Incremental structure prediction during language comprehension: behavioral and neurobiological evidence from Mandarin Chinese and English

Thesis

This thesis investigates the capacity of predictive mechanisms in real-time language comprehension to use subtle linguistic cues. Language comprehension, inherently sequential, involves incremental interpretation of linguistic inputs. However, predictive processing allows comprehenders to anticipate upcoming linguistic structures beyond the linear order, offering potential cognitive efficiency gains, particularly in long-distance dependency constructions. This study aims to understand the extent to which predictive strategies are employed, the level of detail comprehenders can predict, and the condition under which prediction occurs, through both behavioral and neurobiological measurements in Mandarin Chinese and English.

This thesis comprises three independent yet interrelated studies focusing on different linguistic cues that might trigger predictions. Study 1 examines whether Chinese classifiers, which constrain animacy without specifying particular lexical items, elicit semantic feature predictions. Using Representational Similarity Analysis on EEG recordings, Study 1 provides neurobiological evidence that comprehenders can predict abstract semantic features beyond specific lexical items. Study 2 extends these findings by exploring whether animacy-constraining classifiers can guide comprehenders to predict structural elements, i.e., gap sites in head-final relative clauses in Mandarin Chinese. Both eye-tracking and self-paced reading results demonstrate the comprehenders’ability to utilize classifiers to modulate active gap search in the absence of head noun fillers. Study 3 examines the effect of a less understood element, presuppositional constraints, on modulating active gap search. It compares negative island constraint, a presuppositional constraint on filler-gap dependency formation with strong complex NP island constraint, however, the results suggest that negative island constraint cannot be rapidly used by the real-time parser to inhibit active gap filling in the same way strong islands do.

Results across the three studies suggest that predictive structure building in real-time language processing is dynamically modulated by subtle linguistic cues, and cues at different levels might take their effects with different timings: syntactic and semantic cues have more immediate effects while pragmatic cues take longer to compute. These findings contribute to a nuanced understanding of real-time language comprehension, teasing apart various factors affecting predictive structural building in dependency constructions.

Authors: Huang, Z

• DOI: 10.5287/ora-brdjxmz9q
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: classifier; negative island; predictive processing; language comprehension; filler-gap dependency
• Subjects: Neurolinguistics; Psycholinguistics