Lucy Lin, Qualifying Project Presentation

Title: Semantic Measurement of Ideas in Text Corpora

Advisors: Noah Smith and Yejin Choi

Abstract: In this talk, we introduce the problem of semantic measurement: estimating the frequency of a natural language-expressible proposition in a corpus.  We describe a two-stage process involving (1) a fast, high-recall semantic function to filter out unlikely sentences, and (2) a more expensive syntax-based model to score the filtered sentences.  To evaluate our approach, we conduct two case studies: one on framing of policy issues in media and the other in the domain of natural disaster recovery.  Our findings validate our design decisions and motivate future work on semantic measurement applications.