Stolen Probability: A Structural Weakness of Neural Language Models
David
Demeter
author
Gregory
Kimmel
author
Doug
Downey
author
2020-07
text
Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics
Dan
Jurafsky
editor
Joyce
Chai
editor
Natalie
Schluter
editor
Joel
Tetreault
editor
Association for Computational Linguistics
Online
conference publication
Neural Network Language Models (NNLMs) generate probability distributions by applying a softmax function to a distance metric formed by taking the dot product of a prediction vector with all word vectors in a high-dimensional embedding space. The dot-product distance metric forms part of the inductive bias of NNLMs. Although NNLMs optimize well with this inductive bias, we show that this results in a sub-optimal ordering of the embedding space that structurally impoverishes some words at the expense of others when assigning probability. We present numerical, theoretical and empirical analyses which show that words on the interior of the convex hull in the embedding space have their probability bounded by the probabilities of the words on the hull.
demeter-etal-2020-stolen
10.18653/v1/2020.acl-main.198
https://aclanthology.org/2020.acl-main.198
2020-07
2191
2197