1 Homography in Russian and English: An fMRI investigation of the roles of phonological, syntactic, and semantic similarity in situating words in the mental lexicon Vsevolod Kapatsinski and Brian Riordan Indiana University 1. Introduction The nature of the representation of lexical knowledge in the brain has gained increasing attention from cognitive neuroscientists in recent years. Fox example, differences in localization have been observed among semantically distinct members of the same grammatical category, e.g., verbs denoting actions performed with different parts of the body have been shown to elicit activation in different premotor areas (Pulvermueller et al. 1999a, 2001). As another example, there is some evidence for distinct patterns of activation associated with nouns denoting living versus non-living things (e.g., Ishai et al. 1999, Martin and Caramazza 2003). A significant proportion of the research has examined lexical access and lexical ambiguity. In English, words can be both semantically and grammatically ambiguous. An often cited example of semantic ambiguity is the word bank, which can refer to either a financial institution or the natural incline along a river’s edge. Bank is also ambiguous with respect to its grammatical category: in basketball or hockey, to bank a shot has a meaning unrelated to the two noun meanings. Ambiguous words allow researchers to hold the phonological or orthographic form of a word constant while exploring the neural correlates of their lexical entries, i.e., sets of neurons that are especially sensitive to particular lexical items where lexical items are nodes or sets of nodes that link together formal and semantic characteristics of a word. The nature of the semantic aspect of people’s representations of ambiguous words has been explored with both behavioral and electrophysiological techniques. In particular, researchers have begun to explicate the distinction between homonyms and polysemes which has been made in the theoretical lexical semantics literature. Homonyms are words that are characterized as being2 ambiguous between two or more unrelated meanings. An example is bank. Polysemes, on the other hand, are ambiguous between two or more related meanings. In fact, because of the process of meaning extension is so productive, it has been argued that most words are in fact polysemous (Miller, 2001). For instance, the word book can refer to a physical object as in a heavy, hardcover book, or a text as in an interesting book (where the text may not be in the form of the physical object, such as an electronic book, cf. also we can hold a whole book in our long-term memory). In a lexical decision paradigm, Rodd et al (2002) demonstrated that polysemes are processed more quickly than homonymous words. The so-called “ambiguity advantage”, whereby ambiguous words enjoy faster lexical decision latencies than non-ambiguous words, was shown to hold only for polysemes; homonyms were processed more slowly (see also (Klepousniotou, 2002) and discussion of polysemy in (Klein & Murphy, 2001, 2002)). These results were argued to be strong evidence that homonyms have separate lexical entries associated with each of their meanings, while polysemes may have only one lexical entry, which is linked to multiple meanings. This distinction has recently received support in MEG experiments (Beretta, Fiorentino, & Poeppel, 2005; Pylkkänen, Llinas, & Murphy, in press). For instance, Beretta et al showed that the earliest MEG component sensitive to lexical characteristics of the stimuli such as frequency (Embick et al. 2001) and neighborhood density (Pylkkänen et al. 2002), the M350, mirrored the behavioral results obtained by Rodd et al (2002), indicating that even very early processing reflects the distinction between homonyms and polysemes. Grammatical category information (e.g. noun, verb) has also been shown to form a part of words’ lexical representations in the brain. Evidence from aphasia (Shelton & Caramazza, 1999), rTMS (Shapiro, Pascual-Leone, Mottaghy, Gangitano, & Caramazza, 2001), and EEG (Shapiro et al., 2005) suggest that grammatical categories have some neuroanatomical basis. On the other hand, it has been suggested that nouns and verbs activate different areas of the brain because of semantic3 differences between them and not because of differences in syntactic properties (Pulvermueller et al. 1999b). We propose to explore the neural correlates of grammatical category information using ambiguous words. In English in particular, most words lack morphological marking for grammatical category, and thus are ambiguous between, e.g. their noun and verb instantiations. To date Tranel et al (2005) is the only imaging study to explore grammatical category-ambiguous words. In this study, subjects were shown pictures of tools (e.g. comb) or pictures of actions that were homophonic with the tool words (e.g. to comb) and asked to name the tools and actions depicted in the pictures. For comparison, subjects also named pictures of non-homophonic nouns (e.g. camera) and verbs (e.g. juggle). Tranel et al found that the activation patterns for the categories of words formed a gradient pattern: the non-homophonic nouns activated the left inferotemporal region (IT); the homophonic nouns activated IT and the left frontal operculum (FO); the homophonic verbs activated IT, FO, and MT; and the non-homophonic verbs activated all three regions, but more strongly. The authors speculated that this result was due to the homophonic words’ use as both nouns and verbs, forcing a pattern of activation intermediate between the poles of non-homophonic nouns and verbs. However, the picture-naming paradigm of Tranel et al’s study introduces a potential confound: were the activations related to the retrieval of the word to name the picture, or simply the processing of the visual stimulus? The pictures of the actions (e.g. comb, drill, saw) all included the tool with the same name. The visual processing of these pictures is likely to have included activation of the neural representation of the tool as well as the action. This might explain the fact that noun-related areas such as IT were also activated during naming of homophonic verbs. Thus the picture-naming task might not be the most suitable one for exploring the possibly different grammatical category activations of nouns and verbs.4 It may be that the “homonymous”