Lost in the Middle: How Language Models Use Long Contexts

While recent language models have the ability to take long contexts as input, relatively little is known about how well they use longer context. We analyze language model performance on two tasks that require identifying relevant information within their input contexts: multi-document question answering and key-value retrieval. We find that performance is often highest when relevant information occurs at the beginning or end of the input context, and significantly degrades when models must access relevant information in the middle of long contexts. Furthermore, performance substantially decreases as the input context grows longer, even for explicitly long-context models. Our analysis provides a better understanding of how language models use their input context and provides new evaluation protocols for future long-context models.

Anchor Prediction: Automatic Refinement of Internet Links

Internet links enable users to deepen their understanding of a topic by providing convenient access to related information. However, the majority of links are unanchored -- they link to a target webpage as a whole, and readers may expend considerable effort localizing the specific parts of the target webpage that enrich their understanding of the link's source context. To help readers effectively find information in linked webpages, we introduce the task of anchor prediction, where the goal is to identify the specific part of the linked target webpage that is most related to the source linking context. We release the AuthorAnchors dataset, a collection of 34K naturally-occurring anchored links, which reflect relevance judgments by the authors of the source article. To model reader relevance judgments, we annotate and release ReaderAnchors, an evaluation set of anchors that readers find useful. Our analysis shows that effective anchor prediction often requires jointly reasoning over lengthy source and target webpages to determine their implicit relations and identify parts of the target webpage that are related but not redundant. We benchmark a performant T5-based ranking approach to establish baseline performance on the task, finding ample room for improvement.

Evaluating Verifiability in Generative Search Engines

Generative search engines directly generate responses to user queries, along with in-line citations. A prerequisite trait of a trustworthy generative search engine is verifiability, i.e., systems should cite comprehensively (high citation recall; all statements are fully supported by citations) and accurately (high citation precision; every cite supports its associated statement). We conduct human evaluation to audit four popular generative search engines -- Bing Chat, NeevaAI, perplexity.ai, and YouChat -- across a diverse set of queries from a variety of sources (e.g., historical Google user queries, dynamically-collected open-ended questions on Reddit, etc.). We find that responses from existing generative search engines are fluent and appear informative, but frequently contain unsupported statements and inaccurate citations: on average, a mere 51.5% of generated sentences are fully supported by citations and only 74.5% of citations support their associated sentence. We believe that these results are concerningly low for systems that may serve as a primary tool for information-seeking users, especially given their facade of trustworthiness. We hope that our results further motivate the development of trustworthy generative search engines and help researchers and users better understand the shortcomings of existing commercial systems.

Are Sample-Efficient NLP Models More Robust?

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.

Making Heads and Tails of Models with Marginal Calibration for Sparse Tagsets

For interpreting the behavior of a probabilistic model, it is useful to measure a model's calibration--the extent to which it produces reliable confidence scores. We address the open problem of calibration for tagging models with sparse tagsets, and recommend strategies to measure and reduce calibration error (CE) in such models. We show that several post-hoc recalibration techniques all reduce calibration error across the marginal distribution for two existing sequence taggers. Moreover, we propose tag frequency grouping (TFG) as a way to measure calibration error in different frequency bands. Further, recalibrating each group separately promotes a more equitable reduction of calibration error across the tag frequency spectrum.

Identifying the Limits of Cross-Domain Knowledge Transfer for Pretrained Models

There is growing evidence that pretrained language models improve task-specific fine-tuning not just for the languages seen in pretraining, but also for new languages and even non-linguistic data. What is the nature of this surprising cross-domain transfer? We offer a partial answer via a systematic exploration of how much transfer occurs when models are denied any information about word identity via random scrambling. In four classification tasks and two sequence labeling tasks, we evaluate baseline models, LSTMs using GloVe embeddings, and BERT. We find that only BERT shows high rates of transfer into our scrambled domains, and for classification but not sequence labeling tasks. Our analyses seek to explain why transfer succeeds for some tasks but not others, to isolate the separate contributions of pretraining versus fine-tuning, and to quantify the role of word frequency. These findings help explain where and why cross-domain transfer occurs, which can guide future studies and practical fine-tuning efforts.

Can Small and Synthetic Benchmarks Drive Modeling Innovation? A Retrospective Study of Question Answering Modeling Approaches

Datasets are not only resources for training accurate, deployable systems, but are also benchmarks for developing new modeling approaches. While large, natural datasets are necessary for training accurate systems, are they necessary for driving modeling innovation? For example, while the popular SQuAD question answering benchmark has driven the development of new modeling approaches, could synthetic or smaller benchmarks have led to similar innovations? This counterfactual question is impossible to answer, but we can study a necessary condition: the ability for a benchmark to recapitulate findings made on SQuAD. We conduct a retrospective study of 20 SQuAD modeling approaches, investigating how well 32 existing and synthesized benchmarks concur with SQuAD -- i.e., do they rank the approaches similarly? We carefully construct small, targeted synthetic benchmarks that do not resemble natural language, yet have high concurrence with SQuAD, demonstrating that naturalness and size are not necessary for reflecting historical modeling improvements on SQuAD. Our results raise the intriguing possibility that small and carefully designed synthetic benchmarks may be useful for driving the development of new modeling approaches.

Lexical Semantic Recognition

Segmentation and (segment) labeling are generally treated separately in lexical semantics, raising issues due to their close inter-dependence and necessitating joint annotation. We therefore investigate the lexical semantic recognition task of multiword expression segmentation and supersense disambiguation, unifying several previously-disparate styles of lexical semantic annotation. We evaluate a neural CRF model along all annotation axes available in version 4.3 of the STREUSLE corpus: lexical unit segmentation (multiword expressions), word-level syntactic tags, and supersense classes for noun, verb, and preposition/possessive units. As the label set generalizes that of previous tasks (DiMSUM, PARSEME), we additionally evaluate how well the model generalizes to those test sets, with encouraging results. By establishing baseline models and evaluation metrics, we pave the way for comprehensive and accurate modeling of lexical semantics.

Evaluating NLP Models via Contrast Sets

Standard test sets for supervised learning evaluate in-distribution generalization. Unfortunately, when a dataset has systematic gaps (e.g., annotation artifacts), these evaluations are misleading: a model can learn simple decision rules that perform well on the test set but do not capture a dataset's intended capabilities. We propose a new annotation paradigm for NLP that helps to close systematic gaps in the test data. In particular, after a dataset is constructed, we recommend that the dataset authors manually perturb the test instances in small but meaningful ways that (typically) change the gold label, creating contrast sets. Contrast sets provide a local view of a model's decision boundary, which can be used to more accurately evaluate a model's true linguistic capabilities. We demonstrate the efficacy of contrast sets by creating them for 10 diverse NLP datasets (e.g., DROP reading comprehension, UD parsing, IMDb sentiment analysis). Although our contrast sets are not explicitly adversarial, model performance is significantly lower on them than on the original test sets---up to 25\% in some cases. We release our contrast sets as new evaluation benchmarks and encourage future dataset construction efforts to follow similar annotation processes.

Quoref: A Reading Comprehension Dataset with Questions Requiring Coreferential Reasoning

Machine comprehension of texts longer than a single sentence often requires coreference resolution. However, most current reading comprehension benchmarks do not contain complex coreferential phenomena and hence fail to evaluate the ability of models to resolve coreference. We present a new crowdsourced dataset containing more than 24K span-selection questions that require resolving coreference among entities in over 4.7K English paragraphs from Wikipedia. Obtaining questions focused on such phenomena is challenging, because it is hard to avoid lexical cues that shortcut complex reasoning. We deal with this issue by using a strong baseline model as an adversary in the crowdsourcing loop, which helps crowdworkers avoid writing questions with exploitable surface cues. We show that state-of-the-art reading comprehension models perform significantly worse than humans on this benchmark---the best model performance is 70.5 F1, while the estimated human performance is 93.4 F1.