Computational Linguistics Colloquium

Thursday, 19 January 2012, 16:15
Conference Room, Building C7 4

The production and comprehension of reduced pronunciation variants in spontaneous speech

Mirjam Ernestus
Radboud University Nijmegen & Max Planck Institute for Psycholinguistics

In casual speech, words are often pronounced with fewer segments than in formal speech. For instance, the English words "ordinary" and "yesterday" may be pronounced in everyday conversations as "onry" and "yesay". In this talk, I will address the question how speakers and listeners process reduced pronunciation variants, on the basis of evidence from corpus-based research, several production and comprehension psycholinguistic experiments, and experiments with automatic speech recognition systems. I will show that speakers may use reduced pronunciation variants in order to reduce articulatory effort, but that this cannot be the only reason. Further, I will show that both speakers and listeners have representations for reduced pronunciation variants stored in their mental lexicons, but that the comprehension of these variants nevertheless cannot be a case of simple pattern matching. Finally, our data demonstrate that listeners need semantic/syntactic context as well as the acoustic cues in the context to understand reduced pronunciation variants. Our data have important consequences for models of speech production and comprehension. Most importantly, they support hybrid models, which assume both abstract lexical representations and clouds of exemplars.

If you would like to meet with the speaker, please contact Jürgen Trouvain.