An Invariant Learning Characterization of Controlled Text Generation

TL;DR

This paper addresses the challenge of distribution shift in controlled text generation by framing it as an invariant learning problem, proposing methods to improve predictor robustness across diverse text environments.

Contribution

It introduces an invariant learning framework for controlled text generation under distribution shift and proposes heuristics for environment selection.

Findings

Distribution shift significantly impacts controlled generation quality.

Invariant learning methods improve predictor stability across environments.

Empirical results show potential of invariance approaches in real and synthetic data.

Abstract

Controlled generation refers to the problem of creating text that contains stylistic or semantic attributes of interest. Many approaches reduce this problem to training a predictor of the desired attribute. For example, researchers hoping to deploy a large language model to produce non-toxic content may use a toxicity classifier to filter generated text. In practice, the generated text to classify, which is determined by user prompts, may come from a wide range of distributions. In this paper, we show that the performance of controlled generation may be poor if the distributions of text in response to user prompts differ from the distribution the predictor was trained on. To address this problem, we cast controlled generation under distribution shift as an invariant learning problem: the most effective predictor should be invariant across multiple text environments. We then discuss a natural solution that arises from this characterization and propose heuristics for selecting natural environments. We study this characterization and the proposed method empirically using both synthetic and real data. Experiments demonstrate both the challenge of distribution shift in controlled generation and the potential of invariance methods in this setting.