INTERSPEECH 2013
14thAnnual Conference of the International Speech Communication Association

Lyon, France
August 25-29, 2013

How do Multiple Sublexical Cues Converge in Lexical Segmentation? An Artificial Language Learning Study

Odile Bagou, Ulrich Hans Frauenfelder

Université de Genève, Switzerland

This artificial language learning (ALL) study investigates how multiple sublexical cues contribute simultaneously to lexical segmentation in French. Previous research [9] has explored segmentation in lexical identification by presenting conflicting sub- lexical cues in the same experiment. We argue that this method of pitting conflicting cues against each other pairwise does not allow the researcher to discover interactions in cue use. We can obtain a more accurate view of speech segmentation by focusing on how French listeners exploit convergent rather than conflicting sources of segmentation information. To this end, acoustic-phonetic, phonological and prosodic cues were manipulated and combined across conditions. The data clearly showed that the combined effects of multiple sublexical cues are more complex than expected based on previous results. Moreover, phonological cues did not supersede prosodic cues as previously shown for English but rather have cumulative and synergistic effects. Two explanations can be given for these differences in the observed results. First, the processes of lexical segmentation in lexical identification [9] and lexical acquisition studied here may not be the same. Second, the hierarchy of segmentation cues is probably not universal, but rather depends upon the specific phonological and prosodic properties of the language studied.

Full Paper

Bibliographic reference.  Bagou, Odile / Frauenfelder, Ulrich Hans (2013): "How do multiple sublexical cues converge in lexical segmentation? an artificial language learning study", In INTERSPEECH-2013, 2817-2821.