SNLI

Stanford Natural Language Inference

This notebook provides various strategies for tackling the natural language inference given a premise and a hypothesis. The task is to understand if the relationship between the two sentences is entailment, contradiction or neutral. It's essentially a multi-class classification problem. I explore several modelling strategies including some simple baselines. The strategies explored are:

1. Simple word matching between premise and hypothesis. Softmax classifier used
2. All word combinations between premise and hypothesis. Softmax classifier used
3. Glove pre-trained averaged embeddings with softmax
4. Glove pre-trained averaged embeddings with neural network
5. Sentence encoding RNNs with Fasttext embeddings
6. Chained model RNN with Random embeddings
7. Chained model RNN with Fasttext embeddings