Abstract
Natural language interfaces to tabular data must handle ambiguities inherentto queries. Instead of treating ambiguity as a deficiency, we reframe it as afeature of cooperative interaction, where the responsibility of queryspecification is shared among the user and the system. We develop a principledframework distinguishing cooperative queries, i.e., queries that yield aresolvable interpretation, from uncooperative queries that cannot be resolved.Applying the framework to evaluations for tabular question answering andanalysis, we analyze the queries in 15 popular datasets, and observe anuncontrolled mixing of query types neither adequate for evaluating a system'sexecution accuracy nor for evaluating interpretation capabilities. Ourframework and analysis of queries shifts the perspective from fixing ambiguityto embracing cooperation in resolving queries. This reflection enables moreinformed design and evaluation for natural language interfaces for tabulardata, for which we outline implications and directions for future research.