Do Rhythm Measures Tell us Anything about Language Type?
Résumé
In recent language-rhythm studies (Ramus et al. 1999, Low et al., Grabe & Low 2002) a number of different "rhythm measures" have been shown capable of separating languages in a way which appears to confirm empirically the traditional rhythm types (syllable-stress-and mora-timed), where many other instrumental approach have failed. Critically based on the durational and structural variability of syllables within an utterance (cf. also Dauer 1983), these measures can - logically - be expected to increase in reliability with the size of the database from which they are derived. Conversely, measurements, derided from small databases of structurally different speech material, and also possibly from different speakers or material spoken in different styles may expected to vary. The studies so far have not taken these factors into consideration (but cf. Gibbon & Gut for a discussion), and have demonstrated the differentiating power of the measures on rather limited amounts of material, number of speakers etc. Thus it is unclair to what extent the differences between languages found in the studies are real, and whether the apparent differences in discriminatory power between the approaches really reflect significant properties of the languages. This paper endeavours to illuminate the factors influencing the variability in the measures. Three languages - Bulgarian, German and Italian - are examined. The choice of languages is determined partly by their definition as Slav, Germanic and Romance languages, and partly by the availability of segmentally ans (in part) prosodically labelled data. The databases used are: Bulgarian: a) part of the BABEL (project Nr. COP 1304) database (read speach), b) Maptask recordings, c) a recordings of read speech. b) and c) were labelled at IPUS, Saarbrücken; German: the Kiel Corpora of read and spontaneous speech (IPDS 1994; 1995); Italian: the Pisa and Naples part of the AVIP database (ftp://cirass.unina.it). Although the amount of material differs considerably from language to language, the amount and lexical variety as well as the number of speakers exceed that so far reported in the literature by a large multiple. Result obtained for the German data show that overall, the variability observed for German overlaps with rhythm-measure values found for English, Polish and French. Significant shifts are found in both vowel-interval and consonant-interval variability as a function of a) speaker, b) text and c) tempo. Furthermore, differences stemming from tempo change (in sylables per second) can differ for different speakers in the parameter most affected (vowel or consonant variability). Analysis is in progress of the syllable and prosodic structures most contributing to intra-speaker extremes in "rhythm measures" found in the data. The factors considered are length of inter-pause stretch, tempo (phones/sec. and syllables/sec.), relative number of complex syllables and relative number of accented syllables. Parallel analyses are being carried out for Bulgarian and Italian, with the goal of assessing the degree of overlap between the areas of variation within the three languages and thus evaluating the measures as determiners of 'rhythm-type' in the traditional sense.
Loading...