Inferring Implicit Comparison Classes in Natural Language Understanding

点击翻译按钮获取中文摘要

This award was provided as part of NSF's Social, Behavioral and Economic Sciences (SBE) Postdoctoral Research Fellowships (SPRF) program and SBE's Perception, Action, and Cognition program. The goal of the SPRF program is to prepare promising, early career doctoral-level scientists for scientific careers in academia, industry or private sector, and government. SPRF awards involve two years of training under the sponsorship of established scientists and encourage Postdoctoral Fellows to perform independent research. NSF seeks to promote the participation of scientists from all segments of the scientific community, including those from underrepresented groups, in its research programs and activities; the postdoctoral period is considered to be an important level of professional development in attaining this goal. Each Postdoctoral Fellow must address important scientific questions that advance their respective disciplinary fields. Under the sponsorship of Dr. Roger Levy at MIT, this postdoctoral fellowship award supports an early career scientist investigating children's understanding of context-sensitive language. Learning language is challenging because words mean different things in different contexts. The adjective 'big' conveys that the size of an object is greater than some standard, but what that standard is depends on the context (e.g., a big shoe is a lot smaller than a big building). Extant research suggests that by the time children start producing the word 'big', they already understand its context-sensitivity (big shoe vs. big building). Understanding the cues available to and used by a child to form a context-sensitive interpretation of a sentence will inform theories and models of language development. In addition, formalizing with precise mathematical models how context-sensitive language is used and understood will also help us build machines that understanding language in more humanlike ways. The project encompasses a series of experimental, corpus, and computational modeling studies to elucidate the representations that underlie the incredible human capacity to learn and use language flexibly. Under Objective 1, we extend state-of-the-art probabilistic models for interpreting context-sensitive utterances (e.g. 'big') to address how semantically identical but syntactically different utterances (e.g., "That Great Dane is big" vs. "That is a big Great Dane") could provide cues to the relevant standards or comparisons (e.g., big for a dog vs. big for a Great Dane). Under Objective 2, we examine naturalistic video corpora of conversations between young children and their care-givers to uncover the perceptual, linguistic, and referential cues that the child listener could recruit to construct the relevant comparisons for interpreting context-sensitive adjectives like 'big'. Under Objective 3, we test the causal influence of such cues on adult and 4- to 5-year-old's inferences about the relevant comparisons. Testing our hypotheses in both adults and young children sheds light on the developmental origins of understanding context-sensitive language. This work makes new connections between the fields of language acquisition and computational cognitive science while furthering our understanding of both.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.