sounds differentially activate action-related conceptual information. In the present study, we investigated whether words and non-verbal sounds vary in their ability to facilitate visual processing depending on their congruence with an implied action. Participants were presented to an image verification task in which visual stimuli were preceded by either nouns, non-verbal sounds or verbs that were either congruent or incongruent with the action associated with the target stimulus. Reaction times were analyzed using linear mixed-effects models. Results revealed a robust interaction between cue type and action congruence. Non-verbal sounds and verbs produced significantly faster responses when the action was congruent compared to incongruent trials, whereas nouns were optimal only when action was incongruent. These findings suggest that action information plays a role in the representations activated by words and non-verbal sounds. In addition, the evidence shows that verbs appear to activate action-related information similar to that evoked by non-verbal sounds. More broadly, the results contribute to ongoing debates about the mechanisms by which language and non-linguistic cues shape perceptual processing, highlighting the role of action-related cues in predictive conceptual activation.

Evidence shows that there is a bidirectional influence between thought and language (Gleitman and Papafragou, 2005; Gumperz and Levinson, 1991; Levinson, 1997). Among the models of how thought and language interact, one line of research is aimed at testing how different types of cues facilitate visual object recognition, consistently showing that words tend to produce faster recognition of referents than non-verbal sounds. This “label advantage” (e.g. Lupyan and Thompson-Schill, 2012; Edmiston and Lupyan, 2015; Boutonnet and Lupyan, 2015) states that linguistic labels are categorical, which implies that they are more abstract and oblivious to the specificities of exemplars. In contrast, representations activated by sounds are motivated, that is source-correlated, specific and detail-rich representations (see also Casasola, 2005; Waxman and Markow, 1995; Yoshida and Smith, 2005). Experimental evidence further shows that linguistic input can facilitate category learning (Lupyan et al., 2007), although such effects may depend on task demands and contextual constraints (Klemfuss et al., 2012). However, such interpretations often overlook a crucial point: different cues may not necessarily differ in informativeness, but rather in the kind of information they activate. From this perspective, cue effectiveness depends on how well the cue's activation profile aligns with the perceptual information available in the visual stimulus.

This view is consistent with research showing that conceptual activation is multi-layered, multimodal, and context sensitive. Research on verb semantics contributes to this distinction. Verbs differ from nouns not only in grammatical function but also in their semantic architecture, contextual dependence, and link to sensorimotor systems. A growing body of research suggests that semantic representations are not unitary but are composed of multiple types of information derived from both experiential (sensorimotor) and language-based sources (e.g., Ferreira et al., 2008; Huettig and Altmann, 2005; Kousta et al., 2011; Meteyard et al., 2012; Vigliocco et al., 2009, 2004). From this perspective, different cues may preferentially activate different informational components. Nouns are typically associated with stable, category-level information, whereas verbs and environmental sounds are more strongly linked to dynamic, action-related information. This makes verbs particularly sensitive to the presence or absence of action information in the accompanying visual stimulus. When images do not provide action cues, the dynamic information activated by verbs may not align well with the visual input, reducing their facilitation relative to nouns.

In the present study, we focus on action information as a critical dimension of conceptual activation. Action information refers to dynamic properties of a stimulus, including implied motion, posture, and interaction patterns that specify what an agent is doing. We operationalize action congruence as the correspondence between the action implied by an auditory cue (e.g., barking) and the action depicted in the visual stimulus (e.g., a dog barking vs. a dog sleeping). Importantly, although participants were instructed to judge whether images were related to the auditory cue, match decisions did not depend on the presence of action information, but on referential correspondence between cue and image. This allows us to examine how action congruence modulates processing without changing the decision criterion across conditions.

Despite these converging theoretical insights, empirical studies have generally not manipulated action information in pictures in a controlled manner. Most comparisons between words and sounds use images that contain uncontrolled mixtures of static and dynamic features, making it difficult to determine whether observed differences reflect a true advantage of words or simply an advantage of conceptual predictions over event-based predictions under specific conditions. Similarly, verbs are underrepresented in this line of research, and the role of depiction—such as line drawings vs. photographs—remains understudied, despite clear differences in the perceptual richness and action information conveyed by these formats.

Because environmental sounds are produced by dynamic physical interactions—rather than by static objects—they systematically co-occur with action events. Consequently, environmental sounds may naturally activate dynamic conceptual features, even when researchers intend them to activate object-level information. Producing a sound requires an active exchange of energy: a guitar does not emit a strumming sound unless it is being strummed, just as a dog does not produce a bark unless it is actively barking. In general, objects in a steady resting state do not emit sound; sound arises from actions, interactions, or physical changes. Thus, environmental sounds are intrinsically linked to dynamic events, which suggests that they may reliably activate action-related representations.

The foregoing discussion raises the question not only of whether action information plays a role in the concepts activated by auditory cues, but also of how special words are (that is, how effective respect to other types of cues) when controlling for action information. Moreover, since verbs are words that carry action information, and in that respect are in principle more similar to environmental sounds in the information they activate, a closely connected question is how effective verbs are when compared to sounds. Thus, our understanding on how words and non-verbal sounds activate different mental representations can be further refined and improved by explicitly controlling for the action content in sounds and imagery used in a verification task, and by testing not only nouns and non-verbal sounds, but also verbs as auditory cues.

Hence, to test the effect of action information, we use an image verification task in which images and auditory cues were congruent or incongruent with respect to specific actions. For instance, a bark sound is action-congruent with the image of a dog in the act of barking, but it is action-incongruent with the image of a sleeping dog. The efficacy of sounds, according to Lupyan and collaborators' hypothesis (e.g. Lupyan and Thompson-Schill, 2012; Edmiston and Lupyan, 2015; Boutonnet and Lupyan, 2015), should be sensitive to action-congruence, but the efficacy of words should not. However, if words, as we have documented, activate specific action information, it is possible that the efficacy of words also vary depending on action-congruence. Verbs (at least some of them) are words specifically devoted to referring to actions. We test verbs by manipulating the congruence between verbs and images.

We hypothesize that action information is part of the conceptual representation activated by both verbs and environmental sounds. Consequently, action congruence between auditory cues and images should modulate verification times. Nouns, which activate more stable conceptual features, should show a different pattern of sensitivity to action congruence.

Therefore, in order to confirm that words activate special types of representation (contrasting with the specific concepts activated by sounds) we should expect that nouns result in faster verification responses (compared to non-verbal sounds): 1) when presented with incongruent action images, 2) when presented with figurative images and 3) at least similar verification speed when presented with congruent action and realistic images. Regarding verbs, these should cause a faster verification response when images of congruent actions are presented, independently of the depiction.

The experiment was an adaptation of image verification experiments previously conducted by Lupyan and collaborators (Edmiston and Lupyan, 2015; Lupyan and Thompson-Schill, 2012).

We manipulated the congruence between the action information in images and auditory cues; we also manipulated congruence in depiction and used verbs as an additional type of cue. In these manipulations we implemented consistent criteria for sound-image congruence, considering the specificities of images. This had the aim of systematically controlling for relevant features such as action information type of image, and a systematic comparison of nouns, non-verbal sounds and verbs.

Participants

Thirty-seven undergraduate students (24 women; age range 18–22 years, M = 19.4, SD = 1.1) from the Faculty of Human Communication at the Autonomous University of the State of Morelos participated in the experiment. All participants were native speakers of Spanish, reported normal or corrected-to-normal vision and hearing, and provided written informed consent. The study was approved by the university's ethics committee and followed the Declaration of Helsinki guidelines. Participants received course credit for their participation. Accuracy below 75% led to exclusion; however, no participant met this criterion.

The sample size was determined based on previous studies using similar paradigms, which typically include comparable numbers of participants. No formal a priori power analysis was conducted. However, given the within-subject design and the large number of trials per participant, the study provides substantial statistical power for detecting effects at the trial level using linear mixed-effects models.

Materials

Auditory stimuli

Auditory stimuli were created from the basic categories wolf, horse, dog, lion and person. These categories were selected making sure that they have associated auditory cues in the form of nouns, verbs, and non-verbal sounds. For instance, the auditory cues created from the category dog were the noun “Dog”, the verb “Bark” and the characteristic sound of a bark. To test the effect of action information, in addition to the usual basic level categories, in the category person we included four variations consisting of a person performing acts that also have the three types of recognizable auditory cues. The four act-variations comprised a person yelling, clapping, whistling and coughing.

Nouns. The noun cues comprised the words wolf, horse, dog, lion and person.

Verbs. The verb cues comprised the words howl, neigh, bark, roar, yell, whistle, clap and cough.

Sounds. The non-verbal sounds were characteristic and recognizable sounds of the following actions: howl, neigh, bark, roar, yell, whistle, clap and cough.

Labels (nouns and verbs) were produced by a female voice that spoke native Spanish. Non-verbal sounds were obtained from free online digital sound libraries. Auditory cues were normalized in volume and equated to 800 ms in duration, except for the words howl (1,000 ms), horse (1,100 ms), neigh (1,100 ms), and the non-verbal sound for howl (1,000 ms) and neigh (1,100 ms). These exceptions were made to make sure that all the words and sounds were equally intelligible.

Visual stimuli

Thirty-two images were selected. Images were classified according to the combination of two dimensions of congruence with auditory cues: action congruence (the image is congruent with the action producing the sound) and depiction (the image is realistic or figurative). Thus, four groups of eight images depicting actions were created. The four groups, characterized by the combination of the two dimensions, are as follows: realistic depiction and action congruence (8), realistic depiction and action incongruence (8), figurative depiction and action congruence (8) and figurative depiction and action incongruence (8). See Figure 1, for example.

Realistic vs. figurative depiction. This dimension refers to the style in which exemplars are represented in images. It is similar to the controls utilized in Edmiston and Lupyan's (2015) Experiment 3, in which the experimenters asked on-line participants to evaluate the congruence between sound and image. One of the dimensions in which evaluators seemed to rate the congruence of images was depending on whether the image was a realistic photograph or a cartoon drawing (see Experiment 3, Edmiston and Lupyan, 2015, and their Supplementary material).

Based on the latter, we controlled whether the image of the exemplar, (e.g., a lion), is presented in a realistic style (e.g., a photograph of a lion) or in a figurative style (e.g., a cartoon drawing of a lion). Realistic depiction is considered the congruent type of depiction, while figurative depiction is the incongruent type of depiction. By utilizing this second dimension of congruence, we have a common ground with previous studies, and an additional variable to observe the results of manipulation.

Realistic depiction images were downloaded from the Internet. Figurative style images were purposely designed for this study. Images were edited to have a white background that allowed the agent to be visible and without visual distractors. For the figurative depiction, images were purposely drawn and digitally designed for this study with color with clip studio paint software, procuring that the design of the drawn images were equal in the way the action was presented to the photographs. Similar dimensions, angles, positions and colors were considered for each image and photograph of the same agent for all categories.

Sound-action congruence. We use two realistic images (photographs) per category. One of the two images shows an agent performing an action congruent with a sound or a verb. For instance, the image features a lion in the act of roaring, which is congruent with the roar sound and with the verb roar. The other image shows the agent performing an action that is not congruent with the sound or verb. For instance, the image features a lion sleeping, which is not congruent with hearing a roar sound nor with the verb roar. We also used two figurative images showing cartoon drawings of the agents. Following the same logic as with the realistic images, two cartoon images were created, showing congruent or incongruent action. For instance, one cartoon drawing of a lion roaring and one of a lion sleeping. Thus, sound-action congruence/incongruence refers to an image in which the agent is presented performing an action that corresponds to a given verb and non-verbal sound (e.g., lion roaring while hearing a lion's roar or the verb roar) or an agent performing an action different to the one that corresponds to the verb or non-verbal sound (e.g., lion sleeping while hearing a lion's roar or the verb roar).

Stimulus validation

Imaginability. To establish the reliability of the congruence between image and verb cue, an imaginability test was carried out. Twenty-five participants were presented with a verb cue following a question that asked them to imagine the image that corresponded to the action that was previously mentioned. Subsequently, an image was shown, and participants were asked to evaluate the resemblance between the image and what they have imagined, by rating images in a scale from 1 (least resemblance) to 5 (most resemblance).

Familiarity. Words and sounds were presented to 25 participants, and they were asked to rate their familiarity with such stimuli on a scale from 1 to 10.

Discriminability. A discrimination test was performed for both sounds and for images independently. To verify that subjects specifically knew what action corresponded to the non-verbal sound, 25 participants were presented with a sound and were asked to name the action that resembled the sound. A different sample of 15 participants evaluated discriminability of images. Participants were presented with an image and were asked to name the action that resembled the image.

Experimental design

The experiment was an image verification task. Three kinds of auditory cues levels, namely, nouns, non-verbal sounds and verbs, were tested. Visual stimuli comprised the 32 images described in the previous section, varying in the two dimensions of action congruence and depiction.

Each participant completed 240 trials: eight categories × two levels of action congruence × two levels of depiction × three cue types (noun, non-verbal sound and verb) × two repeats.

Procedure

Participants completed the image verification task in a quiet, dimly lit room. On each trial of the verification task each participant listened to a cue signal followed by an image. Each participant was instructed to decide whether the image they saw was associated to the sound they previously heard. A laptop with a touchscreen was used to observe and interact during the task by touching the laptop monitor. Each participant sat at about 15” from the laptop monitor. Cue sounds were heard via headphones.

All trials began with a 250 ms fixation cross, following immediately by an auditory cue. The image appeared in the center of the monitor 400 ms after hearing the auditory cue and remained in the monitor for up to 4 s. Each participant completed 240 trials.

Participants were instructed to decide whether the image corresponded to the auditory cue, responding “yes” when the image was related to the sound and “no” when it was not. Although no explicit instructions were given regarding the level at which this correspondence should be evaluated, the structure of the task constrained the interpretation of “relatedness.”

Match trials always involved images depicting the same referent associated with the cue (e.g., a lion following the noun “lion,” the verb “roar,” or a roaring sound), regardless of the action being performed or the style of depiction. Thus, images showing the same object in different states (e.g., a sleeping lion or a roaring lion) were consistently treated as matches.

Mismatch trials, by contrast, involved images from a different object category (e.g., a dog), ensuring that they were not related to the cue. Therefore, although instructions were general, the design ensured that match vs. mismatch decisions were based on referential correspondence and were comparable across cue types, rather than on the presence or absence of action information.

Participants were instructed to touch the bottom-left quadrant of the touchscreen when images and cue sounds matched, and the bottom right quadrant when image and cue mismatched

Analyses were performed based only on the correct responses of the matching trials that met the following criteria: RTs should not be longer than 2,500 ms (1% of the total) nor shorter than 250 ms (all correct trials met these criteria). Table 1 shows the descriptive statistics of the RTs across conditions.

Reaction times (RTs) were analyzed in a linear mixed-effects model using R (R Core Team, 2025) and the R packages lme4 (Bates et al., 2015). Table 2 shows the summary of the linear mixed-effects model fitted to log-transformed RTs (logRT). The model included the factorial interaction between cue type (noun, sound, verb), action (congruent, incongruent), and depiction (realistic, figurative) as fixed effects. Random intercepts were specified for participants (ID) and stimuli (stimuli). The model was fitted using maximum likelihood estimation. The intercept of the model corresponds to log-transformed reaction times for noun cues in action-incongruent trials with figurative depictions, reflecting the chosen reference levels for all categorical predictors.

The final dataset consisted of 4,874 observations, obtained from 37 participants. Model fit indices indicated adequate fit (AIC = 530.6; BIC = 627.9; logLik = −250.3). Inspection of the residuals did not reveal major deviations from normality. Random-effects estimates showed substantial variability at both the participant level (SD = 0.023) and the stimulus level (SD = 0.068), in addition to residual variance (SD = 0.003).

The analysis revealed a significant main effect of cue. Relative to nouns, both sounds (β = 0.101, SE = 0.022, t = 4.55, p < 0.001) and verbs (β = 0.085, SE = 0.024, t = 3.60, p < 0.001) were associated with slower reaction times in the reference condition (incongruent actions and figurative depictions). This indicates that, under conditions of action incongruence, noun cues elicited faster responses than either sound or verb cues.

There was no significant main effect of action (β = 0.032, SE = 0.034, t = 0.96, p = 0.345) indicating that, when nouns serve as the reference signal, reaction times did not differ globally between congruent and incongruent trials. Similarly, the main effect of depiction did not reach conventional significance levels β = −0.061, SE = 0.033, t = −1.84, p = 0.075), although a trend toward faster responses for realistic depictions was observed.

Interaction effects

A significant interaction between cue and action was observed, indicating that the effect of action congruence on visual processing was statistically different as a function of cue type. Specifically, the interaction between sound cues and action was highly significant (β = −0.190, SE = 0.028, t = −6.84, p < 0.001), as was the interaction between verb cues and action (β = −0.159, SE = 0.029, t = −5.50, p < 0.001).These effects indicate that action congruence modulated reaction times differently across cue types. Given that incongruent action is the reference level, these coefficients indicate a larger reduction in reaction times from incongruent to congruent trials for sounds and verbs relative to nouns. In other words, while nouns showed minimal sensitivity to action congruence, both sounds and verbs exhibited marked facilitation when the action implied by the cue was aligned with the visual stimulus.

The interaction between cue and depiction further revealed dissociable effects across signal types. For sound cues, the interaction with depiction was not significant (β = 0.009, SE = 0.031, t = 0.28, p =0.783), indicating that reaction times to sounds were not reliably modulated by whether the depiction was figurative or realistic.

In contrast, the interaction between verb cues and depiction was statistically significant (β = 0.118, SE = 0.033, t = 3.59, p < 0.001). This interaction indicates that, relative to noun cues, verb processing was differentially affected by depiction. Inspection of the estimated marginal means showed that verb cues were processed more slowly in realistic than in figurative depictions, particularly under action-incongruent contexts.

Importantly, this effect does not reflect a general slowing for realistic depictions, but rather a selective sensitivity of verb cues to changes in depiction.

The interaction between action and depiction was not significant (β = 0.010, SE = 0.047, t = 0.20, p =0.840), indicating that depiction did not modulate the effect of action congruence in a uniform way across cue types. Accordingly, this interaction is not discussed further.

The three-way interactions involving cue × action × depiction did not reach conventional levels of statistical significance. The interaction was marginal for sound cues (β = 0.068, SE = 0.039, t = 1.76, p =0.078) and for verb cues (β = −0.077, SE = 0.040, t = −1.93, p =0.054).

Estimated marginal means and the interaction plots indicated that depiction-related differences for verbs were most pronounced under action-incongruent conditions, whereas noun responses remained comparatively stable across both action and depiction manipulations.

Summary of results

In sum, the results demonstrate that the influence of action congruence on reaction times is strongly modulated by cue type. Responses to images with incongruent action information show that nouns have a significant advantage in facilitating image recognition over verbs and sounds (nouns are the most effective facilitators). These results replicate previous findings in (Lupyan and Thompson-Schill 2012). Recall that such results suggested that words seem to activate special types of representations. Contrastingly, results do not show a systematic advantage of nouns over non-verbal sounds and verbs when the materials exhibit action congruence. Sound and verb cues, but not noun cues, showed substantial facilitation in congruent action contexts. Additionally, verb cues exhibited a selective sensitivity to depiction. No evidence was found for a general effect of depiction or for a uniform interaction between action congruence and depiction across signal types (see Figure 2). We further discuss these findings in the following section.

The present study investigated how different types of pre-visual cues—nouns, verbs, and non-verbal sounds—modulate visual processing as a function of their congruence with action-related information implied by the target stimulus. While previous work has demonstrated that linguistic labels can facilitate visual recognition and categorization, and that words often outperform non-verbal sounds in such tasks, the present results show that this advantage is not unconditional. Instead, facilitation depends critically on the type of conceptual information activated by the cue, particularly with respect to action.

The present findings shows that the relative effectiveness of different auditory cues in visual verification depends critically on the alignment between the type of information activated by the cue and the information available in the visual stimulus. Complementary to the “label advantage” view—which proposes that nouns facilitate recognition because they activate abstract, category-level representations (Lupyan and Thompson-Schill, 2012)—our results reveal that this facilitation is conditional: nouns show a clear advantage when the visual stimulus lacks action information. When images include explicit action cues, the advantage shifts toward verbs and environmental sounds. Aligned with previous studies, our results suggest that words also activate fine-grained perceptual information (Noorman et al., 2018). These findings are consistent with the view that conceptual representations are composed of multiple informational components, and that perceptual facilitation depends on the alignment between the information activated by a cue and the structure of the perceptual input. Nouns tend to activate information relevant for identifying object categories and stable features, whereas verbs and sounds preferentially activate dynamic, event-related representations. Under action-incongruent conditions, dynamic representations prompted by sounds or verbs conflict with the static nature of the image, reducing facilitation. Under action-congruent conditions, however, these dynamic representations provide a better alignment to the visual information, leading to faster verification.

The results also refine interpretations from previous studies. (Edmiston and Lupyan 2015) and (Lupyan and Thompson-Schill 2012) concluded that words possess a privileged representational status, but their materials did not systematically control for action content. Our data suggest that when action information is carefully manipulated, the “power of words” reflects not a unique representational format but the degree of alignment between the cue-driven informational components and the structure of the visual stimulus. The noun advantage disappears when action cues are present, indicating that nouns are not universally superior predictors but rather strong priors whose usefulness decreases when visual evidence strongly favors dynamic interpretations.

Verbs, which have been underexplored in cue–object verification tasks, showed patterns resembling those of environmental sounds. Verbs and sounds showed comparable patterns of facilitation when action cues were congruent, a result that challenges the assumption that lexical cues invariably outperform non-verbal cues. This suggests that verbs activate sensorimotor information that overlaps with the action information encoded in environmental sounds, consistent with evidence showing shared cortical recruitment for linguistic and non-linguistic action representations (Frank et al., 2015). Despite their different symbolic status, both cue types showed strong facilitation when action information was congruent and comparable costs when it was incongruent. Action-related sounds have been shown to activate motor and premotor regions associated with the corresponding actions (Lewis et al., 2005; Kiefer et al., 2007), while verbs—particularly action verbs—are known to recruit sensorimotor representations during comprehension (Hauk et al., 2004; Kemmerer et al., 2008). From this perspective, both verbs and sounds may function as effective cues for anticipating upcoming action-relevant visual information, thereby facilitating perceptual processing when expectations are met.

Representation activated

The interaction between cue type and action congruence is consistent with the idea that auditory cues recruit different representational systems depending on their semantic and grammatical properties. Nouns appear to activate stable, taxonomic information that supports rapid categorization when perceptual input is ambiguous or lacking dynamic structure. Verbs and sounds, by contrast, activate dynamic representations associated with movement, energy exchange, and event structure. Such representations align naturally with visual depictions of action but may interfere with static images, producing slower responses under incongruence. These findings are consistent with long-standing distinctions in psycholinguistics between nouns and verbs, particularly with respect to their semantic organization. Nouns are typically associated with object identity and categorical structure, whereas verbs are more tightly linked to relational and action-based information (Gentner, 1982; Kousta et al., 2011).

Also, this interpretation is consistent with accounts of semantic representation in which conceptual knowledge is composed of multiple types of information, including both experiential and language-based components (Huettig and Altmann, 2005; Ferreira et al., 2008; Vigliocco et al., 2009, 2004), Within this framework, nouns tend to emphasize stable, category-level information, whereas verbs and environmental sounds preferentially activate dynamic, action-related information. From this perspective, the present results suggest that facilitation emerges when the type of information activated by the cue matches the information required by the visual stimulus. When images contain action-related structure, cues that activate dynamic information (verbs and sounds) provide a better match and facilitate processing. When such information is absent or incongruent, cues that emphasize stable object-level information (nouns) may provide a relative advantage.

Importantly, although participants were not explicitly instructed to evaluate correspondence at a specific representational level, the structure of the task constrained match decisions to referential relatedness across all cue types. That is, images depicting the same object were consistently treated as matches regardless of the action being performed, whereas images from different object categories were treated as mismatches. This reduces the likelihood that the observed interaction between cue type and action congruence is driven by differences in decision criteria across conditions and instead supports the interpretation that the effects arise from differences in the type of information activated by each cue and its alignment with the visual stimulus.

The role of depiction

The present experiment included images in a figurative depiction (and realistic depiction). However, interactions involving this factor did not alter the core pattern of results The differential sensitivity of nouns, verbs, and sounds to action congruence was consistent across depiction types, suggesting that the observed effects primarily reflect conceptual activation processes rather than low-level visual properties.Inspection of the estimated marginal means suggested an additional pattern that, while not statistically significant, is theoretically informative. Specifically, in action-incongruent trials, non-verbal sounds tended to elicit slower responses when paired with figurative depictions than with realistic depictions, whereas noun-based cues showed relatively stable response times across depiction. This descriptive pattern is consistent with previous proposals suggesting that linguistic labels, as unmotivated cues, stabilize perceptual representations, whereas non-verbal sounds remain more tightly coupled to perceptual and contextual variability (Edmiston and Lupyan, 2015; Lupyan and Clark, 2015). However, interactions involving depiction did not reach statistical significance in the present data, and this observation should therefore be interpreted with caution and future studies should focus on refining these kinds of distinctions.

Conclusion. What kind of low-level information is activated by words and sounds?

Overall, the study demonstrates that action information plays a central role in modulating the effectiveness of auditory cues in visual verification. Words are not inherently superior to sounds; instead, their facilitation depends on whether the conceptual information they activate matches the visual information available. Nouns excel when images lack action cues, whereas verbs and sounds dominate when action cues are present.

These findings refine previous interpretations of label-driven facilitation and highlight the need for experimental paradigms that explicitly control for the action content of both cues and images. More broadly, they support the view that conceptual activation is flexible, multimodal, and shaped by the alignment between linguistic input and perceptual structure. Future work should examine the temporal dynamics of these interactions and explore whether similar patterns emerge with other grammatical categories and more complex event structures.

Moreover, we suggested that there may exist different types of representations associated to different types of words, and that their activation may occur to achieve the different linguistic functions associated to the different types of words. This, however, does not preclude the possibility that some of the representations associated to words may be similar to representation activated by non-verbal sounds. In this sense, words are not especially distinct from sounds, but rather words may simply co-opt representations adequate to the perform their functions.

Future directions

The present study focused on a specific task and a restricted set of action-related stimuli. Future research could examine whether similar patterns emerge for other conceptual dimensions, such as spatial relations or functional affordances, or in tasks emphasizing learning and adaptation.

Additionally, combining behavioral paradigms with neurophysiological measures could help determine whether the observed convergence between verbs and sounds reflects shared neural substrates or parallel processing routes.