skip to content

Faculty of Arts and HumanitiesLingCologne

Task modulates predictive behaviour in constraining context: Eye-tracking evidence from a prediction error study

Yiling Huoa, Suiping Wangb, and Wing-Yee Chowa

aDivision of Psychology and Language Sciences, University College London, London, UK; bPhilosophy and Social Science Laboratory of Reading and Development in Children and Adolescents (South China Normal University), Ministry of Education, Guangzhou, China

We set out to investigate whether Mandarin Chinese speakers use their tone sandhi knowledge to detect early signs of prediction error in numerals and revise their noun predictions accordingly, using a visual-world eye-tracking paradigm. In Experiment 1, participants saw visual displays (e.g., coffee and cake) while listening to sentences whose constraining context predicted one of the objects (‘At Starbucks, Anne bought…’). While filler trials ended with the expected noun, experimental trials (1/3 of all trials) ended with the other, unexpected object in a numeral-classifier-noun phrase (‘…one piece of cake’). The numeral’s tone either indicated prediction error (Different Tone condition), or was uninformative (Same Tone condition). Participants were asked to select the picture mentioned in the sentence as soon as possible using the keyboard. Results showed no difference in the onset of fixations to the unexpected target between conditions, suggesting no evidence for revising prediction based on tone sandhi.

However, we discovered that despite the constraining context (‘Starbucks…’), participants did not use the context to predict: fixations on the expected object (coffee) were equal to those on the unexpected target (cake) until the noun onset. A follow-up Experiment 2 with the same materials and a comprehension question task revealed significantly more fixations to the expected object than the unexpected target before noun onset. Overall, the experiments suggest listeners showed predictive behaviour during context with comprehension tasks, not with picture selection tasks. We speculate that the picture selection task may have discouraged prediction and led participants to wait for unambiguous bottom-up information.

(1) Experiment design. Lexical tones relevant to the experimental manipulation are spelt using numbers.
Figure 1 (right). Experiment 1, with picture-choosing task (n=38). Proportion of fixations on each object. Sentences containing the numeral ‘liang’ (‘two’), which undergoes the T3 sandhi. Black dots and horizontal error bars indicate the mean onset of looks to the target object and 95% credible intervals.
Figure 2 (left). Experiment 2, with comprehension question task (n=17). Proportion of fixations on each object. Sentences containing the numeral ‘liang’ (‘two’), which undergoes the T3 sandhi. Black dots and horizontal error bars indicate the mean onset of looks to the target object and 95% credible intervals.
References
  • [1] Kamide, Y., Altmann, G. T., & Haywood, S. L. (2003). The time-course of prediction in incremental sentence processing: Evidence from anticipatory eye movements. Journal of Memory and language, 49(1), 133-156.
  • [2] Chow, W.Y., & Huo, Y. (2020) Listeners can use tone information to predict: Evidence from Mandarin Chinese. Poster at the 33rd annual CUNY Human Sentence Processing Conference.