Abstract
Recent advances in deep neural language models combined with the capacity oflarge scale datasets have accelerated the development of natural languagegeneration systems that produce fluent and coherent texts (to various degreesof success) in a multitude of tasks and application contexts. However,controlling the output of these models for desired user and task needs is stillan open challenge. This is crucial not only to customizing the content andstyle of the generated language, but also to their safe and reliable deploymentin the real world. We present an extensive survey on the emerging topic ofconstrained neural language generation in which we formally define andcategorize the problems of natural language generation by distinguishingbetween conditions and constraints (the latter being testable conditions on theoutput text instead of the input), present constrained text generation tasks,and review existing methods and evaluation metrics for constrained textgeneration. Our aim is to highlight recent progress and trends in this emergingfield, informing on the most promising directions and limitations towardsadvancing the state-of-the-art of constrained neural language generationresearch.