''You should probably read this'': Hedge Detection in Text

TL;DR

This study tackles hedge detection by modeling uncertainty cues in text using a joint word-and-POS representation with RNN-based architectures and attention. It demonstrates that POS information provides a predictive signal and that joint input and latent-space models can surpass previous baselines on the CoNLL-2010 Wikipedia dataset, achieving up to 70.24 F1 with domain-specific embeddings. The findings underscore the practical benefit of POS-aware architectures and domain-adapted embeddings for robust hedge detection, with potential applicability to other datasets and evolving NLP models. Overall, the work advances hedge detection by showing how POS-tag signals can be effectively integrated with word embeddings to improve uncertainty detection in short, high-stakes texts.

Abstract

Humans express ideas, beliefs, and statements through language. The manner of expression can carry information indicating the author's degree of confidence in their statement. Understanding the certainty level of a claim is crucial in areas such as medicine, finance, engineering, and many others where errors can lead to disastrous results. In this work, we apply a joint model that leverages words and part-of-speech tags to improve hedge detection in text and achieve a new top score on the CoNLL-2010 Wikipedia corpus.

Figures (2)

• Figure 1: Word & POS Input Joint Model
• Figure 2: Latent Space Joint Model