Christiansen & Chater's (C&C's) proposed account of the manner in which language and its acquisition are constrained by general cognitive mechanisms offers a compelling alternative explanation of language evolution and the relationship between language and the brain. We enthusiastically endorse C&C's proposal that language adapted to fit the general nature of human cognitive and neural processing. However, specification of the mechanisms by which cognitive and learning processes constrain language structure is critical. A particularly promising mechanism is perceptual grounding. Growing evidence suggests that both cognitive and language processing are grounded in multi-modal perceptual-motor representations (Barsalou Reference Barsalou1999; Clark Reference Clark1997; Glenberg & Robertson Reference Glenberg and Robertson2000). We propose that this evidence informs the nature of cognitive processing and its relationship to the structure of language. C&C focus primarily on the morphosyntactic structure of language. Our work extends these arguments to a consideration of lexical processing and the semantic system. In particular, we investigate the structural relationships between sound and meaning at the word or segmental level of analysis. We find that the basic structure of verbal symbols reflect a perceptual-motor grounding of language.
Sound-to-meaning mapping
Arbitrariness has long been considered a hallmark of spoken (and signed) language (de Saussure Reference de Saussure and Baskin1916/1959; Hockett Reference Hockett1977). Within the domain of semantics, the sound sequences that comprise individual words (unlike gestures or pictorial representations, for example) bear no resemblance to their referent, nor do they contain sounds consistently associated with specific meanings. There is no systematic relationship, the account goes, between particular phonemes or phoneme sequences and particular semantic features. Boutique phenomena, such as onomatopoeia, and language-specific phenomena, such as Japanese mimetics (Hamano Reference Hamano1998), have been regarded as notable and forgivable exceptions.
Arbitrariness offers a variety of cognitive advantages that have been highlighted in domain-general accounts of language. The flexibility, abstractness, generativity, referential specificity, and even memory advantages afforded by the arbitrariness of language are significant cognitive constraints that likely facilitate language learning and communicative effectiveness (de Saussure 1916/1959; Gasser Reference Gasser, Forbus, Gentner and Regier2004; Monaghan & Christiansen Reference Monaghan, Christiansen, Sun and Miyake2006). As languages evolve, pressures imposed by the constraints of the cognitive system facilitate gradual convergence on an arbitrary communicative system.
Recent evidence of reliable sound-symbolic aspects of meaning calls into question this notion of a completely arbitrary system of sound-meaning associations (Nuckolls Reference Nuckolls1999). For example, Berlin (Reference Berlin, Hinton, Nichols and Ohala1994) noted a reliable relationship between vowel height and size of animal within the domains of birds and fish in the Huambisa language. In a comprehensive inventory of phoneme-meaning correspondences in Indonesian, McCune (Reference McCune1983) suggested that virtually all words in this language include some sound-meaning correspondences (see also Hutchins [Reference Hutchins1998] and Bergen [Reference Bergen2004] regarding sound-meaning correspondences in English). Classic work by Köhler (Reference Köhler1947) and, more recently, by Ramachandran and Hubbard (Reference Ramachandran and Hubbard2001), Maurer et al. (Reference Maurer, Pathman and Mondloch2006), and Westbury (Reference Westbury2005) suggest that English-speaking adults and children will interpret non-words such as maluma and bouba as referring to round, amoeboid shapes and words like takete and kiki as referring to angular figures. Each of these findings suggests a cognitive association between particular sound sequences and meaning that guide and constrain the evolution of linguistic reference within a particular language.
Sound-to-meaning correspondence and learning
Our work explores (a) whether these sound-meaning correspondences extend cross-linguistically, and (b) the extent to which concordances between sounds and meaning facilitate word learning. This work was inspired in part by a study by Kunihira (Reference Kunihira1971) in which monolingual English speakers were asked to select which of two antonyms (e.g., fast vs. slow) was being expressed by spoken Japanese words. Kunihira found that subjects accurately inferred the correct meaning at above chance rates from the sound structure of these words alone. Our work has extended this basic finding to other languages, including Danish and Russian (Nygaard et al., in preparation).
We demonstrated the functional advantages of these non-arbitrary sound-to-meaning mappings, using a learning task in which monolingual English speakers were taught Japanese words that either had correct word-to-meaning mappings (e.g., ue meaning up), were mapped to the word's antonym (e.g., ue meaning down), or were randomly paired with a meaning (e.g., ue meaning slow). We found a consistent learning advantage for correct over incorrect mappings (Nygaard et al. Reference Nygaard, Cook, Namy, Sloutsky, Love and McRae2008). This finding is consistent with C&C's assumption that properties of language, in this case residual non-arbitrary mappings, evolved because they ease learning and maximize cognitive efficiency.
Recent research from our lab and others also highlights how another type of sound-to-meaning mapping, namely prosody, constrains word meaning. Shintel et al. (Reference Shintel, Nusbaum and Okrent2006) found not only that speakers modulated their speaking style (e.g., fundamental frequency or speaking rate) to reflect properties of a scene they were describing, but also that listeners were sensitive to these modulations and could reliably identify which of two possible scenes were being described based on speaking style. We have recently shown that both adults and children can reliably utilize prosodic information to infer the meaning of a novel word (see also Kunihira Reference Kunihira1971), and that speakers reliably recruit distinct sets of acoustic properties to convey information about different semantic domains (Nygaard et al., in press; Sasso et al. Reference Sasso, Namy and Nygaard2005).
Multiple constraints on the structure of language
We have proposed that sound-meaning correspondences that facilitate learning stem from cross-modal perceptual-motor mappings. Such mappings would enable a speaker or listener to recruit acoustic-phonetic information as a guide to potential word meanings by evoking sensory-motor experiences that transcend the auditory modality. Consistent with C&C, we suggest that compelling cognitive pressures to generate arbitrary sound-meaning correspondences interact with a system that draws on cross-modal sound-meaning correspondences that offer more direct recovery of meaning. We propose that the advantages offered by perceptually grounded representations of spoken language compete with pressures to converge on an abstract, arbitrary system of meaning. Competing constraints placed on the system by the arbitrary versus non-arbitrary aspects of word meaning work in tandem to maintain a system that is optimally efficient given the structure and function of the human brain (Monaghan & Christiansen Reference Monaghan, Christiansen, Sun and Miyake2006; Monaghan et al. Reference Monaghan, Christiansen and Chater2007). These competing constraints are consistent with C&C's assertion that “language will be the outcome of competing selectional forces” (sect. 7, para. 2). We conclude that if, as C&C argue, language has adapted to the brain, and if, as we argue, neural and cognitive constraints reflect a grounding in perceptual-motor systems, then these properties must necessarily permeate the fundamental nature and structure of language.
