Comprehenders predictions in the face of, uh, speech disfluencies
Comprehenders Predictions in the Face of, uh, Speech Disfluencies..docx (111.2Kb)
MetadataShow full item record
Everyday speech is littered with disfluencies. Disfluencies are perceived as negative as it is felt they disrupt conversation flow without adding any communicative function. In this eye tracking experiment the edit interval is directly manipulated making a short edit interval and a longer edit interval. It is hypothesised listeners upon hearing a disfluency would take the disfluency as a sign that an error had been made and as a result would predict what the speaker meant to convey. It was also hypothesised that a longer edit interval would instigate listeners to back track further when making a prediction about a possible repair. Results show a short edit interval leads listeners to predict no change would be made to the reparandum object and a long edit interval leads listeners to predict a colour change would be made to the reparandum object. Listeners seem to take disfluencies to mean that the speaker has made an error and in turn will predict what the speaker meant to convey, however it is apparent that longer edit intervals do not encourage listeners to back track when predicting a repair. Edit interval length was shown to be important in aiding the listener to make a prediction. It is concluded that the longer the edit interval, the more time available for the listener to recognise an error was made and subsequently make a prediction about a possible repair. It is also concluded that listeners would believe the word immediately preceding the disfluency was the error word and therefore would not track back further than the word immediately preceding the disfluency when predicting a repair. Results lend support to the argument that disfluencies, in particular fillers such as uh, serve a communicative function to the listener by indicating an error has been made on behalf of the speaker. The study also highlights the importance of the length of the edit interval when interpreting listeners’ predictions.