Scene Abstraction for Lexical Semantics: Structured Representations of Situated Meaning

TL;DR

This paper introduces Scene Abstraction, a framework for creating structured representations of situated word meanings, supported by a new dataset and experiments showing alignment with human interpretation.

Contribution

The paper presents a novel structured representation framework, a new dataset COCA-Scenes, and empirical evidence demonstrating the effectiveness of scene profiles in capturing human-like word interpretation.

Findings

Scenes are reliably identifiable across human observers with 82.4% accuracy.

Scene profiles align more closely with human interpretation than ATOMIC-based alternatives.

The framework improves understanding of contextual lexical meaning.

Abstract

Coffee and tea share many properties, yet they evoke strikingly different situations, atmospheres, and affective associations. These situated dimensions of word meaning are real and systematic, but they remain implicit in most computational representations of lexical meaning. We propose Scene Abstraction, a framework for constructing structured representations of the interpretive scenes that words participate in across usage contexts. Each scene consists of a Contextual Scene (Events, Entities, Setting) and an expression-centered Expression Profile (Engaged events, Generalizable properties, Evoked emotions), operationalized through few-shot prompting of a large language model. Our contributions are three-fold: (1) a structured representation framework for situated lexical meaning; (2) COCA-Scenes, a dataset of 520 usage instances across 26 keywords for distinct scene identification; and (3) empirical evidence from two experiments suggesting that scenes are reliably identifiable across human observers (82.4% accuracy, +11.8 pp over text-only embeddings) and that our scene profiles more closely align with human interpretation of words in context than ATOMIC-based alternatives (86.4% preference across three semantic dimensions).