15th International Congress of Phonetic Sciences (ICPhS-15)

Barcelona, Spain
August 3-9, 2003

A Goodness Optimization Method for Investigating Phonetic Categorization

Paul Iverson, Bronwen G. Evans

University College London, UK

Existing methods for examining phonetic categorization (i.e., identification or goodness judgments) typically require listeners to give responses on every member of a stimulus set. This article describes a new method that is more efficient for higher-dimensional stimulus sets (i.e., more phonetic detail) in which the number of perceptually distinct stimuli is too large to be played within an experimental session. The method uses a traditional goodness-rating task (i.e., subjects hear individual synthesized stimuli and give goodness ratings on a continuous scale). However, the stimulus selection is determined by an iterative computer algorithm that is designed to find the minimum value of a function within a multidimensional variable space. The method has been applied to map best exemplars of English vowels in a 5- dimensional space including formant movement and duration; best exemplars can be found within the set of 100,700 possible vowels after playing subjects only 35 trials per vowel category.

Full Paper

Bibliographic reference.  Iverson, Paul / Evans, Bronwen G. (2003): "A goodness optimization method for investigating phonetic categorization", In ICPhS-15, 2217-2220.