We convert the inference task to one of simple factoid question answering, allowing us to easily scale up to 16,000 high-quality examples.
The code used to extract assertions and create the dataset is available here.
We used the following embeddings as baselines.
- Dependency-based word embeddings using Google Books [download]
- Relation embeddings using the extracted assertions [download]
The embeddings are saved in NumPy format, and can be used with hyperwords.