answertype

Answertype is a label used in question answering (QA) systems and related datasets to indicate the expected form of the answer. It functions as a metadata category that helps guide answer generation, evaluation, and downstream tasks such as answer extraction and ranking. In many QA annotation schemes, answertype is assigned to a question during data labeling and can be used to filter or partition questions by the form of the correct response.

Common answertype categories include yes/no, numeric, date/time, entity (such as a person, organization, or location), noun phrase, and list or multi-part answers. Some schemes also recognize free-form or multi-sentence answers. Examples: “When was the Declaration of Independence signed?” has the date/time answertype; “Who invented the telephone?” has an entity-type answer; “List the planets in order from closest to farthest from the Sun” has a list-type answer.

In practice, answertype supports model development by providing a target constraint for answer generation and by enabling type-based evaluation. It helps assess whether a model can produce answers in the correct format, complements factual correctness checks, and can be used to diagnose systematic errors (for instance, producing a long sentence when a date is expected).

Limitations include ambiguity in some questions where multiple answer forms are possible, cross-domain variation in labeling schemes, and the potential mismatch between a question’s surface form and its actual required answer type. Related concepts include factoid vs non-factoid questions, and answer extraction or classification tasks.

See also: question answering, answer extraction, annotation schemes, yes-no QA, factoid.