What does % V actually measure ?

The system of classifying languages by their rhythmic patterning traditionally and historically characterizes the patterns in terms of three groups. One group includes the so-called “stress-timed” languages like English, German, and Dutch, which are perceived as having equal intervals between stresses, and are described as sounding rhythmically like Morse code (James 1940). A second rhythmic class includes the “syllable-timed” languages such as Spanish, Italian, and French, described by James (1940) as having a staccato beat, in which syllables are perceived as coming along at regular intervals. Abercrombie (1965) assigned Russian and Arabic to the former category, and Yoruba and Telugu to the latter. Japanese is argued to belong to the third, “mora-timed” class, as investigated phonetically by Port et al. (1987), among others. Several languages, such as Catalan and Polish (Dauer 1987; Nespor 1990) have been claimed to be intermediate languages, exhibiting phonological characteristics of both the stressand syllable-timed classes, and many languages remain unclassified. While these traditional rhythmic classifications are tied to ideas of isochrony, i.e. the temporally regular occurrence either of stresses or of syllables, studies intended to confirm these intuitions by tying rhythmic properties directly to speech patterns, i.e. in terms of measurable intervals between stresses or syllables, have been inconclusive at best (Lehiste 1973, 1977; Roach 1982). Thus, although the perception of regular speech rhythm is often described, it has been exceedingly difficult to tie perceptions of rhythm to quantitative periodicity in any one particular phonetic aspect of a language (i.e. duration); instead, it has been hypothesized that rhythmic differences may at best be captured by a combination of factors, such as possible syllable shapes and the presence or absence of vowel reduction (Dauer 1983). Nevertheless, in recent decades the long-standing intuition that languages differ in their speech rhythm, from staccato ‘machine-gun-like’ strings of rhythmically equivalent syllables to alternating patterns of strong and weak syllables, has received apparent support from both perceptual and acoustic studies. For example, Ramus, Dupoux and Mehler (2003) reported a high success rate for listeners in distinguishing among the rhythm patterns of severely downsampled utterances from languages of different classes, supporting a proposed acoustic measure of the durational ratio of vocalic regions to total acoustic content, %V, which seemed to similarly distinguish among languages of the different proposed rhythmic types. Ramus, Nespor and Mehler (1999), among others, found that languages traditionally described as stress-timed (e.g. Dutch, English, Russian) tended to have lower %V values than syllable-timed languages (e.g. French, Italian, Spanish); and mora-timed Japanese had the highest %V of all. However, subsequent reports using %V as a measure have differed considerably from the original results (White and Mattys 2007), and Arvaniti (2009) showed that sentences containing many open syllables tend to have a higher %V than those that contain many closed and complex syllables, regardless of language. This paper quantifies that relationship by examining the correlation between %V and syllable structure in a set of spoken utterances whose sentence targets were