Prosodic pattern detection based on fuzzy set theory
Nicholas Bacuez · The Journal of the Acoustical Society of America · 2011
This research utilizes fuzzy logic and fuzzy set theory to experimentally abstract the mental representations of prosodic contours otherwise not accessible to speakers or researchers. I have developed a pattern recognition system to extract approximations of these representations and their range of variation from speech datasets. Following the principles of fuzzy set theory, a prosodic contour is defined as a category and the variation range of its components is expressed in terms of degree of membership to the category, from 0 (excluded) to 1 (included). Subsequently, the system analyzes the prosodic contour of new sentences by assessing through fuzzy matching their degree of membership to previously identified categories: 1 is perfect, 0.5 is borderline, and 0.1 is improbable. In the current study, I have used closed questions obtained experimentally from native French speakers. The prosodic contour of closed questions is non problematic and ensures the controllability of the output. The model output is an array storing the ideal curve and its degrees of variation. Fuzzy matching reveals that most closed questions from the dataset are in medium range (0.5). This materializes the idea that vagueness (Russell, Wittgenstein), the linguistic correlate to fuzzy logic, is a natural property of language.