@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",
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.",
}