We propose a simple neural architecture for natural language inference. Our approach uses attention to decompose the problem into subproblems that can be solved separately, thus making it trivially parallelizable. On the Stanford Natural Language Inference (SNLI) dataset, we obtain state-of-the-art results with almost an order of magnitude fewer parameters than previous work and without relying on any word-order information. Adding intra-sentence attention that takes a minimum amount of order into account yields further improvements.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Natural Language Inference | SNLI | % Test Accuracy | 86.8 | 200D decomposable attention feed-forward model with intra-sentence attention |
| Natural Language Inference | SNLI | % Train Accuracy | 90.5 | 200D decomposable attention feed-forward model with intra-sentence attention |
| Natural Language Inference | SNLI | % Test Accuracy | 86.8 | 200D decomposable attention model with intra-sentence attention |
| Natural Language Inference | SNLI | % Train Accuracy | 90.5 | 200D decomposable attention model with intra-sentence attention |
| Natural Language Inference | SNLI | % Test Accuracy | 86.3 | 200D decomposable attention feed-forward model |
| Natural Language Inference | SNLI | % Train Accuracy | 89.5 | 200D decomposable attention feed-forward model |
| Natural Language Inference | SNLI | % Test Accuracy | 86.3 | 200D decomposable attention model |
| Natural Language Inference | SNLI | % Train Accuracy | 89.5 | 200D decomposable attention model |