Source Attribution

The degree to which a generated summary accurately reflects the stated facts of the source document without contradiction or fabrication. In legal contexts, factual consistency is non-negotiable; a single hallucinated precedent or misattributed holding can undermine the entire analysis. Verification typically employs Natural Language Inference (NLI) models that classify each claim as entailed, contradicted, or neutral relative to the source text. Modern legal AI pipelines enforce factual consistency through atomic fact decomposition, breaking summaries into minimal claims for individual verification against the source.

A method for evaluating summary faithfulness by breaking down a generated text into minimal, self-contained factual claims for individual verification against the source. Each atomic fact represents a single, indivisible assertion that can be independently validated or refuted. This granular approach enables precise hallucination rate calculation and provides a structured audit trail for legal professionals. The technique is foundational to source attribution systems, as each atomic fact must be linked to its originating passage.

A task where a model determines if a hypothesis is entailed by, contradicts, or is neutral to a given premise. In legal summarization, NLI serves as the computational backbone for verifying summary faithfulness. A claim extracted from a summary becomes the hypothesis, and the source document serves as the premise. State-of-the-art legal NLI models are fine-tuned on domain-specific datasets to handle the nuanced reasoning patterns found in judicial opinions and contracts.

A metric quantifying the frequency at which a language model generates factually incorrect or unverifiable information not grounded in the source text. In legal AI, hallucination is catastrophic—fabricated case citations, misstated holdings, or invented contractual obligations carry professional liability risks. Source attribution directly combats hallucination by requiring every assertion to be traceable to a specific source location, making unverifiable claims immediately identifiable.

The task of identifying and linking semantically related passages or entities that discuss the same event or fact across a collection of distinct documents. This capability is essential for multi-document fusion and source attribution across case law corpora. Alignment algorithms must resolve coreference across documents—recognizing that 'the defendant' in one opinion refers to the same entity as 'the appellant' in another—to build coherent, cross-referenced summaries.

Automated validation of legal references against a ground-truth authority database. These systems check that cited cases exist, have not been overturned, and actually stand for the proposition attributed to them. When integrated with source attribution pipelines, citation verification ensures that not only is a claim linked to a source, but the source itself is authoritative and correctly interpreted. This dual-layer validation is the gold standard for legal AI integrity.