A Measure-theoretic Characterization of Tight Language Model

Abstract

Language modeling, a central task in natural language processing, involves estimating a probability distribution over strings. In most cases, the estimated distribution sums to 1 over all finite strings. However, in some pathological cases, probability mass can “leak” onto the set of infinite sequences. In order to characterize the notion of leakage more precisely, this paper offers a measure-theoretic treatment of language modeling. We prove that many popular language model families are in fact tight, meaning that they will not leak in this sense. We also generalize characterizations of tightness proposed in previous works.

Publication
Proceedings of the 61th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)

Add the full text or supplementary notes for the publication here using Markdown formatting.