Question Answering (QA)


Question Answering (QA) aims to automatically answer natural language questions posed by human. In MSRA QA team, we are working on QA from two angles: Single-turn QA and Conversational QA.

Single-turn QA takes a single question as input and provides its answer as output without considering any context. We are building state-of-the-art single-turn QA engines based on heterogeneous databases, including Knowledge-based QA, Table/List-based QA, Text-based QA, Community-based QA, and Visual QA. Both Microsoft Bing and Xiaoice have used these technologies already.

Conversational QA, on the other hand, considers not only current input question, but also previous conversation turns. New problems should be addressed in such multi-turn scenario, such as Coreference Resolution, State Tracking, Proactive Asking/Confirming. Conversational QA is critical to intelligent personal assistants and chatbots, such as Microsoft Cortana and Xiaoice.