Calibration, Entropy Rates, and Memory in Language Models
Computation and Language
2019-06-14 v1 Machine Learning
Machine Learning
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 \emph{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.
Cite
@article{arxiv.1906.05664,
title = {Calibration, Entropy Rates, and Memory in Language Models},
author = {Mark Braverman and Xinyi Chen and Sham M. Kakade and Karthik Narasimhan and Cyril Zhang and Yi Zhang},
journal= {arXiv preprint arXiv:1906.05664},
year = {2019}
}