meanWhat

MeanWhat is a term used in linguistic and data-science discourse to denote a proposed metric that captures the average thematic depth of questions beginning with the wh-word what in a given corpus. It is defined as the arithmetic mean of semantic complexity scores assigned to each what-question, where complexity is determined by the breadth of the reference set and the expected length of the answer.

Origin and usage: The term emerged in theoretical discussions around interrogative words and information-seeking behavior, and is sometimes described in the context of evaluating dialogue systems and corpus analyses. The name blends the statistical concept of a mean with the interrogative word What, emphasizing its role as a mean-based measure of what-type questions.

Calculation: For each what-question, assign a complexity score that reflects the scope or difficulty of the expected answer (for example, the number of entities, concepts, or possible referents involved). MeanWhat is calculated as the sum of these scores divided by the number of what-questions. A normalized variant may divide each score by the question’s length to reduce bias from longer questions.

Applications: Researchers use meanWhat to compare questioning patterns across domains, track changes in information-seeking behavior, or evaluate the informativeness of conversational agents when what-questions are prevalent. It can complement metrics such as question frequency, wh-word diversity, and answerability.

Limitations: The usefulness of meanWhat depends on how complexity is defined and annotated, and results can be influenced by dataset composition, task definitions, and labeling conventions. Cross-domain comparisons require standardized scoring schemes.

See also: Mean, Wh- words, Question complexity, Information-seeking behavior.