While research on explaining predictions of open-domain QA systems (ODQA) to users is gaining momentum, most works have failed to evaluate the extent to which explanations improve user trust. While few works evaluate explanations using user studies, they employ settings that may deviate from the end-user's usage in-the-wild: ODQA is most ubiquitous in voice-assistants, yet current research only evaluates explanations using a visual display, and may erroneously extrapolate conclusions about the most performant explanations to other modalities. To alleviate these issues, we conduct user studies that measure whether explanations help users correctly decide when to accept or reject an ODQA system's answer. Unlike prior work, we control for explanation modality, e.g., whether they are communicated to users through a spoken or visual interface, and contrast effectiveness across modalities. Our results show that explanations derived from retrieved evidence passages can outperform strong baselines (calibrated confidence) across modalities but the best explanation strategy in fact changes with the modality. We show common failure cases of current explanations, emphasize end-to-end evaluation of explanations, and caution against evaluating them in proxy modalities that are different from deployment.
Given the increasingly prominent role NLP models (will) play in our lives, it is important to evaluate models on their alignment with human expectations of how models behave. Using Natural Language Inference (NLI) as a case study, we investigated the extent to which human-generated explanations of models' inference decisions align with how models actually make these decisions. More specifically, we defined two alignment metrics that quantify how well natural language human explanations align with model sensitivity to input words, as measured by integrated gradients. Then, we evaluated six different transformer models (the base and large versions of BERT, RoBERTa and ELECTRA), and found that the BERT-base model has the highest alignment with human-generated explanations, for both alignment metrics. Additionally, the base versions of the models we surveyed tended to have higher alignment with human-generated explanations than their larger counterparts, suggesting that increasing the number model parameters could result in worse alignment with human explanations. Finally, we find that a model's alignment with human explanations is not predicted by the model's accuracy on NLI, suggesting that accuracy and alignment are orthogonal, and both are important ways to evaluate models.
Despite its importance to experimental design, statistical power (the probability that, given a real effect, an experiment will reject the null hypothesis) has largely been ignored by the NLP community. Underpowered experiments make it more difficult to discern the difference between statistical noise and meaningful model improvements, and increase the chances of exaggerated findings. By meta-analyzing a set of existing NLP papers and datasets, we characterize typical power for a variety of settings and conclude that underpowered experiments are common in the NLP literature. In particular, for several tasks in the popular GLUE benchmark, small test sets mean that most attempted comparisons to state of the art models will not be adequately powered. Similarly, based on reasonable assumptions, we find that the most typical experimental design for human rating studies will be underpowered to detect small model differences, of the sort that are frequently studied. For machine translation, we find that typical test sets of 2000 sentences have approximately 75% power to detect differences of 1 BLEU point. To improve the situation going forward, we give an overview of best practices for power analysis in NLP and release a series of notebooks to assist with future power analyses.
Many pairwise classification tasks, such as paraphrase detection and open-domain question answering, naturally have extreme label imbalance (e.g., $99.99\%$ of examples are negatives). In contrast, many recent datasets heuristically choose examples to ensure label balance. We show that these heuristics lead to trained models that generalize poorly: State-of-the art models trained on QQP and WikiQA each have only $2.4\%$ average precision when evaluated on realistically imbalanced test data. We instead collect training data with active learning, using a BERT-based embedding model to efficiently retrieve uncertain points from a very large pool of unlabeled utterance pairs. By creating balanced training data with more informative negative examples, active learning greatly improves average precision to $32.5\%$ on QQP and $20.1\%$ on WikiQA.
To avoid giving wrong answers, question answering (QA) models need to know when to abstain from answering. Moreover, users often ask questions that diverge from the model's training data, making errors more likely and thus abstention more critical. In this work, we propose the setting of selective question answering under domain shift, in which a QA model is tested on a mixture of in-domain and out-of-domain data, and must answer (i.e., not abstain on) as many questions as possible while maintaining high accuracy. Abstention policies based solely on the model's softmax probabilities fare poorly, since models are overconfident on out-of-domain inputs. Instead, we train a calibrator to identify inputs on which the QA model errs, and abstain when it predicts an error is likely. Crucially, the calibrator benefits from observing the model's behavior on out-of-domain data, even if from a different domain than the test data. We combine this method with a SQuAD-trained QA model and evaluate on mixtures of SQuAD and five other QA datasets. Our method answers 56% of questions while maintaining 80% accuracy; in contrast, directly using the model's probabilities only answers 48% at 80% accuracy.
Despite excellent performance on many tasks, NLP systems are easily fooled by small adversarial perturbations of inputs. Existing procedures to defend against such perturbations are either (i) heuristic in nature and susceptible to stronger attacks or (ii) provide guaranteed robustness to worst-case attacks, but are incompatible with state-of-the-art models like BERT. In this work, we introduce robust encodings (RobEn): a simple framework that confers guaranteed robustness, without making compromises on model architecture. The core component of RobEn is an encoding function, which maps sentences to a smaller, discrete space of encodings. Systems using these encodings as a bottleneck confer guaranteed robustness with standard training, and the same encodings can be used across multiple tasks. We identify two desiderata to construct robust encoding functions: perturbations of a sentence should map to a small set of encodings (stability), and models using encodings should still perform well (fidelity). We instantiate RobEn to defend against a large family of adversarial typos. Across six tasks from GLUE, our instantiation of RobEn paired with BERT achieves an average robust accuracy of 71.3% against all adversarial typos in the family considered, while previous work using a typo-corrector achieves only 35.3% accuracy against a simple greedy attack.
We present the results of the Machine Reading for Question Answering (MRQA) 2019 shared task on evaluating the generalization capabilities of reading comprehension systems. In this task, we adapted and unified 18 distinct question answering datasets into the same format. Among them, six datasets were made available for training, six datasets were made available for development, and the final six were hidden for final evaluation. Ten teams submitted systems, which explored various ideas including data sampling, multi-task learning, adversarial training and ensembling. The best system achieved an average F1 score of 72.5 on the 12 held-out datasets, 10.7 absolute points higher than our initial baseline based on BERT.
State-of-the-art NLP models can often be fooled by adversaries that apply seemingly innocuous label-preserving transformations (e.g., paraphrasing) to input text. The number of possible transformations scales exponentially with text length, so data augmentation cannot cover all transformations of an input. This paper considers one exponentially large family of label-preserving transformations, in which every word in the input can be replaced with a similar word. We train the first models that are provably robust to all word substitutions in this family. Our training procedure uses Interval Bound Propagation (IBP) to minimize an upper bound on the worst-case loss that any combination of word substitutions can induce. To evaluate models' robustness to these transformations, we measure accuracy on adversarially chosen word substitutions applied to test examples. Our IBP-trained models attain $75\%$ adversarial accuracy on both sentiment analysis on IMDB and natural language inference on SNLI. In comparison, on IMDB, models trained normally and ones trained with data augmentation achieve adversarial accuracy of only $8\%$ and $35\%$, respectively.
Most information extraction methods focus on binary relations expressed within single sentences. In high-value domains, however, $n$-ary relations are of great demand (e.g., drug-gene-mutation interactions in precision oncology). Such relations often involve entity mentions that are far apart in the document, yet existing work on cross-sentence relation extraction is generally confined to small text spans (e.g., three consecutive sentences), which severely limits recall. In this paper, we propose a novel multiscale neural architecture for document-level $n$-ary relation extraction. Our system combines representations learned over various text spans throughout the document and across the subrelation hierarchy. Widening the system's purview to the entire document maximizes potential recall. Moreover, by integrating weak signals across the document, multiscale modeling increases precision, even in the presence of noisy labels from distant supervision. Experiments on biomedical machine reading show that our approach substantially outperforms previous $n$-ary relation extraction methods.
Extractive reading comprehension systems can often locate the correct answer to a question in a context document, but they also tend to make unreliable guesses on questions for which the correct answer is not stated in the context. Existing datasets either focus exclusively on answerable questions, or use automatically generated unanswerable questions that are easy to identify. To address these weaknesses, we present SQuAD 2.0, the latest version of the Stanford Question Answering Dataset (SQuAD). SQuAD 2.0 combines existing SQuAD data with over 50,000 unanswerable questions written adversarially by crowdworkers to look similar to answerable ones. To do well on SQuAD 2.0, systems must not only answer questions when possible, but also determine when no answer is supported by the paragraph and abstain from answering. SQuAD 2.0 is a challenging natural language understanding task for existing models: a strong neural system that gets 86% F1 on SQuAD 1.1 achieves only 66% F1 on SQuAD 2.0.