Metaphoric Extension, Relational Categories, and Abstraction

ISSN: 2327-3798 (Print) 2327-3801 (Online) Journal homepage: http://www.tandfonline.com/loi/plcp21

Metaphoric extension, relational categories, and abstraction

Dedre Gentner & Jennifer Asmuth

To cite this article: Dedre Gentner & Jennifer Asmuth (2017): Metaphoric extension, relational categories, and abstraction, Language, Cognition and Neuroscience, DOI: 10.1080/23273798.2017.1410560 To link to this article: https://doi.org/10.1080/23273798.2017.1410560

Published online: 08 Dec 2017.

Submit your article to this journal

Article views: 59

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=plcp21 LANGUAGE, COGNITION AND NEUROSCIENCE, 2017 https://doi.org/10.1080/23273798.2017.1410560

REGULAR ARTICLE Metaphoric extension, relational categories, and abstraction Dedre Gentnera and Jennifer Asmuthb aDepartment of Psychology, Northwestern University, Evanston, IL, USA; bDepartment of Psychology, Susquehanna University, Selinsgrove, PA, USA

ABSTRACT ARTICLE HISTORY We propose that concepts exist along a continuum of abstraction, from highly concrete to highly Received 18 July 2016 abstract, and we explore a critical kind of abstract category: relational abstractions. We argue Accepted 30 September 2017 that these relational categories emerge gradually from concrete concepts through a process of KEYWORDS progressive analogical abstraction that renders their common structure more salient. This Analogical abstraction; account is supported by recent findings in historical linguistics, language acquisition and progressive alignment; career neuroscience. We suggest that analogical abstraction provides a major route for the of metaphor; concept development of abstractions in language and cognition. development; relational categories

Theories of embodied cognition have emphasised the role Because the idea of abstraction is used in multiple of sensorimotor systems in conceptual representation ways in cognitive science (Burgoon, Henderson, & (Barsalou, 2008;Gibbs,2006; Glenberg & Kaschak, 2003). Markman, 2013), we begin by clarifying our usage. By A core assumption of this view is that conceptual proces- abstract, we mean having few properties that are avail- sing recruits the same neural systems that are engaged able to the senses. For example, a concrete concept during perception and action. Evidence that semantic pro- like cat has multiple sensory properties – it can be seen cessing can be influenced by a person’sconcurrentactions and touched; it has characteristic patterns of movement and perceptions (Kaschak et al., 2005; Kaschak, Loney, & and characteristic sounds, and so on. In contrast, the Borreggine, 2006; Meteyard, Zokaei, Bahrami, & Vigliocco, concept justice cannot be tasted, seen, or heard. Justice 2008; see Vigliocco, Meteyard, Andrews, & Kousta, 2009, cannot be recognised by any perceptual characteristics. for a review) has led some theorists to propose that all rep- We take the process of abstraction to be one of decreas- resentations are modality-specific (e.g. Barsalou, Breazeal, ing the specificity (and thereby increasing the scope) of a & Smith, 2007). This raises the question of how – and concept.1 even whether – abstract concepts could come about. To preview, our thesis is that We contend that people possess abstract concepts as well as concrete, sensorimotor concepts. We propose . Relational concepts are critical to human reasoning further that there are rich connections between con- . Relational concepts often evolve from concrete, crete, embodied concepts and abstract concepts (Gold- embodied concepts via a process of structural align- stone & Barsalou, 1998). Specifically, we propose that ment and abstraction abstract concepts often arise from concrete concepts, . Language use influences the emergence of relational as suggested by the fact that metaphoric systems concepts, both in children’s learning and in language drawn from concrete experience are often used to evolution express abstract ideas (Boroditsky, 2000; Lakoff & . The gradual abstraction of concrete concepts into Johnson, 1980). We further propose that there is a conti- relational concepts is one main route by which nuum of abstraction from highly concrete to highly abstractions arise in language and cognition abstract, such that many relatively abstract concepts still retain aspects of their concrete meaning. We The importance of relational concepts present psychological, historical, and neural findings to support the claim that there is a continuum of abstrac- The ability to represent and reason about relations tion, focusing on a particularly important class of abstract between entities or situations, and to perceive concepts – relational abstractions. common relations across disparate situations, is a key

CONTACT Dedre Gentner [email protected] © 2017 Informa UK Limited, trading as Taylor & Francis Group 2 D. GENTNER AND J. ASMUTH feature of higher-order cognition and one in which we with others, these homesigns did not include stable ges- greatly exceed other species (Christie, Gentner, Call, & tures for spatial relations. Compared to hearing Turkish Haun, 2016; Gentner, 2003, 2010; Penn, Holyoak, & Povi- children (matched by responses on a different spatial nelli, 2008). Gentner, Loewenstein, and Thompson task), the deaf children performed significantly worse (2003), Gentner (2010) and Christie and Gentner, on the spatial mapping tasks – at or only marginally (2010, 2014) have argued that relational thinking is above chance. Gentner et al. concluded that the deaf essential for higher-order cognition, and that the acqui- children were less successful in the spatial mapping sition of relational concepts is critical in cognitive devel- task because, without a consistent set of spatial relational opment (See also Chatterjee, 2010). Relational thinking is terms, they were less likely to represent the arrange- also critical for acquiring expertise within mathematical ments of the boxes in a uniform way and therefore less and scientific domains (e.g. Goldwater & Schalk, 2016; able to align the two arrays. These findings suggest Richland & Simms, 2015). More broadly, relational con- that having explicit language for spatial relations gives cepts are integral to our intellectual and social life; con- children the tools to create corresponding relational sider relations like causation, contradiction, agreement, representations. and dispute, for example. Thus it is important to under- Gentner (2010, 2016; Gentner & Christie, 2010) has stand how relational abstractions are acquired and how proposed that relational language supports the acqui- they evolve within a language. sition and use of relational concepts in several ways One way to trace the cognitive benefits of learning rela- (see also Pyers, Shusterman, Senghas, Spelke, and tional concepts is to observe the effects of learning rela- Emmorey (2010) and Shusterman, Ah Lee, and Spelke tional language. For example, Pruden, Levine, and (2011) for examples of the importance of consistent Huttenlocher (2011) coded children’s production of spatial terms in Nicaraguan Sign Language and in navi- spatial language (including spatial relations such as edge gation through a search task). One key role of relational and side as well as dimensional terms such as tall)during language in acquisition is that common language invites natural family interactions. Children’s spatial language comparison (Gentner & Medina, 1998; Gentner & Namy, between 14 and 46 months predicted their performance 1999). Learning relational concepts from experience is on nonlinguistic tests of spatial ability at 54 months. challenging, because (as we discuss) relations are not More direct evidence that learning and using relational obvious in the world. Specifically, hearing the same rela- language can support relational cognition comes from tional label applied to two different things invites the studies by Loewenstein and Gentner (2005). They gave child to compare them and extract their commonalities. 3–5-year-old children a simple spatial mapping task, in Of course, spatial relations are only one example of which children saw an object hidden in a three-tiered the relational concepts that are central to the human box and had to search for a “winner” object in the corre- experience. For example, even young children acquire sponding place in an identical box. The children were kinship relations (X is the sister of someone), functional more successful if they had heard spatial relations for relations (you can drink X), and causal relations (X can the three tiers (e.g. top, middle, bottom). The relational move things). We next outline some key characteristics language advantage persisted over a two-day delay, of relational concepts. even when no spatial terms were used at test (Loewen- stein & Gentner, 2005) – evidence that spatial relational Relational concepts language influenced children’s representations of the scene (rather than being some sort of immediate Theories of categorisation have often treated all con- priming effect). Loewenstein and Gentner concluded cepts as fundamentally alike. But concepts are not that applying this relational framework helped children uniform in character, and the variations support a delineate the two spaces and see the correspondences. range of different functions (Gentner & Kurtz, 2005; If spatial relational language is instrumental in Kloos & Sloutsky, 2004; Medin, Lynch, & Coley, 1997). forming relational representations, then children who One important difference is that between entities and lack terms for spatial relationships may be seriously dis- relations. In discourse terms, entity concepts are often advantaged in spatial tasks. Gentner, Özyürek, Gurcanli, used to refer to things in the world, and relational con- and Goldin-Meadow (2013) gave the same spatial cepts are used to assert connections between these mapping task to 6-year-old deaf children in Istanbul, referents. Broadly speaking, relational concepts can be whose hearing loss had prevented learning a spoken thought of as those denoted by verbs and prepositions, language, but who not been exposed to a sign language. and entity concepts, as those denoted by concrete Although they had developed their own gesture systems nouns (However, as we will discuss, some nouns (homesigns (Goldin-Meadow, 2003)) to communicate express relational concepts, and not all verbs express LANGUAGE, COGNITION AND NEUROSCIENCE 3 relations.). Therefore, we begin by comparing verbs and contextually driven uses of a given verb can become suf- prepositions with nouns. We then turn to a more subtle ficiently entrenched as to become alternate word senses. contrast within the class of nouns – that between Why are verbs so mutable? Gentner and France (1988) nominal relational concepts and nominal entity suggested that this arises in part from their discourse concepts. function of conveying relations among the things referred to in a sentence. Given a sentence like “The carpet argued with the curtains,” people typically Verbs and other relational form classes assume that the two nouns refer to things in the world Relational classes like verbs and prepositions differ from – namely, a carpet and some curtains – and they adapt concrete noun categories not only in their grammatical the meaning of the verb argued to convey something privileges, but in their semantics (Croft, 1991, 2001; that would be possible for these referents, such as Gentner, 1981, 1982; Gentner & Boroditsky, 2001; Graes- “clashed with each other.” ser & Hopkinson, 1987; Kersten, 1998; Kersten & Earles, Similar patterns hold for prepositions: they convey 2004; Langacker, 1987, 2008; Pavlicic & Markman, 1997; relational meanings, they are contextually mutable, and Vigliocco, Vinson, Druks, Barber, & Cappa, 2011). they are highly polysemous. (Boers, 1996; Brugman, Gentner (1981, 1982, 2005; Gentner & Kurtz, 2005) laid 1988). But verbs and prepositions are not the only rela- out a set of phenomena that differentiate verbs from tional terms. Relational categories may also be named concrete nouns. These include differences in memory, by nouns. We now turn to nominal relational categories, mutability and polysemy. contrasting them with nominal entity categories.

(1) Verbs are less well remembered then nouns (e.g. Nominal relational categories Kersten & Earles, 2004). If people hear a list of sentences, and later are given a recognition task with Research on concepts in cognitive psychology has new sentences, they show poorer recognition largely focused on concepts with common intrinsic prop- memory for verbs than for nouns, especially if the erties, such as cat and trombone. But recent work has verbs are combined with new nouns. investigated a different kind of nominal concept – rela- (2) Verb meanings are more mutable than those of tional categories2 (Asmuth & Gentner, 2017; Gentner, nouns (Gentner & France, 1988). The term mutability 2005; Gentner & Kurtz, 2005; Goldwater & Markman, refers to a word’s propensity to assume different 2011; Markman & Stilwell, 2001; Rehder & Ross, 2001). meanings across varying contexts. Evidence that By relational category we mean a category whose mem- verbs are more mutable than nouns comes from bership is determined by common relational structure, studies by Gentner and France (1988) in which rather than by common intrinsic properties. For instance, people were asked to paraphrase semantically for X to be a bridge, X must connect two other entities or strained sentences such as “The car worshipped.” points; for X to be a carnivore, X must eat animals. Rela- The dominant response was to use a synonym for tional categories contrast with entity categories like cat the noun (roughly preserving the noun’s usual refer- and trombone, whose members share many intrinsic ent) and to alter the meaning of the verb to fit that properties. This distinction can be seen even for more referent: e.g. “The vehicle only responded to him,” abstract categories. For example, all the members of or “Someone’s vehicle was given a rest on a Sunday.” the abstract entity category mammal have warm blood, (3) Verbs are more polysemous than nouns; that is, dic- an internal skeleton, a circulatory system, and so on. In tionary entries for verbs tend to have more meaning contrast, the abstract relational category carnivore3 senses than those for nouns (controlling for word fre- includes cats, spiders, sharks, falcons, frogs, and even a quency) (Gentner, 1982). few plants. They share only the relation that they eat animals. Importantly, we are not saying that concrete These three phenomena are interrelated. Verb mut- entity concepts lack relational structure. Our represen- ability contributes to their poor memory performance, tations of entity concepts like cat typically include rela- because it leads to verbs being encoded in context- tional structure as well as sensorimotor properties. For specific ways (Gentner, 1981; Kersten & Earles, 2004). example, our representation of cat includes relations For example, if a person initially encodes the verb such as that cats eat meat, that they chase mice, and worship in the context of “worshipping a deity” they that they are chased by dogs, as well as intrinsic might fail to recognise the same verb in a sentence like sensory properties such as whiskers, soft fur, and large “Fred worships his girlfriend.” Mutability also contributes eyes. The difference between entity concepts and rela- to the greater polysemy of verbs, in that some tional concepts is not that entity concepts lack relational 4 D. GENTNER AND J. ASMUTH

Table 1. Examples of shift from literal, concrete meaning to abstract, figurative meaning over time.

Literal/concrete figurative/abstract

Sanctuary 1340: A holy place 1374: church or other sacred 1400: A similar place of refuge in a 1445: Metaphorical 1568: figurative usage: a place (in which, by the law of the non-Christian country; an asylum. extension to heart as place of refuge . Þi sanctuary lord þe whilk þi medieval church, a fugitive from . place for faith . hend festynd; lord sall regne That Cytee [Ebron] was also Vsing alwaise soch justice, or a debtor, was entitled . wiþouten end and ouyre. In Sacerdotalle, that is to seyne, Fides thyn herte discrete moderation, as þat sanctuary oure lord sall to immunity from arrest) seyntuarie, of the Tribe of Juda: enbracyth As hir the scholehouse should . be kynge, þat is in all sauyd To whiche Iugement they And it was so fre, that Men propir sanctuary, and be counted a men, wiþouten end. nolden nat obeye but receyved there alle manere of medelith with al thi sanctuarie against defendedyn hem by the Fugityfes of other places, for deedys … feare sikernesse of holy howses, þat here evyl Dedis is to seyn fledden in to sentuarye.

Anchor 880: An appliance for holding a 1225: To secure (the ship) with an 1382: first figurative: That which 1578: To cast anchor, to 1586: Figurative: To fix ship, etc., fixed in a particular anchor; to place at, or bring to, gives the feeling of stability or come to anchor oneself, one’s attention, place anchor security. . Cortez … anckred at thought; take up a . Ðin ancor is git on eorþan . For þi is ancre … vnder . The which as an ankir we han the rivers mouth. position. fæst chirche iancred sikir to the soule . [She] wild me those Tempests of vaine loue to flee: And Anchor fast my selfe on vertues shore.

information; on the contrary, they are often rich in both destruction of the temple). Some relational nouns can relational and featural information. Rather, the difference take a set of distinctive argument markers: for instance, is that relational concepts are low in intrinsic perceptual if Jesse James robbed the gold from the Glendale train, features.4 then we can speak of the robbery of the gold from the The paucity of psychological research on relational train by Jesse James. categories might suggest that such categories are rare. Beyond these syntactic parallels, some of the proces- But in fact they are quite common. Informal ratings of sing differences between verbs vs. nouns also hold for the 100 highest frequency nouns in the British National relational nouns vs. entity nouns. Gentner and Kurtz Corpus revealed that about half of them refer to rela- (2005) have proposed the following analogy as a frame- tional categories (Asmuth & Gentner, 2005). This work for research. suggests that they play a significant role in adult dis- Relational Nouns : Entity Nouns :: Verbs : Nouns course. Nouns naming relational concepts – such as pro- portion, goal, and obstacle – are commonly found in adult Our research (Asmuth & Gentner, 2017) has explored conversation and facilitate concise and effective com- the above analogy further, asking whether the phenom- munication. We would go further, and assert that such ena that differentiate verbs and nouns also hold between terms also allow effective internal processing as well – relational nouns and entity nouns. Specifically, our that relational concepts – including nominal relational studies ask whether relational nouns are more contex- concepts – form a cognitive toolkit that supports tually mutable than entity nouns, and whether this higher-order thinking (Gentner, 2010, 2016; Gentner & leads to poorer memory for relational nouns. We have Christie, 2010). found when people are asked to interpret noun-noun There are substantial parallels between relational combinations consisting of relational nouns and entity nouns and other relational terms like verbs, stemming nouns (e.g. camera response), and are then tested with from the central parallel that they convey extrinsic new phrases, they show more accurate recognition of relationships with other concepts. For example, relational entity nouns than of relational nouns. We suggest that nouns often have arguments attached by prepositional this occurs because people naturally adapt the interpret- or adverbial particles.5 For example, in a distributional ation of the relational noun to fit the entity noun it occurs analysis, Goldwater and Willits (2010) showed that rela- with. For example, if camera response is interpreted as tional nouns are far more likely to be followed by of “trying to look good for Instagram,” it’s unlikely that than are entity nouns (e.g. the brother of Anna; the the person will later recognise response in the phrase LANGUAGE, COGNITION AND NEUROSCIENCE 5 door response (which might be interpreted as “a portal comparison (such as “A child is like a snowflake.”) that opens automatically).” In contrast, because the involves a process of structural alignment between the entity concept tends to be stable, participants are more two representations that renders their common structure likely to recognise camera in a new phrase such as more salient. Because the structure-mapping process camera limitation. favours common relational structure over common One further parallel between relational classes such as surface properties, this process is likely to lead to a rela- verbs and prepositions and nominal relational categories tional abstraction (in this case, perhaps “something deli- is that they tend to be more abstract than entity nouns. cate and ephemeral”). If further analogical comparisons For example, Asmuth and Gentner (2017) noted infor- are made, the relational abstraction may become an mally that when trying to control for level of abstract- alternate word sense of snowflake, creating a conven- ness, it was far easier to find abstract relational tional metaphor. As the metaphoric abstraction categories than to find concrete relational categories, becomes more established, there is a shift from simile and the reverse for entity categories. Intuitively, this fits form (An x is like a y) to metaphor form (An x is a y) with the role of relational concepts as connector; specific (Bowdle & Gentner, 2005), as in “These sophomores are features of the relational concept will often be sus- snowflakes.” That conventional metaphors can be pended to fit the surrounding entity concepts. stated using the same syntax as for categorisation is evi- We next take up our claim that many relational con- dence of an existing abstraction (Glucksberg & Keysar, cepts arise by abstraction from more concrete or embo- 1990). died concepts. As noted above, we take the process of There is considerable evidence for the Career of Meta- abstraction to be one of decreasing the specificity (and phor account of figurative language. As suggested thereby increasing the scope) of a concept. For above, people prefer novel figuratives in simile form example, the concrete concept anchor (meaning a (e.g. “A newspaper is like a telescope” is preferred to “A device attached to a rope that holds a boat in place) newspaper is a telescope”) and conventional figuratives6 has been abstracted over time so that it can now refer in metaphor form (e.g. “A soldier is a pawn” is preferred to a religion, or a loved one, or any other enduring to “A soldier is like a pawn”) (Bowdle & Gentner, 2005, force that provides a sense of security (See Table 1). Expt. 1). More importantly, novel figuratives are compre- This new sense of anchor is more abstract; its referents hended faster in simile form than in metaphor form, and do not need to have the concrete features of a boat the reverse is true for conventional figuratives, which are anchor. However, it preserves the key relational structure comprehended faster in metaphor form than in simile of preventing something from moving in an adverse form (Bowdle & Gentner, 2005, Expt. 2; see also Blank, direction. As this example suggests, there is a natural cor- 1988; Gentner & Wolff, 1997). As proposed in the relation between relationality and abstractness (Asmuth Career of Metaphor, people implicitly assume that & Gentner, 2017). But it is important to note that not all when metaphor form (‘An X is a Y’) is used, the base abstract concepts are relational. For example, the term already has an associated abstraction. When that concept thing is highly abstract in that it has few percep- form is used for a conventional figurative, this expec- tual properties – and indeed, few inherent properties of tation is met, and processing is highly fluent. But meta- any sort – but it is not a relational concept. phor form is infelicitous (and perhaps even misleading) when used for a novel figurative, for which there is no pre-existing abstraction. In these circumstances, the “X How do relational concepts arise? is like a Y” format is more felicitous, because it directs We propose that many – perhaps most – abstract rela- the listener to compare the two literal terms to derive tional concepts (e.g. anchor, sanctuary, signal) come the common structure. Studies by Giora (1997, 1999) about through analogical abstraction of concrete con- indicate that for conventional bases, the abstract cepts. We lay out this proposal using a theoretical frame- meaning is often the default sense, accessed early in pro- work called the Career of Metaphor (Bowdle & Gentner, cessing regardless of context. Finally, Bowdle and 2005; Gentner & Wolff, 1997; see also Chiappe & Gentner (2005, Expt. 3) demonstrated a kind of in vitro Kennedy, 2001; Jamrozik, McQuire, Cardillo, & Chatterjee, conventionalisation: if given a set of similes in which 2016). According to this theory, abstract meanings the same base term is used with different targets, (which are often, though not always, relational) often people later prefer to use metaphor form for figuratives develop gradually over repeated instances of metaphor with that base term. No such shift occurs if people see use. On this account, metaphors (like analogies) are com- literal similarity comparisons with the same base. In prehended by a process of structure-mapping (Gentner, addition to these behavioural studies, the abstraction 1983, 2010). For example, processing a figurative process laid out by the Career of Metaphor idea receives 6 D. GENTNER AND J. ASMUTH support from historical linguistics, neuroscience investi- as “a man with a beard” before developing the full rela- gations and language acquisition. tional meaning used by adults (e.g. the brother of a mother or father) (Clark, 1973). An important process driving children’s abstraction is progressive alignment Language evolution (Gentner & Medina, 1998; Kotovsky & Gentner, 1996). In This account of abstraction via repeated extensions is progressive alignment, a learner initially aligns two consistent with work in historical linguistics (Heine, closely similar concepts, and this paves the way for align- 1997; Traugott, 1978). For example, Heine, Claudi, and ing other pairs with the same relational structure but less Hunnemeyer (1991) examined the historical derivation concrete similarity (Gentner, 2010; Gentner & Medina, of five spatial relational concepts (on, under, front, back 1998). For example, a young child whose family has a and in) across a sample of 125 African languages. They pet dog may fail to understand the term pet as applied found that on was often derived from the term for to the neighbour’s python. But if the child compares head, front from the term for face, and back from the her dog with another pet dog, and perhaps with a pet term for back. Heine et al. (1991) suggest that these cat, she is likely to develop a partial abstraction (that is, abstract relational terms emerge through metaphorical one that still contains many specific details). There is con- extensions of more concrete terms. Consistent with siderable empirical evidence that even these “semi- strong embodiment views, in many cases the human abstractions” can facilitate the further mappings – in body appears to be the initial source of the terms. this case, aligning her pet dog with a pet python. (For However, the human body is not the only source of reviews, see Gentner, 2010; Gentner & Medina, 1998). spatial terms. For cattle-herding groups such as the Maasai of East Africa, the initial source for some preposi- Evidence from neuroscience tions appears to be animal bodies. This is consistent with the idea that metaphoric abstractions tend to be based There is support for the progressive alignment sequence on familiar, easily individuated things, of which the proposed in the Career of Metaphor from studies that human body is but one example. have examined the neural activation of sensorimotor The Career of Metaphor process of metaphorical metaphors. (For excellent reviews, see Chatterjee, 2010; alignment and abstraction also appears to be operative Jamrozik et al., 2016.) For example, Chatterjee and col- for nominal relational concepts. Zharikov and Gentner leagues have found that literal sentences involving (2002) traced the early history of a set of relational action verbs – e.g. The servant ironed out the wrinkles – nouns and found evidence that many of them evolved are processed in left occipito-temporal motion proces- through the figurative extension of concrete physical sing areas. However, metaphorical sentences involving objects (see also Maravilla & Gentner, 2016). For the same verbs – e.g. The boss ironed out the wrinkles – example, bridge once meant a physical structure that led to greater activation in the IFG and more anterior spanned water; it now also possesses an abstract areas of the MTG, suggesting a more abstract interpret- meaning of something that connects two things or ideas, ation of the metaphorical verbs (Chen, Widick, & Chatter- which can range from musical passages to political pos- jee, 2008). Likewise, Raposo, Moss, Stamatakis, and Tyler itions. Table 1 shows examples of the shift from early (2009) report that action verbs presented in literal sen- concrete meaning to more abstract figurative uses. tences generated activation in the motor cortex, as well as in fronto-temporal regions associated with language processing; but presentation of these verbs Development of relational categories within idiomatic sentences (while still associated with Paralleling the historical evolution, relational terms lag fronto-temporal activation) was not associated with acti- behind concrete entity nouns in children’s language vation in the motor or premotor cortex. Desai, Binder, acquisition. Verbs are slower to be acquired than Conant, Mano, and Seidenberg (2011) found that both nouns, not only in English but cross-linguistically (Born- literal and metaphoric action phrases (e.g. the daughter stein & Cote, 2004; Braginsky, Yurovsky, Marchman, & grasped the flowers, the jury grasped the concept) gener- Frank, 2015; Caselli et al., 1995; Gentner, 1982; Gentner ated activation in secondary sensorimotor areas involved & Boroditsky, 2001). Likewise, relational nouns are in action planning. However, activation in the primary acquired later than entity nouns (Gentner, 2005; motor areas was greater for novel phrases than for fam- Gentner & Rattermann, 1991); and when children begin iliar phrases. Further, Desai, Conant, Binder, Park, and Sei- to use relational nouns, they often initially interpret denberg (2013) found that activation in sensorimotor them as entity terms (Gentner, 2005; Gentner & Ratter- areas decreased as the level of abstraction increased. mann, 1991). For example, a child may interpret uncle Consistent with the Career of Metaphor predictions LANGUAGE, COGNITION AND NEUROSCIENCE 7

(Bowdle & Gentner, 2005), as the phrase is interpreted both literal primes (e.g. a scalpel is a razor) and metapho- less literally and more metaphorically (e.g. from The rical primes with alternative senses (e.g. her mind is a instructor is grasping the steering wheel very tightly to razor). This pattern suggests that metaphors with novel The congress is grasping the state of affairs), the emphasis bases are interpreted in terms of the best current align- shifts from characteristics associated with the specific, ment between the literal base and target terms; in con- literal interpretation (physically and firmly holding an trast, metaphors with conventional metaphoric bases object) to the more abstract patterns that hold across (which already have an associated metaphorical abstrac- uses in that metaphoric sense (holding/apprehending a tion) are primed only by metaphors that fit this concept), and this is reflected in the decreased sensori- abstraction. motor activation. There is also evidence consistent with the idea that Conclusion progressive alignment supports abstraction. Using an fMRI adaptation paradigm, Kable and Chatterjee (2006) Relational abstractions are a key component of human demonstrated that the posterior superior temporal cognition. An adequate theory of conceptual structure sulcus, the extrastriate body area, and area MT/MST must include an account of their nature and their evol- showed neural adaptation to repeated actions even ution and acquisition. We have made the case that when the actions were conducted by novel actors. This repeated alignment across multiple exemplars results means that neural circuits within the lateral occipito-tem- in a gradual metaphorical abstraction, and that this poral cortex can become sensitive to specific action pat- process often retains key relational characteristics of a terns. With repeated actions, these circuits may be concept while attenuating more concrete properties. important in shifting attention away from intrinsic fea- Further, a relational concept’s role as a connector tures of the agent to the relational structures indexed among entities requires it to accommodate its meaning by action verbs. Further evidence for the role of progress- to the less mutable entities it relates – resulting in ive alignment in the Career of Metaphor comes from further abstraction. This account is consistent with the research by Cardillo, Watson, Schmidt, Kranjec, and Chat- Career of Metaphor theory (Bowdle & Gentner, 2005) terjee (2012), who traced shifts in the pattern of neural and offers a path from concrete, embodied represen- activation as sensorimotor metaphors become conven- tations to abstract relational concepts. It is supported tionalised. They introduced participants to novel meta- by research in cognitive psychology, language develop- phors and varied the degree of exposure to the ment, historical linguistics, and cognitive neuroscience. phrases via a variety of judgment tasks. They found In sum, natural processes of structure-mapping and pro- decreased activity in the left and right inferior frontal gressive alignment drive the emergence of abstractions gyri as familiarity with the metaphor increased. The left in language and cognition. inferior frontal gyrus (LIFG) is involved in many processes, including the need to choose between compet- Notes ing semantic representations. Cardillo et al. suggest that novel metaphors initially recruit the LIFG in order 1. As used in cognitive science, abstraction generally to suppress an irrelevant literal interpretation of the conveys not just any loss of specificity, but one that base term (see Gernsbacher, Keysar, Robertson, & reveals important or enduring characteristics (Burgoon et al., 2013). However, not all abstractions – even rela- Werner, 2001). As the nonliteral sense of the base term tional abstraction – capture important information. For becomes more familiar and more readily activated, example, things smaller than Jupiter is highly abstract, there is less need to suppress the literal meaning of yet we would not consider it particularly informative. the base. In future research, it would be interesting to 2. A distinction is often made between the terms concept ’ study the neural effects of in vitro conventionalisation, (the intension of a word s meaning) and category (the extension). However, their use in the current literature as in Bowdle and Gentner’s(2005) study, in which the is mixed, and the term “relational categories” is generally same base term is repeatedly used with different targets. used in work on nominal relational concepts. Other support for the Career of Metaphor account 3. Relational categories include temporally bound cat- comes from studies by Chettih, Durgin, and Grodner egories, such as passenger or bride, as well as enduring (2012), which found that for conventional metaphors ones, such as carnivore or ratio. (e.g. an insult is a razor), comprehensibility judgment 4. We are not saying that relational concepts lack concrete features entirely. As discussed later, many relational con- response times were facilitated only by matched-sense cepts are derived historically through gradual abstraction metaphorical primes (e.g. a betrayal is a razor). In con- from concrete concepts, resulting in a continuum of trast, those for novel metaphors were facilitated by concreteness. 8 D. GENTNER AND J. ASMUTH

5. Indeed, some relational nouns are derived from verbs Brugman, C. M. (1988). The story of over: Polysemy, semantics, (e.g. robbery / rob) and preserve the argument structure and the structure of the lexicon. New York, NY: Garland. of the verb (Barker & Dowty, 1993). Burgoon, E. M., Henderson, M. D., & Markman, A. B. (2013). 6. We use the term “conventional figurative” to mean a fig- There are many ways to see the forest for the trees: A tour urative expression for which the base term has a conven- guide for abstraction. Perspectives on Psychological Science, tional metaphoric meaning. Thus the figurative “An X is 8(5), 501–520. (like) a Y,” would count as a conventional figurative as Cardillo, E. R., Watson, C. E., Schmidt, G. L., Kranjec, A., & long as Y has an applicable associated abstraction, Chatterjee, A. (2012). From novel to familiar: Tuning the even if X is novel in combination with Y. brain for metaphors. Neuroimage, 59(4), 3212–3221. doi:10. 1016/j.neuroimage.2011.11.079 Caselli, M. C., Bates, E., Casadio, P., Fenson, J., Fenson, L., Sanderl, Disclosure statement L., & Weir, J. (1995). A cross-linguistic study of early lexical development. Cognitive Development, 10, 159–199. No potential conflict of interest was reported by the authors. Chatterjee, A. (2010). Disembodying cognition. Language and Cognition, 2(1), 79–116. Chen, E., Widick, P., & Chatterjee, A. (2008). Functional-anatom- Funding ical organization of predicate metaphor processing. Brain – This research was supported in part by a grant from the Office and Language, 107(3), 194 202. of Naval Research [award number N00014-16-1-2613], as well Chettih, S., Durgin, F. H., & Grodner, D. J. (2012). Mixing meta- as by the National Science Foundation-Funded Spatial Intelli- phors in the cerebral hemispheres: What happens when gence and Learning Center [grant number SBE-1041707], to careers collide? Journal of Experimental Psychology: – the first author. Learning, Memory, and Cognition, 38(2), 295 311. Chiappe, D. L., & Kennedy, J. M. (2001). Literal bases for metaphor and simile. Metaphor and Symbol, 16, 249–276. References Christie, S., & Gentner, D. (2010). Where hypotheses come from: Learning new relations by structural alignment. Journal of Asmuth, J., & Gentner, D. (2005). Context sensitivity of relational Cognition and Development, 11(3), 356–373. nouns. In B. G. Bara, L. Barsalou, & M. Bucciarelli (Eds.), Christie, S., & Gentner, D. (2014). Language helps children Proceedings of the twenty-seventh annual meeting of the cog- succeed on a classic analogy task. Cognitive Science, 38, nitive science society (pp. 163–168). Mahwah, NJ: Lawrence 383–397. Erlbaum Associates. Christie, S., Gentner, D., Call, J., & Haun, D. B. M. (2016). Asmuth, J., & Gentner, D. (2017). Relational categories are more Sensitivity to relational similarity and object similarity in mutable than entity categories. The Quarterly Journal of apes and children. Current Biology, 26(4), 531–535. Experimental Psychology, 70(10), 2007–2025. doi:10.1080/ Clark, E. V. (1973). What’s in a word? On the child’s acquisition of 17470218.2016.1219752 semantics in his first language. In T. E. Moore (Ed.), Cognitive Barker, C., & Dowty, D. (1993). Non-verbal thematic proto-roles. development and the acquisition of language (pp. 65–110). In A. J. Schafer (Ed.), Proceedings of NELS 23 (pp. 49–62). New York, NY: Academic Press. Amherst, MA: GSLA. Croft, W. (2001). Radical construction grammar: Syntactic theory Barsalou, L. W. (2008). Grounded cognition. Annual Review of in typological perspective. New York, NY: Oxford University Psychology, 59, 617–645. doi:10.1146/annurev.psych.59. Press. 103006.093639 Croft, W. A. (1991). Syntactic categories and grammatical Barsalou, L. W., Breazeal, C., & Smith, L. B. (2007). Cognition as relations: The cognitive organization of information. Chicago, coordinated non-cognition. Cognitive Processing, 8(2), 79–91. IL: University of Chicago Press. Blank, G. D. (1988). Metaphors in the lexicon. Metaphor and Desai, R. H., Binder, J. R., Conant, L. L., Mano, Q. R., & Seidenberg, Symbolic Activity, 3,21–36. M. S. (2011). The neural career of sensory-motor metaphors. Boers, F. (1996). Spatial prepositions and metaphor: A cognitive Journal of Cognitive Neuroscience, 23(9), 2376–2386. doi:10. semantic journey along the up-down and the front-back 1162/jocn.2010.21596 dimensions. Tubingen: Gunter Narr Verlag. Desai, R. H., Conant, L. L., Binder, J. R., Park, H., & Seidenberg, M. Bornstein, M. H., & Cote, L. R. (with Maital, S., Painter, K., Park, S.- S. (2013). A piece of the action: Modulation of sensory-motor Y., Pascual, L. et al.). (2004). Cross-linguistic analysis of voca- regions by action idioms and metaphors. Neuroimage, bulary in young children: Spanish, Dutch, French, Hebrew, 83862–83869. doi:10.1016/j.neuroimage.2013.07.044 Italian, Korean, and American English. Child Development, Gentner, D. (1981). Some interesting differences between verbs 75(4), 1115–1139. and nouns. Cognition and Brain Theory, 4, 161–178. Boroditsky, L. (2000). Metaphoric structuring: Understanding Gentner, D. (1982). Why nouns are learned before verbs: time through spatial metaphors. Cognition, 75(1), 1–28. Linguistic relativity versus natural partitioning. In S. A. Bowdle, B., & Gentner, D. (2005). The career of metaphor. Kuczaj (Ed.), Language development: Vol. 2. Language, Psychological Review, 112, 193–216. thought and culture (pp. 301–334). Hillsdale, NJ: Lawrence Braginsky, M., Yurovsky, D., Marchman, V. A., & Frank, M. C. Erlbaum Associates. (2015, November). Developmental trajectories of vocabulary Gentner, D. (1983). Structure-mapping: A theoretical framework composition across languages. Poster presented at the 40th for analogy. Cognitive Science, 7, 155–170. Annual Boston University Conference on Language Gentner, D. (2003). Why we’re so smart. In D. Gentner, & S. Development. Boston, MA. Goldin-Meadow (Eds.), Language in mind: Advances in the LANGUAGE, COGNITION AND NEUROSCIENCE 9

study of language and thought (pp. 195–235). Cambridge, Goldin-Meadow, S. (2003). The resilience of language: What MA: MIT Press. gesture creation in deaf children can tell us about how all chil- Gentner, D. (2005). The development of relational category dren learn language (J. Werker & H. M. Wellman, Eds.). knowledge. In L. Gershkoff-Stowe, & D. H. Rakison, (Eds.), New York, NY: Psychology Press. Building object categories in developmental time (pp. 245– Goldstone, R. L., & Barsalou, L. W. (1998). Reuniting perception 275). Hillsdale, NJ: Erlbaum. and conception. Cognition, 65, 231–262. Gentner, D. (2010). Bootstrapping the mind: Analogical pro- Goldwater, M. B., & Markman, A. B. (2011). Categorizing entities cesses and symbol systems. Cognitive Science, 34(5), 752–775. by common role. Psychonomic Bulletin and Review, 18(2), Gentner, D. (2016). Language as cognitive tool kit: How 406–413. language supports relational thought. American Goldwater, M. B., & Schalk, L. (2016). Relational categories as a Psychologist, 71(8), 650–657. bridge between cognitive and educational research. Gentner, D., & Boroditsky, L. (2001). Individuation, relativity and Psychological Bulletin, 142(7), 729–757. early word learning. In M. Bowerman, & S. Levinson (Eds.), Goldwater, M., & Willits, J. (2010). The linguistic distribution of rela- Language acquisition and conceptual development (pp. 215– tional categories. In S. Ohlsson & R. Catrambone (Eds.), 256). Cambridge: Cambridge University Press. Proceedings of the 32nd annual conference of the cognitive Gentner, D., & Christie, S. (2010). Mutual bootstrapping between science society (p. 551). Portland, OR: Cognitive Science Society. language and analogical processing. Language and Graesser, A., & Hopkinson, P. (1987). Differences in interconcept Cognition, 2(2), 261–283. organization between nouns and verbs. Journal of Memory Gentner, D., & France, I. M. (1988). The verb mutability effect: and Language, 26, 242–253. Studies of the combinatorial semantics of nouns and verbs. Heine, B. (1997). Cognitive foundations of grammar. New York, In S. L. Small, G. W. Cottrell, & M. K. Tanenhaus (Eds.), NY: Oxford University Press. Lexical ambiguity resolution: Perspectives from psycholinguis- Heine, B., Claudi, U., & Hunnemeyer, F. (1991). tics, neuropsychology, and artificial intelligence (pp. 343– Grammaticalization: A conceptual framework. Chicago, IL: 382). San Mateo, CA: Kaufmann. University of Chicago Press. Gentner, D., & Kurtz, K. (2005). Relational categories. In W. K. Jamrozik, A., McQuire, M., Cardillo, E. R., & Chatterjee, A. (2016). Ahn, R. L. Goldstone, B. C. Love, A. B. Markman, & P. W. Metaphor: Bridging embodiment to abstraction. Wolff (Eds.), Categorization inside and outside the lab (pp. Psychonomic Bulletin & Review, 23(4), 1080–1089. 151–175). Washington, DC: APA. Kable, J. W., & Chatterjee, A. (2006). Specificity of action rep- Gentner, D., Loewenstein, J., & Thompson, L. (2003). Learning resentations in the lateral occipitotemporal cortex. Journal and transfer: A general role for analogical encoding. of Cognitive Neuroscience, 18(9), 1498–1517. Journal of Educational Psychology, 95(2), 393–408. Kaschak, M. P., Loney, R. A., & Borreggine, K. L. (2006). Recent Gentner, D., & Medina, J. (1998). Similarity and the development experience affects the strength of structural priming. of rules. Cognition, 65, 263–297. Cognition, 99(3), B73–B82. Gentner, D., & Namy, L. (1999). Comparison in the development Kaschak, M. P., Madden, C. J., Therriault, D. J., Yaxley, R. H., of categories. Cognitive Development, 14, 487–513. Aveyard, M., Blanchard, A. A., & Zwaan, R. A. (2005). Gentner, D., Özyürek, A., Gurcanli, O., & Goldin-Meadow, S. Perception of motion affects language processing. (2013). Spatial language facilitates spatial cognition: Cognition, 94(3), B79–B89. Evidence from children who lack language input. Kersten, A. W. (1998). An examination of the distinction Cognition, 127(3), 318–330. between nouns and verbs: Associations with two different Gentner, D., & Rattermann, M. J. (1991). Language and the kinds of motion. Memory & Cognition, 26(6), 1214–1232. career of similarity. In S. A. Gelman & J. P. Byrnes (Eds.), Kersten, A. W., & Earles, J. L. (2004). Semantic context influences Perspectives on thought and language: Interrelations in devel- memory for verbs more than memory for nouns. Memory & opment (pp. 225–277). London: Cambridge University Press. Cognition, 32(2), 198–211. Gentner, D., & Wolff, P. (1997). Alignment in the processing of Kloos, H., & Sloutsky, V. M. (2004). Are natural kinds psycholo- metaphor. Journal of Memory and Language, 37, 331–355. gically distinct from nominal kinds? Evidence from learn- Gernsbacher, M. A., Keysar, B., Robertson, R. R. W., & Werner, N. inganddevelopment.InK.Forbus,D.Gentner,&T. K. (2001). The role of suppression and enhancement in Regier (Eds.), Proceedings of the 26th annual conference understanding metaphors. Journal of Memory & Language, of the cognitive science society (pp. 702–707). Mahwah, 45(3), 433–450. NJ: Erlbaum. Gibbs Jr, R. W. (2006). Embodiment and cognitive science. Kotovsky, L., & Gentner, D. (1996). Comparison and categoriz- Cambridge: Cambridge University Press. ation in the development of relational similarity. Child Giora, R. (1997). Understanding figurative and literal language: Development, 67, 2797–2822. The graded salience hypothesis. Cognitive Linguistics, 8, 183– Lakoff, G., & Johnson, M. (1980). Metaphors we live by. Chicago: 120. University of Chicago press. Giora, R. (1999). On the priority of salient meanings: Studies of Langacker, R. W. (1987). Nouns and verbs. Language, 63,53–94. literal and figurative language. Journal of Pragmatics, 31, Langacker, R. W. (2008). Cognitive grammar: A basic introduction. 919–929. New York: Oxford University Press. Glenberg,A.M.,&Kaschak,M.P.(2003). The body’s contribution Loewenstein, J., & Gentner, D. (2005). Relational language and to language. Psychology of Learning and Motivation, 43,93–126. the development of relational mapping. Cognitive Glucksberg, S., & Keysar, B. (1990). Understanding metaphorical Psychology, 50, 315–353. comparisons: Beyond similarity. Psychological Review, 97(1), Maravilla, F., & Gentner, D. (2016, August). Conventionality, rela- 3–18. tionality and aptness in figurative language processing. 10 D. GENTNER AND J. ASMUTH

In J. T. Kao & N. D. Goodman (Chairs), Relationality and Raposo, A., Moss, H. E., Stamatakis, E. A., & Tyler, L. K. (2009). aptness in figurative language processing. Symposium con- Modulation of motor and premotor cortices by actions, ducted at the 38th Annual Conference of the Cognitive action words and action sentences. Neuropsychologia, 47 Science Society, Philadelphia, PN. (2), 388–396. Markman, A. B., & Stilwell, C. H. (2001). Role-governed cat- Rehder, B., & Ross, B. H. (2001). Abstract coherent concepts. egories. Journal of Experimental & Theoretical Intelligence, Journal of Experimental Psychology: Learning, Memory, and 13, 329–358. Cognition, 27, 1261–1275. Medin, D. L., Lynch, E. B., & Coley, J. D. (1997). Categorization Richland, L. E., & Simms, N. (2015). Analogy, higher order think- and reasoning among tree experts: Do all roads lead to ing, and education. Wiley Interdisciplinary Reviews: Cognitive Rome? Cognitive Psychology, 32,49–96. Science, 6, 177–192. doi:10.1002/wcs.1336 Meteyard, L., Zokaei, N., Bahrami, B., & Vigliocco, G. (2008). Shusterman, A., Ah Lee, S., & Spelke, E. S. (2011). Cognitive effects Visual motion interferes with lexical decision on motion of language on human navigation. Cognition, 120(2), 186–201. words. Current Biology, 18(17), R732–R733. Traugott, E. C. (1978). On the expression of spatio-temporal Pavlicic, T., & Markman, A. B. (1997). The structure of the verb relations in language. In J. H. Greenberg (Ed.), Universals of lexicon: Evidence from a structural alignment approach to human language: Vol. 3. Word structure (pp. 369–400). similarity. In M. Shafto & P. Langley (Eds.), Proceedings of Stanford, CA: Stanford University Press. the 19th annual conference of the cognitive science society Vigliocco, G., Meteyard, L., Andrews, M., & Kousta, S. (2009). (pp. 590–595). Hillsdale: Lawrence Erlbaum Associates. Toward a theory of semantic representation. Language and Penn, D. C., Holyoak, K. J., & Povinelli, D. J. (2008). Darwin’s Cognition, 1, 219–247. mistake: Explaining the discontinuity between human and Vigliocco, G., Vinson, D. P., Druks, J., Barber, H., & Cappa, S. F. nonhuman minds. Brain and Behavioral Sciences, 31,109–178. (2011). Nouns and verbs in the brain: A review of behav- Pruden, S. M., Levine, S. C., & Huttenlocher, J. (2011). Children’s ioural, electrophysiological, neuropsychological and spatial thinking: Does talk about the spatial world matter? imaging studies. Neuroscience Biobehavior Research, 35(3), Developmental Science, 14(6), 1417–1430. 407–426. doi:10.1016/j.neubiorev.2010.04.007 Pyers, J. E., Shusterman, A., Senghas, A., Spelke, E. S., & Emmorey, Zharikov, S. S., & Gentner, D. (2002). Why do metaphors seem K. (2010). Evidence from an emerging sign language reveals deeper than similes? In Gray, W. D. & Schunn, C. D. (Eds.), that language supports spatial cognition. Proceedings of the Proceedings of the 24th annual conference of the cognitive National Academy of Sciences, 107(27), 12116–12120. science society (pp. 976–981). Mahwah, NJ: Erlbaum.