2009/06/25

Mattys, S. L. & Melhorn, J. F. (2007). Sentential, lexical, and acoustic effects on the perception of word boundaries. Journal of the Acoustical Society of America, 122(1), 554–567.

Presentation: Shelly

Summary: Sally

In this study, two experiments were conducted to investigate whether there is a trade-off relation among the effects of sentential context, lexical knowledge, and acoustic cues regarding speech segmentation. In Experiment 1, various renditions of near-homophonous phrases (eg. A: plum pie vs. B: plump eye) as well as carrier sentences intentionally read with the semantically appropriate phrase (either A or B) at the end of them were recorded separately from the same native speaker. The sentence-final phrases were then deleted and piloted on 27 native speakers to differentiate the strong/mild level of sentential information. These truncated sentences were then appended with the phrases and used as the stimuli (eg. Strong A: The baker looked at the drawing of a __. Mild A: The girl looked at the drawing of a __. Mild B: The scientist looked at the drawing of a __. Strong B: The surgeon looked at the drawing of a __.). Thirty participants were instructed to decide which of the two words they heard at the end of the utterance. Effects of acoustic strength (strongly or mildly like either phrase), contextual congruency (whether the information carried by the carrier sentences and the following phrase were congruent), and contextual strength (strong vs. mild sentential information) were then analyzed. Results showed that segmentation accuracy was higher when acoustic cues were stronger, and when the sentential context was consistent with the acoustic cues. The interaction between these two factors was also found, showing that the modulation was greater when the acoustic cues were mild. In addition, the assessment of juncture discriminability (d’ score) further revealed that the contextual strength was resulted from a general response bias, rather than a modification of the treatment of acoustic cues.

Following the same logic, Experiment 2 was designed to test whether lexical knowledge influences listeners’ reliance on acoustic cues. In this experiment, the beginning of the near-homophonous phrases were edited so that there was always a lexical bias in favor of one (eg. A:  skum pie rather than skump eye) or the other (eg. B: lump eye rather than lum pie). Another 30 participants were tested for their segmentation accuracy and the results were subjected to an analysis between the effects of acoustic strength and lexical congruency. Results showed that like the sentential effect, segmentation accuracy was higher when the acoustic cues were stronger, and when lexical information was congruent with it; however, unlike the sentential effect, the analysis on the d’ score indicated that the lexical effect not only biased the listeners’ segmentation responses, but  influenced their sensitivity to the acoustic cues. The author finally concluded the study by reiterating the need to study multiple cues on speech segmentation and emphasizing their interactive and compensative nature.