Conversational question answering (CQA) aims to answer a question considering a given document and previous conversations. In CQA, users can obtain desired information concisely through follow-up questions. Therefore, it can be applied to various fields requiring active interaction between users and systems. There are two main challenges to this task: (1) understanding the current question based on conversation history and (2) finding answers to questions in given passages.
Explainable Question Answering
While QA models are capable of generating sentences that correctly answer a given question, it is difficult to deduce why and how the neural network produced such an answer. Our goal is to produce detailed and understandable explanations that justify why the generated answers are correct. We develop QA models which can both answer and explain by (1) generating intermediate inference sentences from facts and (2) constructing a reasoning graph from them.
Multimodal Open Domain Question Answering
Open domain question answering (ODQA) is a task that answers factoid questions when given a large number of source documents. The existing ODQA system mainly focuses on dealing with only textual information. However, real-world knowledge is distributed over different modalities. Our goal is to develop a multi-modal ODQA system that can retrieve and reason over heterogeneous modalities such as text, tables, and images.