ERAS: Evaluating the Robustness of Chinese NLP Models to Morphological Garden Path Errors

TL;DR

This paper introduces ERAS, a benchmark for assessing Chinese NLP models' robustness to morphological garden path errors caused by segmentation ambiguities, revealing vulnerabilities in current models' handling of morphosyntactic context.

Contribution

The paper proposes ERAS, a novel benchmark for evaluating Chinese NLP models' sensitivity to segmentation ambiguities and demonstrates existing models' susceptibility to these errors.

Findings

Word segmentation models make garden path errors on ambiguous sentences.

Sentiment analysis models with character-level tokenization also exhibit garden path errors.

Models often fail to incorporate morphosyntactic context in Chinese text segmentation.

Abstract

In languages without orthographic word boundaries, NLP models perform word segmentation, either as an explicit preprocessing step or as an implicit step in an end-to-end computation. This paper shows that Chinese NLP models are vulnerable to morphological garden path errors: errors caused by a failure to resolve local word segmentation ambiguities using sentence-level morphosyntactic context. We propose a benchmark, ERAS, that tests a model's vulnerability to morphological garden path errors by comparing its behavior on sentences with and without local segmentation ambiguities. Using ERAS, we show that word segmentation models make garden path errors on locally ambiguous sentences, but do not make equivalent errors on unambiguous sentences. We further show that sentiment analysis models with character-level tokenization make implicit garden path errors, even without an explicit word segmentation step in the pipeline. Our results indicate that models' segmentation of Chinese text often fails to account for morphosyntactic context.