Doctoral Dissertation Research: Sources of argument role insensitivity in verb processing

点击翻译按钮获取中文摘要

Humans generally understand utterances quickly and accurately, even in noisy or degraded environments for listening or reading. Many researchers have attributed this success to people’s ability to rapidly predict upcoming words. Previous studies have demonstrated various kinds of evidence for prediction mechanisms, e.g., more predictable words are read more quickly. But less is known about the mechanisms by which predictions are generated. This project investigates these mechanisms, by focusing on situations where people appear to make inappropriate predictions. A useful test case is “role reversed” sentence pairs, such as “the customer that the waitress had served” and “the waitress that the customer served”, in which who did what to whom is reversed. Some psycholinguistic measures of prediction, particularly those involving comprehension, suggest that the verb “served” is equally expected in both sentences, despite being inappropriate in the second. This has been taken as evidence that humans ignore the roles of nouns when generating expectations. However, some other measures of prediction suggest that humans generate appropriate expectations in those same sentences, making full use of role information. This project seeks to resolve this discrepancy. The project combines computational and experimental methods to investigate why different measures indicate a greater or lesser role for semantic roles in moment-by-moment prediction in language. The project will develop a computational model of linguistic prediction that seeks to capture how a shared set of cognitive processes maps onto different experimental measures. The model will be extended based on results from new experiments. In order to understand the time course of predictions and the contributions of different task elements, the experiments will systematically vary whether or not participants are shown anomalous continuations, and what kind of response participants are required to give. The project also develops and refines a scalable pipeline for semi-automatic analysis of spoken language data in psycholinguistic experiments, which can be used by other researchers.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.