Data-to-text Generation with Variational Sequential Planning
We consider the task of data-to-text generation, which aims to create textual output from non-linguistic input. We focus on generating long-form text, i.e., documents with multiple paragraphs, and propose a neural model enhanced with a planning component responsible for organizing high-level information in a coherent and meaningful way. We infer latent plans sequentially with a structured variational model, while interleaving the steps of planning and generation. Text is generated by conditioning on previous variational decisions and previously generated text. Experiments on two data-to-text benchmarks (RotoWire and MLB) show that our model outperforms strong baselines and is sample efficient in the face of limited training data (e.g., a few hundred instances).
PDF Abstract
Datasets
RotoWire
Results from the Paper
| Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Benchmark |
|---|---|---|---|---|---|---|
| Data-to-Text Generation | MLB Dataset | SeqPlan | BLEU | 14.29 | # 1 | |
| Data-to-Text Generation | MLB Dataset (Content Ordering) | SeqPlan | DLD | 22.7 | # 1 | |
| Data-to-Text Generation | MLB Dataset (Content Selection) | SeqPlan | Precision | 43.3 | # 2 | |
| Recall | 53.5 | # 2 | ||||
| Data-to-Text Generation | MLB Dataset (Relation Generation) | SeqPlan | Precision | 95.9 | # 1 | |
| count | 28.9 | # 2 | ||||
| Data-to-Text Generation | RotoWire (Relation Generation) | SeqPlan | count | 46.7 | # 1 | |
| Precision | 97.6 | # 1 |
