5.2.1 Specific Kinds of Abstract Concepts

This section overviews the specificities of the various types of concept. According to Conca, Borsa et al. (Reference Conca, Borsa, Cappa and Catricalà2021), the four concepts that recur across studies are emotional, mental states, social, and numerical concepts. The most often studied are emotional concepts, while researchers often investigate numerical concepts per se rather than comparing them to other concepts.

Emotional words. Whether emotional concepts are abstract concepts or a third, independent category is currently debated (Altarriba et al., Reference Altarriba, Bauer and Benvenuto1999; Altarriba & Bauer, Reference Altarriba and Bauer2004; Kazanas & Altarriba, Reference Kazanas and Altarriba2015; Mazzuca et al., Reference Mazzuca, Barca and Borghi2017). Independently from the effector, in lexical decision – a task in which words must be distinguished from nonwords – emotionally valenced words are processed faster than other words, including other abstract ones, in adults (Siakaluk et al., Reference Siakaluk, Newcombe, Duffels, Li, Sidhu, Yap and Pexman2016; Mazzuca et al., Reference Mazzuca, Lugli, Nicoletti and Borghi2018 ) and children (Ponari et al., Reference Ponari, Norbury and Vigliocco2018; Lund et al., Reference Lund, Sidhu and Pexman2019). The advantage of emotional concepts over other abstract and concrete concepts holds across tasks – associations, ratings, recognition, categorization, and learning tasks – and studies (e.g., Altarriba & Bauer, Reference Altarriba and Bauer2004; Kousta et al., Reference Kousta, Vigliocco, Vinson, Andrews and Del Campo2011; Newcombe et al., Reference Newcombe, Campbell, Siakaluk and Pexman2012; Moffat et al., Reference Moffat, Siakaluk, Sidhu and Pexman2015; Siakaluk et al., Reference Siakaluk, Newcombe, Duffels, Li, Sidhu, Yap and Pexman2016; Mazzuca et al., Reference Mazzuca, Lugli, Nicoletti and Borghi2018; Ponari et al., Reference Ponari, Norbury and Vigliocco2020). Furthermore, with emotional concepts, typically more associated words are produced (e.g., Altarriba et al., Reference Altarriba, Bauer and Benvenuto1999), confirming the richness of their representation.

Social concepts. Feature listing tasks show that social concepts differ from institutional ones (Roversi et al., Reference Roversi, Borghi and Tummolini2013). Results of functional magnetic resonance imaging (fMRI), transcranial magnetic stimulation (TMS), and, more generally, brain imaging studies indicate that social concepts engage different regions, particularly the anterior temporal lobe (ATL), traditionally considered dedicated to social concepts (Mellem et al., Reference Mellem, Jasmin, Peng and Martin2016). They also engage other temporal, occipital, and frontotemporal regions (Conca, Borsa, et al., Reference Conca, Borsa, Cappa and Catricalà2021).

Numerical concepts. I discuss numerical concepts extensively in Chapter 7.2.

Institutional concepts. Institutional concepts, like “norm,” “contract,” and “parliament,” typically refer to institutions or institutional elements, that is, human-made entities that acquire their function through the collective acceptance of some rules (Searle, Reference Searle2010). In a feature production task, my colleagues and I (Roversi et al., Reference Roversi, Borghi and Tummolini2013) contrasted different kinds of concrete and abstract concepts: standard artifacts (e.g., “hammer”), social concepts (e.g., “friendship”), and institutional concepts (e.g., “ownership”). Interestingly, we found that institutional artifacts cannot be assimilated into social entities; differently from the latter, they elicited more normative relations and exemplifications and less contextual considerations (situations, spatiotemporal features). The results support the view that institutional concepts are artifacts, even if they are more symbolic than standard artifacts like hammers and screwdrivers. In a second study (Villani, D’Ascenzo, et al., Reference Villani, D’Ascenzo, Borghi, Roversi, Benassi and Lugli2021), experts and nonexperts in law rated two kinds of abstract concepts (institutional and theoretical, e.g., “acceleration,” “subtraction,” “temperature,” “sum”) and two kinds of concrete ones (natural food, e.g., “banana,” “chestnut,” “potato”; and artifact, e.g., “hammer,” “wheel,” “painting”). Institutional concepts were distinguished into pure-institutional (e.g., “contract,” “state,” “property”) and meta-institutional (e.g., “norm,” “duty,” “justice”), which appeared more abstract. Ratings concerned various dimensions: abstractness-concreteness; imageability; contextual availability; familiarity; age of acquisition; modality of acquisition; social valence; social metacognition; arousal; valence; interoception; metacognition; perceptual modality strength; body-object interaction; and mouth and hand involvement. Compared to theoretical concepts and concrete concepts, institutional concepts evoke more social experiences, and more inner emotional and cognitive states, are more associated with mouth and hearing and less with tact, are low in body object interaction, and are more positive in valence than theoretical concepts but more negative than food ones. Finally, they evoke a specific context (high contextual availability). Interestingly, law experts but not novices associate institutional concepts with interoceptive and emotional experiences; this association highlights that expertise can involve emotional factors.

Overall, ratings showed that Institutional concepts were more related to mouth, hearing, and sociality, likely due to language activation in a social context. My colleagues Caterina Villani, Stefania D’Ascenzo, Corrado Roversi and Luisa Lugli and I tested the hypothesis of a major role of linguistic and social situations for institutional concepts in a priming study. We compared institutional concepts (e.g., “marriage,” contract,”) with abstract theoretical concepts (e.g., “energy,” “multiplication”) and with two kinds of concrete concepts, i.e., food and tools. The names referring to the four kinds of concepts were primed by photos displaying a social-action (dance together), linguistic-social (dialogue), linguistic-textual (reading a book) situations, and a control condition (landscape). The linguistic-social prime selectively interfered with institutional concepts, confirming the relevance of language in social interaction for these concepts and showing they differed from concrete concepts and abstract theoretical concepts. The different priming effects found in experts in law and nonexperts revealed that these concepts have a flexible and variable representation. Intriguingly, the interference effect of linguistic-social primes was more marked for nonexperts, suggesting they conceptualized institutional concepts in terms of negotiation and conflict.

Evaluative concepts. As we have seen, one of the characteristics of abstract concepts is that their meaning is not fully determined, which leaves space for possible negotiations on the word meaning. In this respect, evaluative concepts represent an interesting case. In a recent paper, Fingerhut and Prinz (Reference Fingerhut and Prinz2018) argue that both moral and aesthetic concepts, like “beauty,” are abstract (on the concept of “beauty,” see also Borghi, Reference Borghi2022). The reason why they are abstract might lie in their debatable character (Mazzuca & Santarelli, Reference Mazzuca and Santarelli2022): something can be seen as beautiful by someone, as ugly by somebody else. According to the authors, this debatable character characterizes both morality and aesthetic concepts. What is beautiful and morally acceptable for one individual does not correspond to what is attractive and morally acceptable for another one. What distinguishes the two kinds of concepts are the emotions they evoke. According to Fingerhut and Prinz (Reference Fingerhut and Prinz2018), aesthetic objects are highly emotional, and the emotion of wonder is pivotal. For example, we are thrilled because we see something unexpected and unpredictable. Alternatively, we feel small in front of something big. Finally, in front of art, we tend to contemplate. Consistently, anhedonic individuals (people who are unable to feel pleasure) and alexithymic ones (people who have difficulties expressing emotions) tend to evaluate less valuable artistic products.

Not only aesthetic but also moral concepts have an embodied and emotional character. Many studies that I do not address here in detail show that, for example, the concept of guilt is strongly linked to the notion of purity and to the idea of washing our bodies. Evidence supports the so-called Macbeth effect, that is, the relationship we perceive between physical and moral cleanliness. People who have committed an immoral act, feel a greater need to wash their hands and to choose cleansing products; however, when they find themselves in a clean environment, people evaluate immoral acts as being less immoral. Being in clean versus dirty environments exerts a different influence on moral evaluations. Participants primed with unethical events prefer cleaning products to others (Zhong & Liljenquist, Reference Zhong and Liljenquist2006; Schnall et al., Reference Schnall, Benton and Harvey2008). Although a recent meta-analysis questions the ubiquity of the Macbeth effect, evidence suggests that the relationship between physical and moral cleanliness is present in some conditions (Siev et al., Reference Siev, Zuckerman and Siev2018; Falcinelli et al., Reference Falcinelli, Fini and Borghi2021).

Ecological concepts. In recent years, researchers have investigated concepts belonging to new domains; ecology is a fascinating one. Surprisingly, the literature is scarce regarding how people represent and use ecological concepts, like “climate change,” which can be more concrete and abstract. In recent work, my colleagues Ilenia Falcinelli, Claudia Mazzuca, Chiara Fini, and I found significant differences between ecological and nonecological concepts of close domains, including natural kinds, geopolitical concepts, and technological concepts. Participants rated ecological concepts as being as abstract as the technological ones but at the same time as less familiar. Compared to nonecological concepts, participants were less confident in their meaning and reported needing others more to understand them, even if they did not trust field experts. Also, they rated ecological concepts as more emotionally loaded than technological ones and more likely to generate public debates, even if they evoked fewer social situations. Ecological concepts, particularly abstract ones, are also intriguing because of their indeterminate and ideological character, which can lead to possible conflicts and negotiations.

5.2.2 Comparisons between Kinds of Abstract Concepts

In Section 5.2.1, we saw the particularities characterizing specific concept kinds. Here I briefly review some studies that compared different kinds of concept, aiming to find their similarities and fine-grained differences. Notably, while some studies focused on single concepts, like the emotional, numerical, or institutional ones, some recent works have sought to identify clusters of abstract concepts and their representational structure. Distinctions between different kinds of abstract concepts might emerge through different methods. In some studies, researchers defined a priori the concept kinds (e.g., Setti & Caramelli, Reference Setti and Caramelli2005; Roversi et al., Reference Roversi, Borghi and Tummolini2013; Desai et al., Reference Desai, Reilly and van Dam2018; Villani et al., Reference Villani, Lugli, Liuzza and Borghi2019a). In other cases, concept kinds emerged bottom-up, for example, through cluster analyses, from the ratings or the features produced (e.g., Crutch et al., Reference Crutch, Troche, Reilly and Ridgway2013; Harpaintner et al., Reference Harpaintner, Trumpp and Kiefer2018; Villani et al., Reference Villani, Lugli, Liuzza and Borghi2019b).

A propriety generation task by Harpaintner et al. (Reference Harpaintner, Trumpp and Kiefer2018) revealed that we can distinguish abstract concepts based on their specific semantic features: some are characterized by verbal associations; some by a high proportion of internal/emotional features; while others by a large proportion of sensorimotor features. Abstract concepts also differ as to the bodily parts or effectors they engage. For example, mental state concepts engage the mouth motor system more than other abstract concepts, as shown in a rating and an fMRI study (Ghio et al., Reference Ghio, Vaghi and Tettamanti2013; Dreyer & Pulvermüller, Reference Dreyer and Pulvermüller2018). Numbers – at least low numbers - activate the hand motor system, likely because of finger counting habits (Fischer, Reference Fischer2008; Fischer & Brugger, Reference Fischer and Brugger2011; Fischer et al., Reference Fischer, Kaufmann and Domahs2012)(see Chapter 7.2 for an extended discussion). Emotional concepts have an intermediate status: some evidence indicates that they activate the face and bodily regions through which emotions are expressed (Moseley et al., Reference Moseley, Carota, Hauk, Mohr and Pulvermüller2011; Ghio et al., Reference Ghio, Vaghi and Tettamanti2013). However, there is also contrasting evidence. For example, in two experiments with a lexical decision followed by a recognition task in which either the hand or the mouth were engaged or used to respond, my colleagues and I found that emotional words were processed slower in the mouth than in the hand condition, differently from other abstract words (Mazzuca et al., Reference Mazzuca, Lugli, Nicoletti and Borghi2018). In addition, this activation is less when compared with that of other concepts, such as mental state ones (Ghio et al., Reference Ghio, Vaghi and Tettamanti2013; Dreyer & Pulvermüller, Reference Dreyer and Pulvermüller2018). The involvement of the mouth/face with emotional terms, present but less pronounced than other abstract words, is confirmed by the studies my colleagues and I performed on pacifier use. Prolonged use of the pacifier in infancy interferes less with processing emotional words than other abstract words (Barca et al., Reference Barca, Mazzuca and Borghi2020).

Recently, numerous studies have also focused on the different brain representations of different kinds of abstract concepts. These studies typically compare two or more kinds of concepts. A recent fMRI study focused on abstract concepts that are strongly associated with motor and visual features (e.g., “beauty,” “fight”) and showed that their processing activates brain regions that overlap with modal areas typically engaged during visual and motor tasks (Harpaintner et al., Reference Harpaintner, Sim, Trumpp, Ulrich and Kiefer2020). Catricalà et al. (Reference Catricalà, Conca, Fertonani, Miniussi and Cappa2020) compared abstract social concepts (e.g., “sociability”) and quantity ones (e.g., “immensity”) and investigated through TMS the causal role of right sATL and the right intraparietal sulcus (IPS) in their processing. Participants received a word prime belonging to one of the two categories and had to categorize the target word, which consisted of an exemplar of one of the two categories. When TMS was delivered after the priming, they found a reduction of the response times to unprimed targets, suggesting an involvement of the correspondent cortical area. The reduction of the priming effect was present after ATLs and IPS stimulation for social concepts and only after IPS for quantity ones. In another recent paper, Conca, Catricalà et al. (Reference Conca, Borsa, Cappa and Catricalà2021) used an fMRI-adaptation paradigm with a passive reading task, employing different concrete (living and artifact categories) and abstract categories – emotions (e.g., “fear”), cognitions (e.g., “logic”), attitudes (e.g., “tolerance”), and human actions (e.g., “punishment”). Concrete and abstract concepts activated different areas of the left temporal lobe, with emotions and attitudes adapting to the left middle temporal gyrus and concrete concepts adapting to the left fusiform gyrus. Within abstract concepts, emotions and attitudes (which correspond approximately to the emotional and self-sociality concepts identified by Villani et al., Reference Villani, Lugli, Liuzza and Borghi2019b, and collapsed in Villani, Lugli, et al., Reference Villani, D’Ascenzo, Borghi, Roversi, Benassi and Lugli2021) did not differ from each other. Still, they differed from human actions and cognition, which did not adapt to the left anterior middle temporal gyrus. Further evidence with sentences reveals that the anterior superior temporal gyrus, insula, and supplementary motor area were activated, while the superior temporal gyrus / superior temporal sulcus and left inferior frontal gyrus neural areas are biased towards processing social-emotional concepts over nonsocial concepts (Mellem et al., Reference Mellem, Jasmin, Peng and Martin2016). A recent meta-analysis reveals that concept kinds related to numerical, emotional, moral, and theory of mind (ToM) processing engage both overlapping and different brain areas (Desai et al., Reference Desai, Reilly and van Dam2018). Emotional concepts activate the core regions of the emotional network, with the amygdala and orbitofrontal cortex uniquely activated for them. Numerical concepts recruit areas related to the processing of spatial information and hand movement, i.e., the IPS, superior parietal lobule, and inferior parietal lobule bilaterally. Instead, morality concepts activate various regions, including medial frontal and temporoparietal brain regions, that overlap with areas associated with general semantics, social and episodic memories, and emotion regulation. Similarly, areas recruited by ToM concepts overlap substantially with those engaged by morality and emotional domains.

Notably, new studies comparing concept kinds do not focus only on nouns but also on other syntactic categories. For example, in two studies, Muraki, Sidhu, et al. (Reference Muraki, Cortese, Protzner and Pexman2020) and Muraki et al. (Reference Muraki, Cortese, Protzner and Pexman2020) asked participants to perform a syntactic classification and a memory task of four kinds of verbs: (i) abstract mental (e.g., “assert”), (ii) abstract emotional (e.g., “accuse”), (iii) abstract nonbodily (e.g., “broaden”), and (iv) concrete (e.g., “exercise”). They found differences between verb kinds, showing that non-embodied verbs were processed slower than embodied ones, but results also revealed that the differences between verb kinds might vary across tasks. Event-related potentials analysis confirmed that the nonembodied abstract verbs were the ones that more markedly differed from concrete verbs. Therefore, it is possible that, in the case of verbs, the differences among concept kinds are less marked than with names, while the distinction between more abstract and more concrete verbs remains relevant.

So far, we have seen that feature listing, ratings, and other tasks indicate that different dimensions characterize abstract concepts, which have a different brain representation; we have also noticed that these differences are more pronounced with concept nouns than with verbs. However, it remains to be determined to what extent these dimensions are central for conceptual representation. In a recent preregistered study, my colleagues and I used an interference paradigm to determine which dimension was more relevant for a given kind of concept (Villani, Lugli, et al., Reference Villani, D’Ascenzo, Borghi, Roversi, Benassi and Lugli2021). We used three kinds of concrete concepts, the ones more frequently investigated in the literature – artifacts, natural objects, and food – and three kinds of abstract ones derived from the study by Villani et al. (Reference Villani, Lugli, Liuzza and Borghi2019b). We took the most representative concepts from each of the four clusters and collapsed two clusters – emotional and mental states concepts and social concepts – because of their similarity and partial overlap in the dimensions they captured. The three concept kinds thus consisted of philosophical and spiritual concepts (PS) (e.g., “fate,” “moral”), physical, space, time and quantity concepts (PSTQ) (e.g., “acceleration,” “effort”), and emotional, mental state, and social concepts (EMSS) (e.g., “anxiety,” “friendship”). The task was quite simple: we asked participants to rate word difficulty on a five-point scale and, at the same time to perform a concurrent task – to estimate their heartbeat, squeeze a softball, chew gum, or pronounce a syllable (articulatory suppression). We designed the four conditions based on the study hypotheses and added a baseline in which participants rated the word difficulty without performing any additional task. In the heart-beating condition, participants had to evaluate their heartbeat pace, a method often used in studies on interoception. With the ball-squeezing condition, we intended to test the involvement of the sensorimotor system, particularly of the hand effector. We selected the gum-chewing condition (Topolinski & Strack, Reference Topolinski and Strack2009; Topolinski et al., Reference Topolinski, Lindner and Freudenberg2014) to determine the involvement of language and the mouth motor system. Finally, we designed the articulatory suppression condition to check for the involvement of articulated inner speech. We reasoned that the higher the interference of the concurrent task was, the higher would be the perceived difficulty. To test our hypotheses, we performed Bayesian analyses. Results confirmed that different kinds of concepts tapped into different dimensions.

In keeping with our hypotheses, we found that EMSS activated interoception more than the more concrete PSTQ concepts. The ball-squeezing condition, in which participants manipulated a softball in the rhythm of a metronome, increased the difficulty scores of PSTQ concepts more than the heart and gum condition, even if no difference with the control condition was found. Consistent with this, within concrete concepts, it increased the tools’ difficulty scores. In keeping with our predictions, the gum-chewing condition interfered more with abstract concepts than with the subkinds of concrete concepts. Contrary to our expectations, articulatory suppression did not differentially affect PS concepts and did not have a differential effect on abstract compared to concrete concepts. The absence of the effect of articulatory suppression might be due to the explicit nature of the task or the involvement of nonarticulated inner speech.

Overall, the results allow us to conclude that different kinds of abstract concepts tap into different dimensions (see Figure 5.1). They confirm the role of interoception and mouth motor system involvement for abstract concepts and sensorimotor experience and hand motor system for concrete ones. More specifically, they show that EMSS concepts are grounded in interoceptive experience and confirm that PSTQ concepts are more concrete than other abstract concepts, being less characterized by the interoceptive dimension and more by the sensorimotor one (ball squeezing).

Figure 5.1. Four kinds of abstract concepts identified through cluster analysis (Villani et al., Reference Villani, Lugli, Liuzza and Borghi2019b). For each kind, the upper box reports the content dimensions (concreteness/abstractness, inner grounding, and sensorimotor) identified through principal component analysis (PCA) on various ratings. The lower box reports the pragmatic dimensions that emerged from a rating task and a simulated conversation (Villani, Lugli, et al., Reference Villani, D’Ascenzo, Borghi, Roversi, Benassi and Lugli2021; Villani, Orsoni, et al., Reference Villani, Orsoni, Lugli, Benassi and Borghi2022; Fini et al., Reference Fini, Falcinelli, Cuomo, Era, Candidi, Tummolini and Borghi2023).

Notably, the dimensions that characterize concepts are not stable and fixed but flexible and context-dependent. For example, evidence suggests that the context can modulate the differences and similarities between the various kinds of concept. In a recent study, my colleagues and I collected ratings for more than 100 Italian words in the time period of the first Italian lockdown after the spread of Covid-19 (Mazzuca et al., Reference Mazzuca, Falcinelli, Michalland, Tummolini and Borghi2022). We selected words based on the fact that the pandemic could have influenced their representation; hence, our sample included, among others, words related to family, institutions, diseases, social professions, transportation, and body parts. We also had words derived from two of the clusters found in Villani et al. (Reference Villani, Lugli, Liuzza and Borghi2019b) – emotions and mental states words (EMMS, e.g., “love,” “optimism,” “hope,” “fear,” “anxiety,”) and philosophical and religious words (PS, e.g., “life,” “religion,” “faith,” “destiny,” “death”). Interestingly, the two clusters were not as separate as in the analysis by Villani and colleagues, suggesting that the pandemic might have at least partially modified their representation. Concepts, and particularly abstract ones, are highly variable and flexible.

So far, we have argued that different kinds of abstract – and concrete – concepts tap into different dimensions. But do these different dimensions emerge online when we use concepts conveyed by words in everyday conversations? In a recent online preregistered study (Villani, Orsoni, et al., Reference Villani, Orsoni, Lugli, Benassi and Borghi2022), we presented participants with written sentences related to the abstract concepts PS (e.g., “moral,” “fate,” “salvation”), PSQT (e.g., “energy,” “number”), and EMSS (e.g., “shame,” “joy,” “conflict”), and to concrete concepts related to artifacts, natural objects, and food. An example of two sentences, one related to concrete and one to abstract concepts is: “I made a cake;” “I thought about fate.” Participants were invited to provide a written response simulating a conversation with a familiar person. At the end, they completed the Interpersonal Reactivity Index (IRI; Keaton, Reference Keaton, Worthington and Bodie2017; Italian version adapted by Albiero et al., Reference Albiero, Ingoglia and Lo Coco2006) – we were interested in the relationships between the concepts’ empathic concern and the perspective taking subscale. Later, we coded the produced responses in terms of various dimensions: sensorimotor grounding (properties related to the five senses, materials); thematic relations (spatial, temporal, events, concrete actions, abstract actions); inner grounding (emotions, interoceptions, proprioceptions, beliefs, metacognition); and other (associations, subspecifications, nonperceptual evaluations). The results confirm that PS concepts are the more abstract among abstract concepts, and PSQT the more concrete, while EMSS concepts have a different status. PS concepts evoke more beliefs and introspections (inner grounding) and more abstract actions, while PSQT concepts elicit more visual properties, concrete actions, and temporal relations (sensorimotor grounding, thematic relations). EMSS evoke more abstract actions than PSQT, but less than PS concepts; notably, they yield more interoception features than the other abstract concepts. So far, the results of this study confirm previous ones, extending them because we did not use isolated words but sentences and simulated conversations. Crucially, however, this study allows us to distinguish concept kinds in terms of the conversational dynamics they elicit. The most abstract, – PS concepts – lead to higher uncertainty (more uncertainty expressions like “Mmm …, Explain it to me better,” more questions, and repetitions) and promote more interactive behaviors (more turns) than other abstract concepts. Their higher abstractness level is testified by the fact that they evoke more general statements and elicit more points of view than PSQT concepts. PSQT concepts elicit more “why” questions than other abstract concepts, likely because many of them are scientific terms, and participants might want to understand the underlying mechanisms. Finally, EMSS concepts yield more “who” questions than other concepts, suggesting participants’ interest in the person who experiences emotions.

The last study is critical because it investigates and compares various concept kinds in terms of their conversational dynamics, not only their associated features. Recently Barsalou et al. (Reference Barsalou, Dutriaux and Scheepers2018) stated that we need to study concepts in situated action. This study represents a step in a series of studies that go even further as they investigate concepts in situated interactions (Banks et al., Reference Banks, Borghi, Fargier, Fini, Jonauskaite, Mazzuca and … Woodlinin press; Fini et al., Reference Fini, Falcinelli, Cuomo, Era, Candidi, Tummolini and Borghi2023; see Chapter 8 for more details).

5.3.1 Tenet 1: Grounding and Embodiment

The first and basic tenet of the WAT proposal is that abstract concepts and their varieties, like all other concepts, are grounded in situated experiences and simulations, and are embodied. Different dimensions might be more crucial for the various subkinds.

Subtenet 1: Sensorimotor experiences. Sensorimotor and exteroceptive experiences play a role for abstract concepts, even if it is a less prominent role than for concrete concepts. Compared to concrete concepts, abstract ones activate the acoustic more and the tactile fewer modalities, but they still evoke the senses (Banks & Connell, 2021b) (see Section 5.1).

Subtenet 1.2: Inner experiences. In contrast, inner experiences are particularly relevant for abstract concepts. Studies indicate that abstract concepts, and primarily emotional ones, evoke affections and interoception to a larger extent than concrete concepts (see Section 5.1 for discussion); evidence also shows that they evoke introspection and metacognition (Barsalou & Wiemer-Hastings, Reference Barsalou, Wiemer-Hastings, Pecher and Zwaan2005; Borghi et al., Reference Borghi, Fini, Tummolini, Robinson and Thomas2021).

5.3.2 Tenet 2: Linguistic and Social Experience

Linguistic and social experiences are more crucial for the acquisition, use, and brain representation of abstract concepts than concrete ones.

Subtenet 2.1: Acquisition. Because of the heterogeneous nature of the categorical exemplars, the sole sensorimotor experiences with objects and entities might make it difficult to form categories. An example can clarify. Children experience cups that, although different, share similarities in shape, material, and function – hence, they can form the category of “cup” and then refine and make it more compact thanks to language and the support of others. Others can help them refer to cups – for example, indicating the category exemplars. Things are not so easy with abstract concepts. Take the notion of justice. Children might hear the word “justice” in various situations. Benefiting less from the environmental support – forms of justice are not objects and are not similar in color, shape, and consistency - it is unlikely that they can form a category without strongly relying on linguistic labels, explanations, and the support of others. According to WAT, children – but also adult learners – need the support of caregivers or other people to learn the content of abstract concepts. Notably, my colleagues and I do intend language as a holistic experience, strongly intertwined with the social interaction it promotes, and not as the simple learning of new word-to-referent and word-to-word associations.

Subtenet 2.2: Use. As abstract concepts are more complex, their members are more sparse and perceptually different, and their meaning more undetermined and open to negotiation, they should engage people more in social interaction. Others can provide support in acquiring and helping refine the concept and negotiating its meaning when, for example, two interlocutors intend a concept differently. Hence, abstract concepts would promote interaction. This interaction might occur first with oneself and is aimed at monitoring the knowledge level and finding an interpretation of the word’s meaning through inner speech. We have called this process “inner social metacognition”(Borghi & Fernyhough, Reference Borghi and Fernyhough2023). Then, the need to refer to others to search for support or mutual exchange might promote social interaction and possibly social cohesion. We have called this process “social metacognition” (Borghi et al., Reference Borghi, Barca, Binkofski and Tummolini2018b; Borghi, Reference Borghi2022). I will illustrate these two processes in Chapters 7 and 8 (see Figure 7.1).

Subtenet 2.3: Brain representation. If linguistic experience is crucial for acquiring and using abstract concepts, then their brain representation should reflect this. Compared to concrete concepts, abstract ones should recruit more linguistic and social brain areas and also areas related to inner speech (see Chapters 7 and 8 for discussion). In addition, abstract concepts should engage the mouth motor system more because of its role in articulating words during inner speech. This inner speech recruitment might occur both during self-talk to search for the concept meaning and while preparing to interact with others to ask for the concept meaning. I discuss evidence supporting this claim in Chapters 7 and 8.

5.3.3 Tenet 3: Flexibility across Contexts and Languages

Because of their undetermined character, abstract concepts should be more flexible and variable across participants and contexts (Falandays & Spivey, Reference Falandays and Spivey2019).

Subtenet 3.1: Contextual variability. The meaning of abstract concepts should be more variable across contexts (Davis et al., Reference Davis, Altmann and Yee2020).

Subtenet 3.2: Linguistic diversity. If language is so crucial for their acquisition and use, abstract concepts should be more influenced than concrete concepts by linguistic diversity (Borghi, Reference Borghi2019). I discuss evidence consistent with this claim in Chapter 6.

6.3.1 Concrete Nouns and Concrete Verbs

In an influential study, Malt et al. (Reference Malt, Sloman, Gennari, Shi and Wang1999) demonstrated divergences between knowing, operationalized as the ability to categorize, and naming. When faced with containers of different shapes, the naming pattern of Spanish, English, and Chinese speakers strongly differed. When participants had to sort the containers into categories, the response pattern did not vary across speakers. Hence, with concrete objects, participants seemed to be influenced by the language when thinking for speaking but not when thinking for thinking. This is likely because the environmental constraints strongly affect concepts referring to objects. Many influential studies focus on how different languages classify animals and plants or how they encode tools. Given the various species in different environments, authors have often compared Western and small-scale cultures. Research in folk-biology has investigated how people reason about plants and animals (Medin & Atran, Reference Medin and Atran1999). Despite differences, there are also striking similarities in how these categories are organized. For example, Lopez et al. (Reference Lopez, Atran, Coley, Medin and Smith1997) had American people and people of the traditional Itza-Maya society sort cards with the names of mammals and perform induction tasks, generalizing a property from one mammal to another (e.g., “[Mammal 1] has a disease. [Mammal 2] has another disease. Do you think [Mammal 3] has the disease of [Mammal 1] or the disease of [Mammal 2]?”). In both Western and traditional societies, people used taxonomies of local species of mammals, the members of which were perceptually similar, which differed from scientific taxonomies, and organized them taxonomically in different hierarchical levels. Notably, beyond the commonalities, there were many differences, for example in detecting similarities – foxes, for instance, were sorted together with dogs by Americans, but with cats by Itzai – and in the stronger ecological attention in the Itza population, who took the animals’ environment and way of chasing more into consideration. In an influential research program, Berlin (2002/Reference Berlin2014) analyzed folk classification systems for plants among the Tzeltal Maya of Mexico and the Aguaruna Jivaro of Peru. Remarkably, about 60 percent of labeled classes corresponded to species identified by botanists. These studies suggest that, despite variations, there are striking similarities among languages and cultures in classifying animals, plants, and even artifacts. Can we account for a phenomenon like this by relying on nativist accounts? I do not believe we can. Rather, it is plausible to argue that the biomechanics of our body and the correlational structure of the environment can create some constraints that influence variations across cultures.

Is this phenomenon confined to concrete nouns referring to spatially bounded entities? Although it might be more pronounced with nouns, it does not seem limited to them. Narasimhan et al. (Reference Narasimhan, Kopecka, Bowerman, Gullberg and Majid2012) had speakers of nineteen different languages describe what they saw while watching videos representing everyday verbs like taking and putting (e.g., “putting a cup on the table,” “taking an apple from a bowl”). All languages expressed the agent with the noun and the motion with the verb. However, they disagreed on the path of displacement motion (to a goal vs. from a source) and the spatial relationship between figure and ground (containment, support, etc.). For example, in Yélî Dnye, an isolated language spoken on Rossell Island, a small island offshore in Papua New Guinea, no locative marker comparable to “to” or “from” exists, and it is the verb that encodes whether the action refers to a goal or a source. Also, some languages, like Hungarian, have verbs with general meanings like “putting” while others, like German, do not. With a similar procedure, Majid et al. (Reference Majid, Boster and Bowerman2008) had speakers from twenty-eight languages belonging to thirteen language families and different cultures, from rural to urban, watch videos of cutting and breaking events and describe what the agent did. Even if there were differences across languages, different languages had many commonalities. The main distinction between cutting and breaking events holds across languages, differentiating events where the locus of separation is highly predictable versus unpredictable. For example, in the case of slicing bread but not smashing a plate, it is possible to predict the location of separation from where the knife is placed. Intriguingly, this distinction occurs even in languages and cultures where knives were introduced only recently and cutting tools are hardly used, such as in the Yélî Dnye language. Besides some commonalities, there were also many variations across languages: for example, the number of verbs used varied from fifty in Tzeltal to three in Yélî Dnye; some languages, like Dutch, Swedish, and Mandarin, have a particular verb to refer to cutting with one or more blades; finally, chopping events are either ascribed to cutting or to breaking events depending on the language.

Studies on concrete motion verbs are interesting to verify whether differences in the naming pattern also correspond to conceptual differences. Consider, for example, motion verbs. Malt et al. (Reference Malt, Gennari, Imai, Ameel, Tsuda and Majid2008) had English, Dutch, Spanish, and Japanese speakers observe video clips of a girl locomoting on a treadmill that varied in speed and slope. They had to name what they saw and evaluate the typicality of each clip for some gait terms. Notably, two languages, English and Dutch, encoded manner in the verb, while the other two did not. Despite being high- or low-manner languages, the four languages reflected the discontinuity in the manner of motion between walking and running. Other differences existed, though. For example, languages differed in encoding fine-grained distinctions between actions such as “skipping” and “jumping.” Another study by Gennari et al. (Reference Gennari, Sloman, Malt and Fitch2002) compared English and Spanish motion verbs. Notably, English verbs encode manner (e.g., “stroll”) and use particles to express the path; Spanish verbs encode the path (e.g., “entrar” [enter], “bajar” [descend]), while adverbs or other particles express manner. Participants observed video clips of motion events; in one condition, they had to name the action, in the other, they only watched the video. Then they performed a recognition task, in which they had to decide whether they had previously seen a specific action, and a similarity judgment task. The verbal coding influenced the explicit similarity judgment task but not the recognition task. Hence, “thinking for speaking” did not impact recognition. The results suggest that, although the language has an impact, this impact might be confined and might not strongly affect mental representations of motion events.

Another domain where convergences and divergences across languages can be noted is that of body parts. Universal elements are less than one might predict. For example, some languages do not possess the term “body.” The Jahai language does not possess a term for “face” and “mouth” but has many terms related to small parts of the face (e.g., eyes, upper lip, lower lip, teeth, wrinkles on the side of the eyebrows, etc.). “Soul” and “life force” are considered parts of the body in some languages but not in others (Enfield et al., Reference Enfield, Majid and Van Staden2006). Despite these differences, compared to other domains, body parts seem to be more consistent across languages. Majid et al. (Reference Majid, Jordan and Dunn2015) asked speakers of twenty Germanic languages (e.g., Danish, German, English, Dutch, Frisian, etc.) to name elements of four domains: color (e.g., “blue”), body parts (e.g., “forearm”), containers (e.g., “bowl”), and spatial relations (e.g., “on”). Spatial relations showed the most variation in meaning. The authors interpret the results by contending that grammaticized meanings are more variable than the meanings of open-class lexical items, like nouns. I think this result is also informative for the concreteness/abstractness distinction. More concrete items – containers, body parts, and even object properties – are less variable across languages because they are more subject to constraints given by their referents. Interestingly, in a recent study with Japanese languages, Huisman et al., (Reference Huisman, Hout and Majid2021) found the highest similarity in the face region for bounded parts (i.e., nose, mouth, ear, and eye). This happened when participants both named and colored acoustically presented body parts in a silhouette. In contrast, unbounded parts varied more, especially in the cheek area. Notably, referring to spatially bounded objects is typical of concrete items.

The pattern I have reviewed is quite complex and nuanced. Variations are ubiquitous. However, the results suggest that reliance on objects leads to strong convergence among speakers, in the case of both natural objects and artifacts. Concrete verbs of everyday actions, like “putting/taking” and “cutting/breaking,” have many differences but striking similarities in naming. When there are stronger constraints in the structure of the physical world and biological constraints linked to the specificity of the human perceptual system (Malt & Majid, Reference Levinson and Majid2013), then we are more likely to find cross-linguistic commonalities. These conceptual commonalities might also exist independent of the naming similarities, as the dissociations between the naming and sorting tasks reveal.

6.3.2 Abstract Nouns and Abstract Verbs

If we consider abstract nouns, the variations in meaning become more substantial. Evidence on the abstract concepts of time represents an exciting example. Many studies have investigated the notion of time, based on the conceptual metaphor view (Lakoff & Johnson, Reference Lakoff and Johnson1980; Reference Lakoff and Johnson2008). Hence, the concept of time will refer metaphorically to the concrete concept of space (for an overview of time, space, and number, see Winter et al., Reference Winter, Marghetis and Matlock2015). Overall, many studies demonstrate the cross-linguistic variability of the concept of time and its relationship with the more concrete concept of space (for a review, see Boroditsky, Reference Boroditsky2018).

Languages differ in the metaphors used to refer to the concept of time. For example, different languages use the metaphors of length and quantity. In some languages, such as English and Indonesian, people speak of “long meetings” and a “long time”; in others, such as Greek and Spanish, people say “large time” (largo tiempo). In a study reported by Casasanto (Reference Casasanto2008), English, Indonesian, Greek, and Spanish participants estimated how much the length of a line or the quantity of water in a container would grow and remain on the screen. Results showed, in line with a Whorfian view, that participants performed this nonlinguistic task relying on the metaphors of time in their languages. This representation is not stable. For example, while Swedish and Spanish people relied respectively on the length and quantity metaphors, bilingual people flexibly shifted between the two representations depending on the used language (Bylund & Athanasopoulos, Reference Bylund and Athanasopoulos2017).

How do people represent the relationship between future, present, and past? It depends on the spoken language. Some languages, such as English, use horizontal metaphors to think about time. People associate the past with the left and the future with the right. When Spanish participants judged whether written words refer to the past or the future, they responded faster to future terms by pressing a right key and to past words by pressing a left key (Santiago et al., Reference Santiago, Lupáñez, Pérez and Funes2007). Other languages use multiple mappings, with a dominant one. For example, Chinese Mandarin speakers arrange pictures vertically and gesture along the vertical axes to indicate the progression in time. A study by Boroditsky (Reference Boroditsky2001) was the first to reveal with an implicit priming task that the vertical metaphor to refer to time is more frequent in Chinese Mandarin speakers, while for English speakers, time flows horizontally from left to right. Interestingly, the effect can shift in bilingual people, who use either the vertical or the horizontal mapping depending on the language they are using (Boroditsky et al., Reference Boroditsky, Fuhrman and McCormick2011).

Writing direction also impacts how people think of time. In cultures that adopt the left–right mapping preferentially, the writing direction influences the direction of the mapping. Hebrew and English speakers had to make temporal order judgments after seeing two pictures representing an earlier and a later phase of an event by pressing a left or a right key; for example, they saw the picture of an intact banana and a peeled banana. Hebrew participants responded faster when the right key represented “earlier” responses, whereas English speakers did the opposite, consistent with the opposite writing directions, right-to-left, and left-to-right (Fuhrman & Boroditsky, Reference Fuhrman and Boroditsky2010). When deciding whether words were related to time, Spanish and German participants responded faster by pressing the right key for words about the future and the left about the past, while Yiddish participants did the opposite (Santiago et al., Reference Santiago, Lupáñez, Pérez and Funes2007; Ulrich & Maienborn, Reference Ulrich and Maienborn2010). Interestingly, even the reading direction of blind people (who read with their hands) influences how they conceive time (Hendricks & Boroditsky, Reference Hendricks and Boroditsky2015).

Many languages use the front–back axis to describe the relationship between past and future. In most cases, people see the past as behind the ego, as conveyed by the English expression “Let’s put the past behind us.” However, this mapping is not universal. For example, the Aymara language speakers represent the past in front of them and the future behind them. Thus, Aymara speakers point the index finger on the forehead to refer to old generations (Núñez & Sweetser, Reference Núñez and Sweetser2006). Similar to Aymara, in Vietnamese (Sullivan & Bui, Reference Sullivan and Bui2016), time also flows from the front (past) to the back (future) because events of the past events can be “seen,” but future events are still uncertain. In other languages, people represent time along cardinal directions. For example, Pormpuraaw, an Australian Aboriginal community, represents time as flowing from East to West (Boroditsky & Gaby, Reference Boroditsky and Gaby2010), while for speakers of Yupno, in Papua New Guinea, the past is construed as downhill, and the future as uphill (Núñez et al., Reference Núñez, Cooperrider, Doan and Wassmann2012).

An intriguing proposal to account for divergences in conceiving the relationship between past and future is the temporal focus hypothesis (de la Fuente et al., Reference De la Fuente, Santiago, Román, Dumitrache and Casasanto2014). Cultures vary in whether they are past-oriented or future-oriented, depending on the value they place on tradition and progress. According to the temporal focus hypothesis, this should impact people’s representation of time: the future is ahead in future-oriented cultures, behind in past-oriented societies. In keeping with this hypothesis, a study showed that Spanish people saw the future ahead differently from Moroccans who saw the past ahead; interestingly, the Spanish elderly had a time representation that was intermediate between the Spanish and Moroccan ones. A recent study showed that, despite one exception (South Africa), the distinction between past- and future-oriented societies could explain data taken from ten Western (sub)cultural groups, ten subcultural groups from China and Vietnam, and Great Britain (Callizo-Romero et al., Reference Callizo-Romero, Tutnjević, Pandza, Ouellet, Kranjec, Ilić and Casasanto2020).

We have seen that different space-time mappings exist and that languages greatly vary in the mappings they adopt. Intriguingly, there are languages where no clear space-time mapping is present. In Yélî Dnye language, no calendar time is present – no names of the days, months, or years. People refer to time through gestures, indicating the sun. When comparing Yélî Dnye speakers with Dutch speakers in locating verbal (e.g., yesterday, today, tomorrow) and nonverbal sequences (e.g., maturation of an organism) in space, Dutch people used a left-to-right mapping. In contrast, no stable space-time mapping characterized Yélî Dnye speakers’ responses – they used left-to-right, east-to-west, and forward–away axes. Not casually, Levinson and Majid (Reference Levinson and Majid2013) called Rossell “the island of time.” These results cast doubts on the universality of space-time mapping and suggest that space is used when the complexity of time organization and structuring requires it. Other languages that do not seem to use linear representations of time are the Yucatec Maya (Le Guen & Pool Balam, Reference Le Guen and Pool Balam2012), who conceive time as a succession of events not spatially organized, or the Amondawa in Amazonia (Sinha et al., Reference Sinha, Sinha, Zinken and Sampaio2011).

This short overview reveals that the abstract concept of time is highly variable across cultures and languages and that even the mapping with time, which occurs across many languages, is likely not universal. Linguistic and nonlinguistic tasks consistently show that many different mappings are present and might concur. Furthermore, the reading direction has a significant impact. Being the representation of time flexible, it might shift in bilingual people. In addition, evidence shows that people can also be trained and can learn new linguistic relations between space-time concepts, and that these new relations can impact their thought (Hendricks & Boroditsky, Reference Hendricks and Boroditsky2017).

I briefly consider a further example of concepts that people typically rate as abstract (albeit less abstract than concepts such as time), namely, emotional terms. In a recent study, Jackson et al. (Reference Jackson, Watts, Henry, List, Forkel, Mucha and Lindquist2019) examined the semantics of twenty-four emotion terms across 2,474 spoken languages of twenty language families. Notably, when compared with other abstract concepts, emotional ones are more embodied, that is, they are generally evaluated as being more concrete and imageable, being acquired earlier, and evoking interoception and sensory modalities like vision and touch to a larger extent (Villani et al., Reference Villani, Lugli, Liuzza and Borghi2019a). The domain of emotions is fascinating because of the debate revolving around their representation. Even though this is outside the scope of the present volume, it might be worth remembering the contrast between influential views, according to which more basic emotions will be universal (e.g., Ekman, Reference Ekman1992), and constructivist approaches, which put the accent on the process of building emotions (e.g., Barrett, Reference Barrett2017). Here we focus on emotional concepts expressed by language, that is, how people speak about emotions, which does not necessarily reflect how they perceive emotions (Majid, Reference Majid2019). The authors analyze the cases of colexification, when multiple concepts use the same form within a language – for example, the Persian word “ænduh” expresses both “grief” and “regret.” Across languages, valence and neurophysiological activation are good predictors allowing the differentiation of emotion from neutral words. Despite these universal aspects, the data reveal a striking cross-linguistic variability. Even if often equated in dictionary translation, emotional words vary consistently and significantly across languages. Emotions vary more than the semantics of color, known for its variability. For example, the word “anger” is associated with “envy” among Nakh-Daghestanian languages, but it is more associated with “hate,” “bad,” and “proud” among languages. Language spoken in contiguous geographical communities has more similar colexification patterns.

So far, we have seen that, although there are many differences across languages, linguistic variations affect less concrete than abstract concepts. As examples of concrete concepts, I chose spatially bounded objects and entities, such as artifacts (containers), plants, and animals. As an example of an abstract concept, I used the notion of time and have briefly addressed the concept of emotion (I discussed the abstract concept of number in Chapter 7.2). How about concepts that lie in between abstract and concrete ones? I briefly discuss the examples of odor and gender concepts. Odor is intriguing because it does refer to a sensory modality but an ineffable one and does not have a physical object as a referent. Odors cannot be seen, heard, or touched, but the source they originate from can be seen and touched. In a recent study on twenty different languages, Majid et al. (Reference Majid, Burenhult, Stensmyr, De Valk and Hansson2018) examined the codability (agreement of speakers, length of the produced speech, and specificity of the response) of words referring to the five senses. They found marked cross-cultural differences; intriguingly, the order of sensory modalities generally did not correspond to the Aristotelian hierarchy of senses – sight, hearing, touch, taste, and smell – but different cultures emphasized different senses. Across languages, however, the odor has a lower codability score. Is odor “the muted sense”(Olofsson & Gottfried, Reference Olofsson and Gottfried2015) weakly related to language across all cultures? It is less verbally expressed than other senses, but there are exceptions. Comparative analyses of a free naming of smell and color showed that, while English speakers had few words referring to odors, Jahai speakers from a hunter-gatherer population from the Malay Peninsula named odors easily, with the same level of interspeaker agreement as color (Majid & Burenhult, Reference Majid and Burenhult2014). Do these differences depend on different ways of perceiving color, or are they due to different naming patterns? Perception of color pleasantness does not vary across cultures, as revealed by a rating study on nine non-Western cultures (Arshamian et al., Reference Arshamian, Gerkin, Kruspe, Wnuk, Floyd, O’Meara and Majid2022). Consistent with this, analyses of Dutch and Jahai speakers’ facial expressions showed that emotional reactions to odor did not differ (Majid, Burenhult, et al., Reference Majid, Burenhult, Stensmyr, De Valk and Hansson2018). However, the naming pattern did: Jahai speakers used more abstract words to refer to odors (e.g., “musty”), and Dutch participants more concrete words referring to the source of the aroma (e.g., “smells like lemon”)(Majid, Burenhult, et al., Reference Majid, Burenhult, Stensmyr, De Valk and Hansson2018). Hence, across cultures, odor pleasantness is perceived similarly, and, with some exceptions, such as the Jahai culture, it is also less variable because, across languages, it is less encoded in language. How can we reconcile this finding with the fact that, among the senses, the smell is the most ineffable and thus apparently more abstract one? One possible way to account for these differences is to rely on the abstractness of the words used. Across most cultures and languages, people mainly name odors using concrete words that refer to their sources (e.g., “smells of liquorice”). This reliance on the concrete source of smell might explain the lower cross-linguistic variability of odors across languages.

Another example of a concept the abstractness of which might vary across cultures and languages is gender, an interesting concept since it includes both embodied, biological components and cultural ones. Mazzuca et al. (Reference Mazzuca, Majid, Lugli, Nicoletti and Borghi2020; Mazzuca, Borghi et al. Reference Mazzuca, Borghi, van Putten, Lugli, Nicoletti and Majid2022) asked Italians, Dutch, and English speakers to perform a free listing task and a typicality rating task. Across tasks, a major difference emerged between Italians, who focused more on sociocultural factors (e.g., “identity”), and Dutch participants, who focused more on biological elements (e.g., “genitals”) and adopted a more essentialist stance. The difference might occur because gender in Italy is at the center of political debates, rendering sociocultural features more salient. Consistent with this, Italians rated abstract features as more related to gender, while Dutch and English participants, especially Dutch, associated gender with more concrete aspects (see Figure 6.1).

Figure 6.1 Conceptualization of gender by Dutch, English, and Italian participants (Mazzuca, Borghi et al., Reference Mazzuca, Borghi, van Putten, Lugli, Nicoletti and Majid2022). Panel A gives cluster analysis results for Italian and Dutch clusters, showing that Italian clusters were more focused on sociocultural factors, and Dutch clusters on biological aspects. Panels B and C report the ratings of the produced words. Panel B shows that the more words are related to gender, the more they are abstract, but only for Italians. The effect is reversed for Dutch participants, while English participants are in the middle. Panel C shows the opposition only between Italians and Dutch participants: for the former, gender typicality is positively correlated with abstractness; for the latter, the pattern is the opposite.

So far, I have contended, in light of examples of concrete, abstract, and intermediate concepts, that the meaning of abstract words varies more across languages than that of concrete terms. In the literature, there is some evidence contrasting with the hypothesis. Analyzing semantic neighborhoods of 1,010 meanings in 41 languages, Thomson et al. (2022) showed that words that aligned most were not the most concrete ones but those that belonged to highly structured domains, such as number, time, and kinship. Some of these concepts are abstract. However, they belong to the most concrete among abstract concepts, i.e., PSTQ ones, and importantly they are concepts the meaning of which is conventional and less debatable. Furthermore, this study does not investigate people’s mental representation of time but whether they use similar words (e.g., “day” and “month”) across languages. However, this evidence is important and challenging. It also induces reflections on whether concreteness-abstractness ratings should be combined with other information (e.g., age and modality of acquisition, negotiable character of the meaning, etc.). My colleagues and I directly tested the hypothesis that abstract concepts are more variable across cultures and languages in two studies. In the first, Italian and Iranian participants performed “sensibility” judgments of sentences referring literally or metaphorically to motor actions (e.g., “s/he grasps the cup/a concept”). Concurrently they observed a video displaying a movement congruent or incongruent with the one mentioned in the sentence. For example, in the incongruent condition they saw a video of someone pouring tea, while the sentence referred to grasping a cup. Compared to the other conditions (congruent abstract, incongruent concrete, and abstract sentences), congruent concrete sentences were facilitated in Italian and interfered with Persian (Ghandhari et al., Reference Ghandhari, Fini, Da Rold and Borghi2020). The results suggest a stronger integration between gestures and language in Italians. We interpreted the result on Iranians referring to the tendency to avoid movement when talking, which is common in the Persian culture. In this task, we found stronger variations in concrete than in abstract verbs; however, we ascribe the result to the specific paradigm and the privileged relations of concrete concepts with actions. More intriguingly, in a study we are performing together with Federico Da Rold, Claudia Mazzuca, Chiara Fini, and Mina Ghandhari, we used a sorting task to test the hypothesis that abstract concepts might differ more by comparing participants of three different cultures, both Western and Eastern, namely, Italian, Iranian, and Israeli. Participants used an appositely created app and had to sort forty names of concrete and abstract concepts into two, four, six, and an indeterminate number of groups by dragging words on a tablet. Concrete words included the most frequently studied categories in the literature, that is, tools, animals, and food (Warrington & Shallice, Reference Warrington and Shallice1984; Rumiati & Foroni, Reference Rumiati and Foroni2016). Abstract words included members of the three groups identified by Villani et al. (Reference Villani, Lugli, Liuzza and Borghi2019b): philosophical and spiritual (PS); emotional, mental state, and social concepts (EMSS); and physical, space, time, and quantity concepts (PSTQ). As shown in Figure 6.2, within each language, concrete concepts cluster more easily into three categories, while abstract concepts are sparser and heterogeneous. More crucially, the variability of sorting across cultures is higher for abstract than concrete words, both when we consider abstract and concrete concepts as unitary categories and when we divide them into subclusters each, as Figure 6.2 shows.

Figure 6.2 Results of a principal component analysis performed on a sorting task. Iranian, Israeli, and Italian participants had to divide twenty concrete (Figure 6.2a) and twenty abstract (Figure 6.2b) words into two, four, six, or a free number of categories.

7.1.1 Abstract Concepts, Language, and Mouth Motor System Activation

Moving the mouth during conceptual processing might signal the presence of inner language. Across different studies, my colleagues and I consistently found that, while concrete concepts activated the hand motor system more, the mouth and face motor systems are more engaged when processing abstract concepts. We started with studies mimicking the acquisition of concepts. We performed four experiments asking participants either to manipulate or to observe new objects on a computer screen (Borghi et al., Reference Borghi, Flumini, Cimatti, Marocco and Scorolli2011). Concrete elements were three-dimensional (3D) objects varying in shape and color and rich in perceptual features; participants had to move them around by dragging the mouse and then inserting them into a hole. Abstract concepts did not have a single object as a referent but multiple ones; these objects interacted in different ways. For example, after two or more objects collided, one moved in a straight line while the other spiraled away at a different speed. The interacting objects were all cylinders and sky-blue, and participants had to observe them interacting but could not move them using the mouse. First, they had to decide whether or not two elements belonged to the same category. As expected, abstract concepts yielded more errors than abstract ones. Then participants learned the category name. Invented names (e.g., “CALONA,” “NOROLO”) had the same number of letters; half of them finished with A, half with O (feminine vs. masculine in Italian). Labeling facilitated the categorization of both concrete and abstract concepts. Hence, when participants had to match words to images, abstract concepts remained more difficult than concrete ones. This result replicates the well-known concreteness effect, that is, the processing advantage of concrete concepts compared to abstract ones with artificial categories. A further property generation task testified that our operationalization was correct: participants produced more perceptual properties for concrete items and more abstract and relational properties for abstract ones, similar to what happens with natural categories. In further experiments, to investigate the role linguistic support plays in facilitating conceptual acquisition, we added to half of the invented names an explanation of their meaning. Explanations for concrete items mentioned perceptual elements (e.g., “CALONA: “a figure having a concavity”); for abstract concepts, the explanations pointed to the interaction between the elements (e.g., “PANIFA: two 3D moving figures. After the contact, one of them moves in a straight line; the other one executes a turning movement with a different velocity”).

Interestingly, in a subsequent words-to-images matching task, the presence of explanations reduced the difference between responses to concrete and abstract concepts. Explanations seemed to support the acquisition of abstract concepts but did not impact concrete ones, likely because perceptual information, together with the label, was sufficient to form the category.

To address whether language activation with abstract concepts has a motor counterpart, we introduced a property verification task manipulating the response effector (mouth vs. hand). Participants had to decide whether a property, like the color, was true of a given category (e.g., “Is the FUSAPO red?”). For half of the experiment, they responded “yes” using the microphone (mouth); for the other half, they pressed a “yes” key on the keyboard. In a previous study, we had found faster response times with the microphone than key-pressing responses with sentences related to mouth-related actions (e.g., sucking a candy) (Scorolli & Borghi, Reference Scorolli and Borghi2007). Across two experiments, responses to concrete words were faster with the keyboard, responses to abstract words faster with the microphone. This finding indicates that acquiring novel abstract concepts involves language, thus engaging the mouth effector.

A possible limitation of the study is that the referents of concrete and abstract categories were perceptually different. In everyday life, what really differs is how we assemble what we perceive to form concrete and abstract concepts. Hence, in a further study (Granito et al., Reference Granito, Scorolli and Borghi2015), we presented participants with categories created with Lego bricks. Through a sorting task, we assessed which categories participants had formed. As predicted, sorting was faster for concrete than for abstract concepts. Then the experimenter taught the participants a label and explained each category’s criterion. The same component could be part of a concrete or abstract category, but concrete labels pointed to single objects, and abstract ones referred to relations (Gentner & Boroditsky, Reference Gentner and Boroditsky2001). The explanations of concrete concepts contained information on the item’s color and shape; the explanations of abstract concepts included spatial relations. Here are two examples, the first of a concrete and the second of an abstract category: “A jagged stack-shaped object with a yellow protrusion”; “The two objects have one contact point and form a concavity.” Accuracy in a subsequent categorical recognition task was higher with concrete than abstract concepts, confirming the concreteness effect. However, linguistic training (label and linguistic explanation) improved the performance, especially with abstract stimuli. Linguistic training also sped up the mouth responses in an image-word match experiment. Indeed, participants who had received the linguistic training responded faster with the mouth (microphone) than those who hadn’t. Analyzing the strategies adopted, we found that linguistic training was particularly crucial for participants whose initial categories were quite distant from the ones they later adopted; it enhanced convergence.

This work extends the previous one, highlighting the importance of linguistic support (labels and explanations) in abstract concept acquisition. It also supports its results, showing the relationship between linguistic activation and mouth motor system engagement.

Studies with artificial categories have many disadvantages and some advantages. One of the main advantages of these two studies is that they allowed us to separate concepts from words; participants first formed the category, then received its name. Overall, results revealed that abstract concepts are harder to create and abstract words harder to learn than concrete ones, requiring more linguistic support.

The disadvantages of studies with artificial categories are notoriously many. How can we be sure that the way we operationalize concrete and abstract concepts is correct? Do the results obtained with artificial categories extend to everyday ones?

As to the first question, it is certainly possible that our operationalization did not capture all aspects that abstract concepts typically involve. In the study by Borghi et al. (Reference Borghi, Flumini, Cimatti, Marocco and Scorolli2011), we considered two distinctive aspects of concrete versus abstract concepts: richness versus poverty in perceptual aspects; and the presence of one versus many referents. In the study by Granito et al. (Reference Granito, Scorolli and Borghi2015), we considered the distinction between objects and relations. I do not think that the operationalizations we made were incorrect. For example, in the first study, the feature production task revealed that concrete concepts elicited more perceptual features, while abstract concepts yielded more relational ones, as in real life. As to the second study, relations are certainly more abstract than objects. Hence, these operationalizations are not wrong, but they certainly do not capture all aspects of abstractness.

Now consider the second question: Do the results obtained with artificial categories extend to everyday ones? Can we demonstrate that abstract concepts involve language more than concrete ones and that this engages the mouth more than the hand motor system?

The first response comes from an experiment in which my colleagues and I controlled whether the higher involvement of the mouth with abstract concepts was simply due to their heterogeneous character and low-dimensional character (Granito et al., Reference Granito, Scorolli and Borghi2015). To this aim, we selected real Italian abstract words, together with concrete words whose members were either heterogeneous (e.g., “tools”) or not (e.g., “penguin”). We asked participants to evaluate how much the hand and mouth effectors were involved in possible actions with the target items. We found that participants associated concrete concepts, be they compact or heterogeneous, with the hands and abstract ones with the mouth.

Other responses come from various further experiments performed with real words in our and other labs. I first review the ones conducted by me and my colleagues. Borghi and Zarcone (Reference Borghi and Zarcone2016) had participants read abstract and concrete definitions on the computer screen. Concrete definitions described the object’s color and shape and included thematic relations; abstract definitions included taxonomic relations (for example, definitions for “hen” included “Its verse is “cackle,” it has two legs and is located in farms,” and “Pet birds, it belongs to the family of chickens”; definitions for “logics” included “You use it when, from a clue, you make correct reasoning,” and “Philosophical discipline that investigates the laws of reasoning”). An abstract or concrete word followed each definition. The task consisted of determining whether the definition was appropriate for the word. If appropriate, participants had to press either a keyboard button with the dominant hand’s index or a device with the teeth, depending on the block. If it was not appropriate, they had to refrain from responding. We found an overall advantage of concrete over abstract stimuli, thus confirming and extending the concreteness effect. However, the most important result was the predicted interaction between the target words and the response effectors. Responses with the mouth were overall slower due to the device, but the mouth’s disadvantage over the hand decreased with abstract compared to concrete concepts. Hence, responses were facilitated when participants responded by pressing the device with the mouth to abstract words. Computing the differences in response time between hand and mouth responses, we can see that, for example, words like “logic,” “reason,” and “motive” scored very high in abstractness ratings, and the mouth–hand difference in response time was very high.

In contrast, words like “sand,” “statue,” “boot,” and “ice” scored low in abstractness, and the mouth–hand difference was minimal. Further ratings in which participants had to determine to what extent each word was associated with a hand or a mouth action confirmed the result.

Having demonstrated facilitation of mouth responses with real abstract words, the question arose as to whether this mouth facilitation with abstract concepts extended to other tasks, including tasks that implied a lower level of processing, like the lexical decision. The lexical decision is a task in which participants have to decide whether or not a word belongs to a given language – in this case, Italian. According to a lexical decision study by Mazzuca et al. (Reference Mazzuca, Lugli, Nicoletti and Borghi2018), mouth responses with abstract concepts have no facilitation. But this facilitation occurs with other tasks, such as recognition. My colleagues and I conducted two experiments in which participants performed first a lexical decision, then a recognition task. In one experiment, they responded by pressing a key with their hand or a device with their teeth, as in Borghi and Zarcone’s (Reference Borghi and Zarcone2016) study; in the other, they responded with their foot to the critical trials (Italian words) while they had to press the button with the hand or teeth only to catch-trials (Italian words with a bold letter). Processing abstract concepts required more time (concreteness effect) across the experiments and tasks. More crucially, we did not find the expected interaction between the kind of concept and response effectors in the lexical decision task. Still, we found it in the recognition tasks in which abstract concepts were facilitated more in the mouth than in the hand condition. In this study, together with concrete and abstract concepts, we also included emotional concepts. Their fluctuation pattern suggests that we can neither assimilate them into concrete nor abstract concepts.

Overall, the present results prove that facilitation occurs when participants respond to abstract concepts using the mouth effector and concrete ones using the hand effector. However, the task flexibly modulated this facilitation. Hence, facilitation is present in tasks requiring deep processing, like the definition-matching and recognition tasks, but not in tasks involving shallow processing, such as the lexical decision. However, some issues remain unsolved. First, one could argue that the activation of different effectors is simply a nonconstitutive byproduct of conceptual activation (Mahon & Caramazza, Reference Mahon and Caramazza2008). Second, it is unclear whether the kind of considered concepts influences the involvement of effectors.

To address the first issue, we designed some interference paradigms. The underlying reasoning is simple: interfering with hand/mouth actions should create processing problems if the motor system’s involvement is pivotal. In recent work (Fini et al., Reference Fini, Zannino, Orsoni, Carlesimo, Benassi and Borghi2021), my colleagues and I had participants decide whether words were concrete or abstract while squeezing a softball or pronouncing a syllable (articulatory suppression). In a first experiment run with participants, squeezing a ball slowed response times with concrete concepts, while pronouncing a syllable slowed response times with abstract concepts. In a second experiment run with participants, we found that articulatory suppression interfered with processing with both concepts but more markedly with abstract ones; no interference from ball squeezing was present. Hence, the results suggest that the mouth motor system’s involvement is crucial for abstract concepts, at least for this task. In another study (Villani et al., Reference Villani, Lugli, Liuzza, Nicoletti and Borghi2021), we asked participants to rate the difficulty of different kinds of concrete and abstract concepts while performing a concurrent dual-task. Two concurrent tasks involved the mouth, i.e., a gum-chewing task and an articulatory suppression task. Two further tasks required participants to manipulate a softball (hand motor system involvement) and estimate their heartbeat pace – a task typically used to determine interoceptive awareness. We reasoned that the perceived difficulty should increase if the second task interfered with the first. As predicted, gum-chewing made processing concrete concepts easier compared to all the other conditions. Specifically, the gum-chewing condition interfered more with abstract than concrete concepts of animals, tools, and food. The result confirms that holding something in the mouth can interfere more the higher the word abstractness is. Surprisingly, however, we did not find any selective effect with articulatory suppression, which interfered similarly with concrete and abstract concepts. The discrepancy with Fini et al. (Reference Fini, Zannino, Orsoni, Carlesimo, Benassi and Borghi2021) results could be due to the explicit character of the rating task. Results of the online interference paradigms suggest that mouth involvement might be crucial for comprehending word meaning. Recently, my colleagues Johannes Nedergaard, Marta Kapielska, and I tried to prevent participants from using language while categorizing images related to concrete and abstract concepts. Participants observed three images, two representing the same abstract concept (e.g., “calm”), and had to choose the odd one (odd-one-out paradigm). We contrasted a baseline with two interference conditions: visuospatial and verbal. Contrary to our predictions, visual interference had a stronger effect on abstract than on concrete concepts, likely because the relationships between concepts and visual images is not very tight, and it is more difficult to find a match between the image and the concept. Intriguingly, responses were not slower under verbal than visuospatial interference with abstract concepts overall, but they were with the more abstract philosophical-spiritual (PS) ones (e.g., “religion,” “logic,” “salvation,” “enigma”), compared to the other categories of abstract concepts identified by Villani et al., (Reference Villani, Lugli, Liuzza and Borghi2019) i.e., emotional, mental state, and social concepts (EMMS), physical, space, time, and quantity (PSTQ), and self and sociality (SS) abstract concepts. In sum, impeding participants from using language seems to especially affect the processing of very abstract concepts.

On a parallel thread of research, we asked ourselves whether the interference due to the mouth’s involvement might interfere with the acquisition of abstract concepts. We reasoned that many children used the pacifier in the phase of language acquisition, and we asked ourselves whether this might have impacted the acquisition of abstract concepts selectively. We started investigating the long-term effect of the pacifier’s use. In the first study, my colleagues and I asked first-grade children to define abstract, concrete, and emotional concepts (Barca et al., Reference Barca, Mazzuca and Borghi2017). Children were divided into groups depending on whether and how long they had used the pacifier – we assessed this information by asking their parents. First, two judges evaluated the definitions produced, determining whether they were correct, wrong, or partially correct. Pacifier use did not influence the correctness of the definitions. Then we analyzed the conceptual relations produced by children; two judges coded them, and a third one intervened in case of disagreements. The most common were perceptual, including perceptual properties and parts (e.g., “helicopter: something that has a propeller”), emotional (e.g., “heart: something that is inside us and makes us kind”), experiential (e.g., “brush: when the teacher tells me to paint something, and I paint with the brush”), taxonomic (e.g., “banana: it’s a fruit”), thematic (spatial: “library: where the books are”; action-function: e.g., “box: you put something inside”), and free association relations (e.g., “culture: when in the morning you have to go to school and have to wear an apron”). Results showed that children who had used the pacifier for longer than three years produced fewer experiential relations. More crucially, if you considered the conceptual relations used, their distinction between concrete and abstract concepts and concrete and emotional concepts was less marked than those of the other children. The longer pacifier use might have interfered with the acquisition of abstract concepts at an age characterized by a consistent increase in vocabulary, especially of abstract terms.

Which mechanism might underlie the differences in abstract concept representation, as reflected by the produced definitions, following the pacifier use? We hypothesize that mouth involvement plays a significant role. Still, other explanations are possible: for example, the pacifier partially hides facial expressions, and this might render adults less responsive to children due to the reduction in empathy. If this was the case, we should find a similar pattern of results with emotional and abstract concepts. But this was not the case. In the following experiment, we had 8-year-olds perform a categorization task. They read words and had to decide whether the words referred to animals or not. If they didn’t, they had to press a button. The nonanimal words were concrete, abstract, or emotional. Children who had used the pacifier for more than three years had slower responses to abstract concepts than to emotional and concrete concepts; furthermore, the longer they had used the pacifier, the slower their response times with abstract words were. This result suggests that abstract and emotional words cannot be assimilated but that their processing subtends different underlying mechanisms. Hence, we hypothesize that the empathy-based explanation works more for emotional concepts. In this vein, Niedenthal et al. (Reference Niedenthal, Augustinova, Rychlowska, Droit-Volet, Zinner, Knafo and Brauer2012) showed that prolonged pacifier use could impact emotional development, especially in male children. However, we believe that the selective slowing down in the processing of abstract concepts depends on the involvement of the mouth while using the pacifier. So far, I have addressed long-term effects of pacifier use. In a recent study performed by Laura Barca, Claudia Mazzuca and me, we investigated whether the pacifier effect is present when children learn their first words. We tested 24–36-month-old children and asked parents to complete the MacArthur-Bates Communicative Development Inventory (the Italian version) (Caselli et al., Reference Caselli, Bello, Rinaldi, Stefanini and Pasqualetti2015). We found that the months of pacifier use did not affect the total number of words produced by children or the different abstractness of words. This null result leads us to hypothesize a possible developmental course of the pacifier’s hindering effects: these effects likely occur after three years of age, when children learn most abstract words.

So far, I have illustrated various results with adults and children, showing an association between the involvement of the mouth and abstract concepts. Responses with the mouth are facilitated, while the active hindering of mouth movement determines abstract concepts processing costs. The results of other laboratories are in line with ours. For example, Ghio et al.(Reference Ghio, Vaghi and Tettamanti2013) demonstrated that participants associate mental state abstract concepts with mouth actions, emotional concepts with mouth and hand actions, and math concepts with hand actions, likely owing to finger-counting habits. Dreyer and Pulvermüller (Reference Dreyer and Pulvermüller2018) showed that mental state concepts, but not emotional ones, activate the face/mouth area with functional magnetic resonance imaging (fMRI).

A question now arises: Why would the mouth be involved during abstract concept acquisition and use? Over the years, my colleagues and I have thought of some possible responses. Multiple mechanisms, which are not necessarily mutually exclusive, might explain the mouth motor system engagement. In both the phases of concept acquisition and use, a unique role might be played by inner speech.

7.1.2 Abstract Concepts and Inner Speech

Inner speech might be crucial during both the acquisition and the processing of abstract concepts. During the learning of novel concepts, inner speech can help internal rehearsal of the linguistic labels to support memory consolidation (Borghi & Fernyhough, Reference Borghi and Fernyhough2023). This process, which characterizes the acquisition of all kinds of words, might be particularly relevant for abstract ones because of the critical role that labels assume in keeping together and soldering the connection between their heterogeneous and different category exemplars.

Inner speech might also be crucial during the processing and use of abstract words. Here, various mechanisms might be at play. First, because the acquisition of abstract concepts mainly occurs linguistically, people might reenact this acquisition process during their comprehension and use. Second, people might need to search for a possible word meaning by rehearsing associated words. This search would explain why participants process abstract concepts slower than concrete concepts (concreteness effect). This search might occur through inner speech, and would be compatible with the engagement of the inferior frontal gyrus, linked to phonological search in working memory (Binder et al., Reference Binder, Frost, Hammeke, Bellgowan, Springer, Kaufman and Possing2000). Aside from inner searching for words, during this monitoring phase, Charles Fernyhough and I have proposed that people might even have a dialogue with themselves to find the word’s possible meaning; we called this process “inner social metacognition” (Borghi & Fernyhough, Reference Borghi and Fernyhough2023). Intriguingly, through this inner dialogue, people might even be able to find inner solutions that they were not aware of possessing (Alderson-Day et al., Reference Alderson-Day, Weis, McCarthy-Jones, Moseley, Smailes and Fernyhough2016). Suppose, however, people cannot find an answer or are not satisfied with the solution found. Their uncertainty and low confidence in their knowledge of abstract words will induce people to resort to others who can help them understand the meaning. My colleagues and I have called this mechanism “social metacognition” (Borghi et al., Reference Borghi, Barca, Binkofski and Tummolini2018, Reference Borghi, Barca, Binkofski, Castelfranchi, Pezzulo and Tummolini2019). The term “metacognition” refers to the fact that people might assess their knowledge and determine its limits (Shea, Reference Shea2018); the term “social” is used to indicate that people might revert to others, particularly authoritative others (Prinz, Reference Prinz2012), for support. Hence, we will activate the mouth to prepare ourselves to ask for information from more competent others. However, we might also revert to others to discuss and negotiate the word meaning (Mazzuca & Santarelli, Reference Mazzuca and Santarelli2022; hence, the interaction might not be asymmetric. Finally, because the abstract concept meaning is indeterminate, we might revert to others to understand what they have in mind in order to maximize our exchange with them. Our view of social metacognition is in keeping with one influential proposal focusing more generally on metacognition (Frith, Reference Frith2012b; Shea et al., Reference Shea, Boldt, Bang, Yeung, Heyes and Frith2014). According to this proposal, humans make use of “system two metacognition.” System two metacognition occurs when two or more individuals have to coordinate sensorimotor systems, enabling them to share the metacognitive representations they had previously developed individually (Shea et al., Reference Shea, Boldt, Bang, Yeung, Heyes and Frith2014). This form of metacognition enables individuals to understand the mental states of others but also to share their evaluations on their current task, for example, exchanging their confidence judgments. It allows individuals to solve problems more easily thanks to cooperation with others; consistent evidence indicates that sharing confidence judgments increases performance in joint action (Fusaroli et al., Reference Fusaroli, Bahrami, Olsen, Roepstorff, Rees, Frith and Tylén2012).

Section 7.4 will deal more extensively with the relationship between abstract concepts and social interaction. Here it is essential to mention that the processes of inner social metacognition and social metacognition might involve inner speech. Furthermore, as argued in Chapter 2, different kinds of inner speech might be at work (Alderson-Day et al., Reference Alderson-Day, Mitrenga, Wilkinson, McCarthy-Jones and Fernyhough2018). Forms of inner speech related to working memory might help us search for associated words; evaluative inner speech might contribute to monitoring self-knowledge; dialogic inner speech might help us find an explanation of a word’s meaning (Borghi & Fernyhough, Reference Borghi and Fernyhough2023). Furthermore, inner speech can help individuals in phase two metacognition, that is, in inner prepariations and articulation of questions and observations to submit to others (see Figure 7.1).

Figure 7.1 Abstract words as tools: how the processes of inner social metacognition and social metacognition might work. People might feel uncertain about a word’s meaning and, in phase 1, either inner search for words or engage in a dialogic inner speech to find a solution internally. In phase two, people might revert to others, either asking for information, in case of asymmetric expertise level, or trying to understand what the other intends with a given concept and negotiating with the other the word meaning. Inner speech can assist in phase 1 and phase 2, for example, inner articulating the questions and observations to present to others. The two phases/processes might not necessarily be in sequence but co-occur.

7.4.1 Abstractness and Autism Spectrum Condition

Autistic spectrum condition (ASC) is a neurodevelopmental condition characterized by impairments in social interaction and repetitive patterns of behavior (Roehr, Reference Roehr2013) (for a recent special issue, see Narzisi, 2020). I prefer the term ASC to the more common autism spectrum disorder (ASD) because it is broader and less connoted in a clinical sense. Studies on ASC can widely contribute to building and validating a theory on abstract concepts. Impairments in social interaction, in particular, characterize ASC. Furthermore, even if results are sometimes scattered and not always consistent, ASC individuals might have difficulties with abstraction and abstractness. To date, it has been impossible to establish a causal connection between the abstraction and abstractness ability and social ability. Still, there are various reasons why investigating ASC is critical for a grounded theory of abstract concepts that ascribes a crucial relevance to social interaction.

I first briefly review studies pointing to ASC characteristics that are indirectly relevant for the formation and use of abstract concepts, such as work on visualization, ToM, conversation and narratives, and inner speech. I focus on the literature on categorization.

7.4.2 Abstractness and Schizophrenia

In this section, I concentrate on schizophrenia. I discuss only a small fraction of studies, considering only the aspects of schizophrenia that might be relevant for a theory of abstract concepts that ascribes a crucial role to language and social interaction.

Schizophrenia is a heterogeneous psychiatric disorder. According to DSM 5, positive symptoms (e.g., delusions and hallucinations) and negative symptoms (e.g., reduced emotional expression) characterize schizophrenia. Notably, schizophrenic patients have overall language impairment in both discourse production and comprehension (Rochester, Reference Rochester2013; Swaab et al., Reference Swaab, Boudewyn, Long, Luck, Kring, Ragland and Solomon2013), communicative and pragmatics problems (Bambini et al., Reference Bambini, Arcara, Bechi, Buonocore, Cavallaro and Bosia2016), and difficulties in Theory of Mind (Parola et al., Reference Parola, Berardinelli and Bosco2018) and the social sphere. I shortly describe the problems of schizophrenic patients with pragmatics, ToM, and inner speech and illustrate one of the core features of schizophrenia, concretism.

7.4.3 Abstractness and Aphasia

I believe that we need to be able to form flexible categories in order to learn abstract categories. Take, for example, extremely flexible categories such as ad hoc and goal-derived ones (Barsalou, Reference Barsalou1983; Reference Barsalou and Bower1991). An example of an ad hoc category is “ways to escape from the mafia” (Barsalou, Reference Barsalou1983). Goal-derived categories are more stable in memory than ad hoc ones, but their members are generally collected uniquely based on a common goal. Consider the goal-derived category “birthday presents”: it can include members of different taxonomic categories, such as animals, plants, and different kinds of artifacts (books, clothes, toys, records). Members of ad hoc and goal-derived categories generally aren’t perceptually similar; these categories are low dimensional (Lupyan & Mirman, Reference Lupyan and Mirman2013; Henningsen-Schomers & Pulvermüller, Reference Henningsen-Schomers and Pulvermüller2021), that is, they cohere only on a single or a few dimensions – in this case, the common goal. Low-dimensional categories can also share perceptual dimensions – for example, when we put together all red objects. Forming low-dimensional categories implies abstracting from many perceptual and nonperceptual features and isolating a single dimension, focusing on it. Low-dimensional categories might also have thematic similarities, generally co-occuring in the same theme or situation, although not visual similarities (Kalénine et al., Reference Kalénine, Mirman and Buxbaum2012a; Kalénine et al., Reference Kalénine, Mirman, Middleton and Buxbaum2012b). This is the case, for example, of “birthday presents” and “things to take from one’s home during a fire” (Langland-Hassan et al., Reference Langland-Hassan, Faries, Gatyas, Dietz and Richardson2021).

Abstract concepts are typically low-dimensional categories (Borghi, Reference Borghi2022a; see also the distinction between sparse and dense categories in Sloutsky, Reference Sloutsky2010). They do not have many perceptual common traits and share only a few elements. In a recent paper, I have addressed the differences between concrete concepts, goal-derived categories, and abstract concepts (Borghi, Reference Borghi and Djebbara2022b). Concrete concepts (e.g., “table”) typically assemble perceptually similar exemplars with common affordances. Ad hoc and goal-derived categories (e.g., “things to take to the seaside”) collect perceptually dissimilar exemplars, endowed with different affordances but referring to common sociocultural practices. Finally, abstract concepts (e.g., “truth”) collect dissimilar exemplars, the affordances of which might be based on common sociocultural practices. For example, being sincere and telling the truth are encouraged in various cultures. It is possible to hypothesize that learning ad hoc and goal-derived categories, which do not refer to similar elements and are low-dimensional categories, might bootstrap the later acquisition of abstract concepts (see Chapter 8.1.2, Figure 8.1).

Because low-dimensional categories have only a few common features, individuals might need more online linguistic support to form them. A way to test this hypothesis is to investigate categorization in aphasic people. Aphasia is a condition that impairs the ability to comprehend and produce language, typically due to a stroke, head injury, or neurodegenerative disease.

Lupyan and Mirman (Reference Lupyan and Mirman2013) found that individuals with aphasia performed worse than matched controls when they had to categorize objects that shared only one dimension but not objects that shared multiple dimensions. Importantly, the adopted task did not require verbal responses: they had to press a button to select, within a given array, pictures that respected a certain criterion, for example, all the blue things versus all the pictured fruit. The performance of aphasic participants was selectively correlated with their difficulties in naming, not by severity of their aphasia, the size and localization of the lesion leading to aphasia, and their overall semantic performance. This result supports the idea that language contributes to reifying categories – labels can make categories more compact and cohesive. More crucial is the recent work by Langland-Hassan et al. (Reference Langland-Hassan, Faries, Gatyas, Dietz and Richardson2021) on aphasia. The authors had aphasic people and age-, gender- and age-matched controls perform on a trial concreteness task. Items consisted of one target image and four possible matching ones and were normed for visual similarity and common setting (which of the four images goes best with the top image?) to obtain a measure of trial concreteness. Trial concreteness does not correlate with conceptual concreteness, as assessed through imageability and concreteness/abstractness ratings: it represents an independent measure of abstract thought obtained using nonverbal stimuli. During the experiment, participants had to decide which of the four presented images matched better with the target; measures were accuracy and response times. Confidence ratings and response times were also collected. Lower trial concreteness resulted in lower accuracy, lower confidence, and longer response times. Aphasic individuals scored lower than controls in both accuracy and response times. More crucially, aphasic individuals scored lower than controls in response times, at the decrease of the trial concreteness level. However, the effect did not extend to confidence scores and response times, suggesting that aphasic patients might have metacognitive impairments. Notably, however, their confidence response time negatively correlated with their linguistic ability, meaning that people might feel less confident when unable to find an appropriate label. Overall, the results support the hypothesis that language provides vital support for abstract thought and metacognition about abstract thought.

7.4.4 Summary: Autism, Schizophrenia, Aphasia

Linguistic and social experiences are crucial for all concepts, but they become more critical at increasing conceptual abstractness. Research on autistic and schizophrenic individuals suggests a pivotal role for social interaction, and inner speech, in acquiring and developing the capability to use abstract concepts. Research on aphasia highlights the importance of language in supporting our ability to use abstract thought. New data are necessary to come to definite conclusions. However, the neuropsychological and psychiatric studies I have overviewed indicate that both language and social interaction play a paramount role in developing abstract concepts.

8.1.1 First Abstract Concepts, Language, and Social Interaction in Infants

Abstract words are typically acquired later than concrete ones, as the date of age of acquisition reveals. Analyses have revealed that concrete words represent 75 percent of the most frequent words of children (from first to eighth grade) and only 28 percent of the most used words of adults (Brown, Reference Brown1957, reported in Schwanenflugel, Reference Schwanenflugel1991). During childhood, the number of abstract words increases dramatically: age of acquisition ratings suggest that at the age of four, less than 10 percent of known words are abstract, while they amount to more than 40 percent of all words at age twelve (Ponari et al., Reference Ponari, Norbury and Vigliocco2018).

Studies on infants specifically focusing on abstract concepts are few. An important one investigates the comprehension of abstract and concrete words in 6–16-month-old infants (Bergelson & Swingley, Reference Bergelson and Swingley2013). Parents said the name of one of two videos shown to the infants. Results revealed that 10-month-olds, but not younger infants, looked at the video effectively named by their parents; thus, already at ten months, infants showed understanding of simple abstract words like “uh-oh” and “all present.” Children’s success with abstract words significantly increased around fourteen months. Notably, infants demonstrate comprehension of concrete words at around six months of age. The greater difficulty of abstract concepts might be due to their referential uncertainty, enhanced by the fact that parents pronounce the concept nouns in the presence of their referents less often than with concrete concepts. Hence, abstract words are harder to learn because it is more difficult for children to establish referents. What changes at ten and fourteen months allow the acquisition of abstract concepts? Notably, two crucial modifications in social cognition occur at around ten and fourteen months. The ability to follow the gaze of others at ten months might be particularly decisive in selecting and forming categories based on elements that do not have common perceptual features. Even more crucial might be the emergence of a developed form of joint attention at fourteen months, in which the infant understands what both she and the adult know together. Interestingly, there was no correspondence between parental reports and naturalistic data. While parental reporting underlined continuity in word acquisition, in line with an accumulator model, naturalistic data on comprehension obtained in the lab revealed a boosting effect between 10–12 and 14 months old (Bergelson, Reference Bergelson2020), corresponding to the development of critical social abilities and the emergence of the first produced words. Hence, parents do not seem to notice the boosting effect on their children’s language acquisition; they also use abstract words less consistently with their referents than they do with concrete words. Similar findings are reported by Katharina Rohlfing and collaborators (Grimminger et al., Reference Grimminger, Rohlfing, Lüke, Liszkowski and Ritterfeld2020), who showed that parents already used decontextualized talk when speaking to infants in their first year of life. We can consider this decontextualized talk more abstract. For example, observing parent–child interactions, the authors found that when children were 12 months old, almost 16 percent of all parental utterances were decontextualized, with an individual variation from 3 to 33 percent. Most of these decontextualized utterances focused on nonpersonal stories, past events, and comparisons of something present to something known by the child. Notably, whether or not decontextualized, that is, more abstract speech, was used, had no differential impact on later language development, as assessed at twenty-four months.

Studies on pacifier use (see Chapter 7.1.1) have shown that language widely contributes to the acquisition of abstract concepts. These studies have indicated that pacifier use, influencing the mouth movements, impacts both the conceptual relations produced in a definition task and the response times in a categorization task, leading to longer response times with abstract compared to concrete and emotional words (Barca et al., Reference Barca, Mazzuca and Borghi2017; Reference Barca, Mazzuca and Borghi2020; Barca, Reference Barca2021). However, these studies investigated the long-term effect of pacifier use in first- and third-grade children. As I saw with Laura Barca and Claudia Mazzuca in a recent study with 24–36-month-old children, instead, that pacifier use does not impact the pattern of acquired words in infants. Many possible reasons underlie this result. First, parental reports might be less reliable than direct observation. Second, and more crucially, the investigated words differ consistently in the studies my colleagues and I performed on school-age children and this study on infants. To access conceptual knowledge in infants, we used a questionnaire compiled by parents, the Primo Vocabolario del Bambino (Caselli et al., Reference Caselli, Bello, Rinaldi, Stefanini and Pasqualetti2015); adapted from MacArthur-Bates Communicative Development Inventories (CDIs). Hence, for example, we asssessed whether infants use, among others, words related to sounds of nature, vehicles, food, body parts, people, routines, and verbs. Six independent researchers evaluated the concreteness/abstractness of all the words on the questionnaire. Results showed that the number of produced words did not vary as a function of pacifier use. In addition, there was no clear pattern, indicating that prolonged pacifier use might influence more abstract than concrete concepts. Aside from the fact that parental reports might not be entirely reliable and that we used ratings obtained by adults and not by infants, the terms defined as abstract were not particularly complex. For example, we classified objects (e.g., “box,” “toothbrush,” “pillow”) as more concrete than temporal expressions (e.g., “tomorrow,” “morning,” “late”). Yet, even these temporal expressions might not be “hard words” (Schwanenflugel, Reference Schwanenflugel1991; Gleitman et al., Reference Gleitman, Cassidy, Nappa, Papafragou and Trueswell2005) to learn, that is, words in order to understand the meaning of which, people need to use inner speech or revert to others to ask for information. Therefore, children might recruit the mouth motor system later, when they have to learn harder and more complex concepts like “justice.” Inner reexplaining of the meaning, talking to oneself, and reverting to others because of the low confidence and uncertainty on the word meaning might occur more plausibly for words like “freedom” than for words like “to carry,” even though “to carry” is more abstract than “table.”

So, children learn abstract concepts later than concrete ones. But why do people learn abstract categories, and what is their function? A recent study by Lewis et al. (Reference Lewis, Colunga and Lupyan2021) based on parentally completed checklists of children’s vocabulary suggests that children who learn earlier superordinate categories later experience a more robust vocabulary growth; the study focuses on the period between 19 and 26 months old. To be precise, children who have acquired more general nouns and verbs (e.g., “animal” instead of “fox”; “move” instead of “run”) learn more words associated with the previously acquired superordinate term. The authors rule out the result being due to general intelligence. However, they cannot exclude the effect being due to the children’s exposition to a culturally richer milieu, where a greater number of different words are used.

One potential problem of work on infants is that most studies make use of parental reports and parental ratings of children’s production (e.g., Frank et al., Reference Frank, Braginsky, Yurovsky and Marchman2017); while this has the advantage of reaching many children and parents, it might also limit the reliability of the results (Lewis et al., Reference Lewis, Colunga and Lupyan2021). Ecological methods, in which children freely interact with their parents, are crucial to understanding how learning the first abstract concepts occurs. Specifically, these paradigms allow researchers to consider the role of memory and investigate how learning happens in time. Wonderful examples concern analyses of learning new words in children, in which the children wear a head-mounted camera, allowing online checking of their interactions with the surrounding objects and people (e.g., Yu et al., Reference Yu, Suanda and Smith2019, Reference Yu, Zhang, Slone and Smith2021; Slone et al., Reference Slone, Smith and Yu2019). Recent evidence on a toy session in which dyads composed of toddlers (mean age: 21 months) and parents interacted suggests that toddlers and parents represent a system endowed with memory. Hence, they do not simply perform curiosity-driven actions but engage in interactions characterized by a story-like structure (Karmazyn-raz & Smith, Reference Karmazyn-Raz and Smith2023). For example, toddlers interact with toys, followed by parents who interact with the same toys and talk. Toddlers return to their preferred toy and explore new toys and situations, and parents and toddlers always maintain tight time coordination of manual actions with objects and talk. Their actions derive from the previous ones and influence the next, making optimal conditions for learning, passing from working memory to stable memories. Increasing the knowledge of the relations between toys, their predictive errors on the action that objects afford decreases. Notably, this study focuses on a section of free play with toys, where parents and toddlers coordinate in interacting with objects. Nevertheless, one could predict that the role of memory and time can be even more critical in learning abstract concepts, where the social tuning between the actors might be particularly crucial (see Section 8.1.2).

Ecological studies investigating children’s interaction with their parents or peers have a further advantage: they allow the direct tracking of children’s vocabulary in use, even if the samples are not large. In a recent study (Bellagamba et al., Reference Bellagamba, Borghi, Mazzuca, Pecora, Ferrara and Fogel2022), my colleagues and I analyzed the emergence of abstract concepts in interactive mother–child situations. We used a longitudinal database on infant–mother interactions made available by our colleague Alan Fogel. The transcription of children’s vocabulary production was based on a previous study by Camaioni et al. (Reference Camaioni, Aureli, Bellagamba and Fogel2003). We coded the words produced by eight children, four boys, and four girls, at ten months and then at twenty-four months. We distinguished the words in three degrees of abstractness: low, intermediate, and high. For example, words with a low level of abstractness included, among others, objects that children could manipulate, visible parts of the bodies, and words referring to their mother, who was present. Words of intermediate level included names of relatives and body parts that were not present or visible (e.g., “grandma,” “eyes”), questions, descriptive words (e.g., “nice”), locations, and internal physiological states (e.g., “hot”). Finally, words of high abstractness level encompassed pronouns, quantifiers (“more,” ”some”), abstract routings (e.g., “all done!”), and internal states referring to emotions, volition, cognition, and morality (e.g., “love,” “need,” “pretend,” and “bad”). Results clearly showed that the abstract concepts increased with age, with a significant shift between 12 and 15 months old and between 22 and 24 months old. While there was a general increase in abstract concepts, a fine-grained analysis of the various age groups allowed us to record the emergence of their different kinds. Within abstract concepts, abstract routines were already present from the first age group (12–15 months); negations, personal pronouns, and numbers emerged around 16–18 months; while abstract assertions and quantities emerged around 19–21 months. Our results also suggest why creating and using abstract concepts might be important. Indeed, we found that the larger the vocabulary size, the higher the number of abstract concepts produced. This result can have two different interpretations. First, it might suggest that children must master language proficiently to use abstract concepts (Gleitman et al., Reference Gleitman, Cassidy, Nappa, Papafragou and Trueswell2005) since language is essential to learning new abstract concepts. Alternatively, this finding shows that both superordinate and abstract concepts promote language learning, extending the results of Lewis et al. (Reference Lewis, Colunga and Lupyan2021). Whatever the correct interpretation, in both cases, these results highlight a tight relationship between language learning, social interaction, and abstract concept learning.

To summarize: studies on infants, realized both with questionnaires and naturalistic observations, show that abstract concepts, even the simplest ones, emerge later than concrete ones. Evidence suggests that such emergence might be motivated by significant social changes, such as the ability to follow the gaze of others and to engage with them in complex forms of joint attention and joint action. While they talk to infants, parents use words that often do not coexist with their referents; this contributes to making the social abilities to follow their gaze and understand what they know crucial. Besides pointing out the importance of social interaction for acquiring abstract concepts, I have addressed why people learn to master abstraction and abstractness. Some studies suggest that learning superordinate and abstract concepts is strictly linked with the richness of linguistic vocabulary. Further research should determine whether learning superordinate and abstract concepts plays a causal role in facilitating linguistic acquisition in infants or whether they are simply correlated to the broadness of the productive vocabulary. I will now review some studies on social interaction in children at an age at which they master more abstract concepts.

8.1.2 Abstract Concepts, Language, and Social Interaction in Children

As I have mentioned, abstract concepts grow from 10 percent of a child’s production at 4 years old to 40 percent at 12 years old. Why such a marked increase? What happens at age 12? The school context undoubtedly has an influence, but there might be maturational factors that allow children to develop their capability for abstractness. I hypothesize that children learn to rely more on other people and become more sophisticated in their ability to choose their informants.

In a recent study, my colleagues and I tested the hypothesis that 5–7-year-old children will rely more on others when comprehending abstract rather than concrete concepts (Paoletti, Fini et al., Reference Paoletti, Fini, Filippini, Massari, D’Abundo, Merla and Borghi2022). We designed a very simple task similar to a lexical decision: children listened to names and were required to decide whether they existed; if the word existed, they had to press a button. We presented concrete and abstract words and nonwords. We thus recorded response times and accuracy and, through thermal imaging, the nasal tip’s temperature. Analysis of response times and errors revealed that children performed better with concrete than with abstract concepts – for the first time, we demonstrated the concreteness effect with response times in very young children (see also Lund et al., Reference Lund, Sidhu and Pexman2019). Notably, we found a higher increase in the nasal tip’s temperature with abstract concepts. This increase in temperature in specific facial regions correlates with the emotional state and the signals of the sympathetic and parasympathetic systems. The parasynthetic system activation leads to vascular relaxation and an increase in temperature (Aureli et al., Reference Aureli, Grazia, Cardone and Merla2015) and indicates prosocial behavior. This finding suggests that, with abstract concepts, children might have been more uncertain, as the pattern of response times and errors suggests. This uncertainty leads to a more substantial reliance on others. Whether these others are the people in the experimental room or hypothetical others that might support the children, even if not directly present in the lab, cannot be determined based on current data. The mechanism of relying on others may be activated, as one of the possible outcomes of an internal monitoring process, when one experiences uncertainty and little confidence (see Borghi et al., Reference Borghi, Fini, Tummolini, Robinson and Thomas2021; Fini & Borghi, Reference Fini and Borghi2019).

Provided that, with abstract concepts, individuals rely more on others, it is important to achieve the ability to select the “right” others, that is, the others who are competent and can provide rich information. What must children develop to be able to choose the right experts? According to Kominsky et al. (Reference Kominsky, Zamm and Keil2018), people need to have some notion of causal mechanisms to ask for help from others. Kominsky et al. (Reference Kominsky, Zamm and Keil2018) performed experiments with 7–10-year-olds and adults and demonstrated that the more causally complex an object is, the more likely individuals are to ask for help. For example, children are more likely to revert to others when interacting with objects as complex as a microscope, which has various diverse components. Even 5-year-olds have a sense of causal complexity and use it in asking for help. For example, 5- and 6-year-olds, but not 4-year-olds, expect complex artifacts to be internally complex and are sensitive to the number of internal parts. Yet, only older children are sensitive to the differences and interconnections between these parts (Ahl et al., Reference Ahl, Amir and Keil2020). In fact, the ability to detect when something is complex develops more clearly around 7–9 years of age. Children of 7–9-years-old and adults, but not 5–6-year-olds, were able to determine the complexity of an object and decide whether they needed help to fix it. To do this, they were able to select relevant information (e.g., the difficulty in building an object or in fixing it) over irrelevant information (e.g., the difficulty in spelling its name or in seeing it at night) (Ahl et al., Reference Ahl, DeAngelis, Stephenson, Joo, Keil and Keil2018). The difficulty of younger children thus consists in not being able to filter and select the right pieces of information. Hence, the humans’ ability to judge whether they need help and select the right informants develops gradually, with a major change after age six. In another study on numerical competence (Kominsky et al., Reference Kominsky, Langthorne and Keil2016), the authors investigated whether children distinguish between ostentatious confidence and real competence. Kindergarten and first-graders did not choose informants who told them that they did not know, for example, the number of all the leaves of all the trees in the world, and preferred informants who pretended to know. After fourth grade, children were able to identify competent people on whom to rely, that is, people who told them it was possible to know the number of windows of the White House but not to foresee the most popular boy name on October 2224 (see Chapter 3.2.2). Despite these age differences, even younger children could distinguish what information could be knowable and what could not. However, they were either not able to integrate this knowledge with that about the informants, or they tended to believe what other people told them anyway.

Some studies show that the acquisition of emotional concepts precedes that of purely abstract concepts (see Chapter 5). Researchers have proposed that emotional terms can provide a bootstrapping mechanism – being the first words children learn without an object as a referent, emotional terms can help children learn more abstract concepts. Ponari et al. (Reference Ponari, Norbury and Vigliocco2018) offer evidence consistent with this. They performed an auditory lexical decision task with children from age 6 to 12 and found the facilitation of positive words over negative ones in 8- and 9-year-olds. Valence did not affect children of the other age groups, likely because younger children did not know many words and older children had already acquired neutral abstract words. Lund et al. (Reference Lund, Sidhu and Pexman2019) employed an auditory lexical decision to test 5-, 6-, and 7-year-olds; overall, they found a response times facilitation of positive over negative words; they also found an interaction showing that 6-year-olds processed faster positive abstract words than neutral abstract words. Kim et al. (Reference Kim, Sidhu and Pexman2020) also found effects of valence in a recognition task with 7–8-year-old children: accuracy was higher for negative than for neutral words, but only for abstract words, while no effect of valence was present in concrete words.

My colleagues and I propose that goal-derived categories might also contribute to learning more abstract concepts (Borghi et al., Reference Borghi, Barca, Binkofski, Castelfranchi, Pezzulo and Tummolini2019; Borghi, Reference Borghi and Djebbara2022b). The members of goal-derived categories are generally assembled based on common goals and do not have many common properties (Barsalou, Reference Barsalou1983; Reference Barsalou1985). For example, the goal-derived category “birthday presents” can include records, books, animals, and plants – elements that belong to disparate taxonomic categories. Hence, they violate the correlational structure of the environment. Unlike other low-dimensional categories (Lupyan & Mirman, Reference Lupyan and Mirman2013), such as “red objects,” the feature they have in common is not a perceptual one. In this respect, they strongly resemble abstract concepts. Differently from the exemplars of abstract categories, the members of goal-derived categories can be objects (e.g., “birthday presents”), and goal-derived categories based on shared goals. However, goal-derived categories, like abstract concepts, can also refer to activities (e.g., “things to do during the weekend”). According to an influential view (Lucariello & Nelson, Reference Lucariello and Nelson1985; Nelson, Reference Nelson1988;), slot-filler categories, based on a shared function, are the basis of the formation of superordinate categories. Hence, preschoolers will not possess taxonomic categories but event-based categories, where each slot can be filled by various elements. For example, the event “sleeping” includes a slot related to possible sleep locations and another to sleeping time. Initially, each slot can be filled by only one value (e.g., “bed”), but with further experience, other values might fill it (e.g., “armchair,” “train”). Slot-filler categories posit the foundations for the development of taxonomic categories; for example, the superordinate category “food” is composed of the slot-filler categories “food for breakfast,” “food for lunch,” etc. As I have explained, the creation of superordinate concepts pertains to the process I called abstraction, which is related to but different from abstractness (for a similar distinction, see Bolognesi, Reference Bolognesi2020; Bolognesi & Caselli, Reference Bolognesi and Caselli2022). Even if no evidence has been collected so far, one can hypothesize that children become able to form abstract concepts only if they can apply the slot-filling mechanism and create low-dimensional categories, especially if the feature(s) the various members share is not a perceptual one (Figure 8.1). Across development, the timeline of word acquisition reflects this pattern: abstract concepts are acquired later than concrete ones and later than superordinate ones.

Figure 8.1 The mechanism that might lead from taxonomic categories to goal-derived categories to abstract concepts. While the members of taxonomic categories are perceptually similar, those of goal-derived ones may belong to different taxonomic categories and are flexibly assembled based on common goals. Similarly, the members of abstract categories are heterogeneous, and this partly explains why linguistic explanations facilitate their acquisition.

In sum, in this section, I have outlined some mechanisms that might lead children to learn abstract concepts. A precondition to learning abstract concepts like those used by adults is, in my view, the development of the ability to appropriately interact with others, understanding if they are competent or are pretending to know what they do not know. Notably, sociality is fundamental both to acquiring concrete concepts, either basic or superordinate ones, and to learning abstract concepts. However, social support is needed for different reasons. When infants learn the word “bottle,” parental support is needed to establish reference. Social support might be even more necessary when children learn superordinate words, like “animal,” which are more general, and goal-derived categories, like “things to take on holiday” (Bellagamba et al., Reference Bellagamba, Borghi, Mazzuca, Pecora, Ferrara and Fogel2022). However, I hope to have clarified that abstractness, although related, is different from abstraction. Superordinate words often refer to bounded referents, whereas abstract concepts do not; they instead refer to a heterogeneous set of situations, events, and states. When children acquire the abstract concept of “freedom,” other people might help them think of the different and sparse referents to which the category points. But this might not be the whole story. To form and use abstract concepts, children have to become able to go above what they perceive through the senses. For example, they should see a scene in which someone runs on the grass not only as an exemplar of the event “running” but also as an exemplar of the category “freedom.” Here, the others might not only be needed to establish reference (Dove, Reference Dove2016) but also to scaffold children to engage in abstract thought, a special kind of thought that involves fantasy, imagination, emotions, etc. Others might thus become “intellectual reference points”(Fini & Borghi, Reference Fini and Borghi2019). Hence, for future research, it will be paramount to investigate the acquisition and use of concrete concepts of different hierarchical levels, goal-derived concepts, and abstract concepts in different social contexts. In this framework, the role played by social competencies becomes fundamental for the learning and use of abstract concepts. I propose that children start to acquire and use many abstract concepts when they can rely on other people and select the right informants. Among possible mechanisms that bootstrap the acquisition of abstract concepts, an important role is played by the acquisition of emotional concept (the first concepts that do not have an object as referent) and goal-derived categories (the first concepts whose exemplars are not assembled on the basis of perceptual similarity). Acquiring these categories can give them a sense of the possible flexibility of concepts and acquire the awareness that concepts do not necessarily have a stable referent, which is crucial to using abstract concepts and words. The acquisition of emotional and goal-derived concepts can pave the way for more sophisticated abstract concepts, for which they have to engage in a novel form of thinking, a thinking that “goes beyond.”

Once abstract concepts have been acquired, what happens over the lifespan? Due to space limitations, I will not address this topic in-depth but just give some glimpses into it. In the literature, there is controversial evidence (for a review, see Borghi & Setti, Reference Borghi and Setti2017). Studies on recall indicate that the advantage of concrete over abstract words (concreteness effect) declines with age due to the lower ability of older adults to rely on sensorimotor properties and their preserved linguistic capabilities, especially comprehension ones. Results on lexical decision seem to contradict this finding, showing that the concreteness effect is maintained in the elderly. However, careful analysis suggests that this capability relies on the compensatory role of linguistic, and especially phonological, neural networks. These results are consistent with a recent proposal that my colleagues Matthew Costello, Jennifer MacCormack, Paek Eun Jin, Uma Jallow, and I advanced to account for the fact that, with age, people experience reduced sensorimotor, interoceptive, and emotional abilities but no strong decline of language capabilities (Costello & Bloesch, Reference Costello and Bloesch2017). We propose that even if language is never fully detached from its sensorimotor basis, with age, it allows a more detached mode of engagement, compensating for the reduced embodied experiences. This is consistent with the role the elderly play in our societies, in which they are appreciated for their wisdom and their ability to evaluate with detachment from the here and now. Hence, abstractness capability emerges slowly in children, is preserved in the elderly, and can cover an essential social role.

8.2.1 Abstract Words and Social Contexts

Concepts are not stable, and people update them continuously in light of the current situations and new experiences. In the 1980s and 1990s, the early work of Larry Barsalou highlighted this flexibility, viewing concepts as flexible and variable constructions in working memory (e.g., Barsalou, Reference Barsalou1993). And yet, a version of concepts as stable entities has prevailed, motivating the choice of experimental paradigms in which the context is hardly considered (for a critique, see Yee & Thompson-Schill, Reference Yee and Thompson-Schill2016). Studies on abstract concepts are no exception.

Most studies on abstract concepts are performed using linguistic stimuli, consisting of the words that express the concepts. Typical studies employ single words, nouns or verbs, or simple sentences, often without subordinates. The most common tasks are ratings, feature listing, lexical decisions, and property verification. These methods have led the researchers to capture some of the most critical characteristics of concepts and identify possible abstractness hallmarks. However, investigating concepts by examining words in isolation might lead to distortions. As convincingly argued by Barsalou et al. (Reference Barsalou, Dutriaux and Scheepers2018), statistical regularities:

An example of a method that investigates concepts in action is the visual word paradigm: participants’ eye movements are recorded as they hear or produce sentences while looking at real objects or pictures in a display (Tanenhaus et al., Reference Tanenhaus, Spivey-Knowlton, Eberhard and Sedivy1995; Huettig et al., Reference Huettig, Rommers and Meyer2011).

Curiously, however, the role of context for abstract concept processing has been recognized for a long time. Classic context availability scores revealed that, typically, people evaluate abstract concepts as associated with more contexts but linked less to a specific context (Schwanenflugel & Stowe, Reference Schwanenflugel and Stowe1989). Along similar lines, Davis et al. (Reference Davis, Altmann and Yee2020) propose the notion of situational systematicity, defined as “the extent to which the same objects and relations constitute the concept (i.e., schema) regardless of the situation” (p. 2). The authors contend that abstract concepts have lower situational systematicity, that is, members and relations composing abstract concepts are more dispersed in space and time. For example, people can instantiate the concept of “justice” by referring to a process that might occur over the years. Due to their low situational systematicity, abstract concepts are not processed bottom-up, starting from sensory experiences, but top-down, starting from previous knowledge.

Despite this long-standing interest in the associations between abstract words and contexts, there are few studies in which abstract words are embedded within contexts. This evidence indicates that the context facilitates conceptual processing. For example, McRae et al. (Reference McRae, Nedjadrasul, Pau, Lo and King2018) had participants perform a lexical decision task with abstract words preceded by pictures of situations. The pictures could be related to the word (e.g., two girls who eat the same cob of corn primed the concept “share”) or unrelated. The related prime speeded up response times, provided that participants had sufficient time to process the picture (one second) efficiently. There was also facilitation of related over unrelated primes when participants had to focus on the picture to decide whether it depicted a normal situation. This evidence suggests that abstract concepts are related to real events and situations in more complex ways than concrete concepts.

Investigating concepts in isolation has many limitations, but these limitations are particularly noteworthy when considering abstract concepts. As argued in Chapter 5, concepts are not entirely abstract or concrete, and both their meaning and abstractness level can consistently shift depending on the context. Crucially, abstract concepts are primarily associated with specific contexts related to interaction with others and oneself (for a broad overview, see Troyer and McRae, Reference Troyer and McRae2021). In Section 8.1, we saw how studies on conceptual acquisition reveal the importance of the social context for abstract concepts’ acquisition. In Chapter 7, I introduced the notions of “inner social metacognition” and “social metacognition,” arguing that people need more scaffolding support from others to process abstract than concrete concepts. In two phases, which are not necessarily sequentially organized, people will search for the meaning of abstract concepts through inner dialogue; in a second phase, they will ask for support or seek a confrontation with others. Thus, the social context is determinant for the acquisition and for the processing of abstract concepts, both in asymmetric situations, when we learn from others, and in symmetric ones, when we have to discuss and negotiate meaning with others. Notably, in neither case, is the support of others guaranteed. We might search for others’ help and receive a refusal. Various social dynamics might be at play, from cooperation to conflict (Di Paolo et al., Reference Di Paolo, Cuffari and De Jaegher2018). And yet, we might cooperate in maintaining a dialogue, an interaction (Pickering & Garrod, Reference Pickering and Garrod2021). In this sense, words, particularly abstract words, can be considered social tools.

To my knowledge, only a few studies investigate the relationship between abstract concepts and social context. In a recent study submitted for publication, my colleagues Daniele Nico, Elena Daprati, Luca Tummolini, and I intended to test whether different social contexts, favorable and unfavorable, modulate abstract concepts’ processing. Participants’ task was to decide whether an abstract or concrete definition matched an abstract or concrete target word. Correct responses gained them points. The instructions told half the participants that they were playing against an opponent (who did not exist) and the other half that they were playing alone. In the condition where a competitor was present, participants were told that the experimenter could decide how many trials to assign to them. The socially favorable condition was created by the experimenter allowing participants to play in two-thirds of the trials, and hence to gain many points; in the socially unfavorable condition, participants could play in only one-third of the trials. There was also an intermediate condition, in which participants could play in half the trials. Dependent variables were the time spent reading the definition and the time required to decide whether it matched the word. The most exciting result showed a selective slowing of responses to abstract words in the socially favorable competitive condition compared to when the participants believed themselves to be playing alone. Different interpretations are possible. One is that participants responded more slowly when their effort was greater and felt more under pressure due to the competition; yet, this does not explain that no difference exists between socially favorable condition and the intermediate and unfavorable conditions. It is also possible that, because people especially need the support of others with abstract concepts, lacking such support might explain the longer response times found with abstract concepts when an opponent was present. However, this explanation does not clarify why this would happen in a favorable social context. A further possibility is that, in this situation, participants tended to feel more pressure because they were favored; hence their performance decreased. There is an additional, intriguing interpretation that I prefer. We have seen that, because of the higher complexity of abstract than concrete concepts and the higher uncertainty they generate, people will turn to others more in order to grasp abstract concepts than to grasp concrete concepts (Borghi, Reference Borghi2022a). Consequently, the presence of others should enhance abstract conceptual processing under the condition that people perceive others as potentially well disposed towards them. Following this reasoning, participants should have performed better when the experimenter assigned them more possibilities (two-thirds of the trials). This was not the finding; rather, the results showed that participants performed worse in the competitive situation precisely in this condition. The competitive situations impaired performance with abstract concepts, especially when the participants had more to accomplish. A final intriguing explanation is that the favorable social situation and the processing of abstract concepts recruit similar mechanisms and resources, resulting in an interference effect, thus leading to longer response times. Whatever the explanation, the results clearly show that manipulating the social context has a selective effect on the processing of abstract concepts.

In another recent preregistered study, my colleagues Caterina Villani, Stefania D’Ascenzo, Michele Ubertone, Mariagrazia Benassi, Corrado Roversi, Luisa Lugli, and I intended to investigate whether social context might prime the processing of concrete and abstract concepts. Legal experts and nonexperts performed a go-no-go task. When they read a word referring to a planet or atmospheric and astronomical phenomenon, they had to refrain from responding; otherwise, they had to press a button. Target words referred to institutional and theoretical abstract concepts (e.g., “contract,” “energy”) and food and tool concrete concepts (e.g., “bread,” “knife”). Before seeing the words, participants saw an image prime. Primes belonged to four different types: social-action (e.g., dancing), linguistic-action (e.g., talking), linguistic-textual (e.g., reading), and control (e.g., the image of a landscape). Importantly, the social and linguistic primes but not the control prime impaired the processing of abstract but not concrete concepts, likely because of a conflict between the linguistic and social situations evoked by the prime and those elicited by the target words. Interestingly, the effect was more marked with primes evoking interpersonal communication (linguistic-social) than purely social-action and linguistic-textual situations. Hence, this priming study confirms that social and linguistic contexts modulate abstract concepts’ processing.

The short review in this section shows that context facilitates the processing of abstract concepts. More specifically, it reveals that contexts that refer to social interactions are a more effective prime for abstract than concrete concepts and that the social dynamics present in the experimental setting – favorable, competitive, etc. – selectively influence their processing.

8.2.2 Abstract Words and Real-time Contexts

We have seen that a significant step forward in research on concepts involves investigating them in situated contexts. However, conceiving words as social tools requires rethinking the methods for addressing abstract concepts even further. Due to the extreme variability of abstract concepts (Falandays & Spivey, Reference Falandays and Spivey2019; Glenberg, Reference Glenberg2019), I believe that research should investigate them online, in real-time interactions during their use (Villani, Orsoni, et al., Reference Villani, Orsoni, Lugli, Benassi and Borghi2022). At the same time, it is essential to have experimentally controlled situations. One should find a combination of ecological methods and experimental control.

To my knowledge, there are only a few studies investigating concepts in real-time situations. Zdrazilova et al. (Reference Zdrazilova, Sidhu and Pexman2018) had participants play the taboo task: participants had to explain the meaning of a concrete or abstract word to a partner, avoiding using the word itself. Zdrazilova and colleagues recorded and coded the descriptions and the produced gestures. They found that, with abstract words, participants more often produced person (agent) and introspective utterances; with concrete words, they more often produced object/entity utterances; taxonomic relations pervaded across both categories. Interestingly, the results on the linguistic production were very similar to those obtained by Barsalou and Wiemer-Hastings (Reference Barsalou, Wiemer-Hastings, Pecher and Zwaan2005) in a feature listing task: abstract concepts elicited more social and introspective features. Importantly, participants rarely used synonyms, but, especially with abstract concepts, they produced complex descriptions that made use of different modalities. As to gestures, concrete and abstract concepts evoked distinctive gestures. With abstract concepts, participants used more metaphoric and beat gestures. Metaphoric gestures represent abstract ideas, like moving the hands forward to talk about the future, while beat gestures don’t have a specific semantic meaning but emphasize the speech’s prosody. With concrete concepts, participants used more iconic gestures, representing physical objects or events. The presence of iconic gestures with concrete concepts and metaphoric gestures with abstract ones is compatible with the conceptual metaphor theory (Lakoff, Reference Lakoff2006; Lakoff & Johnson, Reference Lakoff and Johnson2008; Winter et al., Reference Winter, Marghetis and Matlock2015; see also Chapter 6.2 of this volume). However, it is possible that the meaning of concepts cannot be exhausted through metaphoric and iconic gestures. Beat gestures can contribute to filling this gap; the presence of beat gestures can also attest to the critical role of verbal information, which they highlight and underline, for abstract concepts. Crucially, communicative gestures, for example, moving toward partners to invite them to complete a sentence or clapping hands at the end of the other’s turn, although very frequent, did not differ between concrete and abstract concepts. According to the social metacognition hypothesis, the authors should have found more communicative gestures with abstract concepts, but this was not the case. One possible explanation is that because both members had to reach an explicit common aim, they could take for granted that the other would collaborate; thus, the cooperation didn’t have to be enhanced with abstract concepts. To understand better what was going on, one could analyze whether there were differences between the gestures of the two partners, the one who had to explain and the partner who had to guess the word meaning. I would predict more communicative gestures performed by the second partner because the other would be more necessary as support to accomplish the task, especially to guess abstract concepts.

Imagine being in front of a photograph; your task consists of guessing the concrete/abstract concept to which it refers, e.g., “fleet” or “friendship.” You are first asked to guess abstract concepts (block 1), then concrete concepts (block 2); the order of the two blocks is counterbalanced. If you are not able to guess, you are allowed to ask for suggestions from two experimenters, a different one for the block of concrete and abstract concepts. After each block, you are invited to perform a joint action task, grasping a bottle with two avatars, each guided by a different experimenter. Importantly, you are informed that another guessing session will follow after the joint action session. My colleagues and I adopted this procedure in a recently published study (Fini et al., 2020). Kinematics analyses revealed that the movement synchrony between the participant and the avatar, that is, the time delay between the contact times of their index and thumb with the bottle, was higher when the avatar was guided by the experimenter offering hints on abstract concepts.

In other words, abstract concepts enhanced synchrony, promoting collaborative behavior. What is the possible cause? Likely the greater difficulty of abstract concepts rendered the experimenter’s help being more necessary in order to guess them. In line with this interpretation, both an objective index, namely, the number of required suggestions, and participants’ subjective perception revealed that others were more crucial to guessing abstract concepts. My colleagues and I interpret this result in the framework of the WAT theory as support of the notion of social metacognition. With abstract concepts, people are aware of the limitations of their knowledge – in this case, of their difficulties in matching images and words – and think of other people as possible sources of guidance and advice.

This study suggests a possible link between processing abstract concepts and prosocial behavior. However, lots of questions remain open. Does this result apply only to this task, or can it be generalized? Does it apply primarily to specific kinds of abstract concepts? Understanding whether abstract concepts’ processing impacts joint action represents a new line of research, rich in theoretical and possibly societal implications.

A different way to investigate concepts in interaction is by studying concepts mediated by words during a particular form of joint action, namely, dialogue. How do people use concepts in conversation? What differences between concrete and abstract concepts emerge? These questions motivate a recent study my colleagues and I performed (Villani, Orsoni, et al., Reference Villani, Orsoni, Lugli, Benassi and Borghi2022). Imagine you read a sentence including a concrete word or an abstract one, for example, “I made a cake,” or “I thought about destiny.” Your task involves reacting to this sentence as if you were chatting to an acquaintance. The study was performed online, and participants received written sentences and replied with written responses. The responses were then coded in terms of the content of the produced sentences (e.g., reference to the five senses, parts, thematic relations, taxonomic ones, emotions, and inner bodily aspects) and their pragmatic characteristics. Results showed interesting differences in content between concrete and abstract concepts and their subkinds, reflecting and expanding results in the literature (see Chapter 5.2). For example, concrete concepts evoked more sensorimotor properties, abstract concepts more inner ones (e.g., emotions, beliefs, etc.). More crucially, concrete and abstract concepts differed in the dynamic and interactive aspects induced by the simulation of a conversation. Compared to concrete concepts, with abstract concepts, participants produced more expressions revealing uncertainty (e.g., “mmm,” “what do you mean?”) and engaged in more interactive exchanges – they asked more questions, particularly “why” and “how” questions, and replied asking the other for more specifications. Even if they had only one sentence to respond to, they relaunched the conversation, asking for more turns. Significantly, not only do abstract concepts differ from concrete ones, but also the different kinds of abstract concepts present differences in the way they promote interactive exchanges. Specifically, philosophical-spiritual (PS) concepts (e.g., “salvation,” “religion”) differed from the other abstract concepts: consistent with their higher degree of abstractness (Villani et al., Reference Villani, Lugli, Liuzza and Borghi2019), they were characterized by more frequent signs of uncertainty (more frequent repetitions, uncertainty expressions, and questions) and the tendency to engage the interlocutor more (more turns) and to produce general statements. Physical, space, time and quantity concepts (PSTQ) concepts (e.g., “reflex”), which are more “concrete” and linked to sensorimotor properties than the other abstract concepts, elicited more “why” questions, likely because many of them are scientific terms. Finally, emotional mental state and social concepts (EMSS) (e.g., “love”), which yield more interoceptive and inner properties than the other concepts, elicited more “who” questions, attesting to the interest in the agent who experiences or is an object of the emotion. Hence, these results support the hypothesis that abstract concepts generate more uncertainty and likely involve the interlocutor more in the dialogue. Yet, this study focused on a simulated and written conversation, while it would be essential to focus on real-time conversations. It is also important to understand possible differences in conversations with a single person, chatting in front of the other or online, or public discourse, as transmitted, for example, through social media. In an ongoing study, my colleagues Claudia Mazzuca, Caterina Villani, Daniele Nico, Elena Daprati, and I asked participants to write a post on Facebook or Twitter. They had to write the post starting from different kinds of concepts: concrete tool concepts (e.g., “lamp,” “knife,” and “pencil”), PS abstract concepts (e.g., “moral,” “destiny,” and “logics”), and EMMS (e.g., “calm,” “revenge,” and “shame”). Results so far confirm that with PS people produce more questions, at least on Facebook. They also reveal that tools activate more interaction with the public, probably because speaking about them involves fewer personal matters. Does this mean that people engage more often in conversation with abstract concepts than concrete ones and find it easier? Not necessarily. My colleagues and I recently asked Italian and English people to rate how easy it is to start a conversation on different kinds of abstract concepts and concrete ones (Fini et al., in press). We found that people prefer to start a conversation with abstract rather than with concrete concepts, especially with self-sociality ones (e.g., “kindness”). When familiarity and pleasantness were controlled, the difference between concrete and abstract concepts was less marked. Basically, people prefer to start conversing about abstract concepts, that is, with concepts whose meaning is more debatable. However, they prefer to start with self-sociality abstract concepts – familiar, pleasant, and light topics – not with the more complex ones. At the same time, people want to continue conversing about abstract concepts, PS ones, which require more intellectual support or intellectual exchanges and score higher in social metacognition (Figure 8.2).

Figure 8.2 Abstract concepts and social interaction.

Panel 1: Results on social metacognition. A rating study on 124 Covid-related words showed that, with words that scored lower in Body Object Interaction, hence are more abstract, participants’ perceived confidence in their own knowledge was lower, and the need for others’ help stronger (social metacognition) (Mazzuca et al., Reference Mazzuca, Falcinelli, Michalland, Tummolini and Borghi2022); a kinematics study showed that participants were more synchronous in performing a joint action task with the experimenter helping them guess to which images abstract (rather than concrete) concepts referred, because of the stronger perceived need for her help (Fini et al., Reference Fini, Era, da Rold, Candidi and Borghi2021).

Panel 2: Results on abstract concepts in conversation. A rating study showed that participants perceived it easier to start a conversation with abstract concepts than with concrete concepts, particularly with self-sociality ones (e.g., kindness) (Fini et al., in press); A study in which participants responded to sentences revealed that they used more expression signaling uncertainty, more “why” and “how” questions, and were more frequently willing to continue the conversation with sentences involving abstract than concrete concepts. The effect was stronger with the more complex abstract concepts, i.e., philosophical-spiritual ones (e.g., “logic,” “religion”) (Villani, Orsoni, et al., Reference Villani, Orsoni, Lugli, Benassi and Borghi2022).

In current times, interaction within dyads often occurs virtually, for example, through online chatting. In a recent study with my colleagues Chiara Fini, Giovanna Cuomo, Vanessa Era, Ilenia Falcinelli, Mattia Gervasi, Claudia Mazzuca, Matteo Candidi, and Bodo Winter, we created an online interactive situation in which participants wrote sentences starting from either a concrete or an abstract word. We randomly assigned participants to two conditions: social and individual. In the social condition, they chatted in dyads starting from the cue words; in the individual condition, they wrote on the chat, knowing that another person was doing the same at the same time and that later they could read what the other had written. Subsequent measures of psychological distance through the Inclusion of Other in the Self (IOS) scale (Aron et al., Reference Aron, Aron and Smollan1992) revealed that, after chatting, participants felt closer to the other than if they had performed the task individually. Analyzing the social condition, we found that participants evaluated as more demanding conversations starting from abstract concepts, and felt they contributed more than when they chatted on concrete concepts. Crucially, with abstract concepts, but not concrete ones, their psychological closeness to the other was higher the more they felt the other contributed to the conversation. The effect did not depend on the valence and pleasantness of concrete and abstract words, which were balanced (see also Reitsma-van Rooijen et al., Reference Reitsma-van Rooijen, Semin and Van Leeuwen2007, who showed that physical closeness increased after reading a positive abstract sentence and decreased after reading a negative abstract sentence; see also Lahnakoski et al., Reference Lahnakoski, Forbes, McCall and Schilbach2020).

Overall, these studies with interactive methods reveal that when people have to chat or discuss an abstract concept, they engage more in the relationship; they feel closer to the other and are more synchronous in movement – hence, they are keener to collaborate. I think this phenomenon occurs because individuals need others who might become their intellectual referents (Fini & Borghi, Reference Fini and Borghi2019). Notice that this does not necessarily imply an asymmetric relationship between individuals. In the kinematics study I described (Fini et al., 2020), the experimenter was the holder of expertise and delivered suggestions; hence she was more expert than the participant. In contrast, in the online chatting study (Fini et al., in press), the other was not qualified as an expert. Yet, the more they contributed to the conversation, the closer they were perceived. Thus, even if expertise is not crucial, it appears that individuals consider the level of engagement of another person in the interaction. Think of scientists who discuss the novel definition of a concept: each of them will feel obligated to the others because all contribute to the final outcome. Abstract concepts offer this space to negotiate and contextualize more than concrete concepts because they have fewer physical constraints.

Mazzuca and Santarelli (Reference Mazzuca and Santarelli2022) have recently provided an example of this contestability of abstract concepts, examining the concept of gender. The authors overview the characteristics of political concepts. First, political concepts are partially indeterminate (Koselleck, Reference Koselleck2004), that is, they are open to a possible future, general, ambiguous, since they might have multiple referents, and conflictual, a potential source of disagreements. Strategies to politicize a concept are to stress its abstractness and more contestable aspects. A recent study by Mazzuca et al. (Reference Mazzuca, Majid, Lugli, Nicoletti and Borghi2020) exemplified this process. Normative (monosexual, cisgender) and non-normative individuals (e.g., queer, gender diverse, and plurisexual) produced free associations to the word “gender.” While normative individuals produced more frequently perceptual, biological, and binary features (e.g., female–male), non-normative ones focused more on experiential, social, and political aspects (e.g., “queer,” “discrimination,” “construct”). Highlighting the abstract characteristics of the concept of gender, non-normative individuals enhance its debatability and its indeterminacy without anchoring it to stable, solid, and universal perceptual grounds.

In sum, most studies on abstract concepts focus on single words and simple sentences. Instead, I have argued the importance of using interactive methods to investigate concepts during their use. I have also described some examples of studies that attempt to incorporate an interactive dimension. To my knowledge, no study has investigated the emergence and use of abstract concepts employing a dual brain neuroscience approach that allows capturing the dynamics of real-time interactions (see Chapter 3.1). Interactive methods are particularly important because of the tight relationship between abstract concepts and the social dimension. Implementing both interactive behavioral methods like the ones I have illustrated, together with studies like the hyper scanning ones, which address forms of synchronization among participants, will allow researchers to capture the complex dynamics of acquisition and negotiation that abstract concepts elicit. Because of the openness of abstract concepts, people might either learn their meaning from others or negotiate their meaning with others. Hence, the presence and contribution of others can be paramount for abstract concepts’ representation. In Section 8.3, I will discuss some possible causes for why abstract concepts are so widespread.