Sentence Simplification with Deep Reinforcement Learning

By Xingxing Zhang, Mirella Lapata. Conference on Empirical Methods in Natural Language Processing, 2017

438 citations recorded by the source corpus.

Sentence simplification aims to make sentences easier to read and understand. Most recent approaches draw on insights from machine translation to learn simplification rewrites from monolingual corpora of complex and simple sentences. We address the simplification problem with an encoder-decoder model coupled with a deep reinforcement learning framework. Our model, which we call DRESS (as shorthand for Deep REinforcement Sentence Simplification), explores the space of possible simplifications while learning to optimize a reward function that encourages outputs which are simple, fluent, and preserve the meaning of the input. Experiments on three datasets demonstrate that our model outperforms competitive simplification systems.

Subfields: Artificial Intelligence

View Paper: https://www.semanticscholar.org/paper/31f5864ada5fb08b69da74b6d5ad99e385dcc737

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Seq2Edits: Sequence Transduction Using Span-level Edit Operations (2020, methodology): “Our text simplification training set (the WikiLarge corpus (Zhang and Lapata, 2017)) consists of 296K examples, the smallest amongst all our training cor-”
An Overview of Natural Language State Representation for Reinforcement Learning (2020, background): “Both the hidden state of the encoder and the decoder comprised the state in sentence simplification (Zhang & Lapata, 2017).”
Machine Comprehension by Text-to-Text Neural Question Generation (2017, methodology): “Finally, we relate the recent body of works that apply reinforcement learning to natural language generation, such as Li et al. (2016); Ranzato et al. (2016); Kandasamy and Bachrach (2017); Zhang and Lapata (2017).”
BLEU is Not Suitable for the Evaluation of Text Simplification (2018, background): “for TS evaluation (e.g., Xu et al., 2016; Nisioi et al., 2017; Zhang and Lapata, 2017; Ma and Sun, 2017), only few works tested its correlation with human judgments.”
Split and Rephrase (2017, methodology): “In sentence simplification approaches, rephrasing is performed either by a machine translation (Coster and Kauchak, 2011; Wubben et al., 2012; Narayan and Gardent, 2014; Xu et al., 2016; Zhang and Lapata, 2017) or by a probabilistic model (Zhu et al.”
Bootstrapping Generators from Noisy Data (2018, methodology): “, 2016) or to inject task-specific knowledge (Zhang and Lapata, 2017).”
ASSET: A Dataset for Tuning and Evaluation of Sentence Simplification Models with Multiple Rewriting Transformations (2020, background): “ing SS models, several system outputs on this data are already publicly available (Zhang and Lapata, 2017; Zhao et al., 2018; Martin et al., 2020).”
Improving Automatic Source Code Summarization via Deep Reinforcement Learning (2018, background): “Text generation can also be formulated as a decision-making problem and there have been several reinforcement learning-based works on this specific tasks, including image captioning [38], dialogue generation [20] and sentence simplification [52].”
Control Prefixes for Parameter-Efficient Text Generation (2021, methodology): “For these experiments, we integrate the method with a sequence-to-sequence model trained for text simplification on the WikiLarge (Zhang and Lapata, 2017) dataset.”
Ranking Sentences for Extractive Summarization with Reinforcement Learning (2018, background): “The space of sentential representations is vast and fairly unconstrained compared to other prediction tasks operating with fixed vocabularies (Li et al., 2016; Paulus et al., 2017; Zhang and Lapata, 2017).”
RL-LIM: Reinforcement Learning-based Locally Interpretable Modeling (2019, neutral): “Instead, motivated by its many successful applications (Ranzato et al., 2015; Zaremba & Sutskever, 2015; Zhang & Lapata, 2017), we use REINFORCE algorithm (Williams, 1992) such that the selection action is rewarded by the performance of its impact.”
Evaluating Factuality in Text Simplification (2022, methodology): “Many existing text simplification systems attempt to address the problem of meaning preservation by using a semantic similarity score either directly in their loss/reward function or in a candidate ranking step (Zhang and Lapata, 2017; Kriz et al., 2019; Zhao et al., 2020; Maddela et al., 2021).”
FELIX: Flexible Text Editing Through Tagging and Insertion (2020, methodology): “Training is performed on WikiLarge, (Zhang and Lapata, 2017a) a large simpliﬁca-tion corpus which consists of a mixture of of three Wikipedia simpliﬁcation datasets collected by (Kauchak, 2013; Woodsend and Lapata, 2011; Zhu et al., 2010).”
Neural CRF Model for Sentence Alignment in Text Simplification (2020, positive): “, 2015) and WIKILARGE (Zhang and Lapata, 2017), were created by automatically aligning sentences between compa-”