Training LLMs to Recognize Hedges in Spontaneous Narratives

TL;DR

This paper explores training large language models to identify hedges in spontaneous narratives, comparing fine-tuned BERT and prompting methods, and improving annotation quality through LLM-in-the-Loop analysis.

Contribution

It introduces a new annotated corpus of hedges in spontaneous speech and evaluates multiple LLM-based approaches for hedge detection, advancing conversational signal understanding.

Findings

Fine-tuned BERT outperformed prompting methods.

Few-shot GPT-4o achieved competitive results.

LLM-in-the-Loop improved annotation quality and revealed ambiguous hedges.

Abstract

Hedges allow speakers to mark utterances as provisional, whether to signal non-prototypicality or "fuzziness", to indicate a lack of commitment to an utterance, to attribute responsibility for a statement to someone else, to invite input from a partner, or to soften critical feedback in the service of face-management needs. Here we focus on hedges in an experimentally parameterized corpus of 63 Roadrunner cartoon narratives spontaneously produced from memory by 21 speakers for co-present addressees, transcribed to text (Galati and Brennan, 2010). We created a gold standard of hedges annotated by human coders (the Roadrunner-Hedge corpus) and compared three LLM-based approaches for hedge detection: fine-tuning BERT, and zero and few-shot prompting with GPT-4o and LLaMA-3. The best-performing approach was a fine-tuned BERT model, followed by few-shot GPT-4o. After an error analysis on the top performing approaches, we used an LLM-in-the-Loop approach to improve the gold standard coding, as well as to highlight cases in which hedges are ambiguous in linguistically interesting ways that will guide future research. This is the first step in our research program to train LLMs to interpret and generate collateral signals appropriately and meaningfully in conversation.