7th International Conference on Spoken Language Processing

September 16-20, 2002
Denver, Colorado, USA

Multi-Dimensional Analysis of Sonority: Perception, Acoustics, and Phonology

Masahiko Komatsu (1), Shinichi Tokuma (2), Won Tokuma (3), Takayuki Arai (1)

(1) Sophia University, Japan; (2) Chuo University, Japan; (3) Seijo University, Japan

Sonority is an important notion in phonology, but its definition has been controversial. Our analysis showed that sonority can be located in a multi-dimensional perceptual space, and that the dimensions of the space have correspondence to both acoustic parameters and phonological features. In the experiment, a confusion matrix was calculated from the results of consonant perception test for LPC residual signals made from /Ca/ syllables. It is considered that LPC residual signals contain only suprasegmental information and thus the confusion pattern for the signals indicates the consonantsí similarities in suprasegmental domain. This confusion matrix was analyzed with MDS. The result showed that the consonants can be modeled in a 3-dimensional perceptual space according to their sonority. Its dimensions could be related to acoustic measurements (length, pitch, RMS, HNR) and phonological features ([voice], [sonorant], [continuant]). The result also showed that sonority can be mostly defined within the suprasegmental domain.


Full Paper

Bibliographic reference.  Komatsu, Masahiko / Tokuma, Shinichi / Tokuma, Won / Arai, Takayuki (2002): "Multi-dimensional analysis of sonority: perception, acoustics, and phonology", In ICSLP-2002, 2293-2296.