The University of Texas at Dallas HLTRI's Participation in EPIC-QA: Searching for Entailed Questions Revealing Novel Answer Nuggets

TL;DR

This paper presents SER4EQUNOVA, a neural IR system that leverages entailment relations and question generation to improve COVID-19 question answering, achieving promising results in the EPIC-QA benchmark.

Contribution

Introduces a multi-phase neural IR system combining BM25, BERT, T5, and entailment graphs for improved question answering in COVID-19 domain.

Findings

Outperformed baseline methods in EPIC-QA Expert QA task.

Effectively identified novel answer nuggets using entailment relations.

Demonstrated the utility of question entailment graphs for answer re-ranking.

Abstract

The Epidemic Question Answering (EPIC-QA) track at the Text Analysis Conference (TAC) is an evaluation of methodologies for answering ad-hoc questions about the COVID-19 disease. This paper describes our participation in both tasks of EPIC-QA, targeting: (1) Expert QA and (2) Consumer QA. Our methods used a multi-phase neural Information Retrieval (IR) system based on combining BM25, BERT, and T5 as well as the idea of considering entailment relations between the original question and questions automatically generated from answer candidate sentences. Moreover, because entailment relations were also considered between all generated questions, we were able to re-rank the answer sentences based on the number of novel answer nuggets they contained, as indicated by the processing of a question entailment graph. Our system, called SEaRching for Entailed QUestions revealing NOVel nuggets of Answers (SER4EQUNOVA), produced promising results in both EPIC-QA tasks, excelling in the Expert QA task.