A Survey of Knowledge-enhanced Text Generation

The goal of text-to-text generation is to make machines express like a human in many applications such as conversation, summarization, and translation. It one the most important yet challenging tasks natural language processing (NLP). Various neural encoder-decoder models have been proposed achieve by learning map input text output text. However, alone often provides limited knowledge generate desired output, so performance still far from satisfaction real-world scenarios. To address this issue, researchers considered incorporating (i) internal embedded (ii) external outside sources base graph into system. This research topic known knowledge-enhanced . In survey, we present comprehensive review on over past five years. main content includes two parts: general methods architectures for integrating generation; specific techniques according different forms data. survey can broad audiences, practitioners, academia industry.