# Learning Outside the Box: Discourse-level Features Improve Metaphor Identification

## TL;DR

This paper demonstrates that incorporating discourse-level features significantly enhances metaphor identification accuracy, achieving near state-of-the-art results without complex models by leveraging broader contextual information.

## Contribution

It introduces the use of discourse-level features and document embeddings for metaphor detection, moving beyond sentence-level analysis.

## Key findings

- Broader discourse features improve metaphor identification.
- Gradient boosting classifiers with document embeddings achieve near state-of-the-art results.
- Qualitative analysis confirms the importance of wider context in metaphor processing.

## Abstract

Most current approaches to metaphor identification use restricted linguistic contexts, e.g. by considering only a verb's arguments or the sentence containing a phrase. Inspired by pragmatic accounts of metaphor, we argue that broader discourse features are crucial for better metaphor identification. We train simple gradient boosting classifiers on representations of an utterance and its surrounding discourse learned with a variety of document embedding methods, obtaining near state-of-the-art results on the 2018 VU Amsterdam metaphor identification task without the complex metaphor-specific features or deep neural architectures employed by other systems. A qualitative analysis further confirms the need for broader context in metaphor processing.