An NLP task that identifies the semantic roles of words in a sentence — who did what to whom.
Semantic Role Labeling (SRL) is a natural language processing task that identifies the predicate-argument structure of sentences — determining who did what to whom, when, where, and how. For each predicate (typically a verb) in a sentence, SRL identifies the semantic roles of the associated arguments: the Agent (who performed the action), the Patient (what was affected), the Instrument (how it was done), the Location, the Time, and other roles defined by semantic frameworks like PropBank or FrameNet.
Semantic Role Labeling is a critical component of enterprise information extraction pipelines. Before events, actions, and relationships can be stored in knowledge graphs or used to answer questions about who did what, SRL must identify the semantic structure of source documents. In 2026, SRL is used in legal document analysis, financial event extraction, clinical trial processing, and supply chain event monitoring.
Modern SRL uses transformer-based models fine-tuned on annotated corpora like PropBank (which annotates verb arguments) or FrameNet (which annotates semantic frames — structured representations of event types). The model identifies predicate spans, then classifies the role of each argument relative to each predicate. State-of-the-art SRL models achieve over 87% F1 on standard benchmarks.
Processing the sentence 'Apple acquired Beats Electronics for $3 billion in 2014': SRL identifies 'acquired' as the predicate, 'Apple' as the Agent (acquirer), 'Beats Electronics' as the Patient (acquired entity), '$3 billion' as the Price, and '2014' as the Time. This structured representation is stored in a knowledge graph, enabling queries like 'What companies has Apple acquired for more than $1 billion?'