A crosslinguistic corpus phonetic analysis of intrinsic vowel duration

This study investigated the crosslinguistic distribution of intrinsic vowel duration, the observation that high vowels tend to be shorter than low vowels, using a large-scale corpus phonetic approach. Previous studies have reported a durational difference but have been limited by small speaker samples and narrow language coverage. Nevertheless, this observation has been debated to reflect an automatic consequence of the vowel height difference, or to be under speaker control. We reconsider this dichotomy in terms of competing pressures of target uniformity and enhancement in phonetic realization. Target uniformity would be consistent with a systematically shorter duration for high than low vowels and no systematic asymmetry between front and back vowels. In contrast, enhancement would predict larger, more variable durational differences between high and low vowels. To address the empirical gap, we analyzed duration contrasts among /i/, /u/, and /a/ in inter-obstruent contexts from over 60 languages, 80 language–length contrasts, and 16 language families in the VoxCommunis Corpus using automatic and manual vowel segmentations. Linear mixed-effects models assessed duration differences across vowel height and backness, incorporating rate-normalized and raw measures, while by-speaker analyses probed individual consistency. High vowels were shorter than low vowels in approximately 90% of language contrasts and significantly so for almost 80%. Crucially, no consistent durational bias emerged between high front and high back vowels. These findings suggest that intrinsic vowel duration is better understood as a statistical universal emerging from the balance between uniformity and enhancement in phonetic realization, rather than as a purely automatic universal.