Variability in hesitations in Punjabi semi-spontaneous narrative speech: An automatic clustering based analysis

Abstract

This research offers a first analysis of hesitations in Punjabi, an under-researched language, in conjunction with a crosslinguistic comparison. We show speaker related variation in the frequency of hesitations in Punjabi. Variability was also observed in the form of filled pauses which comprised vowels or vowel-consonant sequences with nasals or obstruents. The vowels in filled pauses differed based on their segmental context and individual speakers. Automatic clustering showed that (lexicalized) filled pauses were grouped by F0 register, instead of F0 contour. These results (1) have cross-linguistic significance and (2) provide insights for modeling hesitations in speech technological systems.