Eq. 4, we can derive a decomposed form: g (p Eq. 7 indicates that we assign high probability to The statistics of the English-EWT -Dep dataset is as follows: Table 7: Statistics of the English-EWT -Dep dataset. It's worth noting that our reported performance of sentence summarization can be further improved We use a single layer LSTM with single direction. C.4 Data-to-T ext Generation In this part, we describe more details about the experiments of data-to-text generation task. The settings of data-to-text generation is similar with that of fact-based text editing. The entities given in the source inputs may not has the same form in the references.

However, it is usually difficult to manually constrain the generation process of these models.

However, it is usually difficult to manually constrain the generation process of these models.