Learning which features matter: RoBERTa acquires a preference for linguistic generalizations (eventually)

Alex Warstadt, Yian Zhang, Haau Sing Li, Haokun Liu, Samuel R. Bowman

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    Abstract

    One reason pretraining on self-supervised linguistic tasks is effective is that it teaches models features that are helpful for language understanding. However, we want pretrained models to learn not only to represent linguistic features, but also to use those features preferentially during fine-turning. With this goal in mind, we introduce a new English-language diagnostic set called MSGS (the Mixed Signals Generalization Set), which consists of 20 ambiguous binary classification tasks that we use to test whether a pretrained model prefers linguistic or surface generalizations during fine-tuning. We pretrain RoBERTa models from scratch on quantities of data ranging from 1M to 1B words and compare their performance on MSGS to the publicly available RoBERTaBASE. We find that models can learn to represent linguistic features with little pretraining data, but require far more data to learn to prefer linguistic generalizations over surface ones. Eventually, with about 30B words of pretraining data, RoBERTaBASE does demonstrate a linguistic bias with some regularity. We conclude that while self-supervised pretraining is an effective way to learn helpful inductive biases, there is likely room to improve the rate at which models learn which features matter.

    Original languageEnglish (US)
    Title of host publicationEMNLP 2020 - 2020 Conference on Empirical Methods in Natural Language Processing, Proceedings of the Conference
    PublisherAssociation for Computational Linguistics (ACL)
    Pages217-235
    Number of pages19
    ISBN (Electronic)9781952148606
    StatePublished - 2020
    Event2020 Conference on Empirical Methods in Natural Language Processing, EMNLP 2020 - Virtual, Online
    Duration: Nov 16 2020Nov 20 2020

    Publication series

    NameEMNLP 2020 - 2020 Conference on Empirical Methods in Natural Language Processing, Proceedings of the Conference

    Conference

    Conference2020 Conference on Empirical Methods in Natural Language Processing, EMNLP 2020
    CityVirtual, Online
    Period11/16/2011/20/20

    ASJC Scopus subject areas

    • Information Systems
    • Computer Science Applications
    • Computational Theory and Mathematics

    Fingerprint

    Dive into the research topics of 'Learning which features matter: RoBERTa acquires a preference for linguistic generalizations (eventually)'. Together they form a unique fingerprint.

    Cite this