LangSci Lunch Talk Yogarshi Vyas (Computer Science)

Title: Identifying cross-lingual semantic divergences for better multilingual natural language processing

Abstract : Building natural language processing (NLP) tools that can operate in a multilingual world, and can reason about cross-lingual content, often requires understanding how words in different languages relate to each other. In this talk, we discuss our work on automatic algorithms for identifying lexico-semantic relations between words in two languages. Specifically, we focus on the identifying the asymmetric hypernymy relation which relates a specific word (the hyponym) with a broader word (the hypernym) using an “is-a” relationship (red is a color, dog is an animal). We start by proposing a machine learning framework for this task that requires no labeled hypernym-hyponym data, and can thus be applied to any language. We then enrich this framework using syntactic information from the languages that we are operating in. Finally, we discuss the robustness of the enriched framework when exposed to low resource scenarios, as are likely to arise when dealing with multilingual content.

_______

Food and ideas bring people together. Our weekly lunch talk series provides students and faculty with the opportunity to present their in-progress work to a supportive, interdisciplinary audience.

Lunch is served starting at 12:15pm.