When Perceptual Representations Defer to Grammar: Conflicting Linguistic and Perceptual Cues in Cantonese Classifiers

A growing body of evidence suggests semantic representations for linguistic expressions are grounded in perception, and that these representations guide online sentence processing. However, semantic distinctions that superficially seem grounded in perception are sometimes partially or wholly grammaticalized. To date, these cases have not been investigated in studies of real-time comprehension. We consider the case of shape classifiers in Cantonese Chinese - prenominal expressions that arguably encode size/shape characteristics of their associated nouns, but whose occurrence with nouns is determined by grammar. E.g., the classifier "tiu" typically precedes nouns denoting long-narrow-flexible things (e.g., ropes, snakes...). However, "tiu" occurs with the noun for [goldfish] even though goldfishes are not prototypically long-narrow-flexible. Conversely, although a stocking is normally long-narrow-flexible, this noun cannot occur with "tiu". Thus, grammatical rules governing classifier-noun pairings can conflict with perceptually-based meanings of classifiers. An offline test of Cantonese speakers' intuitions showed they considered classifier-noun pairings to be rule-based 97.6% of the time, rather than being dependent on prototypical size/shape features of noun referents. However, perceptual information encoded by classifiers might be computed unconsciously during online comprehension, somewhat like the processing of fictive motion [1]. To assess this possibility, we employed an eye-tracking methodology known to be sensitive to perceptual representations associated with words [2]. Cantonese listeners followed spoken instructions containing a classifier-noun pairing that named a target object (e.g., "tiu" [snake]). Displays also contained a "competitor" item that either (i) matched both the classifier's grammatical and perceptual parameters (e.g., [rope]), (ii) matched the grammatical but not the canonical perceptual parameters (e.g., [goldfish]); or (iii) matched the perceptual but not the grammatical parameters (e.g., [stocking]). Of interest was how often the competitor attracted fixations prior to eventual fixation on the target. Our findings indicated that only grammatically-legitimate competitors attracted attention as the noun phrase unfolded; the match/mismatch with classifier-denoted perceptual parameters had no effect. Thus, perceptual information encoded by classifiers did not seem to be computed during online interpretation. A second eye-tracking experiment tested whether perceptually-grounded meanings might become more apparent when target objects do not possess the canonical perceptual features conveyed by their associated classifier (e.g., a goldfish for "tiu"), potentially highlighting the relevance of perceptual features. Competitor pictures either matched the perceptual but not the grammatical parameters of the classifier (e.g., [stocking]), or matched both the perceptual and the grammatical parameters (e.g., [snake]). As before, the latter type of competitor attracted fixations during the initial stages of processing. However, unlike the earlier experiment, "grammatically illegitimate" competitors matching the classifier's perceptual parameters were now observed to attract attention. Overall, the results show perceptual features of classifiers are computed at a relatively shallow level during comprehension, and that these computations are normally dominated by rule-like grammatical constraints. Implications for theories claiming a perceptual/embodied basis for linguistic representations will be discussed. References Richardson & Matlock (2005). The integration of figurative language and static depictions. Cognition, 102, 129-138. Huettig & Altmann (2007). Visual-shape competition and control of eye fixation during the processing of words. Visual Cognition, 15, 985-1018.