The IconQA dataset consists of 3 different sub-tasks:

• Multi-image-choice questions: choose from 2-5 image choices
• Multi-text-choice questions: choose from 2-5 text choices
• Filling-in-the-blank questions: short textual answers

The IconQA dataset involves questions that require a variety of cognitive reasoning and arithmetic reasoning types. We group these reasoning skills into 13 categories: Geometry, Counting, Comparing, Spatial, Scene, Pattern, Time, Fraction, Estimation, Algebra, Measurement, Commonsense, and Probability.

The chart on the left shows the most common three-word combinations that start the sentences. The chart on the right shows the most freqent words in the questions.