Adversarially Regularising Neural NLI Models to Integrate Logical Background Knowledge

By Pasquale Minervini, Sebastian Riedel. 2018

126 citations recorded by the source corpus.

Adversarial examples are inputs to machine learning models designed to cause the model to make a mistake. They are useful for understanding the shortcomings of machine learning models, interpreting their results, and for regularisation. In NLP, however, most example generation strategies produce input text by using known, pre-specified semantic transformations, requiring significant manual effort and in-depth understanding of the problem and domain. In this paper, we investigate the problem of automatically generating adversarial examples that violate a set of given First-Order Logic constraints in Natural Language Inference (NLI). We reduce the problem of identifying such adversarial examples to a combinatorial optimisation problem, by maximising a quantity measuring the degree of violation of such constraints and by using a language model for generating linguistically-plausible examples. Furthermore, we propose a method for adversarially regularising neural NLI models for incorporating background knowledge. Our results show that, while the proposed method does not always improve results on the SNLI and MultiNLI datasets, it significantly and consistently increases the predictive accuracy on adversarially-crafted datasets – up to a 79.6% relative improvement – while drastically reducing the number of background knowledge violations. Furthermore, we show that adversarial examples transfer among model architectures, and that the proposed adversarial training procedure improves the robustness of NLI models to adversarial examples.

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

DROP: A Reading Comprehension Benchmark Requiring Discrete Reasoning Over Paragraphs (2019, background): “Adversarial dataset construction: We continue a recent trend in creating datasets with adversarial baselines in the loop (Paperno et al., 2016; Minervini and Riedel, 2018; Zellers et al., 2018b; Zhang et al., 2018; Zellers et al., 2018a).”
Dynabench: Rethinking Benchmarking in NLP (2021, background): “Text classiﬁcation models have been shown to be sensitive to single input character change (Ebrahimi et al., 2018b) and ﬁrst-order logic inconsistencies (Minervini and Riedel, 2018).”
Make Up Your Mind! Adversarial Generation of Inconsistent Natural Language Explanations (2019, background): “Finally, Minervini and Riedel [24] attempted to ﬁnd inputs where a model trained on SNLI violates a set of logical constraints.”
Logic-Guided Data Augmentation and Regularization for Consistent Question Answering (2020, methodology): “One can select the data based on a model’s prediction inconsistencies (Minervini and Riedel, 2018) or randomly sample at each epoch (Kang et al., 2018).”
There is Strength in Numbers: Avoiding the Hypothesis-Only Bias in Natural Language Inference via Ensemble Adversarial Training (2020, background): “Minervini and Riedel (2018) propose searching for violations of constraints, such as the symmetry of contradiction and transitivity of entailment , for identifying where NLI models make mistakes.”
Look at the First Sentence: Position Bias in Question Answering (2020, background): “For instance, sentence classiﬁcation models in natural language inference and argument reasoning comprehension suffer from word statistics bias (Poliak et al., 2018; Minervini and Riedel, 2018; Kang et al., 2018; Belinkov et al., 2019; Niven and Kao, 2019).”
Adversarial Attacks on Deep Learning Models in Natural Language Processing: A Survey (2019, methodology): “These representative works received the interests of evaluation with adversarial examples recently [35, 54, 55, 92, 104, 112, 118, 130, 156].”
Generating Data to Mitigate Spurious Correlations in Natural Language Inference Datasets (2022, background): “• LI-TS: text-fragment swap test by swapping the premise and hypothesis (Wang et al., 2019; Minervini and Riedel, 2018).”
Natural Language Inference with a Human Touch: Using Human Explanations to Guide Model Attention (2021, methodology): “, 2020), using data-augmentation (Min et al., 2020; Minervini and Riedel, 2018), fine-tuning on minority examples (Yaghoobzadeh et al.”
A Logic-Driven Framework for Consistency of Neural Models (2019, methodology): “The closest related work to this paper is probably that of Minervini and Riedel (2018), which uses the Gödel t-norm to discover adversarial examples that violate constraints.”
Maieutic Prompting: Logically Consistent Reasoning with Recursive Explanations (2022, methodology): “Following previous works (Minervini and Riedel, 2018; Wang et al., 2019), we convert the NLI labels into logical relations as following:”
BeliefBank: Adding Memory to a Pre-Trained Language Model for a Systematic Notion of Belief (2021, background): “, (Ribeiro et al., 2019; Minervini and Riedel, 2018; Li et al., 2019; Asai and Hajishirzi, 2020), they have not solved the problem.”
Enriching a Model's Notion of Belief using a Persistent Memory (2021, background): “Others have trained models with (small) sets of examples with known constraints between them, and included an additional loss term reflecting inconsistency among set members during training (Minervini and Riedel, 2018; Li et al., 2019; Asai and Hajishirzi, 2020).”
The Curse of Performance Instability in Analysis Datasets: Consequences, Source, and Suggestions (2020, background): “, 2019), some also involved logical-related study (Minervini and Riedel, 2018; Wang et al., 2019).”
Beyond Leaderboards: A survey of methods for revealing weaknesses in Natural Language Inference data and models (2020, background): “…re-weighting the loss function to incorporate the bias in the data (Schuster et al., 2019; Zhang et al., 2019b), parameter regularisation (Sagawa et al., 2020) and the use of external resources, such as linguistic knowledge (Zhou et al., 2019; Wu et al., 2019) or logic (Minervini and Riedel, 2018).”
Syntactic Data Augmentation Increases Robustness to Inference Heuristics (2020, methodology): “Data augmentation is often employed to increase robustness in vision (Perez and Wang, 2017) and language (Belinkov and Bisk, 2018; Wei and Zou, 2019), including in NLI (Minervini and Riedel, 2018; Yanaka et al., 2019).”
Learning Credible Deep Neural Networks with Rationale Regularization (2019, methodology): “The DNN model can succeed by simply recognize patterns that only happen to be predictive on instances over the test set [42].”
Towards Robustifying NLI Models Against Lexical Dataset Biases (2020, background): “Minervini and Riedel (2018) generate adversarial examples and regularize models based on ﬁrst-order logic rules.”
On Adversarial Removal of Hypothesis-only Bias in Natural Language Inference (2019, methodology): “…is driving a ﬁre truck A man is doing tricks on a skateboard Nobody is doing tricks Although recent work has applied adversarial learning to NLI (Minervini & Riedel, 2018; Kang et al., 2018), this is the ﬁrst work to our knowledge that explicitly studies NLI models designed to ignore…”