ÓÈÎïÊÓƵ

About the speaker

is a final-year PhD candidate at Saarland University, and the 2024 recipient of the Independent Postdoc Fellowship at the Heidelberg Institute for Theoretical Studies. Their research focuses on social and technical aspects of trustworthy natural language processing (NLP). Their more technical work involves proposing linguistically-grounded methods and evaluations in NLP, drawing from a variety of subfields. On the social side, they investigate how we define abstract concepts in NLP like ‘fairness’ and ‘reasoning’, as well their societal implications. Before their PhD, they worked on speech recognition in industry, as a computational linguist at Dialpad, and as a research assistant at the Phonological Processing Lab and the Discourse Processing Lab at ÓÈÎïÊÓƵ.

Title

Linguistics for Trustworthy Natural Language Processing

Abstract

From spam filtering and automatic translation to speech recognition and language models, natural language processing (NLP) systems have increasingly become a part of our daily lives. However, despite their success and ubiquity, NLP systems are also prone to failures that erode our trust and cause harms. The overarching goal of my research is to use linguistics to identify these failures and build trustworthy NLP systems. In this talk, I present two examples of my work under this umbrella. The first draws from discourse to evaluate stereotypes and pronominal reasoning with language models, and the second uses traditional computational linguistics approaches to improve the reliability of automatic question-answering systems. I will close with a high-level overview of where I see trustworthy NLP and my work going in the next 10 years, and the critical role of linguistics in creating future systems that are fair to everyone, and faithful to context and input. 

Joining via Zoom

Please email lingcomm@sfu.ca to receive the Zoom link.