Research Focus Phonological typicality and sentence processing Michael K. Tanenhaus 1 and Mary Hare 2 1 Department of Brain and Cognitive Sciences, Meliora Hall, River Campus, University of Rochester, Rochester, NY 14627, USA 2 Department of Psychology, Bowling Green State University, Bowling Green, OH 43403, USA In studies of language, it is widely accepted that the form of a word is independent of its meaning and syntactic category. Thus, the relationship between phonological form and grammatical class would not be expected to affect reading time. However, Farmer et al. have now shown that the phonological typicality of a noun or verb influences how rapidly it is read. This finding has implications for both sentence processing and the interpretation of fixation patterns in reading. Introduction To a first approximation, natural languages separate form from content [1,2], enabling a finite set of forms to combine to refer to an infinite set of concepts [3]. For example, the phonology of a word determines neither its meaning nor its syntactic category (e.g. ‘cat’, ‘fat’ and ‘sat’ are phonologically similar, but their meanings are unre- lated and they belong to different syntactic categories). Moreover, attempts to define syntactic categories in terms of semantic notions (e.g. a noun is a person, place or thing) are notoriously unsuccessful. The form–content distinc- tion is honored both in formal analyses of language structure and in models that show how people build linguistic structure during comprehension. Therefore, phonological factors that are correlated with grammatical category would seem unlikely to influence reading times during sentence processing. However, Farmer et al. [4] have recently provided evidence that the phonological typicality of a word with respect to its syntactic category does affect reading. Determining phonological typicality Farmer et al. measured the phonological typicality of a large set of monosyllabic words that were classified unam- biguously as nouns or verbs by translating the phonemic structure of each word into a vector that had a fixed number of phoneme slots, with each phoneme represented by a set of distinctive features. They computed the mean Euclidean distance (a measure of distance in multidimen- sional space) of each word vector to all nouns and all verbs. Nouns and verbs were closer to words of their own category in terms of their phonology, but individual items varied in terms of how closely they matched the prototypical pattern of their class (Figure 1). The authors then conducted a series of experiments that demonstrated that readers exploit these distributional patterns. Typicality and reading times In Experiment 1, the authors reanalyzed an existing database of naming latencies [5] and found that phonolo- gical typicality is a significant predictor of time taken to read the word aloud, even after accounting for known factors such as word frequency, word length and ortho- graphic neighborhood size. Because the mapping between sound and spelling is quasi-regular in English, we suggest that the phonological and orthographical typicality of a word for a syntactic category will be highly correlated. Experiments 2–4 showed that phonological typicality influences reading comprehension. In Experiment 2, participants in a self-paced reading task were presented with structurally unambiguous sentences containing verbs that have strong statistical tendencies to be followed by noun phrases (e.g. ‘The boy saved...’). Participants were presented with sentence frames that contained phonologi- cally prototypical (noun-like) nouns and non-typical (verb- like) nouns. Mean reading times were significantly faster for the noun-like nouns than for the verb-like nouns. Experiment 3 tested the same effects in verbs, using sentence frames that generated strong expectations for a verb (e.g. ‘The girl tried to...’). In this case, reading times were faster for phonologically typical verbs than for noun- like verbs. Finally, Experiment 4 showed that phonological factors influence the interpretation of noun–verb homonyms such as ‘needs’. Noun–verb homonyms were embedded in sen- tences that resolved the ambiguity towards a noun interpretation (e.g. ‘...needs were not being met’) or a verb interpretation (e.g. ‘...needs to be more focused’). When the sentence contained a noun-like homograph, there was a significantly greater increase in reading times when the ambiguity was resolved towards a verb interpretation than when it was resolved towards a noun interpretation, which indicates that phonological typicality biased readers to interpret the ambiguous word as a noun. Similarly, for verb-like homonyms, reading times were significantly greater when the ambiguity was resolved towards a noun interpretation rather than towards a verb interpretation. Moreover, readers were significantly more accurate on comprehension questions when the interpretation of the sentence was consistent with the phonological typicality of the ambiguous word. Implications for sentence processing Are these results so surprising? Recent work has shown that syntactic, semantic, pragmatic and real-world infor- mation have rapid effects on syntactic-ambiguity resol- ution. In addition, adults and infants are remarkably Update TRENDS in Cognitive Sciences Vol.11 No.3 Corresponding author: Tanenhaus, M.K. (mtan@bcs.rochester.edu). Available online 4 January 2007. www.sciencedirect.com