Calibration, Entropy Rates, and Memory in Language Models
Mark Braverman, Xinyi Chen, Sham Kakade, Karthik Narasimhan, Cyril Zhang, Yi Zhang
Abstract
Building accurate language models that capture meaningful long-term dependencies is a core challenge in natural language processing. Towards this end, we present a calibration-based approach to measure long-term discrepancies between a generative sequence model and the true distribution, and use these discrepancies to improve the model. Empirically, we show that state-of-the-art language models, including LSTMs and Transformers, are miscalibrated: the entropy rates of their generations drift dramatically upward over time. We then provide provable methods to mitigate this phenomenon. Furthermore, we show how this calibration-based approach can also be used to measure the amount of memory that language models use for prediction.
BibTeX
@InProceedings{pmlr-v119-braverman20a,
title = {Calibration, Entropy Rates, and Memory in Language Models},
author = {Braverman, Mark and Chen, Xinyi and Kakade, Sham and Narasimhan, Karthik and Zhang, Cyril and Zhang, Yi},
booktitle = {Proceedings of the 37th International Conference on Machine Learning},
pages = {1089--1099},
year = {2020},
editor = {III, Hal Daumé and Singh, Aarti},
volume = {119},
series = {Proceedings of Machine Learning Research},
month = {13--18 Jul},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v119/braverman20a/braverman20a.pdf},
url = {https://proceedings.mlr.press/v119/braverman20a.html},
abstract = {Building accurate language models that capture meaningful long-term dependencies is a core challenge in natural language processing. Towards this end, we present a calibration-based approach to measure long-term discrepancies between a generative sequence model and the true distribution, and use these discrepancies to improve the model. Empirically, we show that state-of-the-art language models, including LSTMs and Transformers, are miscalibrated: the entropy rates of their generations drift dramatically upward over time. We then provide provable methods to mitigate this phenomenon. Furthermore, we show how this calibration-based approach can also be used to measure the amount of memory that language models use for prediction.}
}