BoolQ (Boolean Questions) is a question answering dataset included in SuperGLUE, consisting of naturally occurring Yes/No questions derived from Google Search queries — unlike artificial questions, BoolQ queries are often ambiguous or unstated, requiring the model to infer the answer from a paired Wikipedia passage.

Characteristics

• Source: Real user queries "Is the knicks game on tv tonight?"
• Context: A Wikipedia paragraph that may or may not explicitly contain the answer.
• Difficulty: Often requires implicit reasoning. "Does France have a king?" (Passage: France is a Republic... implies No).

Why It Matters

• Realism: Tests the ability to answer the most common type of human query (verification).
• Inference: The answer is rarely a simple span extraction ("Yes" or "No" is not in the text).
• SuperGLUE: A core component of the SuperGLUE benchmark for difficult NLU.

BoolQ is yes or no? — testing whether models can determine the truth value of a statement based on evidence text.