Defining abstract concepts and words is not an easy task. It is noteworthy that the term “abstract” is represented in LIS by a sign located near the head and referring to something that cannot be touched and grasped, to something that is not material and concrete but that rather fades away.

Here we will adopt a rather broad operational definition of abstract terms. We define abstract the words and the signs that, differently from concrete ones, do not refer to single, concrete and manipulable items, but are rather grounded in situations, events, mental states, etc. Abstract words are typically rated as less imaginable as concrete ones, they are more complex than concrete words since they often refer to relations between elements rather than to single objects/entities, and they are characterized by higher intersubjective and intra-subjective variability (see Borghi and Binkofski, 2014, for clarifications on this definition). Notice however that the opposition between concrete and abstract concepts might not be a dichotomy but rather a continuum. Ratings asking people to judge the concreteness of large sets of words showed that concrete and abstract concepts are distributed in a bimodal way, falling into two big clusters (according to features, such as tangibility or visibility); within each cluster, however, the entities had different concreteness degrees (Nelson and Schreiber, 1992; Wiemer-Hastings et al., 2001).

Despite the difficulty in finding a shared definition, embodied theories of abstract concepts are numerous; below we will briefly illustrate the most important ones.

According to classical Embodied Cognition (EC) theories of abstract words there would not be a substantial difference between concrete and abstract words, since both are grounded in perception, action and emotional systems. For example, the abstract concepts of number would be grounded in action due to finger counting experience (for a review, see Fischer and Brugger, 2011). Further evidence in support of this view is obtained by studies that link words to action, for example by evidence on the Action-sentence Compatibility Effect (ACE). Results showed that judging the sensibility of sentences which describe the transfer of both concrete objects and abstract information (e.g., “giving the pizza” vs. “giving the information”) requires less time when the action implied by the sentence matches the action required to make the response (Glenberg and Kaschak, 2002; Glenberg et al., 2008a,b). This finding suggests that the mechanisms underlying transfer of abstract concepts (e.g., “the information”) are the same as those underlying transfer of concrete ones (e.g., “the pizza”) (see also Guan et al., 2013).

The other EC theories we will illustrate posit that abstract and concrete concepts and words are represented differently. The most influential one is probably the Conceptual Metaphor Theory, which states that abstract concepts are represented by image schemas derived from concrete domains. Evidence supporting this theory has shown for example that similarity is represented as closeness, categories as containers, and that the abstract notion of time is mapped onto the concrete domain of space (e.g., Lakoff and Johnson, 1980; Gibbs and Steen, 1999; Boroditsky and Ramscar, 2002; Casasanto and Boroditsky, 2008; Boot and Pecher, 2010, 2011; Casasanto et al., 2010; Flusberg et al., 2010; Lai and Boroditsky, 2013).

Further theories identify differences in content between concrete and abstract concepts. According to Barsalou and Wiemer-Hastings (Barsalou, 1999; Barsalou and Wiemer-Hastings, 2005), abstract concepts differ from concrete concepts as the first activate situations and introspective relationships more frequently. Evidence in favor of this approach is based mainly on results of feature generation tasks, showing that, whereas with concrete concepts, such as “bottle,” people tend to produce mostly properties referring to perceptual characteristics such as color, size, shape, matter, parts (e.g., “green,” “plastic,” “neck”), abstract concepts such as “freedom” evoke more frequently situations, events, introspective states (e.g., “running on the grass,” “exiting from prison,” etc.).

A novel proposal advanced by Vigliocco and colleagues (Kousta et al., 2011; Vigliocco et al., 2014) states that abstract concepts differ from concrete ones in content, since they rely more on emotional experience. Analyzing a large database Kousta et al. (2011) demonstrated that, when imageability was kept constant, emotional valence was a significant predictor of concreteness ratings. Recent brain imaging evidence (Vigliocco et al., 2014) further supports this view.

Other recent approaches, such as the Language and Situated Simulation Theory (LASS) (Barsalou et al., 2008; Simmons et al., 2008), the Symbol Interdependence Theory (Louwerse and Connell, 2011), the proposal by Dove (2011, 2014) and the Words As social Tools (WAT) proposals (Borghi and Cimatti, 2009; Borghi, 2014; Borghi and Binkofski, 2014; evidence in Borghi et al., 2011; Scorolli et al., 2011, 2012; Sakreida et al., 2013), argue that both linguistic and sensorimotor information are crucial for conceptual representation. LASS does not specifically focus on abstract concepts, but on conceptual representation more generally. According to LASS, both the linguistic and the simulation system are activated during conceptual processing; the linguistic system is faster and more superficial, while the simulation system is engaged for understanding of meaning. In some situations using the linguistic system represents a shortcut as it allows to respond immediately to a task (particularly to linguistic tasks) without necessarily accessing to conceptual meaning (Pecher and Boot, 2011). In a similar vein, Louwerse's Symbol Interdependency Theory states that shallow linguistic representations precede deeper perceptual representations (Louwerse, 2011; Louwerse and Connell, 2011; Connell and Lynott, 2012).

Compared to the other multiple representation theories, WAT (Borghi and Cimatti, 2009; Borghi and Binkofski, 2014) and Dove's view (Dove, 2014) focus specifically on the difference between concrete and abstract concepts and words. According to both views abstract concepts representation relies more on language than representation of concrete words. In his proposal on abstract concepts Dove (2011, 2014) stresses the important scaffolding role language can play and the fact that the abilities acquired thanks to language allow its use not only as a means of communication but of thought as well. The main tenets of WAT are the following: a. both concrete and abstract concepts are embodied and grounded in perception and action systems, b. for abstract concepts linguistic information plays a more crucial role than for concrete ones, c. this is due to the different acquisition modality of concrete and abstract words; d. this distributional difference is reflected in the representation in the brain of concrete and abstract concepts, e. given that representation of abstract concepts is more influenced by language, linguistic diversity has a major impact on abstract concepts representation. An important principle of the WAT proposal concerns the acquisition mechanism of the two kinds of words: with concrete words, the concrete entities (e.g., book) can be perceived together with their linguist labels. In the case of abstract words, the linguistic experience might be more important, because typically abstract words do not have a single concrete referent and also because they usually refer to exemplars differing to a great extent. Verbal labels are hence used to assemble a set of quite sparse and diverse sensorimotor experiences (e.g., we probably put together different experiences of freedom once we have learned the word “freedom”). Evidence in support of this proposal is multifaceted (for review see Borghi and Binkofski, 2014). Brain imaging studies demonstrated greater engagement of the verbal system for processing of abstract concepts and greater engagement of the perceptual and motor system for concrete concepts (e.g., Binder et al., 2005; Sabsevitz et al., 2005; Rüschemeyer et al., 2007; Desai et al., 2010; Sakreida et al., 2013), and behavioral research has shown a high cross-linguistic variability with abstract words (e.g., Boroditsky, 2011). Notably, acquisition evidence has shown that the process of acquisition of the two kinds of words might differ (e.g., Wauters et al., 2003; Borghi et al., 2011). In particular, studies on Mode of Acquisition (MOA) (e.g., Wauters et al., 2003) have shown that children acquire the meaning of concrete words, such as “bottle,” associating the word with its referent, the bottle, or with an action typically performed with or on the bottle by themselves or by another individual (Capirci et al., 2005). The meaning of abstract words like “grammar” or “philosophy,” instead, has to be explained by means of language. Finally, the meaning of a word like “tundra” can be acquired in both ways, depending on the environment where it is learned. MOA ratings, which correlate but are not totally explained by age of acquisition, concreteness and imageability, gradually change with age: initially acquisition is mainly perceptual, later it is mainly linguistic.

The question theorists adopting an EC approach have to ask is the following: is it possible to account for abstract words with a unified framework? Isn't it possible, instead, that the domain of abstract words is not homogeneous, and that the different subsets of abstract words have to be explained relying on different mechanisms? Recent studies showing fine-grained differences between subsets of abstract words (e.g., Ghio et al., 2013; Roversi et al., 2013) suggest that this might be the case. For example, abstract words as diverse as “category,” “truth,” and “risk” could rely on different mechanisms: the first could metaphorically evoke a container (Boot and Pecher, 2010), the second could evoke linguistic information and the third might activate situations. If this is true, this would lead us to abandon the overall notion of abstractness and to partition the domain into sub-domains of abstract words.

One intriguing way to understand the way in which abstract words are represented and to deal with the challenge abstract words pose to the EC perspective is to analyze how they are dealt with in sign languages. In our opinion, the way in which sign languages encode concepts into signs can help us understand how abstract linguistic items are represented, and which theory among those on abstract concepts can better account for their meaning.

Linguistic research undertaken since Stokoe's (1960) seminal work on American Sign Language (ASL) has led to the discovery and description of a very large number of national sign languages, now widely recognized by the scientific community as full-fledged, natural languages, which include Italian Sign Language or LIS (Volterra, 1987; Pizzuto and Corazza, 1996). In the last edition of the Ethnolog database 137 Sign Languages (SL) are listed. It has been shown that, even though these languages are perceived and produced in the visual-gestural (rather than in the vocal-auditory) modality, they satisfy the communicative and expressive needs of a community and possess all the basic linguistic components including phonological, lexical, syntactic and grammatical systems. Just as words of a spoken language are formed on the basis of phonemes in various combinations, all signs of a signed language are formed by combining a defined number of formational parameters (called also as cheremes). More precisely, a sign can be broken down into four basic parameters: the form or configuration taken on by the hand; the orientation the hand takes on while making the sign; the location in which the sign is performed; the movement the hand describes.

As Penny Boyes Braem pointed out already in 1981, signed lexical units are often made up of formal features visually motivated and thereby iconic. Their visual motivation is not idiosyncratic, it derives from regularities at the level of formational parameters. Handshapes, for example, are often linked to features of a sign's meaning via reference to some peculiar visual forms (Pizzuto et al., 1995; Pietrandrea and Russo, 2007). The same holds true for location and often for movement (for a comprehensive analysis of the iconicity of the LIS parameters, see Pietrandrea, 2002). In spite and beyond important structural resemblances between Sign Languages and Vocal Languages, equally relevant structural differences need to be taken in due account (Sutton-Spence, 2005; Cuxac and Sallandre, 2007; Pizzuto et al., 2007; Perniss et al., 2010; West and Sutton-Spence, 2010; Boyes Braem et al., 2012; Meurant et al., 2013; Perniss and Vigliocco, 2014). The grammar and the syntax of a sign language are expressed in various ways, including use of space, modulation of movement, facial expression and position of the trunk and shoulders. A great deal of research has been carried out on the signs used by the Deaf Italian community (a complete bibliography on LIS is available at biblioLIS http://www.istc.cnr.it/sites/default/files/u182/bibliolis_arg_2011.pdf).

To our knowledge the relationship between sign languages and abstract concepts has been investigated in a few studies so far (e.g., West and Sutton-Spence, 2010). In 2005 the Journal “Sign Language Studies” devoted a Special Issue to a crosslinguistic analysis of SL in the metaphorical domains of thought and communication. Linguists studying different sign languages (British, American, Catalan, and Italian) examined the mappings involved in SL metaphors, showing the process of embodiment active in metaphorical structures. Some structures share similarities across sign languages but there are also some interesting differences. Russo (2005) suggests that signed language metaphors are intrinsically related to aspects of the linguistic and cultural dimensions of a specific deaf community. More recently Roush (2011) has addressed the issue of the cognitive representation of abstract terms in sign languages. The author analyzed how a number of abstract words are represented in American Sign Language (ASL). Roush (2011) applied a specific linguistic-cognitive framework, the Conceptual Metaphor Theory, to investigate how the area of (im)politeness is conceptualized through metaphors and reflected and iconically represented in ASL. Our approach shares with Roush the view that using sign languages is an important perspective helping understand the way in which concepts are represented, however the ultimate aim why we use sign languages for investigating cognitive issues is slightly different. While Roush focuses on a specific theory we move from a variety of embodied theories struggling to account for abstract concepts representation. Specifically, our investigation is aimed at analyzing how abstract concepts belonging to different domains are represented in LIS, assuming that this analysis will allow us to understand whether the category of abstract terms is homogeneous or whether it needs to be re-organized into different sub-sets.

We advance the following hypotheses. First, in line with all embodied theories we predict that all the considered abstract concepts are at least in part grounded in the sensorimotor system. This guarantees the fact that the problem of symbol grounding (Harnad, 1990) is not present, since symbols used to represent abstract concepts are not arbitrarily linked to their referents.

At the same time, however, we predict that theories taking into account only sensorimotor nonlinguistic information will not be able to explain all examples we provide. In our view a unified framework, either based only on sensorimotor (for a review, see Pecher et al., 2011) or only on linguistic information (e.g., Paivio, 1986) will not be able to account for the differences between kinds of abstract concepts. In line with multiple representation theories we predict, instead, that to account for some abstract concepts a combination of sensorimotor, emotional, and linguistic information will be necessary. With “linguistic information” we intend any kind of exploitation of forms derived from any kind of language, be the same sign language or a different sign or spoken language. An example is the concept of “causation”: it is grounded in sensorimotor information since it might activate a variety of situations in which, for example, one element determines an effect on another one (e.g., a ball hurting another ball and provoking its movement, a handle being pressed to open a door etc.); at the same time, however, to acquire the concept children might rely on explanations of what causation is provided by others, such as parents or teachers, or by authoritative written sources, such as dictionaries, encyclopedias, etc. Another example highlighting how the formation of abstract concepts can rely on linguistic sources is the concept of “linguistics,” which originates from and refers to the more concrete concept of “language.” Specific examples pertaining SLs, such as linguistics, language, truth, etc., are discussed later in the paper. To highlight the role of linguistic information we have selected on purpose concepts where the role of linguistic elements is particularly evident, even if sensorimotor information still plays a role. This combination of sensorimotor and linguistic information is what we mean when we speak of “different levels of embodiment.”

First, our analysis indicates that, even if in sign languages the body is always involved to convey meanings, this involvement occurs at different levels. Skeptics of an embodied cognition perspective might object that it is not completely surprising that sign languages would provide evidence of grounding, given their visual nature and in particular the large amount of iconicity utilized by the language. In sign languages the coupling between language processing and sensori-motor processing becomes indeed more evident than in spoken languages. The body is always involved in spoken languages, for example through vocal articulators but in Sign languages the body, the hands and facial expressions become the main articulators. For example, the hands used for everyday activities such as pointing, enumerating or manipulating objects are also used for representing the same activities.

At the same time, however, it is possible to detect different levels of embodiment through a sign language analysis. The continuity between praxis, gesture and sign is easily recognizable at different levels of SLs structure: formational parameters, lexicon, morphology and syntax (see below for a more detailed discussion of this point). Despite this special characteristic of SLs has been widely recognized (e.g., Sandler and Lillo-Martin, 2006), only a few studies have explored the relationship between sign language and embodied theories, stressing the role of iconicity in sign languages (e.g., Pizzuto and Volterra, 2000; Boyes Braem et al., 2002; Morgan et al., 2008; Perniss et al., 2010). Iconicity can provide an additional mechanism for the grounding of language in sensorimotor systems; in SLs the presence of iconicity is pervasive, as a consequence SLs can be considered a special open window to better understand how language can be grounded. For example, according to Taub's (2001) cognitive-linguistic view, iconicity “is not an objective relationship between image and referent; rather, it is a relationship between our mental models of image and referent.” She claims that the creation of an iconic sign involves four successive stages: conceptualization, image selection, schematization, and sign encoding. The choice of the mental image is always mediated by cultural conventions, modality factors and language-specific conventions. This explains also why there is not an “Universal Sign Language” but rather many different Sign Languages. In a recent paper, Perniss and Vigliocco (2014) have highlighted the role of iconicity in both spoken and sign languages considering iconicity as a major vehicle for linking language and human sensory-motor experience. According to their perspective, iconicity represents the key to understand language evolution, development and processing providing a mechanism for displacement, referentiality and embodiment. They have also distinguished different types of iconic mapping, from a form of iconicity based more on imitative resemblance between the sign and the referent to a form of iconicity requiring more abstract mapping of features.

The novelty of our work, that recognizes the special and more evident role played by iconicity in Sign Languages, consists in focusing not only on the different levels of abstraction of the sign-referent mapping, but in identifying and examining a special case of referents, those of abstract concepts. Analyzing how signs can express abstract concepts in different ways (or through different iconic and not iconic mechanisms) provides some contributions to the debate on how different theories may account for abstract words representation. LIS can indeed provide interesting insights on the different degrees in which the various parameters of the signs are linked to the expressed concepts. In many cases specific locations assume an iconic meaning (for example, the majority of signs for mental activity are performed on the forehead), in other cases also the configuration and/or the movement performed are salient (for example, the sign clear is performed with an open hand configuration moving away from the eyes; a grasping movement characterizes the sign understand) (Pietrandrea, 2002).

More crucially to the aim of the present paper, our work provides some insights and has a number of theoretical implications for the debate on how embodied and grounded theories might account for abstract concepts and words (see also Dove, 2009, 2011). The novelty of our work consists in investigating whether signs can provide support for the different embodied theories of abstract concepts.

In line with the previous literature on Conceptual Metaphor Theory, we found that many signs convey a metaphorical meaning and are based on underlying metaphors (e.g., the metaphors of knowing as seeing, of the head as container of mental activities, of the chest as container of feelings and emotions), in keeping with the view that abstract concepts are represented through a metaphorical mapping mechanism. However, in contrast with previous studies we have seen that this is not the whole story, for two main reasons.

The first is that our data support further embodied cognition theories according to which action, situations and emotions are important for abstract concepts representation. Some signs (e.g., the sign for impossible_h−fff) provide evidence in favor of the view according to which abstract concepts are grounded on situations; other signs (e.g., the sign to constrain) offer support to the ACE view and other signs (e.g., the sign to express oneself) provide evidence favoring the emotion theory of abstract concepts. At a theoretical level the complex framework we obtained cast doubts on the possibility that a single explanation, for example based on a metaphorical mapping mechanism, is valid for the entire domain of abstract concepts and terms (See Prinz, 2002, 2012, for a similar view, according to which different abstract concepts can be explained referring to situation, to metaphors, to action as well as to linguistic information). At the same time, it confirms the necessity to perform fine-grained analyses of the differences between kinds of abstract concepts, analyses which some authors have started to conduct (e.g., Ghio et al., 2013; Roversi et al., 2013).

The second conclusion we can make is that, even if the analysis on LIS we performed provides support to all the aforementioned theories, at the same time it highlights their limitations. All these theories together are not able to fully account for the whole variety of signs we described. More importantly, they are not able to account for signs expressing some abstract concepts, such as truth.

We think that one of the main contributions of the present work consists in showing that, for some abstract concepts (e.g., the name of a discipline such as “linguistics,” a concept such as “truth,” etc.), LIS exploits linguistic information. This linguistic information could derive from different sources: from the same sign language (e.g., the LIS impossible_aa sign derives from the LIS sign possible_aa), from a foreign sign language as ASL (e.g., language/lingua and linguistics) or from spoken/written Italian (e.g., true). This finding challenges many current embodied theories of abstract concepts and clearly supports the WAT view. More generally, it supports multiple representation views according to which not only sensorimotor but also emotional and especially linguistic information, differently distributed, characterize abstract concepts representation (beyond the WAT theory, see also Barsalou et al., 2008; Louwerse, 2011; see Kousta et al., 2011, for a multiple representation view stressing the role of emotions for abstract concepts and Dove, 2014, for a multiple representation view stressing the importance of language, similarly to WAT).

Finally, a methodological note. LIS has proved to be an interesting and powerful mean to access how concepts are represented. We hope we have been able to suggest that the study of sign languages represents a fruitful and promising research line to investigate issues crucial for embodied and grounded cognition perspectives, in particular whether different degrees of embodiment exist (Taub, 2001) and whether they vary depending on the domain. Other studies have already demonstrated the importance of the study of sign languages for an embodied and grounded perspective. However, to our knowledge the present study is the first in which examples from a sign language are used to test and validate different theories on abstract concepts. Obviously a certain caution should be used, since, even though they are performed with the body, signs are, like words, arbitrary, so it is difficult to argue that they reflect directly the way concepts are represented. However, they are surely more grounded and to a certain extent more “visible” than words, thus they certainly represent an important cue to help understand conceptual representation. The present paper, being a theoretical paper rather than an experimental one, intends to indicate a possible direction of work. In order to perform a more systematic and thorough analysis, one would need to ask LIS signers to rate different kinds of signs in terms of abstractness, and then select a subset of signs evaluated as abstract and analyze them. Future work is planned to perform such an analysis.

Overall, we think our work provides some hints for how to address issues related to the future of embodied cognition and to the notion of body. Our LIS analyses suggest that, even if the signs we described always involve the body, different degrees of embodiment might be present. Furthermore, our results suggest that to account for abstract concepts not only sensorimotor and emotional experience should be called into play, but that also linguistic information plays a major role. This might appear in conflict with an embodied approach. We believe it is not, since language is not a disembodied activity but an important part of our total human experience. A challenge for future research is to identify sub-sets of abstract concepts, and to determine whether linguistic information becomes progressively more relevant, the higher the degree of concepts abstractness is.

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