Comparing Human and Large Language Model Interpretation of Implicit Information

TL;DR

This paper introduces the task of Implicit Information Extraction (IIE) to evaluate how well Large Language Models interpret implicit meanings compared to humans, revealing coverage limitations and contextual differences.

Contribution

It proposes an LLM-based IIE pipeline for extracting structured knowledge and evaluates its performance against human judgments across datasets.

Findings

Humans agree with most model triplets but add many, indicating limited LLM coverage.

Models are more conservative about implicit inferences in socially rich contexts.

Humans become more conservative in shorter, fact-oriented contexts.

Abstract

The interpretation of implicit meanings is an integral aspect of human communication. However, this framework may not transfer to interactions with Large Language Models (LLMs). To investigate this, we introduce the task of Implicit Information Extraction (IIE) and propose an LLM-based IIE pipeline that builds a structured knowledge graph from a context sentence by extracting relational triplets, validating implicit inferences, and analyzing temporal relations. We evaluate two LLMs against crowdsourced human judgments on two datasets. We find that humans agree with most model triplets yet consistently propose many additions, indicating limited coverage in current LLM-based IIE. Moreover, in our experiments, models appear to be more conservative about implicit inferences than humans in socially rich contexts, whereas humans become more conservative in shorter, fact-oriented contexts. Our code is available at https://github.com/Antonio-Dee/IIE_from_LLM.