Ankur P. Parikh, Xuezhi Wang, Sebastian Gehrmann, Manaal Faruqui, Bhuwan Dhingra, Diyi Yang, Dipanjan Das
We present ToTTo, an open-domain English table-to-text dataset with over 120,000 training examples that proposes a controlled generation task: given a Wikipedia table and a set of highlighted table cells, produce a one-sentence description. To obtain generated targets that are natural but also faithful to the source table, we introduce a dataset construction process where annotators directly revise existing candidate sentences from Wikipedia. We present systematic analyses of our dataset and annotation process as well as results achieved by several state-of-the-art baselines. While usually fluent, existing methods often hallucinate phrases that are not supported by the table, suggesting that this dataset can serve as a useful research benchmark for high-precision conditional text generation.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Text Generation | ToTTo | BLEU | 44 | BERT-to-BERT |
| Text Generation | ToTTo | PARENT | 52.6 | BERT-to-BERT |
| Text Generation | ToTTo | BLEU | 41.6 | Pointer Generator |
| Text Generation | ToTTo | PARENT | 51.6 | Pointer Generator |
| Text Generation | ToTTo | BLEU | 19.2 | NCP+CC (Puduppully et al 2019) |
| Text Generation | ToTTo | PARENT | 29.2 | NCP+CC (Puduppully et al 2019) |
| Data-to-Text Generation | ToTTo | BLEU | 44 | BERT-to-BERT |
| Data-to-Text Generation | ToTTo | PARENT | 52.6 | BERT-to-BERT |
| Data-to-Text Generation | ToTTo | BLEU | 41.6 | Pointer Generator |
| Data-to-Text Generation | ToTTo | PARENT | 51.6 | Pointer Generator |
| Data-to-Text Generation | ToTTo | BLEU | 19.2 | NCP+CC (Puduppully et al 2019) |
| Data-to-Text Generation | ToTTo | PARENT | 29.2 | NCP+CC (Puduppully et al 2019) |