What the Question Parser Extracts from a User String: Keywords, Scope, Shape, Decomposition, Clarification

A user's question within a RAG system is parsed to extract more than just its words, including the expected answer shape, where to look in a document, and whether the question is compound or too vague. The parser populates a structured data row with five field families: keywords, answer shape/type, scope, decomposition, and clarification. Keywords are collected from multiple sources to bridge the vocabulary gap between the user's query and the document's language. These sources include direct user input, LLM-generated rewrites to handle synonyms, and an expert-curated dictionary of concepts and their keyword variants.