Origin of Consonant Duration Patterns

Segment duration varies with syllable structure (Lehiste, 1970), resulting in word-level duration patterns. These patterns provide important phonetic cues to syllable and word boundaries in English (Christie, 1977; Boucher, 1988; Tuller & Kelso, 1991), and possibly in other languages (Maddieson, 1985). For instance, English-speaking listeners syllabify the same sequence differently depending on the relative duration of the participating segments. An obstruent-sonorant sequence is syllabified as an onset cluster when the obstruent is longer than the sonorant, but split when the consonants are similar in length (Christie, 1977). The relationship between segment duration patterns and syllable perception could indicate that syllables are units of speech production. Such a conclusion is consistent with most theories of speech production, which invoke the syllable explicitly or implicitly as a unit in speech motor programming (e.g., Stetson, 1951; Kozhevnikov & Chistovich, 1965; Lehiste, 1970; MacNeilage, 1970; Fujimura, 1981; Shattuck-Hufnagel, 1983; Fowler, 1987; Dell, Burger, Svec, 1997). A major reason that the syllable has been chosen as an appropriate unit in speech motor programming is because suprasegmental sound patterns, such as the aforementioned consonant duration patterns, are best characterized with respect to the syllable. But the duration patterns themselves—long for consonantal onsets, short for consonantal offsets, long-short for onset clusters, and so on—are arbitrary. That is, there is nothing about the phonological syllable that can explain why an onset cluster is produced with a long-short duration pattern instead of short-long, long-long or any other kind of pattern. Since we know that speakers control duration to convey other sorts of boundary information, e.g., lexical boundaries (Turk & Shattuck-Hufnagel, 2001), it is reasonable to suggest that the duration patterns are so produced because they convey syllable structure information. But if the claim is that the same duration patterns are universally associated with the same syllable structures (Lehiste, 1970; Maddieson, 1985), then we might wonder how such arbitrary patterns have become language universal. The seeming arbitrariness of the segment duration patterns and their universality might be explained, however, if the link between the patterns and the syllable is broken, at least in speech production. If segment duration patterns are delinked from syllables in production, then the patterns are free to be explained according to other, more basic, production constraints on the sequential articulation of segments. The patterns may then be relinked to the syllable in perception. Syllable boundaries can be extrapolated from word boundaries, which are also characterized by the same suprasegmental phonetic and phonological patterns. For illustration of a basic production explanation for segment duration patterns, consider the longshort pattern typically associated with onset clusters. Cross-language frequency data indicate that onset clusters are usually constructed of an obstruent-sonorant sequence that precedes a vowel (Bell & Hooper, 1978). This type of sequence suggests a coarticulatory origin for the long-short duration pattern based on jaw movement. In general, obstruents are articulated with a more closed jaw than sonorants, and sonorants are articulated with a more closed jaw than vowels (Lindblom, 1983; Keating, Lindblom, Lubker, Kreiman, 1994). An obstruent-sonorant-vowel sequence is therefore articulated with the jaw moving from a closed position to an open position. The sonorants may be shortened when the segments are sequentially articulated because jaw movement is continuous in speech. Sonorants are only initiated after obstruent release, when the jaw is already in a lower position, and must be terminated when the continually downward-moving jaw makes for inefficient consonantal