INSCIT: Information-Seeking Conversations with Mixed-Initiative Interactions
Published
2023-05-24
Zeqiu Wu
,
Ryu Parish
,
Hao Cheng
,
Sewon Min
,
Prithviraj Ammanabrolu
,
Mari Ostendorf
,
Hannaneh Hajishirzi
Zeqiu Wu
University of Washington
Ryu Parish
University of Washington
Hao Cheng
Microsoft Research
Sewon Min
University of Washington
Prithviraj Ammanabrolu
Allen Institute for AI
Mari Ostendorf
University of Washington
Hannaneh Hajishirzi
University of Washington
Allen Institute for AI
Abstract
In an information-seeking conversation, a user may ask questions that are under-specified or unanswerable. An ideal agent would interact by initiating different response types according to the available knowledge sources. However, most current studies either fail to or artificially incorporate such agent-side initiative. This work presents INSCIT, a dataset for Information-Seeking Conversations with mixed-initiative Interactions. It contains 4.7K user-agent turns from 805 human-human conversations where the agent searches over Wikipedia and either directly answers, asks for clarification, or provides relevant information to address user queries. The data supports two subtasks, evidence passage identification and response generation, as well as a human evaluation protocol to assess model performance. We report results of two systems based on state-of-the-art models of conversational knowledge identification and open-domain question answering. Both systems significantly underperform humans, suggesting ample room for improvement in future studies.
Presented at ACL 2023
Article at MIT Press