Learning to Attend On Essential Terms: An Enhanced Retriever-Reader Model for Open-domain Question Answering

Abstract

Open-domain question answering remains a challenging task as it requires models that are capable of understanding questions and answers, collecting useful information, and reasoning over evidence. Previous work typically formulates this task as a reading comprehension or entailment problem given evidence retrieved from search engines. However, existing techniques struggle to retrieve indirectly related evidence when no directly related evidence is provided, especially for complex questions where it is hard to parse precisely what the question asks. In this paper we propose a retriever-reader model that learns to attend on essential terms during the question answering process. We build (1) an essential term selector which first identifies the most important words in a question, then reformulates the query and searches for related evidence; and (2) an enhanced reader that distinguishes between essential terms and distracting words to predict the answer. We evaluate our model on multiple open-domain multiplechoice QA datasets, notably performing at the level of the state-of-the-art on the AI2 Reasoning Challenge (ARC) dataset.

Fine-Grained Relevance Annotations for Multi-Task Document Ranking and Question Answering

“…In addition to retrieval focused models, the research community also proposed multi-task models for the QA-at-scale or sometimes referred to open-world-QA task, where systems need to retrieve candidates and select answers. Models such as the Retriever-Reader model from Ni et al [22], the multi-task Retrieve-and-Read models [23,24], or models based on phrase indices [29], are mainly evaluated on re-purposed QA-collections that lack dense retrieval judgements. FiRA presents a perfect fit for the evaluation of such models, as it can cover both retrieval and question answering aspects equally well.…”

Answering Event-Related Questions over Long-Term News Article Archives

“…Wang et al [7] introduce R 3 model, where IR component and MRC component are trained jointly by reinforcement learning. Ni et al [10] propose ET-RR model, which improves IR part by identifying essential terms of a question and reformulating the query.…”

Alignment over Heterogeneous Embeddings for Question Answering

“…Some of these works also rely on structured knowledge bases (Zhong et al, 2018a;Ni et al, 2018) such as ConceptNet (Speer et al, 2017). Some approaches use query expansion methods in addition to the above methods (Musa et al, 2018;Nogueira and Cho, 2017;Ni et al, 2018). For example, Musa et al (2018) used a sequence to sequence model (Sutskever et al, 2014) to generate an enhanced query for ARC which retrieves better supporting passages.…”