A Mutual Information Maximization Perspective of Language Representation Learning

TL;DR

This paper presents a unified mutual information maximization framework for language representation learning, connecting classical and modern models, and introduces a new self-supervised task inspired by vision methods.

Contribution

It unifies various language embedding models under a mutual information perspective and proposes a novel self-supervised objective for improved sentence representation learning.

Findings

The framework provides theoretical insights into existing models.

A new self-supervised task based on mutual information is introduced.

The approach facilitates cross-domain transfer of representation learning techniques.

Abstract

We show state-of-the-art word representation learning methods maximize an objective function that is a lower bound on the mutual information between different parts of a word sequence (i.e., a sentence). Our formulation provides an alternative perspective that unifies classical word embedding models (e.g., Skip-gram) and modern contextual embeddings (e.g., BERT, XLNet). In addition to enhancing our theoretical understanding of these methods, our derivation leads to a principled framework that can be used to construct new self-supervised tasks. We provide an example by drawing inspirations from related methods based on mutual information maximization that have been successful in computer vision, and introduce a simple self-supervised objective that maximizes the mutual information between a global sentence representation and n-grams in the sentence. Our analysis offers a holistic view of representation learning methods to transfer knowledge and translate progress across multiple domains (e.g., natural language processing, computer vision, audio processing).