Abstract, emotional and concrete concepts and the activation of mouth-hand effectors

Method

Materials

We selected 90 Italian words from the Della Rosa et al. database (Della Rosa et al., 2010), composed by 30 concrete words, 30 abstract and 30 emotional words. The selected words were balanced in Familiarity (mean = 590; SD = 148.09). We considered as concrete words the words that scored high on Concreteness (max. 700, min. 596; mean of Concreteness = 660.42; mean of Abstractness = 138.87), and considered as abstract the words with high Abstractness scores (max. 635, min. 375; mean of Abstractness = 535.24; mean of Concreteness = 213.28). We considered as emotional words those words that had intermediate scores of Abstractness and Concreteness (mean of Concreteness = 400; mean of Abstractness = 372.74) and that according to the experimenters had high emotional valence.

Because the emotional value was not present as a parameter in Della Rosa et al. (2010), we performed an on-line pre-test in which 25 participants (13 females; mean of age = 30.6; SD = 14.5) judged the emotional value of each word on a 7-points Likert scale (1 was rated as non-emotional and 7 as completely emotional). Since in the literature it is debated whether emotional words can be considered as a subset of abstract concepts or represent a kind of concept different from both concrete and abstract concepts (Altarriba, Bauer & Benvenuto, 1999; Kousta et al., 2011), the pre-test also aimed at clearly distinguish abstract, concrete and emotional words, to avoid overlaps between abstract and emotional words.

Among the original 90 words we selected 48 words: 16 more concrete, 16 more abstract and 16 rated as more emotional. Characteristics of the three categories of selected words in terms of dimensions and psycholinguistic variables are shown in Table 1(A).

We performed a T-Student test for independent samples by items and we calculated the Effect size (Cohen’s d), in order to verify if concrete, abstract and emotional words we chose differed in Concreteness, Abstractness and Emotional value.

All the categories (abstract, concrete and emotional words) resulted to differ in Concreteness (p_s < .001), Abstractness (p_s < 001) and Emotional value (p_s < .001). Emotional words were rated as more emotional than both abstract and concrete words (p_s < .001); also abstract words were rated as more emotional than concrete words (p < .001).

We also verified whether abstract, concrete and emotional words differed along a series of dimensions that, according to different theories, are considered crucial to distinguish concrete and abstract words, i.e., Imageability (IMG; Paivio, 1986), Age of Acquisition (AoA; Gilhooly & Logie, 1980), Context Availability (CA; Schwanenflugel, Akin & Luh, 1992) and Modality of Acquisition (MoA; Wauters et al., 2003). We decided to avoid controlling for these variables while selecting abstract and concrete words, because this would result in using a very reduced number of abstract and concrete concepts. More crucially, it would lead us to use concepts that are weird and scarcely familiar. We instead decided to select “good” abstract and concrete words: when compared to concrete words, “good” abstract words are typically less imageable (IMG), they are acquired later (AoA) and mainly through language rather than through perceptual modalities (MoA), and they are less associated to contexts (CA) (see e.g., Borghi et al., 2018b; C Villani, L Lugli, MT Liuzza, AM Borghi, 2018, unpublished data).

The abstract, concrete and emotional words we selected thus differed not only in Concreteness and Abstractness but also in Imageability, Age of Acquisition, Context availability, and Modality of Acquisition.

As to psycholinguistic variables, we took care in balancing words for Familiarity, thus the three categories didn’t differ on this dimension (all p_s = .9).

The abstract and emotional words we selected were more frequent than concrete ones (p_s < .05). We chose highly frequent abstract words because we wanted to avoid stimuli that were unfamiliar for participants which could have lead to abstractness driven effects. Emotional and abstract words did not differ for frequency (p = .4).

Concrete words were also shorter than abstract and emotional words (p_s < 04), while emotional and abstract words did not differ (p = .2).

Results of the comparisons between the three categories of selected words in terms of psycholinguistic variables and of all the relevant dimensions are shown in Table 1(B).

The effect of these variables on the results was determined through Generalized Linear Mixed Models (GLMMs) performed on both experiments (1 and 2) and tasks (Lexical Decision and Recognition) with accuracy and RTs as dependent variables. Together with Type of Concept and Effector as main factors, we added Imageability, Age of Acquisition, Contex Availability, Modality of Acquisition, Frequency and Number of Letters as covariates (see ‘Results’ section).

To complete our stimuli we subsequently added 48 pseudowords, created by modifying one letter at the beginning, in the middle or at the end of concrete, abstract and emotional words in the same proportion as the critical words. Then we created 24 words to be used as catch-trials: they were Italian words with a bold letter, at the beginning, in the middle or at the end of the word. Finally, we selected 24 new words, maintaining the proportion between abstract, concrete and emotional words for the recognition task. Words that can directly activate hand or mouth (e.g., tools or food related words) were excluded from the list. Stimuli are shown in Table 2.

The experiment consisted of two tasks, a lexical decision task and a recognition task, that were presented in sequence; the lexical decision task always preceded the recognition one. Two separate lists of words were created for the two tasks: for the lexical decision task 24 critical words (8 concrete, 8 abstract, 8 emotional), and 24 pseudo-words. For the recognition task list, 24 critical words (8 concrete, 8 abstract and 8 emotional) and 24 new words were used.

Procedure

Participants were tested individually, and were instructed to respond as quickly and accurately as possible to each trial using a response box connected with a pedal and a button (see Fig. 1A). They were given the instructions on the computer screen and were trained at the beginning of every task. In no case further instruction from the experimenter was needed; she only needed to specify how to use the button for the mouth responses, and she made sure that participants used their dominant hand for hand’s responses. Participants were not aware of the subsequent recognition task; during the instructions they were just informed that they would complete an experiment composed by two phases. Only the participant and the experimenter were present in the room; after the training the experimenter sat outside the lab in order to avoid any kind of interference with the experiment. Testing took place on a Pc (resolution: 1,024 × 768 pixel) running EPrime2 Professional software. Participants sat on a comfortable chair in front of a computer screen, at a distance of about 60 cm.

Words appeared at the center of a computer screen at maximum 6.20° of visual angle. Each trial began with a centred black fixation cross for 500 ms, followed by the presentation of the word. Words remained on the screen for a time of maximum 1.5 s. After 1 s the next trial started (see Fig. 1B).

Lexical decision task. The task was divided into two experimental blocks, each preceded by a training block of 12 trials (6 words and 6 pseudo-words). Depending on the block, participants kept the button to be pressed with their dominant hand or with the mouth. The order of the blocks was counterbalanced across participants. A set of 48 words was presented on the computer screen (24 critical words, 24 pseudo-words). Participants were asked to press the button with the hand or with the mouth, depending on the instructions, if they read an Italian word, and to refrain from responding if the word they read was not an Italian one.

Recognition task. The task was divided into two experimental blocks, each preceded by a training block of 6 trials (3 words, 3 new words). Depending on the block, participants were required to keep the button in their dominant hand or in the mouth, between the teeth. A set of 48 words was presented in each block, composed by 24 critical words and 24 new words. The order of the blocks was counterbalanced across participants. Participants were asked to press the button with the hand or with the mouth, depending on the condition, in case they recognized the word on the screen as a word already presented in the previous task, or to refrain from responding if they read a new word.

Results

The data were analyzed with Generalized linear mixed models (GLMM). Generalized linear mixed models (GLMM) incorporate random effects into the linear predictor of a generalized linear model (GLM). They include fixed linear predictor variables, within subject measures and random variances to account for cluster-related correlations in the data. GLMMs consider all the available data points, including those which occasionally failed on some trials. Furthermore, GLMMs can handle non-normal outcome distributions, which is important to account for accuracy, as it is likely to be highly skewed (Hedeker, 2005).

The effects of Type of Concept (abstract, concrete and emotional), of the Effector (mouth vs. hand) and of their interaction were analyzed on the response times and on the accuracy score (error) with two types of GLMM using Frequency, Number of Letters, Imageability, Age of Acquisition, Context Availability and Modality of Acquisition as covariates. We introduced these covariates to be sure that they did not affect our main results—for example, that the differences we found in processing the kinds of concepts were not due to a linguistic dimension or due to their different frequency.

In the first type of GLMM model, the response time was assumed to be normally distributed, therefore the Identity link function was used, whereas in the second type of model we adopted a logistic function because the accuracy was considered as having a binomial distribution (presence/absence of error). The two types of models were applied for the Lexical Decision and Recognition tasks of both Experiments 1 and 2.

Table 3 displays the GLMMs results in the different experiments showing the effect of Type of Concept (abstract, concrete and emotional) and the Effector (mouth vs hand) and their interaction as well as the covariates effect on the response time and accuracy.

Means of response times as a function of Type of Concept and Effector for both tasks (Lexical Decision and Recognition) and experiments (1 and 2) are displayed in Table 4.

Lexical decision task. All erroneous trials (2.96%) were removed before the analysis of RTs. In the lexical decision task, we found that the factor Type of Concept affected the response time [Wald (2) = 11.04; p = .004]. Abstract concepts (mean: 738.59 ms, SE = 10062.11) were responded to slower than emotional ones (mean: 697.27 ms, SE = 10058.79) and concrete ones (mean: 685.18 ms, SE = 10048.57). The analysis also showed an advantage for the hand (mean: 669.44 ms, SE = 10056.26) over the mouth effector (mean: 744.58 ms, SE = 10056.70) [Wald (1) = 77.43; p < .001].

All these effects are corrected for Frequency, Number of Letters, Imageability, Age of Acquisition, Context Availability and Modality of Acquisition, however only Frequency [Wald (1) = 33.84; p < .001; B = −.51], Number of Letters [Wald (1) = 69.04; p < .001; B = 18.36] and Age of Acquisition [Wald (1) = 9.46; p = .002; B = .31] affected the time of response in the lexical decision task. The predicted interaction between the two main factors did not reach the significance.

The factor Type of Concept also affected the accuracy of responses [Wald (2) = 6.82; p = .03]. The mean rates of errors corrected by the covariates were computed from the logistic model for each effect. Mean rates of errors are reported in terms of percentage of errors [abstract = 96%, SE = 13.15; concrete = 99%, SE = 4.87; emotional = 98%, SE = 5.38]. All the effects are corrected for the abovementioned covariates, however only Frequency [Wald (1) = 11.06; p = .001; B = .01] affected the accuracy of responses. No other main effect or interaction reached the significance.

Recognition task. All erroneous trials (28.30%) were removed before the analysis of RTs. In the analysis of response times, we found an advantage of the hand (mean: 726 ms, SE = 14405.92) over the mouth effector (mean: 777.34 ms, SE = 14396.75) [Wald (1) = 22.68; p < .001]. All the effects were corrected using Frequency, Number of Letters, Imageability, Age of Acquisition, Context Availability and Modality of Acquisition as covariates, however only Number of Letters [Wald (1) = 25.17; p < .001; B = 13.60] affected the response time in the recognition task. No other main effect or interaction reached the significance.

Nonetheless, a qualitative inspection of the interaction between the two main factors Type of Concept and Effector revealed that the pattern was in trend with our hypothesis: the advantage of the hand over the mouth responses was namely smallest with abstract words. Emotional words were the most affected by the effector, showing 70 ms of advantage of the hand over the mouth responses. They were followed by concrete words that showed a facilitation of 50 ms of the hand over the mouth responses, while abstract words showed the smallest advantage with the hand (30 ms of difference), as shown in Table 4 and Fig. 2.

In the analysis on the accuracy, the interaction between the two main factors Type of Concept and Effector was significant [Wald (2) = 7.16; p = .02] (see Fig. 3). All the effects are corrected for the abovementioned covariates, however only Number of Letters [Wald (1) = 4.06; p = .04; B = −.05], Frequency [Wald (1) =25.20; p < .001; B = −.006] and Modality of Acquisition [Wald (1) = 6.17; p = .01; B = −.003] resulted to affect the accuracy. No other main effect reached the significance.

Discussion

The results show that in the lexical decision task abstract words were processed slower than emotional words and in trend slower than concrete ones, confirming the well-established concreteness effect (Paivio, 1986). In the lexical decision task, Frequency and Number of Letters, together with Age of Acquisition appeared to be important variables, that may have affected the results since in general the selected abstract words were longer and acquired later. As to emotional words, their pattern seems to be similar to the pattern of concrete words: in general, they yielded shorter response times than abstract words, and they seem to be slightly advantaged in the hand condition.

Across the two tasks, responses with the mouth were slower than those with the hand, independently from the concept kind; this effect is not worth discussing since it was likely due to the fact that the button to hold among the teeth was harder to press than the key to press with the hand (see also Borghi & Zarcone, 2016).

As to the activation of the hand and mouth effectors in relation to the different type of concepts, in the lexical decision task we found no significant interaction. A possible explanation is that effectors are differently activated only in tasks that require a deeper processing level. The interaction between Type of Concept and Effector was instead significant in the recognition task, in the analysis on accuracy, in which the general disadvantage with the mouth effector was less pronounced with abstract words with respect to concrete ones, while emotional words showed an advantage with the mouth effector. It is worth noting that, beside Frequency and Number of Letters, Modality of Acquisition impacted significantly the results on accuracy of the recognition task. This could suggest that, in a task requiring a deeper level of processing the way in which we acquire words is more relevant than other psycholinguistic variables.

Method

Materials

Materials were the same as Experiment 1, except for 16 catch-trials that were added. Catch-trials were Italian words with a bold letter.

As in Experiment 1, the experiment consisted of two tasks, a lexical decision task and a recognition task, that were presented in sequence; the lexical decision task always preceded the recognition one. Two separate lists of words were created for the two tasks: for the lexical decision task 24 critical words (8 concrete, 8 abstract, 8 emotional) and 24 pseudo-words were used. For the recognition task list, the stimuli consisted of 24 critical words (8 concrete, 8 abstract and 8 emotional) and of 24 new words.

Procedure

The procedure was the same as that of Experiment 1.

Lexical decision task. The task was divided into two experimental blocks, each preceded by a training block of 16 trials (8 words and 8 catch-trials). A set of 64 words was presented on the computer screen (24 critical words, 24 pseudo- words and 16 catch-trials). The words were arranged in two different lists, one for each block. Depending on the block, participants were required to keep a button in their dominant hand or in the mouth, between the teeth, and to respond to catch-trials by pressing it; they were instead asked to press the pedal to respond to critical stimuli. The order of the blocks was counterbalanced across participants. Participants were asked to press the pedal if they read an Italian word, and to refrain from responding if the word they read was not an Italian one. They were also required to respond to catch-trials by pressing the button with the hand or mouth, depending on the condition. Hence the mouth and the hand were not the direct response effectors, but depending on the condition either the mouth or the hand were occupied during the execution of the task.

Recognition task. The task was divided into two experimental blocks, each preceded by a training block of 8 trials (3 words, 3 new words and 2 catch-trials). A set of 62 words was presented in each block, composed by 24 critical words, 24 new words and 12 catch-trials. The order of the blocks was counterbalanced across participants. Depending on the block, participants were required to keep the button in their dominant hand or in the mouth, between the teeth. As to the critical stimuli, participants were asked to press the pedal in case they recognized that the word on the screen had already been presented in the previous task, or to refrain from responding if the read a new word. When catch-trials were presented, they had to respond by pressing the button with the hand or mouth, depending on the block.

Results

Lexical decision. All erroneous trials (4.58%) were removed before the analysis of RTs. In the lexical decision task, the factor Type of Concept affected the response time [Wald (2) = 15.56; p < .001], showing that abstract words (mean: 876.14 ms, SE = 10707.86) were processed slower than both concrete (mean: 771.67 ms, SE = 10692.50) and emotional words (mean: 821.95 ms, SE = 10704.06). All the effects are corrected using Frequency, Number of Letters, Imageability, Age of Acquisition, Context Availability and Modality of Acquisition as covariates, however only Frequency [Wald (1) = 10.82; p = .001; B = −.33], Number of Letters [Wald (1) = 33.68; p < .001; B = 13.75] and Imageability [Wald (1) = 7.37; p = .007; B = .31] affected the response time. No other main effect or interaction reached the significance.

In the analysis on accuracy, the hand condition elicited more errors than the mouth condition. The mean rates of errors corrected by the covariates were computed from the logistic model for each effect. Mean rates of errors are reported in terms of percentage of errors (mouth: 95%, SE = 11.28; hand: 97%, SE = 7.09); (Effector [Wald (1) = 4.88; p = .02]. All the effects are corrected for the linguistic dimension of the stimulus, however only Imageability [Wald (1) = 4.24; p = .03; B = −.005] had an impact on the accuracy. No other main effect or interaction reached the significance.

Recognition task. All erroneous trials (26.1%) were removed before the analysis of RTs. In the analyses on response times, no main effect or effects of the covariates reached the significance. However, the interaction between the two main factors Type of Concept and Effector was significant [Wald (2) = 7.72; p = .02] showing the predicted interaction: abstract words were responded to faster with the mouth than with the hand, while concrete and emotional words were responded faster with the hand than with the mouth (see Table 4 and Fig. 4).

In the analysis on accuracy no main effect or interaction reached the significance, nor did the effect of the covariates.

Discussion

Across the two tasks, abstract words were processed slower than both concrete and emotional words. As in Experiment 1, our findings confirm the concreteness effect and at the same time suggest that emotional words cannot be properly assimilated neither to concrete nor to abstract words. In line with the results of Experiment 1, we did not find a significant interaction between Type of Concept and Effector in the lexical decision task. The effect of the covariates Frequency, Number of Letters and Imageability in the lexical decision task, but not in the recognition task, suggests that these parameters could have affected more incisively a relatively shallow task, especially in the speed of processing. In the recognition task, instead, the interaction reached significance in the analysis of response times. Abstract words yielded faster RTs when processed in the mouth condition, while concrete and emotional words were responded to faster in the hand condition, as predicted by the WAT (Words As social Tools) view.

Overall processing differences between concept kinds

Overall, abstract concepts were processed generally slower than concrete ones in the lexical decision task (E1 and E2 lexical decision). Our results confirm the well-established concreteness effect (Paivio, 1986; but see exceptions to the effect when controlling stimuli for valence: Kousta et al., 2011; Barca, Burani & Arduino, 2002), that shows that abstract words are slower than concrete ones, and extended it showing that in the lexical decision task they are also slower than emotional words (see also Ponari, Norbury & Vigliocco, 2017). Interestingly, however, such effect reached significance only in the lexical decision task.

As to emotional words, our results cast doubts on the assimilation of emotional to abstract concepts: across experiments and tasks, the pattern of responses elicited by emotional words differed from that of abstract words and occasionally from that of concrete words too. In the lexical decision task of both experiments emotional words were processed faster than abstract words and did not significantly differ from concrete words. Our results are in line with those of a study by Siakaluk et al. (2016) showing that valenced words were processed faster than other words in lexical decision task. Overall, emotional words differed in processing from both concrete and abstract concepts, confirming the views according to which they represent a third kind of concept (Altarriba, Bauer & Benvenuto, 1999; Setti & Caramelli, 2005).

Processing differences between concept kinds in relation to the effectors

The experiments we designed were driven from the hypothesis that abstract concepts would activate more the mouth motor system. Furthermore, we wanted to explore whether the two effectors, mouth and hand, would be differently activated with emotional words. We thus expected to find a Type of Concept x Effector interaction.

If we consider lexical decision, in neither experiment was the predicted Type of Concept x Effector interaction significant. Results thus seemed to suggest that the lexical decision task did not lead to a differential activation of the hand and mouth effectors depending on the concept kind, likely because of the superficial processing level it implied. This interpretation is supported by the fact that in the lexical decision task typical psycholinguistic variables such as Frequency and Number of Letters resulted to impact the speed of responses in both Experiments and also the accuracy in Experiment 1.

The results consistently differed if we consider the Recognition task. In Experiment 1 the interaction reached significance in accuracy and was present in trend in RTs, while in Experiment 2 the interaction was significant in RTs. In Experiment 1, the general disadvantage caused by the mouth device was smaller for abstract words with respect to concrete words in both the accuracy and the response times. In Experiment 2, in which the hand and mouth were occupied but responses were provided in the same manner, namely pressing a pedal with the foot, responses to abstract words were faster in the mouth than in the hand condition, while the opposite was true for both concrete and emotional words. These results clearly confirm our hypothesis, indicating that abstract words were facilitated when the mouth was activated, and extend previous results, showing that such a facilitation occurred not only when the mouth was the direct response effector but also when the mouth was occupied with a device.

This confirms the predictions of the WAT proposal, according to which abstract concepts re-enact linguistic and social experience more than concrete concepts, hence determining a higher activation of the mouth. Three possible mechanisms can underline this activation (see for further discussion Borghi & Zarcone, 2016; Borghi et al., 2018b): (a) the re-enactment of the acquisition experience, which is mainly linguistic and occurs in a social context; (b) the inner speech used to re-explain to us the meaning of abstract concepts; (c) the meta-cognitive awareness that our conceptual knowledge is inadequate followed by the motor preparation to ask to others information on words meaning (social-metacognition, Borghi et al., 2018b). The present study does not allow us to determine which of the three mechanisms is responsible of the effects; further research is needed in order to disentangle them.

The pattern of results of emotional concepts was also rather consistent across the two experiments, and clearly different from that of abstract concepts. In the recognition task of both Experiments 1 and 2, emotional words were processed slightly slower in the mouth than in the hand condition, differently from abstract words. Finally, the fact that abstract but not emotional words selectively engage only the mouth effector is in keeping with recent experimental results. Ratings results showed that emotional concepts activate both the mouth and the hand effectors, while mental states concepts activate more selectively the mouth (Ghio, Vaghi & Tettamanti, 2013). fMRI results clearly demonstrated that while the face/mouth motor system in the brain is more activated by “pure” abstract concepts as mental state concepts than by emotional ones, which activate hand and face motor cortex to similar degrees (Dreyer & Pulvermüller, 2018). Overall, this finding is in line with views according to which no strict dichotomy between abstract and concrete concepts exists. Our results rather suggest that emotional concepts cannot be assimilated neither to concrete nor to abstract concepts, (see Barca, Mazzuca & Borghi, 2017; Mazzuca, Barca & Borghi, 2017, for further discussion), and are in line with the proposal according to which emotional concepts, being more grounded than other abstract concepts, provide a bootstrapping mechanism to learn them (Ponari, Norbury & Vigliocco, 2017).

While we found that abstract concepts processing was facilitated with the mouth, the results are less marked than in a previous study (Borghi & Zarcone, 2016). We ascribe this difference to two factors: first, to the fact that in the previous study participants were provided with a context and not only with single words, and second, to the fact that the task was a deep processing one.

It can be objected that effectors effects have previously been found with lexical decision tasks. However, it is difficult to directly compare our results with those of previous studies with lexical decision, such as those by Pulvermüller (2005) and Neininger & Pulvermüller (2003), because these studies employed as stimuli action words directly related to the effectors (Pulvermüller, 2005) and strongly perceptual related words (Neininger & Pulvermüller, 2003).

Even if we did not find the predicted interaction with effectors, our findings are in line with those of previous lexical decision studies with concrete and abstract words. For example, Dreyer et al. (2015) comparing abstract emotional words with tool, food, animal words and effector related words, showed that in the control group abstract-emotional words yielded longer RTs than food and animal related words, while there was no difference in the accuracy. Moreover, in the same study (always in the control group), Dreyer et al. found that hand-related verbs were processed faster than verbs related to face and leg. Our results on lexical decision are in line with the evidence reported, showing a concreteness effect and a general advantage of responses given with the hand.

In sum, the present study adds important information to previous studies on concepts and effectors activation: it suggests that the mouth and hand effectors can be differently activated depending on the task and on the depth level it implies. The different effectors did not influence results in the lexical decision task, but they had an impact on a subsequent recognition task.