Anticipatory deaccenting in language comprehension

Abstract

We evaluated the hypothesis that listeners can generate expectations about upcoming input using anticipatory deaccenting, in which the absence of a nuclear pitch accent on an utterance-new noun is licensed by the subsequent repetition of that noun (e.g. Drag the SQUARE with the house to the TRIangle with the house). The phonemic restoration paradigm was modified to obscure word-initial segmental information uniquely identifying the final word in a spoken instruction, resulting in a stimulus compatible with two lexical alternatives (e.g. mouse/house). In Experiment 1, we measured participants' final interpretations and response times. In Experiment 2, we used the same materials in a crowd-sourced gating study. Sentence interpretations at gated intervals, final interpretations and response times provided converging evidence that the anticipatory deaccenting pattern contributed to listeners' referential expectations. The results illustrate the availability and importance of sentence-level accent patterns in spoken language comprehension.

Keywords:

• language comprehension
• prosody
• deaccenting
• phoneme restoration
• sentence gating

Acknowledgements

We thank Sarah Glacken and Rebekah Goldstein for contributions to a pilot study, and Dana Subik for lab assistance. We are grateful to Michael Wagner for providing supplemental acoustic analyses and for providing arguments that convinced us that the eye-movement data were problematic for the reasons discussed in Note 3.

Funding

This research was supported by NIH grants [HD27206 and HD073890] to MKT and an NSF graduate research fellowship to MB.

Notes

1. It is also possible that accent shape differences arise along a continuum rather than being as strictly categorical as the distinction between “contrastive” and “standard” accents implies. The point about deaccenting is unaffected.

2. One reason that a speaker might not consistently use anticipatory deaccenting in language production is because of variation in the amount of upcoming material a speaker has planned when he or she begins to speak. For example, when a speaker has not yet planned the second phrase in (6), the trigger for anticipatory deaccenting (i.e. the upcoming repetition of fugues) is not available when the word ‘fugues’ is first produced. Therefore, repetition across parallel phrases sometimes elicits anticipatory deaccenting, as in (3), and sometimes does not, as in (6).

3. We also recorded eye movements in Experiment 1, using an SRI EyeLink 2 eye-tracker. We had hoped that eye movements would provide information about the time course of interpretation. However, because the task was to click and drag the first shape, most fixations remained on that shape, only shifting to the target shape late in the utterance. Therefore, the fixation data were not sensitive to any potential effects of deaccenting prior to the onset of the final word.

Additional information

Funding

Funding: This research was supported by NIH grants [HD27206 and HD073890] to MKT and an NSF graduate research fellowship to MB.