The property of a search query being grammatically natural and unambiguous — a signal used to distinguish full-sentence questions from fragmentary keyword strings, with implications for query expansion and intent classification.
A well-formed query reads like a natural sentence or question — it has clear grammatical structure, states its information need explicitly, and would make sense spoken aloud. A poorly-formed query is a keyword fragment: short, telegraphic, and relying on the search engine to infer intent from context. "What are the best noise-cancelling headphones for travel?" is well-formed. "best headphones travel" is not.
Well-formedness is a meaningful signal for query classification. Well-formed queries tend to have clearer intent, narrower scope, and stronger signals for informational or conversational content. Poorly-formed queries are more ambiguous and often transactional or navigational. Knowing which type a query belongs to lets a system route it to the right retrieval strategy, expansion logic, or answer format.
In AI search, the distinction is amplified: query fan-out systems and grounding pipelines behave differently depending on query structure. A well-formed question generates more targeted sub-queries and retrieves more relevant grounding passages than an equivalent keyword fragment.
At DEJAN we built a well-formedness classifier on top of the ALBERT architecture, trained on Google's publicly released query well-formedness dataset supplemented with client data from a cycling apparel manufacturer. The model achieves 80% accuracy — a 10% improvement over Google's original LSTM-based classifier — and has been deployed in our query processing pipeline to flag ambiguous queries as candidates for expansion before they are classified for intent. The model and its weights are available for download from the DEJAN research pages.
