Why is constrained neural language generation particularly challenging?

TL;DR

Constrained neural language generation addresses the gap between powerful unconditional NLG models and the need to satisfy user- and task-specific requirements. The paper formally distinguishes conditions on inputs from constraints on outputs and surveys a spectrum of constraint types (lexical, format, semantic, syntactic, utility) and associated tasks (MT, dialogue, summarization, style transfer, QA, narrative/poetry). It categorizes methods into decoding, fine-tuning, discriminative, edit-based, model adaptation, and prompting, highlighting trade-offs between constraint satisfaction, fluency, and efficiency, as well as evaluation challenges. The authors identify open problems—expressivity, non-differentiable constraints, dataset scarcity, and evaluation gaps—and propose directions like RLHF, mechanistic interpretability, and constraint-aware prompting to advance robust, safe, and controllable constrained NLG with real-world impact.

Abstract

Recent advances in deep neural language models combined with the capacity of large scale datasets have accelerated the development of natural language generation systems that produce fluent and coherent texts (to various degrees of success) in a multitude of tasks and application contexts. However, controlling the output of these models for desired user and task needs is still an open challenge. This is crucial not only to customizing the content and style of the generated language, but also to their safe and reliable deployment in the real world. We present an extensive survey on the emerging topic of constrained neural language generation in which we formally define and categorize the problems of natural language generation by distinguishing between conditions and constraints (the latter being testable conditions on the output text instead of the input), present constrained text generation tasks, and review existing methods and evaluation metrics for constrained text generation. Our aim is to highlight recent progress and trends in this emerging field, informing on the most promising directions and limitations towards advancing the state-of-the-art of constrained neural language generation research.