How Language Affects Thought in a Connectionist Model

How Language Affects Thought in a Connectionist Model Katia Dilkina (knd@{andrew.cmu.edu, stanford.edu}) Stanford University, Department of Psychology Jordan Hall (Bldg 420), 450 Serra Mall, Stanford, CA 94305 USA James L. McClelland (jlm@psych.stanford.edu) Stanford University, Department of Psychology Jordan Hall (Bldg 420), 450 Serra Mall, Stanford, CA 94305 USA Lera Boroditsky (lera@psych.stanford.edu) Stanford University, Department of Psychology Jordan Hall (Bldg 420), 450 Serra Mall, Stanford, CA 94305 USA kinds of sensory-motor (including linguistic) input. Within this framework, linguistic information about an entity (its name, grammatical gender, other relevant grammatical and syntactic markings, etc.) is treated the same way as other applicable kinds of information – information about what that entity looks like (visual), what it sounds like (auditory), what it feels like (tactile), how it moves (motoric), and so on. The semantic system involves a large network of modality-specific distributed representations (as suggested by imaging studies; Martin & Chao, 2001). Importantly, we believe that there is an additional representation sensitive to both within- and between-modality covariation that serves to link the modality-specific information together (Damasio, 1989; Rogers et al., 2004). These representations combine sensory-motor information with linguistic information and provide the substrate where effects of meaning similarity in picture and word naming and recognition tasks as well as linguistic relativity findings arise (for a related perspective see Vigliocco and colleagues, 2003, 2004, 2005, 2006). We have used a connectionist implementation of this theory to account for the relationship of semantic and lexical deficits in semantic dementia patients (Dilkina & McClelland, 2006). Furthermore, the theory has implications for a wide range of research areas including language comprehension and production, bilingualism, and conceptual representations and processing (note that for the purposes of this paper, semantics is synonymous with conceptual knowledge; it does not refer to the meaning of words per se). In the current paper, we will focus on its implications for findings of linguistic relativity, and more specifically on effects of grammatical gender categories. Abstract How do the languages we speak shape the way we think? In a series of studies, Boroditsky, Schmidt, and Phillips (2003) investigated the effect of grammatical gender on people’s responses to questions about the properties and similarity relations among objects. Here, we use a connectionist network to simulate these findings and find a possible mechanism for linguistic relativity effects (such as effects of grammatical categorization). The model’s behavior paralleled the effects seen in the human data. The results also suggest that the within- and between- category similarity relations among objects may play a role in generating these effects. Keywords: semantics; conceptual knowledge; language; language and thought; linguistic relativity; connectionism. Introduction How do the languages we speak shape the way we think? The hypothesis that aspects of language may influence the way we think is most strongly associated with Sapir (1921) and Whorf (1956). Recently, evidence supporting this hypothesis has come from experimental studies documenting that speakers of different languages perform differently on (non-linguistic) tasks such as categorization, perceptual discrimination, and similarity judgments (e.g., Boroditsky, 2001; Boroditsky, Schmidt, & Phillips, 2003; Davidoff, Davies & Roberson, 1999, 2000; Levinson, 1996; Lucy, 1992; Sera et al., 2002; Slobin, 1992, 1996). In this paper we ask how cross-linguistic differences may arise, and attempt to provide a computational mechanism through which aspects of linguistic representations may influence thinking. In our view, cross-linguistic differences arise as a result of differences in experience. Speakers of various languages are exposed to distinct patterns of linguistic input characterized by its specific statistical properties and associations with other kinds of information. Throughout development, conceptual knowledge about objects in the world is acquired by integrating different Grammatical Gender In English, only persons are referred to with the gender- marked pronouns he and she. However, in many other languages, including Spanish and German, all nouns are marked for gender – even nouns referring to inanimate objects. For example, the word for ‘key’ is feminine in

[1]  John B. Carroll,et al.  Language, Thought, and Reality (Selected Writings of Benjamin Lee Whorf). , 1957 .

[2]  L. Vygotsky,et al.  Thought and Language , 1963 .

[3]  D. Slobin The Crosslinguistic Study of Language Acquisition , 1987 .

[4]  D. Slobin Thinking for Speaking , 1987 .

[5]  Antonio R. Damasio,et al.  The Brain Binds Entities and Events by Multiregional Activation from Convergence Zones , 1989, Neural Computation.

[6]  J. Lucy,et al.  Grammatical categories and cognition: References , 1992 .

[7]  S. Levinson Frames of reference and Molyneux's question: Cross-linguistic evidence , 1996 .

[8]  S. Levinson,et al.  LANGUAGE AND SPACE , 1996 .

[9]  D. Slobin From “thought and language” to “thinking for speaking” , 1996 .

[10]  S. Levinson,et al.  Rethinking Linguistic Relativity , 1991, Current Anthropology.

[11]  J. Davidoff,et al.  Colour categories in a stone-age tribe , 1999, Nature.

[12]  D. Slobin,et al.  The Crosslinguistic Study of Language Acquisition , 2000 .

[13]  J. Davidoff,et al.  Color categories are not universal: replications and new evidence from a stone-age culture. , 2000, Journal of experimental psychology. General.

[14]  L. Boroditsky Does Language Shape Thought?: Mandarin and English Speakers' Conceptions of Time , 2001, Cognitive Psychology.

[15]  Alex Martin,et al.  Semantic memory and the brain: structure and processes , 2001, Current Opinion in Neurobiology.

[16]  M. Sera,et al.  When language affects cognition and when it does not: an analysis of grammatical gender and classification. , 2002, Journal of experimental psychology. General.

[17]  Lauren A. Schmidt,et al.  Sex, syntax and semantics. , 2003 .

[18]  S. Cappa,et al.  The breakdown of semantic knowledge: Insights from a statistical model of meaning representation , 2003, Brain and Language.

[19]  C. Habel,et al.  Language , 1931, NeuroImage.

[20]  James L. McClelland,et al.  Structure and deterioration of semantic memory: a neuropsychological and computational investigation. , 2004, Psychological review.

[21]  M. Garrett,et al.  Representing the meanings of object and action words: The featural and unitary semantic space hypothesis , 2004, Cognitive Psychology.

[22]  G. Vigliocco,et al.  Grammatical gender effects on cognition: implications for language learning and language use. , 2005, Journal of experimental psychology. General.

[23]  A Connectionist Account of the Pattern of Deficits Across Semantic and Lexical Tasks in Five Semantic Dementia Patients , 2006 .

[24]  Gabriella Vigliocco,et al.  Language-specific properties of the lexicon: Implications for learning and processing , 2006 .