May the Force Be with Your Copy Mechanism: Enhanced Supervised-Copy Method for Natural Language Generation
Recent neural sequence-to-sequence models with a copy mechanism have achieved remarkable progress in various text generation tasks. These models addressed out-of-vocabulary problems and facilitated the generation of rare words. However, the identification of the word which needs to be copied is difficult, as observed by prior copy models, which suffer from incorrect generation and lacking abstractness. In this paper, we propose a novel supervised approach of a copy network that helps the model decide which words need to be copied and which need to be generated. Specifically, we re-define the objective function, which leverages source sequences and target vocabularies as guidance for copying. The experimental results on data-to-text generation and abstractive summarization tasks verify that our approach enhances the copying quality and improves the degree of abstractness.
PDF Abstract arXiv 2021 PDF arXiv 2021 Abstract
Datasets
CNN/Daily Mail
RotoWire
Results from the Paper
| Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Benchmark |
|---|---|---|---|---|---|---|
| Data-to-Text Generation | MLB Dataset | Force-Copy | BLEU | 10.5 | # 4 | |
| Data-to-Text Generation | MLB Dataset (Content Ordering) | Force-Copy | DLD | 21.16 | # 3 | |
| Data-to-Text Generation | MLB Dataset (Content Selection) | Force-Copy | Precision | 49.39 | # 1 | |
| Recall | 50.89 | # 3 | ||||
| Data-to-Text Generation | MLB Dataset (Relation Generation) | Force-Copy | Precision | 84.50 | # 3 | |
| count | 21.05 | # 4 | ||||
| Data-to-Text Generation | RotoWire | Force-Copy | BLEU | 17.26 | # 3 | |
| Data-to-Text Generation | RotoWire (Content Ordering) | Force-Copy | DLD | 17.26% | # 4 | |
| BLEU | 15.8 | # 3 | ||||
| Data-to-Text Generation | Rotowire (Content Selection) | Force-Copy | Precision | 34.34% | # 2 | |
| Recall | 48.85% | # 4 | ||||
| Data-to-Text Generation | RotoWire (Relation Generation) | Force-Copy | count | 27.37 | # 4 | |
| Precision | 95.40% | # 3 |
