keywords: NLP CJKmainfont: KaiTi –-

# Paraphrase Generation

## Paper List

• Paraphrase Generation with Deep Reinforcement Learning
• Neural paraphrase generation recently draws attention in different application scenarios. The task is often formalized as a sequence-to-sequence (Seq2Seq) learning problem. Prakash et al. (2016) employ a stacked residual LSTM network in the Seq2Seq model to enlarge the model capacity. Cao et al. (2017) utilize an additional vocabulary to restrict word candidates during generation. Gupta et al. (2018) use a variational auto-encoder framework to generate more diverse paraphrases. Ma et al. (2018) utilize an attention layer instead of a linear mapping in the decoder to pick up word candidates. Iyyer et al. (2018) harness syntactic information for controllable paraphrase generation. Zhang and Lapata (2017) tackle a similar task of sentence simplification withe Seq2Seq model coupled with deep reinforcement learning, in which the reward function is manually defined for the task. Similar to these works, we also pretrain the paraphrase generator within the Seq2Seq framework. The main difference lies in that we use another trainable neural network, referred to as evaluator, to guide the training of the generator through reinforcement learning.
- There is also work on paraphrasing generation in different settings. For example, Mallinson et al. (2017) leverage bilingual data to produce paraphrases by pivoting over a shared translation in another language. Wieting et al. (2017); Wieting and Gimpel (2018) use neural machine translation to generate paraphrases via back-translation of bilingual sentence pairs. Buck et al. (2018) and Dong et al. (2017) tackle the problem of QA-specific paraphrasing with the guidance from an external QA system and an associated evaluation metric.
• Semantic Parsing via Paraphrasing
• Canonical utterance construction Given an utterance x and the KB, we construct a set of candidate logical forms Zx, and then for each z 2 Zx generate a small set of canonical natural language utterances Cz. Our goal at this point is only to generate a manageable set of logical forms containing the correct one, and then generate an appropriate canonical utterance from it. This strategy is feasible in factoid QA where compositionality is low, and so the size of Zx is limited (Section 4)
• Paraphrasing We score the canonical utterances in Cz with respect to the input utterance x using a paraphrase model, which offers two advantages. First, the paraphrase model is decoupled from the KB, so we can train it from large text corpora. Second, natural language utterances often do not express predicates explicitly, e.g., the question “What is Italy’s money?” expresses the binary predicate CurrencyOf with a possessive construction. Paraphrasing methods are well-suited for handling such text-to-text gaps.

text simplification, SAMSA, and the first evaluation experiments that directly target the structural simplification component, separately from the lexical component.