Database of Dutch diphone perception
A common view in psycholinguistic models of spoken-word recognition is that speech decoding
is essentially phonemic: the speech signal is mapped onto intermediate phonemic
representations, which in turn are used to access phonemically coded lexical representations. Several
types of evidence indicate, however, that acoustic-phonetic information is passed to the mental lexicon
continuously, as it becomes available in the speech signal. This is not to say, however, that
phonemes have no role to play in lexical access. The view that speech decoding is essentially
phonemic captures the important fact that differences in the input which signal the intentions of speakers are phonemic.
The process and representations involved in lexical access must thus be sensitive to phonemic
differences. It is perhaps for this reason that psycholinguistic models of spoken-word recognition
have assigned a greater role in lexical access to phonemic distinctions than to the continuous
unfolding of phonemically relevant information.
The aim of the present research was to provide temporally detailed perceptual data concerning the unfolding of phonetic information over time. For this purpose portions of natural utterances were presented to listeners for classification. To measure phoneme activations in all left and right phonetic contexts, gated versions of all possible Dutch diphones were used.
This database is described in: Smits, R., Warner, N., McQueen, J.M. & Cutler (2003), Unfolding of phonetic information over time: A database of Dutch diphone perception, Journal of the Acoustical Society of America, 113, 563-574. A preprint of this paper is available.