Faithful Temporal Question Answering over Heterogeneous Sources
Keywords: Question Answering, Faithfulness, Explainability, Temporal
TL;DR: This paper introduces a novel approach for faithful temporal question answering, incorporating heterogeneous sources into the answering process.
Abstract: Temporal question answering (QA) aims to return crisp answers to user questions that involve temporal constraints, with phrases such as "...in 2019" or "... before COVID". In the former, time is an explicit condition, in the latter it is implicit. State-of-the-art methods for temporal QA have limitations along three dimensions. First, when QA systems rely on neural inference, time constraints are often merely soft-matched, giving room to invalid or inexplicable answers. Second, questions with implicit time are poorly supported. Third, most systems tap into one source of information only, typically either a knowledge base (KB) or a text corpus. We propose a temporal QA system that addresses these shortcomings. First, it explicitly identifies and enforces temporal constraints for faithful answering with tangible evidence. Second, it includes techniques for properly handling implicit questions. Third, it operates over heterogeneous sources, covering KB, text, and also web tables in a unified manner. The method has three stages: (i) understanding the question and its temporal conditions, (ii) retrieving evidence from all sources, consistent with the temporal constraints, and (iii) faithfully answering the question from these pieces of evidence. As implicit questions are largely underrepresented in established benchmarks, we introduce a principled method for generating diverse questions of this kind in a systematically controllable way from heterogeneous sources. Experiments on an existing and the new benchmark show superior performance over a suite of baselines.
Track: Semantics and Knowledge
Submission Guidelines Scope: Yes
Submission Guidelines Blind: Yes
Submission Guidelines Format: Yes
Submission Guidelines Limit: Yes
Submission Guidelines Authorship: Yes
Student Author: No
Submission Number: 1367
Loading