Similarity relations among spoken words: The special status of rimes in English

BRUNO DE CARA 0 USHA GOSWAMI 0 0 University College London , London, England This paper presents an analysis of the distribution of phonological similarity relations among monosyllabic spoken words in English. It differs from classical analyses of phonological neighborhood density (e.g., Luce & Pisoni, 1998) by assuming that not all phonological neighbors are equal. Rather, it is assumed that the phonological lexicon has psycholinguistic structure. Accordingly, in addition to considering the number of phonological neighbors for any given word, it becomes important to consider the nature of these neighbors. If one type of neighbor is more dominant, neighborhood density effects may reflect levels of segmental representation other than the phoneme, particularly prior to literacy. Statistical analyses of the nature of phonological neighborhoods in terms of rime neighbors (e.g., hat /cat), consonant neighbors (e.g., hat /hit), and lead neighbors (e.g., hat /ham) were thus performed for all monosyllabic words in the Celex corpus (4,086 words). Our results show that most phonological neighbors are rime neighbors (e.g., hat /cat) in English. Similar patterns were found when a corpus of words for which age-of-acquisition ratings were available was analyzed. The resultant database can be used as a tool for controlling and selecting stimuli when the role of lexical neighborhoods in phonological development and speech processing is examined. - Recent theories of phonological development suggest that, as language is acquired, the growing number of similarsounding words in the mental lexicon ( phonological neighborhood density; henceforth, N ) creates a pressure to represent words in a phonologicallywell specified manner to support efficient discrimination (Metsala & Walley, 1998). The proposal is that phonological awareness (the ability to manipulate components of spoken words in tasks including word segmentation and sound categorization) may emerge as the result of spoken vocabulary growth and associated changes in interitem phonologicalsimilarity relations (lexical restructuring theory, or LRT; see Metsala, 1999; Metsala & Walley, 1998). When vocabulary size is small, phonological similarity between words is thought unlikely to interfere with efficient access, and so it is assumed that there is no need to represent words in a phonologically detailed manner. Early word representations are thus claimed to be holistic (i.e., to represent global phonological characteristics; e.g., Ferguson, 1986; Jusczyk, 1986, 1993; Walley & Flege, 1999). As vocabulary grows, children need to distinguish between more and more words that sound similar to each other, and this eventually creates a developmental pressure to represent smaller segments of Support for this research was partly provided by Fyssen Foundation and Marie Curie Foundation awards to B.D.C. Preparation of this paper was supported by ESRC small grant (RN000223153)to U.G. We thank Ronald Peereman for helpful discussions concerning our approach. Correspondence concerning this article should be addressed to U. Goswami, Behavioural and Brain Sciences Unit, Institute of Child Health, University College London, 30 Guilford St., London WC1N 1EH, England (e-mail: ). speech, such as syllablesand, ultimately, phonemes (Fowler, 1991; Metsala & Walley, 1998; Walley, 1993). By adulthood, it is assumed that all words are represented as linear sequences of phonemes (e.g., prince is represented as /p/ /r/ /i/ /n/ /s/ ). Neighborhood density effects in speechprocessing tasks in adults (typically, words from sparser neighborhoods are recognized more quickly) are usually taken as evidence for such phoneme-based representations. The developmental lexical restructuring process postulated by Metsala and Walley (1998) is thoughtto be relatively word specific, depending on such factors as overall vocabulary size and the number of similar-sounding words in the lexicon. For example, words with many similar-sounding neighbors (words with dense N ) are thought to experience more pressure for phoneme-level restructuring than do words with few similar-sounding neighbors (words with sparse N ). Hence, early in development, words with dense N should be processed more accurately in speech-based tasks. Consistent with this prediction, Logan (1992) found that 2-year-olds were better at identifying (by pointing to pictures) familiar words from dense neighborhoods than those fro m sparse neighborhoods. This density effect had disappeared by age 4. In addition, Metsala (1999) found that 3- and 4-year-old children performed significantly better in a simple phoneme-blending task when the target words were from dense neighborhoods, rather than from sparse neighborhoods. The broad picture of phonological development characterized by LRT is probably correct. However, there are two problems with LRT as a developmentalhypothesis.First, the theory goes from syllable to phoneme without postulating a strong developmental role for intrasyllabic units like onset/rime (e.g., /k//{t/ for cat). This is surprising given the importance of onset-rime units in phonological development prior to literacy (Goswami & Bryant, 1990; Treiman, 1988). For example, preliterate children usually perform relatively well in onset-rime tasks (e.g., segmenting cat into/k//{t/ ) while doing rather poorly in phonemic awareness tasks (e.g., segmenting cat into/k//{//t/; see Goswami & Bryant, 1990). The representation of phonemes in words is largely dependent on literacy acquisition, not on vocabulary acquisition (e.g., illiterate adults have poor phoneme awareness; Morais, Cary, Alegria, & Bertelson, 1979). Some authors have also reported a significant relationship between vocabulary development and rimelevel phonological skills in young children (e.g., Maclean, Bryant, & Bradley, 1987), but not between vocabulary development and phoneme-level skills (e.g., Hulme et al., 2002). Second, LRT does not allocate any special role to the subtypes of the neighbors that a word has. Metsala and Walley (1998) considered the overall number of phonological neighbors as a causal factor in LRT, but not the nature of these neighbors. This is important, because the lexicon of spoken word forms may have psycholinguistic structure at levels other than the phoneme. For example, given the psychological salience of the rime to young children, it seems possible that many phonological neighbors in English are rime neighbors. Traditional studies of speech processing describe some of these similarity relations in terms of phonotactic probabilities (i.e., possible combinations between phonemes; e.g., Bailey & Hahn, 2001; Vitevitch & Luce, 1999). The focus of these studies in the adult auditory-processing literature is always on the phoneme. We propose that such a focus is appropriate only for literate participants. It may not be appropriate for preliterate participants. For preliterate participants (usually young children), (...truncated)