@inproceedings{li-hao-2025-eras,
    title = "{ERAS}: Evaluating the Robustness of {C}hinese {NLP} Models to Morphological Garden Path Errors",
    author = "Li, Qinchan  and
      Hao, Sophie",
    editor = "Chiruzzo, Luis  and
      Ritter, Alan  and
      Wang, Lu",
    booktitle = "Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)",
    month = apr,
    year = "2025",
    address = "Albuquerque, New Mexico",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/fix-sig-urls/2025.naacl-long.159/",
    pages = "3100--3111",
    ISBN = "979-8-89176-189-6",
    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 morphological 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."
}