Undersensitivity in Neural Reading Comprehension

@article{Welbl2020UndersensitivityIN,
  title={Undersensitivity in Neural Reading Comprehension},
  author={Johannes Welbl and Pasquale Minervini and Max Bartolo and Pontus Stenetorp and Sebastian Riedel},
  journal={ArXiv},
  year={2020},
  volume={abs/2003.04808}
}
Current reading comprehension methods generalise well to in-distribution test sets, yet perform poorly on adversarially selected data. Prior work on adversarial inputs typically studies model oversensitivity: semantically invariant text perturbations that cause a model’s prediction to change. Here we focus on the complementary problem: excessive prediction undersensitivity, where input text is meaningfully changed but the model’s prediction does not, even though it should. We formulate an… 
Contrasting Human- and Machine-Generated Word-Level Adversarial Examples for Text Classification
TLDR
This work reports on crowdsourcing studies in which humans are tasked with iteratively modifying words in an input text, while receiving immediate model feedback, with the aim of causing a sentiment classification model to misclassify the example.
Exploring Strategies for Generalizable Commonsense Reasoning with Pre-trained Models
TLDR
This paper investigates what models learn from commonsense reasoning datasets and shows that fine-tuning performs best, by learning both the content and the structure of the task, but suffers from overfitting and limited generalization to novel answers.
DuReaderrobust: A Chinese Dataset Towards Evaluating the Robustness of Machine Reading Comprehension Models
TLDR
The experimental results show that MRC models based on the pre-trained language models perform much worse than human does on the robustness test set, although they perform as well as human on in-domain test set.
DuReader_robust: A Chinese Dataset Towards Evaluating Robustness and Generalization of Machine Reading Comprehension in Real-World Applications
TLDR
This work introduces a real-world Chinese dataset – DuReader_robust – designed to evaluate the MRC models from three aspects: over-sensitivity, over-stability and generalization, and shows that M RC models do not perform well on the challenge test set.
Model Agnostic Answer Reranking System for Adversarial Question Answering
TLDR
This work presents a simple model-agnostic approach to question answering that can be applied directly to any QA model without any retraining, and outperforms state-of-the-art defense techniques.
Neural Conversational QA: Learning to Reason vs Exploiting Patterns
TLDR
This paper presents a simple yet effective model that learns embedding representations to incorporate dialog history along with the previous answers to follow-up questions that outperforms existing methods on all metrics and is more robust in dealing with spurious patterns.
An Understanding-Oriented Robust Machine Reading Comprehension Model
TLDR
This paper proposes an understanding-oriented machine reading comprehension model to address three kinds of robustness issues, which are over sensitivity, over stability and generalization, which is integrated with a multi-task learning based method.
Robust Question Answering Through Sub-part Alignment
TLDR
This work model question answering as an alignment problem, decomposing both the question and context into smaller units based on off-the-shelf semantic representations, and align the question to a subgraph of the context in order to find the answer.
A Survey on Machine Reading Comprehension Systems
TLDR
It is demonstrated that the focus of research has changed in recent years from answer extraction to answer generation, from single to multi-document reading comprehension, and from learning from scratch to using pre-trained embeddings.
A survey of methods for revealing and overcoming weaknesses of data-driven Natural Language Understanding
TLDR
This structured survey provides an overview of the evolving research area by categorising reported weaknesses in models and datasets and the methods proposed to reveal and alleviate those weaknesses for the English language.
...
...

References

SHOWING 1-10 OF 41 REFERENCES
Robust Machine Comprehension Models via Adversarial Training
TLDR
A novel alternative adversary-generation algorithm is proposed, AddSentDiverse, that significantly increases the variance within the adversarial training data by providing effective examples that punish the model for making certain superficial assumptions, and can make a state-of-the-art model significantly more robust.
Adversarial Examples for Evaluating Reading Comprehension Systems
TLDR
This work proposes an adversarial evaluation scheme for the Stanford Question Answering Dataset that tests whether systems can answer questions about paragraphs that contain adversarially inserted sentences without changing the correct answer or misleading humans.
Adversarially Regularising Neural NLI Models to Integrate Logical Background Knowledge
TLDR
This paper reduces the problem of automatically generating adversarial examples that violate a set of given First-Order Logic constraints in Natural Language Inference by maximising a quantity measuring the degree of violation of such constraints and using a language model for generating linguistically-plausible examples.
Generating Natural Language Adversarial Examples
TLDR
A black-box population-based optimization algorithm is used to generate semantically and syntactically similar adversarial examples that fool well-trained sentiment analysis and textual entailment models with success rates of 97% and 70%, respectively.
Universal Adversarial Triggers for NLP
Adversarial examples highlight model vulnerabilities and are useful for evaluation and interpretation. We define universal adversarial triggers: input-agnostic sequences of tokens that trigger a
Pathologies of Neural Models Make Interpretations Difficult
TLDR
This work uses input reduction, which iteratively removes the least important word from the input, to expose pathological behaviors of neural models: the remaining words appear nonsensical to humans and are not the ones determined as important by interpretation methods.
Know What You Don’t Know: Unanswerable Questions for SQuAD
TLDR
SQuadRUn is a new dataset that combines the existing Stanford Question Answering Dataset (SQuAD) with over 50,000 unanswerable questions written adversarially by crowdworkers to look similar to answerable ones.
Adversarial Over-Sensitivity and Over-Stability Strategies for Dialogue Models
We present two categories of model-agnostic adversarial strategies that reveal the weaknesses of several generative, task-oriented dialogue models: Should-Not-Change strategies that evaluate
Towards Verified Robustness under Text Deletion Interventions
TLDR
A novel technique for formal verification of this specification for models based on the popular decomposable attention mechanism is developed by employing the efficient yet effective interval bound propagation (IBP) approach.
AdvEntuRe: Adversarial Training for Textual Entailment with Knowledge-Guided Examples
TLDR
This work proposes knowledge-guided adversarial example generators for incorporating large lexical resources in entailment models via only a handful of rule templates and proposes the first GAN-style approach for training it using a natural language example generator that iteratively adjusts to the discriminator’s weaknesses.
...
...