Entailment Relation Aware Paraphrase Generation

We introduce a new task of entailment relation aware paraphrase generation which aims at generating conforming to given (e.g. equivalent, forward entailing, or reverse entailing) with respect input. propose reinforcement learning-based weakly-supervised paraphrasing system, ERAP, that can be trained using existing and natural language inference (NLI) corpora without an explicit task-specific corpus. A combination automated human evaluations show ERAP generates paraphrases the specified are good quality as compared baselines uncontrolled systems. Using for augmenting training data downstream textual improves performance over introduces fewer artifacts, indicating benefit control during paraphrasing.