Distributional properties of language and sub-syllabic processing units

During the past few years, several studies have drawn attention to the role of distributional characteristics in the segmentation of speech into word units. In the present chapter, we examine the role of distributional constraints in the acquisition of sub-syllabic language processing units. In particular, we suggest that specific processing units such as the rime might emerge from the distributional properties of sequences of vowels and consonants. We then summarize recent work showing that speakers are indeed sensitive to statistical distributions between vocalic and consonantal segments within the syllable. Finally, a simulation study using the Parser model (Perruchet & Vinter, 1998) indicates how probabilistic constraints can influence the acquisition of representational units. We conclude that what can be viewed as processing units probably represents only the tip of the iceberg of broader knowledge about statistical properties of language and that the acquisition of such representational units is, at least partially, shaped by general learning processes. 3 In most contemporary models, language recognition and use are supposed to hinge on a set of intermediate representational processing units of various sizes. For example, it has been claimed that syllables constitute processing units in auditory word recognition, at least for syllable-timed languages (e. have all been proposed as underlying reading processes. Two key characteristics of these proposals are that the processing units envisaged often correspond to units of linguistic description, and that their acquisition is generally not considered. In this chapter, we focus on the role of distributional properties of languages as one critical determinant in the specification of language processing units. We propose that what can be viewed as processing units probably represents only the tip of the iceberg of broader knowledge about statistical properties of language and that the emergence of such representations is, at least partially, shaped by general learning processes. During the last decade, the role of distributional properties of language has mainly addressed the problem of word segmentation in the speech stream. Several studies suggest that young children are able to segment the continuous speech flow into individual words by using prosodic and segmental cues. Learning of the statistical regularities of the language is also suggested by the observation that 9-month-old children listen longer to words that include frequent phonetic sequences than to have showed that adults exposed to an artificial language composed of continuous tri-syllabic sequences without signification were able to learn the language by using the …