Segmentation cues in conversational speech: robust semantics and fragile phonotactics

Area of Science:

• Psycholinguistics
• Speech Perception
• Phonetics

Background:

• Listeners use multiple cues to segment connected speech into words.
• Previous studies often used read speech, which may not reflect natural conversational settings.
• Speaker's articulatory effort (hyperarticulation vs. hypoarticulation) might influence cue salience in conversation.

Purpose of the Study:

• To investigate how speech style (spontaneous conversation vs. read speech) affects listeners' interpretation of word segmentation cues.
• To examine the interplay between semantic likelihood and phonotactic cues in spontaneous and read speech.
• To test the hypothesis of a compensatory relationship between acoustic-phonetic cues and other information sources.

Main Methods:

• Used a map task to elicit spontaneous conversational speech.
• Systematically varied semantic likelihood and cross-boundary phonotactics in two-word target phrases (landmarks).
• Employed cross-modal identity priming to assess listeners' segmentation cue interpretation in both spontaneous and re-recorded read speech.

Main Results:

• An interaction between cue valence and cue type was observed, independent of speech style.
• Semantic plausibility significantly affected segmentation, while cross-boundary phonotactics did not.
• Speech style differences were evident but did not alter the main patterns of listener behavior.

Conclusions:

• Lexical-semantic information is primary for word segmentation in natural conversational speech.
• The importance of phonotactic cues for segmentation may have been overestimated in previous studies.
• Read speech research reliably predicts segmentation patterns in spontaneous conversation, despite acoustic differences.