Legal Query Expansion

Maps colloquial or layperson terms to formal legal terms of art to bridge the semantic gap. A query for 'car accident' is expanded to include 'motor vehicle collision,' 'automobile tort,' and 'vehicular negligence.' This process relies on curated legal thesauri and domain-specific word embeddings trained on judicial corpora to ensure that the retrieval system surfaces documents using the precise statutory language, even when the user does not know it.

Automatically appends the formal legal citation strings of known key precedents and statutes to the query. For example, a query about 'employment discrimination standards' is expanded to include '42 U.S.C. § 2000e-2' and 'McDonnell Douglas Corp. v. Green, 411 U.S. 792 (1973).' This technique leverages a citation knowledge graph to identify the most authoritative and frequently cited documents for a given legal concept, directly boosting the ranking of binding primary authority.

Expands a query by traversing a legal ontology to include related doctrinal concepts and their constituent elements. A search for 'self-defense' is expanded to include 'duty to retreat,' 'stand your ground,' 'reasonable belief of imminent harm,' and 'proportionality of force.' This ensures the retrieval system does not just find documents mentioning the keyword 'self-defense' but also those discussing its specific legal tests and sub-doctrines, enabling comprehensive multi-hop reasoning.

Uses dense vector embeddings from a domain-adapted legal model to find semantically related terms not present in a static thesaurus. The query embedding is used to probe the vector space for nearest-neighbor terms that co-occur in similar legal contexts. This captures nuanced relationships, such as expanding 'piercing the corporate veil' to include 'alter ego doctrine' and 'instrumentality rule,' which are functionally identical but lexically distinct doctrines.

Transforms a natural language query into a structured lexical search string with explicit Boolean connectors and proximity operators. The query 'liability for slip and fall on icy sidewalk' is rewritten as '(slip OR trip) /5 (fall) AND (ice OR snow OR wintry) /10 (walkway OR sidewalk OR pavement) AND (premises liability).' This hybrid approach combines the precision of classic legal research syntax with the ease of natural language input, optimizing for both dense and sparse retrieval indexes.

Resolves polysemy in legal language by analyzing the query context to select the correct expansion path. The term 'consideration' is ambiguous; it could refer to contract law (bargained-for exchange) or a judicial opinion (analysis of an issue). The system uses a lightweight classifier to detect the surrounding context—such as the presence of terms like 'offer' or 'acceptance'—and expands the query with the correct set of synonyms, preventing the retrieval of irrelevant documents from the wrong practice area.