Abstract
Frame Semantics offers a compelling framework for representing meaning as structured conceptual scenarios (frames). FrameNet operationalizes this theory as a lexical resource linking words to frames and their conceptual roles (frame elements, FEs). However, applying Frame Semantics to arbitrary text has historically required either laborious manual annotation or complex, task-specific classification pipelines that presuppose known targets and/or frames. Here we present an open-source method that leverages Generative AI (GenAI) to automatically parse semantic frames and their FEs from any input text, without requiring pre-specification of frames or target spans, or task-specific model training. The tool combines a retrieval step, in which candidate frames are identified via embedding-based semantic similarity, with a selection step, in which a Large Language Model (LLM) is prompted, using zero- or few-shot examples, to determine which frames are evoked, which spans trigger them, and which spans fill their FEs, returning a structured, machine-readable output. We detail installation, execution, and output interpretation, and provide a built-in evaluation module supporting exact, semantic, and graph-based matching against FrameNet's human-annotated data. This freely available tool lowers the barrier to frame-semantic analysis for researchers across linguistics, who wish to explore structured semantic content in naturalistic text, including in relation to behavioral and neural measures of language comprehension.