Adversarial Example Generation with Syntactically Controlled Paraphrase Networks

By Mohit Iyyer, John Wieting, Kevin Gimpel, Luke S. Zettlemoyer. 2018

786 citations recorded by the source corpus.

We propose syntactically controlled paraphrase networks (SCPNs) and use them to generate adversarial examples. Given a sentence and a target syntactic form (e.g., a constituency parse), SCPNs are trained to produce a paraphrase of the sentence with the desired syntax. We show it is possible to create training data for this task by first doing backtranslation at a very large scale, and then using a parser to label the syntactic transformations that naturally occur during this process. Such data allows us to train a neural encoder-decoder model with extra inputs to specify the target syntax. A combination of automated and human evaluations show that SCPNs generate paraphrases that follow their target specifications without decreasing paraphrase quality when compared to baseline (uncontrolled) paraphrase systems. Furthermore, they are more capable of generating syntactically adversarial examples that both (1) “fool” pretrained models and (2) improve the robustness of these models to syntactic variation when used to augment their training data.

Subfields: Artificial Intelligence, Natural Language Processing

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

In Natural Language Processing

Citations in the corpus

Tailor: Generating and Perturbing Text with Semantic Controls (2021, methodology): “Recent works incorporating syntactic structures for paraphrasing (Iyyer et al., 2018; Chen et al., 2019; Bao et al., 2019; Kumar et al., 2020; Sun et al., 2021; Huang and Chang, 2021) or discrete semantic signatures for diverse generation (Weir et al., 2020) are similar to T ailor in their…”
A Survey of Data Augmentation Approaches for NLP (2021, background): “Other approaches include syntactic or controlled paraphrasing (Iyyer et al., 2018; Kumar et al., 2020), document-level paraphrasing (Gangal et al., 2021), augmenting misclassiﬁed examples (Dreossi et al., 2018), BERT cross-encoder labeling of new inputs (Thakur et al., 2021), guided generation…”
Polyjuice: Generating Counterfactuals for Explaining, Evaluating, and Improving Models (2021, background): “For example, adversarial generators aim to maintain semantics while changing model predictions (Ribeiro et al., 2018; Iyyer et al., 2018; Li et al., 2021), whereas concurrent work to our own (Madaan et al.”
Analysis Methods in Neural Language Processing: A Survey (2018, background): “, 2018) Black 7 Char Text classification, sentiment (Jia and Liang, 2017) Black 7 Sentence Reading comprehension (Iyyer et al., 2018) Black 7 Syntax Sentiment, entailment (Shi et al.”
Universal Adversarial Triggers for Attacking and Analyzing NLP (2019, methodology): “Other attack methods are based on generative (Iyyer et al., 2018) or human-in-the-loop approaches (Wallace et al.”
Learning the Difference that Makes a Difference with Counterfactually-Augmented Data (2019, background): “For example, otherwise state-of-the-art models have been shown to be vulnerable to synthetic transformations such as distractor phrases [11, 27], to misclassify paraphrased task [10, 22] and to fail on template-based modifications [25].”
BLEURT: Learning Robust Metrics for Text Generation (2020, methodology): “Generating synthetic data via para-phrases and perturbations has been commonly used for generating adversarial examples (Jia and Liang, 2017; Iyyer et al., 2018; Belinkov and Bisk, 2018; Ribeiro et al., 2018), an orthogonal line of research.”
Semantically Equivalent Adversarial Rules for Debugging NLP models (2018, background): “Table 5b), which recent work on paraphrasing (Iyyer et al., 2018) or generation using GANs (Zhao et al.”
Evaluating the Factual Consistency of Abstractive Text Summarization (2019, methodology): “Paraphrases were produced by backtranslation using Neural Machine Translation systems (Iyyer et al., 2018).”
PAWS: Paraphrase Adversaries from Word Scrambling (2019, background): “Most text generation models rely on large amount of training data (Iyyer et al., 2018; Guu et al., 2018; Gupta et al., 2018; Li et al., 2018), which is unfortunately not available in our case.”
A Survey on Data Collection for Machine Learning: A Big Data - AI Integration Perspective (2018, neutral): “Synthetic Data Generative Adversarial Networks [57], [58], [59], [60], [61], [62] Policies [63], [64] Image [65], [66], [67], [68], [69], [70], [71] Text [72], [73], [74]”
HotFlip: White-Box Adversarial Examples for Text Classification (2017, background): “Interest in understanding vulnerabilities of NLP systems is growing (Jia and Liang, 2017; Zhao et al., 2018; Belinkov and Bisk, 2018; Iyyer et al., 2018).”
Pathologies of Neural Models Make Interpretations Difficult (2018, background): “The robustness of NLP models has been studied extensively (Papernot et al., 2016; Jia and Liang, 2017; Iyyer et al., 2018; Ribeiro et al., 2018), and most studies define adversarial examples similar to the image domain: small perturbations to the input lead to large changes in the output.”
Good-Enough Compositional Data Augmentation (2019, background): “An alternative line of work on paraphrase-based data augmentation (Ganitkevitch et al., 2013; Iyyer et al., 2018) uses external, text-only resources to encourage robust interpretation of new inputs corresponding to known outputs.”
PAWS-X: A Cross-lingual Adversarial Dataset for Paraphrase Identification (2019, background): “, 2018), and text classification (Alzantot et al., 2018; Iyyer et al., 2018).”
Analyzing Dynamic Adversarial Training Data in the Limit (2021, background): “, 2018) or syntax (Iyyer et al., 2018) whereas DADC examples are only limited by human creativity.”
FreeLB: Enhanced Adversarial Training for Natural Language Understanding (2019, methodology): “Iyyer et al. (2018) propose a neural paraphrase model based on back-translated data to produce paraphrases that have different sentence structures.”
ParaNMT-50M: Pushing the Limits of Paraphrastic Sentence Embeddings with Millions of Machine Translations (2017, methodology): “PARANMT-50M has also been used for syntactically-controlled paraphrase generation; this work is described in detail by Iyyer et al. (2018). A syntactically controlled paraphrase net-”