ACL 2023long12 citations

MixCE: Training Autoregressive Language Models by Mixing Forward and Reverse Cross-Entropies

Shiyue Zhang, Shijie Wu, Ozan Irsoy, Steven Lu, Mohit Bansal, Mark Dredze, David Rosenberg

Abstract

Autoregressive language models are trained by minimizing the cross-entropy of the model distribution Q relative to the data distribution P – that is, minimizing the forward cross-entropy, which is equivalent to maximum likelihood estimation (MLE). We have observed that models trained in this way may “over-generalize”, in the sense that they produce non-human-like text. Moreover, we believe that reverse cross-entropy, i.e., the cross-entropy of P relative to Q, is a better reflection of how a human would evaluate text generated by a model. Hence, we propose learning with MixCE, an objective that mixes the forward and reverse cross-entropies. We evaluate models trained with this objective on synthetic data settings (where P is known) and real data, and show that the resulting models yield better generated text without complex decoding strategies.

BibTeX
@inproceedings{zhang-etal-2023-mixce,
    title = "{M}ix{CE}: Training Autoregressive Language Models by Mixing Forward and Reverse Cross-Entropies",
    author = "Zhang, Shiyue  and
      Wu, Shijie  and
      Irsoy, Ozan  and
      Lu, Steven  and
      Bansal, Mohit  and
      Dredze, Mark  and
      Rosenberg, David",
    editor = "Rogers, Anna  and
      Boyd-Graber, Jordan  and
      Okazaki, Naoaki",
    booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2023",
    address = "Toronto, Canada",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2023.acl-long.502/",
    doi = "10.18653/v1/2023.acl-long.502",
    pages = "9027--9050"
}
MixCE: Training Autoregressive Language Models by Mixing Forward and Reverse Cross-Entropies · ACL 2023