Definition
SNLI is a benchmark dataset containing over 500,000 labeled sentence pairs annotated with three classes: entailment, contradiction, and neutral. It was created to advance research in natural language inference (NLI), which involves determining whether a hypothesis is true given a premise. SNLI has become a standard evaluation metric for models’ ability to understand logical relationships between sentences, influencing the development of transformer-based architectures like BERT.
Summary
Stanford Natural Language Inference Corpus, a large dataset of English sentences paired with human-written textual entailment labels.
Key Concepts
- Natural Language Inference
- Textual Entailment
- Sentence Pairs
- Benchmark
Use Cases
- Evaluating semantic similarity models
- Training NLI classifiers
- Researching logical reasoning in NLP