Entailment Relation Aware Paraphrase Generation
This addresses the need for controlled paraphrase generation in NLP applications, offering incremental improvements over existing methods for tasks like data augmentation in textual entailment.
The paper tackles the problem of generating paraphrases that adhere to a specified entailment relation (e.g., equivalent, forward entailing) with respect to an input, proposing ERAP, a reinforcement learning-based weakly-supervised system. Results show ERAP produces high-quality, relation-conforming paraphrases and improves downstream textual entailment task performance by 1-2% over baselines while reducing training artifacts.
We introduce a new task of entailment relation aware paraphrase generation which aims at generating a paraphrase conforming to a given entailment relation (e.g. equivalent, forward entailing, or reverse entailing) with respect to a given input. We propose a reinforcement learning-based weakly-supervised paraphrasing system, ERAP, that can be trained using existing paraphrase and natural language inference (NLI) corpora without an explicit task-specific corpus. A combination of automated and human evaluations show that ERAP generates paraphrases conforming to the specified entailment relation and are of good quality as compared to the baselines and uncontrolled paraphrasing systems. Using ERAP for augmenting training data for downstream textual entailment task improves performance over an uncontrolled paraphrasing system, and introduces fewer training artifacts, indicating the benefit of explicit control during paraphrasing.