3 ms·Decomposable neural attention for natural language inference in spaCy and Keras3 points by syllogism 10y ago