Neural Data-to-Text Generation: An Encoder-Decoder Structure with Multi-Candidate-based Context Module

The data-to-text generation task mainly uses the encoder-decoder architecture, in which the context module provides the information that the decoder wants to observe at the moment. However, there are multiple entities and elements in a single sentence. We conjecture that the architecture has room for improvement to be more suitable for the data-to-text generation task. This paper proposes the Multi-Candidate-based Context Module, using the concept of multiple candidates to simultaneously observe multiple entities and their records. The experiment confirms the effectiveness of our multi-candidate concept and the improvement over the state-of-the-art on the recently released Rotowire dataset.