Detecting negatives (such as non-entailment relationships, unanswerable questions, and false claims) is an important and challenging aspect of many natural language understanding tasks. Though manually collecting challenging negative examples can help models detect them, it is both costly and domain-specific. In this work, we propose Self-labeled Counterfactuals for Extrapolating to Negative Examples (SCENE), an automatic method for synthesizing training data that greatly improves models' ability to detect challenging negative examples. In contrast with standard data augmentation, which synthesizes new examples for existing labels, SCENE can synthesize negative examples zero-shot from only positive ones. Given a positive example, SCENE perturbs it with a mask infilling model, then determines whether the resulting example is negative based on a self-training heuristic. With access to only answerable training examples, SCENE can close 69.6% of the performance gap on SQuAD 2.0, a dataset where half of the evaluation examples are unanswerable, compared to a model trained on SQuAD 2.0. Our method also extends to boolean question answering and recognizing textual entailment, and improves generalization from SQuAD to ACE-whQA, an out-of-domain extractive QA benchmark.
In-context learning (ICL) enables large language models (LLMs) to perform new tasks by prompting them with a sequence of training examples. However, ICL is very sensitive to the choice of training examples: randomly sampling examples from a training set leads to high variance in performance. In this paper, we show that curating a carefully chosen subset of training data greatly stabilizes ICL performance. We propose two methods to choose training subsets, both of which score training examples individually and then select the highest-scoring ones. CondAcc scores a training example by its average ICL accuracy when combined with random training examples, while Datamodels learns a linear proxy model that estimates how the presence of each training example influences LLM accuracy. On average, CondAcc and Datamodels outperform sampling from the entire training set by 7.7% and 6.3%, respectively, across 5 tasks and two LLMs. Our analysis shows that stable subset examples are no more diverse than average, and are not outliers in terms of sequence length and perplexity.
In many task settings, text classification models are likely to encounter examples from novel classes on which they cannot predict correctly. Selective prediction, in which models abstain on low-confidence examples, provides a possible solution, but existing models are often overly confident on OOD examples. To remedy this overconfidence, we introduce Contrastive Novelty-Augmented Learning (CoNAL), a two-step method that generates OOD examples representative of novel classes, then trains to decrease confidence on them. First, we generate OOD examples by prompting a large language model twice: we prompt it to enumerate relevant novel labels, then generate examples from each novel class matching the task format. Second, we train our classifier with a novel contrastive objective that encourages lower confidence on generated OOD examples than training examples. When trained with CoNAL, classifiers improve in their ability to detect and abstain on OOD examples over prior methods by an average of 2.3% AUAC and 5.5% AUROC across 4 NLP datasets, with no cost to in-distribution accuracy.
For vision-and-language reasoning tasks, both fully connectionist, end-to-end methods and hybrid, neuro-symbolic methods have achieved high in-distribution performance. In which out-of-distribution settings does each paradigm excel? We investigate this question on both single-image and multi-image visual question-answering through four types of generalization tests: a novel segment-combine test for multi-image queries, contrast set, compositional generalization, and cross-benchmark transfer. Vision-and-language end-to-end trained systems exhibit sizeable performance drops across all these tests. Neuro-symbolic methods suffer even more on cross-benchmark transfer from GQA to VQA, but they show smaller accuracy drops on the other generalization tests and their performance quickly improves by few-shot training. Overall, our results demonstrate the complementary benefits of these two paradigms, and emphasize the importance of using a diverse suite of generalization tests to fully characterize model robustness to distribution shift.
In order to reliably process natural language, NLP systems must generalize to the long tail of rare utterances. We propose a method to create challenging benchmarks that require generalizing to the tail of the distribution by re-splitting existing datasets. We create 'Likelihood splits' where examples that are assigned lower likelihood by a pre-trained language model (LM) are placed in the test set, and more likely examples are in the training set. This simple approach can be customized to construct meaningful train-test splits for a wide range of tasks. Likelihood splits are more challenging than random splits: relative error rates of state-of-the-art models on our splits increase by 59% for semantic parsing on Spider, 77% for natural language inference on SNLI, and 38% for yes/no question answering on BoolQ compared with the corresponding random splits. Moreover, Likelihood splits create fairer benchmarks than adversarial filtering; when the LM used to create the splits is used as the task model, our splits do not adversely penalize the LM.
Recent work has observed that pre-trained models have higher out-of-distribution (OOD) robustness when they are exposed to less in-distribution (ID) training data (Radford et al., 2021). In particular, zero-shot models (e.g., GPT-3 and CLIP) have higher robustness than conventionally fine-tuned models, but these robustness gains fade as zero-shot models are fine-tuned on more ID data. We study this relationship between sample efficiency and robustness -- if two models have the same ID performance, does the model trained on fewer examples (higher sample efficiency) perform better OOD (higher robustness)? Surprisingly, experiments across three tasks, 23 total ID-OOD settings, and 14 models do not reveal a consistent relationship between sample efficiency and robustness -- while models with higher sample efficiency are sometimes more robust, most often there is no change in robustness, with some cases even showing decreased robustness. Since results vary on a case-by-case basis, we conduct detailed case studies of two particular ID-OOD pairs (SST-2 -> IMDb sentiment and SNLI -> HANS) to better understand why better sample efficiency may or may not yield higher robustness; attaining such an understanding requires case-by-case analysis of why models are not robust on a particular ID-OOD setting and how modeling techniques affect model capabilities.
Real-world natural language processing (NLP) models need to be continually updated to fix the prediction errors in out-of-distribution (OOD) data streams while overcoming catastrophic forgetting. However, existing continual learning (CL) problem setups cannot cover such a realistic and complex scenario. In response to this, we propose a new CL problem formulation dubbed continual model refinement (CMR). Compared to prior CL settings, CMR is more practical and introduces unique challenges (boundary-agnostic and non-stationary distribution shift, diverse mixtures of multiple OOD data clusters, error-centric streams, etc.). We extend several existing CL approaches to the CMR setting and evaluate them extensively. For benchmarking and analysis, we propose a general sampling algorithm to obtain dynamic OOD data streams with controllable non-stationarity, as well as a suite of metrics measuring various aspects of online performance. Our experiments and detailed analysis reveal the promise and challenges of the CMR problem, supporting that studying CMR in dynamic OOD streams can benefit the longevity of deployed NLP models in production.
Question answering (QA) over real-world knowledge bases (KBs) is challenging because of the diverse (essentially unbounded) types of reasoning patterns needed. However, we hypothesize in a large KB, reasoning patterns required to answer a query type reoccur for various entities in their respective subgraph neighborhoods. Leveraging this structural similarity between local neighborhoods of different subgraphs, we introduce a semiparametric model with (i) a nonparametric component that for each query, dynamically retrieves other similar $k$-nearest neighbor (KNN) training queries along with query-specific subgraphs and (ii) a parametric component that is trained to identify the (latent) reasoning patterns from the subgraphs of KNN queries and then apply it to the subgraph of the target query. We also propose a novel algorithm to select a query-specific compact subgraph from within the massive knowledge graph (KG), allowing us to scale to full Freebase KG containing billions of edges. We show that our model answers queries requiring complex reasoning patterns more effectively than existing KG completion algorithms. The proposed model outperforms or performs competitively with state-of-the-art models on several KBQA benchmarks.
In Dynamic Adversarial Data Collection (DADC), human annotators are tasked with finding examples that models struggle to predict correctly. Models trained on DADC-collected training data have been shown to be more robust in adversarial and out-of-domain settings, and are considerably harder for humans to fool. However, DADC is more time-consuming than traditional data collection and thus more costly per example. In this work, we examine if we can maintain the advantages of DADC, without suffering the additional cost. To that end, we introduce Generative Annotation Assistants (GAAs), generator-in-the-loop models that provide real-time suggestions that annotators can either approve, modify, or reject entirely. We collect training datasets in twenty experimental settings and perform a detailed analysis of this approach for the task of extractive question answering (QA) for both standard and adversarial data collection. We demonstrate that GAAs provide significant efficiency benefits in terms of annotation speed, while leading to improved model fooling rates. In addition, we show that GAA-assisted data leads to higher downstream model performance on a variety of question answering tasks.
We study the robustness of machine reading comprehension (MRC) models to entity renaming -- do models make more wrong predictions when answer entities have different names? Such failures would indicate that models are overly reliant on entity knowledge to answer questions, and therefore may generalize poorly when facts about the world change or questions are asked about novel entities. To systematically audit model robustness, we propose a general and scalable method to replace person names with names from a variety of sources, ranging from common English names to names from other languages to arbitrary strings. Across four datasets and three pretrained model architectures, MRC models consistently perform worse when entities are renamed, with particularly large accuracy drops on datasets constructed via distant supervision. We also find large differences between models: SpanBERT, which is pretrained with span-level masking, is more robust than RoBERTa, despite having similar accuracy on unperturbed test data. Inspired by this, we experiment with span-level and entity-level masking as a continual pretraining objective and find that they can further improve the robustness of MRC models.