A question about a document can be about its text, an embedded image, or a table -- and answering it well usually means retrieving from the specific modality it's actually asking about, rather than searching everything indiscriminately every time.
Write route_modality_for_question(question, modality_keywords, default_modality), where modality_keywords maps a modality name to a list of keywords. Return the first modality (in dict order) with a case-insensitive keyword match in the question, or default_modality if none match.
No results yet
Run your code to check it against the examples, or submit to run every test.