Discrete Word Embedding for Logical Natural Language Understanding