We explore and enhance the ability of neural language models to generate novel scientific directions grounded in literature. Work on literature-based hypothesis generation has traditionally focused on binary link prediction—severely limiting the expressivity of hypotheses. This line of work also does not focus on optimizing novelty. We take a dramatic departure with a novel setting in which models use as input background contexts (e.g., problems, experimental settings, goals), and output natural language ideas grounded in literature. We present SciMON, a modeling framework that uses retrieval of “inspirations” from past scientific papers, and explicitly optimizes for novelty by iteratively comparing to prior papers and updating idea suggestions until sufficient novelty is achieved. Comprehensive evaluations reveal that GPT-4 tends to generate ideas with overall low technical depth and novelty, while our methods partially mitigate this issue. Our work represents a first step toward evaluating and developing language models that generate new ideas derived from the scientific literature. Code, data, and resources are publicly available for research purposes: https://github.com/eaglew/clbd.

We introduce the task of automatically revising scientific papers based on peer feedback and release ARIES, a dataset of review comments and their corresponding paper edits. The data is drawn from real reviewer-author interactions from computer science, and we provide labels linking each reviewer comment to the specific paper edits made by the author in response. We automatically create a high-precision silver training set, as well as an expert-labeled test set that shows high inter-annotator agreement. In experiments with 10 models covering the state of the art, we find that they struggle even to identify which edits correspond to a comment—especially when the relationship between the edit and the comment is indirect and requires reasoning to uncover. We also extensively analyze GPT-4’s ability to generate edits given a comment and the original paper. We find that it often succeeds on a superficial level, but tends to rigidly follow the wording of the feedback rather than the underlying intent, and lacks technical details compared to human-written edits.

Extracting fine-grained experimental findings from literature can provide dramatic utility for scientific applications. Prior work has developed annotation schemas and datasets for limited aspects of this problem, failing to capture the real-world complexity and nuance required. Focusing on biomedicine, this work presents CARE—a new IE dataset for the task of extracting clinical findings. We develop a new annotation schema capturing fine-grained findings as n-ary relations between entities and attributes, which unifies phenomena challenging for current IE systems such as discontinuous entity spans, nested relations, variable arity n-ary relations and numeric results in a single schema. We collect extensive annotations for 700 abstracts from two sources: clinical trials and case reports. We also demonstrate the generalizability of our schema to the computer science and materials science domains. We benchmark state-of-the-art IE systems on CARE, showing that even models such as GPT4 struggle. We release our resources to advance research on extracting and aggregating literature findings.

Topic pages aggregate useful information about an entity or concept into a single succinct and accessible article. Automated creation of topic pages would enable their rapid curation as information resources, providing an alternative to traditional web search. While most prior work has focused on generating topic pages about biographical entities, in this work, we develop a completely automated process to generate high-quality topic pages for scientific entities, with a focus on biomedical concepts. We release TOPICAL, a web app and associated open-source code, comprising a model pipeline combining retrieval, clustering, and prompting, that makes it easy for anyone to generate topic pages for a wide variety of biomedical entities on demand. In a human evaluation of 150 diverse topic pages generated using TOPICAL, we find that the vast majority were considered relevant, accurate, and coherent, with correct supporting citations. We make all code publicly available and host a free-to-use web app at: https://s2-topical.apps.allenai.org.

Commonsense capabilities of pre-trained language models dramatically improve with scale, leading many to believe that scale is the only winning recipe. But is it? Here, we investigate an alternative that a priori seems impossible: can smaller language models (e.g., GPT-2) win over models that are orders of magnitude larger and better (e.g., GPT-3), if powered with novel commonsense distillation algorithms?The key intellectual challenge is to design a learning algorithm that achieve a competitive level of commonsense acquisition, without relying on the benefits of scale. In particular, we study generative models of commonsense knowledge, focusing on the task of generating generics, statements of commonsense facts about everyday concepts, e.g., birds can fly. We introduce I2D2, a novel commonsense distillation framework that loosely follows the Symbolic Knowledge Distillation of West et al. but breaks the dependence on the extreme-scale teacher model with two innovations: (1) the novel adaptation of NeuroLogic Decoding to enhance the generation quality of the weak, off-the-shelf language models, and (2) self-imitation learning to iteratively learn from the model’s own enhanced commonsense acquisition capabilities. Empirical results suggest that scale is not the only way, as novel algorithms can be a promising alternative. Moreover, our study leads to a new corpus of generics, Gen-A-tomic, that is the largest and highest quality available to date.

Generics express generalizations about the world (e.g., birds can fly) that are not universally true (e.g., newborn birds and penguins cannot fly). Commonsense knowledge bases, used extensively in NLP, encode some generic knowledge but rarely enumerate such exceptions and knowing when a generic statement holds or does not hold true is crucial for developing a comprehensive understanding of generics. We present a novel framework informed by linguistic theory to generate exemplars—specific cases when a generic holds true or false. We generate ~19k exemplars for ~650 generics and show that our framework outperforms a strong GPT-3 baseline by 12.8 precision points. Our analysis highlights the importance of linguistic theory-based controllability for generating exemplars, the insufficiency of knowledge bases as a source of exemplars, and the challenges exemplars pose for the task of natural language inference.

Entity linking (EL) is the task of linking a textual mention to its corresponding entry in a knowledge base, and is critical for many knowledge-intensive NLP applications. When applied to tables in scientific papers, EL is a step toward large-scale scientific knowledge bases that could enable advanced scientific question answering and analytics. We present the first dataset for EL in scientific tables. EL for scientific tables is especially challenging because scientific knowledge bases can be very incomplete, and disambiguating table mentions typically requires understanding the paper’s text in addition to the table. Our dataset, Scientific Table Entity Linking (S2abEL), focuses on EL in machine learning results tables and includes hand-labeled cell types, attributed sources, and entity links from the PaperswithCode taxonomy for 8,429 cells from 732 tables. We introduce a neural baseline method designed for EL on scientific tables containing many out-of-knowledge-base mentions, and show that it significantly outperforms a state-of-the-art generic table EL method. The best baselines fall below human performance, and our analysis highlights avenues for improvement.

Learned representations of scientific documents can serve as valuable input features for downstream tasks without further fine-tuning. However, existing benchmarks for evaluating these representations fail to capture the diversity of relevant tasks. In response, we introduce SciRepEval, the first comprehensive benchmark for training and evaluating scientific document representations. It includes 24 challenging and realistic tasks, 8 of which are new, across four formats: classification, regression, ranking and search. We then use this benchmark to study and improve the generalization ability of scientific document representation models. We show how state-of-the-art models like SPECTER and SciNCL struggle to generalize across the task formats, and that simple multi-task training fails to improve them. However, a new approach that learns multiple embeddings per document, each tailored to a different format, can improve performance. We experiment with task-format-specific control codes and adapters and find they outperform the existing single-embedding state-of-the-art by over 2 points absolute. We release the resulting family of multi-format models, called SPECTER2, for the community to use and build on.

Various NLP tasks require a complex hierarchical structure over nodes, where each node is a cluster of items. Examples include generating entailment graphs, hierarchical cross-document coreference resolution, annotating event and subevent relations, etc. To enable efficient annotation of such hierarchical structures, we release CHAMP, an open source tool allowing to incrementally construct both clusters and hierarchy simultaneously over any type of texts. This incremental approach significantly reduces annotation time compared to the common pairwise annotation approach and also guarantees maintaining transitivity at the cluster and hierarchy levels. Furthermore, CHAMP includes a consolidation mode, where an adjudicator can easily compare multiple cluster hierarchy annotations and resolve disagreements.

Despite growing interest in applying natural language processing (NLP) and computer vision (CV) models to the scholarly domain, scientific documents remain challenging to work with. They’re often in difficult-to-use PDF formats, and the ecosystem of models to process them is fragmented and incomplete. We introduce PaperMage, an open-source Python toolkit for analyzing and processing visually-rich, structured scientific documents. PaperMage offers clean and intuitive abstractions for seamlessly representing and manipulating both textual and visual document elements. PaperMage achieves this by integrating disparate state-of-the-art NLP and CV models into a unified framework, and provides turn-key recipes for common scientific document processing use-cases. PaperMage has powered multiple research prototypes of AI applications over scientific documents, along with Semantic Scholar’s large-scale production system for processing millions of PDFs. GitHub: https://github.com/allenai/papermage

Real-world applications of neural language models often involve running many different models over the same corpus. The high computational cost of these runs has led to interest in techniques that can reuse the contextualized embeddings produced in previous runs to speed training and inference of future ones. We refer to this approach as embedding recycling (ER). While multiple ER techniques have been proposed, their practical effectiveness is still unknown because existing evaluations consider very few models and do not adequately account for overhead costs. We perform an extensive evaluation of ER across eight different models (17 to 900 million parameters) and fourteen tasks in English. We show how a simple ER technique that caches activations from an intermediate layer of a pretrained model, and learns task-specific adapters on the later layers, is broadly effective. For the best-performing baseline in our experiments (DeBERTa-v2 XL), adding a precomputed cache results in a 90% speedup during training and 87-91% speedup for inference, with negligible impact on accuracy. Our analysis reveals important areas of future work.

Recent work has shown that infusing layout features into language models (LMs) improves processing of visually-rich documents such as scientific papers. Layout-infused LMs are often evaluated on documents with familiar layout features (e.g., papers from the same publisher), but in practice models encounter documents with unfamiliar distributions of layout features, such as new combinations of text sizes and styles, or new spatial configurations of textual elements. In this work we test whether layout-infused LMs are robust to layout distribution shifts. As a case study we use the task of scientific document structure recovery, segmenting a scientific paper into its structural categories (e.g., “title”, “caption”, “reference”). To emulate distribution shifts that occur in practice we re-partition the GROTOAP2 dataset. We find that under layout distribution shifts model performance degrades by up to 20 F1. Simple training strategies, such as increasing training diversity, can reduce this degradation by over 35% relative F1; however, models fail to reach in-distribution performance in any tested out-of-distribution conditions. This work highlights the need to consider layout distribution shifts during model evaluation, and presents a methodology for conducting such evaluations.

Systems that automatically define unfamiliar terms hold the promise of improving the accessibility of scientific texts, especially for readers who may lack prerequisite background knowledge. However, current systems assume a single “best” description per concept, which fails to account for the many ways a concept can be described. We present ACCoRD, an end-to-end system tackling the novel task of generating sets of descriptions of scientific concepts. Our system takes advantage of the myriad ways a concept is mentioned across the scientific literature to produce distinct, diverse descriptions oftarget concepts in terms of different reference concepts. In a user study, we find that users prefer (1) descriptions produced by our end-to-end system, and (2) multiple descriptions to a single “best” description. We release the ACCoRD corpus which includes 1,275 labeled contexts and 1,787 expert-authored concept descriptions to support research on our task.

Self-rationalization models that predict task labels and generate free-text elaborations for their predictions could enable more intuitive interaction with NLP systems. These models are, however, currently trained with a large amount of human-written free-text explanations for each task which hinders their broader usage. We propose to study a more realistic setting of self-rationalization using few training examples. We present FEB—a standardized collection of four existing English-language datasets and associated metrics. We identify the right prompting approach by extensively exploring natural language prompts on FEB. Then, by using this prompt and scaling the model size, we demonstrate that making progress on few-shot self-rationalization is possible. We show there is still ample room for improvement in this task: the average plausibility of generated explanations assessed by human annotators is at most 51% (with GPT-3), while plausibility of human explanations is 76%. We hope that FEB and our proposed approach will spur the community to take on the few-shot self-rationalization challenge.

How to usefully encode compositional task structure has long been a core challenge in AI. Recent work in chain of thought prompting has shown that for very large neural language models (LMs), explicitly demonstrating the inferential steps involved in a target task may improve performance over end-to-end learning that focuses on the target task alone. However, chain of thought prompting has significant limitations due to its dependency on huge pretrained LMs. In this work, we present compositional fine-tuning (CFT): an approach based on explicitly decomposing a target task into component tasks, and then fine-tuning smaller LMs on a curriculum of such component tasks. We apply CFT to recommendation tasks in two domains, world travel and local dining, as well as a previously studied inferential task (sports understanding). We show that CFT outperforms end-to-end learning even with equal amounts of data, and gets consistently better as more component tasks are modeled via fine-tuning. Compared with chain of thought prompting, CFT performs at least as well using LMs only 7.4% of the size, and is moreover applicable to task domains for which data are not available during pretraining.

Abstractive summarization systems today produce fluent and relevant output, but often “hallucinate” statements not supported by the source text. We analyze the connection between hallucinations and training data, and find evidence that models hallucinate because they train on target summaries that are unsupported by the source. Based on our findings, we present PINOCCHIO, a new decoding method that improves the consistency of a transformer-based abstractive summarizer by constraining beam search to avoid hallucinations. Given the model states and outputs at a given step, PINOCCHIO detects likely model hallucinations based on various measures of attribution to the source text. PINOCCHIO backtracks to find more consistent output, and can opt to produce no summary at all when no consistent generation can be found. In experiments, we find that PINOCCHIO improves the consistency of generation by an average of 67% on two abstractive summarization datasets, without hurting recall.

Pretrained contextualized language models such as BERT and T5 have established a new state-of-the-art for ad-hoc search. However, it is not yet well understood why these methods are so effective, what makes some variants more effective than others, and what pitfalls they may have. We present a new comprehensive framework for Analyzing the Behavior of Neural IR ModeLs (ABNIRML), which includes new types of diagnostic probes that allow us to test several characteristics—such as writing styles, factuality, sensitivity to paraphrasing and word order—that are not addressed by previous techniques. To demonstrate the value of the framework, we conduct an extensive empirical study that yields insights into the factors that contribute to the neural model’s gains, and identify potential unintended biases the models exhibit. Some of our results confirm conventional wisdom, for example, that recent neural ranking models rely less on exact term overlap with the query, and instead leverage richer linguistic information, evidenced by their higher sensitivity to word and sentence order. Other results are more surprising, such as that some models (e.g., T5 and ColBERT) are biased towards factually correct (rather than simply relevant) texts. Further, some characteristics vary even for the same base language model, and other characteristics can appear due to random variations during model training.1

Accurately extracting structured content from PDFs is a critical first step for NLP over scientific papers. Recent work has improved extraction accuracy by incorporating elementary layout information, for example, each token’s 2D position on the page, into language model pretraining. We introduce new methods that explicitly model VIsual LAyout (VILA) groups, that is, text lines or text blocks, to further improve performance. In our I-VILA approach, we show that simply inserting special tokens denoting layout group boundaries into model inputs can lead to a 1.9% Macro F1 improvement in token classification. In the H-VILA approach, we show that hierarchical encoding of layout-groups can result in up to 47% inference time reduction with less than 0.8% Macro F1 loss. Unlike prior layout-aware approaches, our methods do not require expensive additional pretraining, only fine-tuning, which we show can reduce training cost by up to 95%. Experiments are conducted on a newly curated evaluation suite, S2-VLUE, that unifies existing automatically labeled datasets and includes a new dataset of manual annotations covering diverse papers from 19 scientific disciplines. Pre-trained weights, benchmark datasets, and source code are available at https://github.com/allenai/VILA.

The capabilities of today’s natural language processing systems are typically evaluated using large datasets of curated questions and answers. While these are critical benchmarks of progress, they also suffer from weakness due to artificial distributions and incomplete knowledge. Artifacts arising from artificial distributions can overstate language model performance, while incomplete knowledge limits fine-grained analysis. In this work, we introduce a complementary benchmarking approach based on SimPlified Language Activity Traces (SPLAT). SPLATs are corpora of language encodings of activity in some closed domain (we study traces from chess and baseball games in this work). SPLAT datasets use naturally-arising distributions, allow the generation of question-answer pairs at scale, and afford complete knowledge in their closed domains. We show that language models of three different architectures can answer questions about world states using only verb-like encodings of activity. Our approach is extensible to new language models and additional question-answering tasks.

Conversations aimed at determining good recommendations are iterative in nature. People often express their preferences in terms of a critique of the current recommendation (e.g., “It doesn’t look good for a date”), requiring some degree of common sense for a preference to be inferred. In this work, we present a method for transforming a user critique into a positive preference (e.g., “I prefer more romantic”) in order to retrieve reviews pertaining to potentially better recommendations (e.g., “Perfect for a romantic dinner”). We leverage a large neural language model (LM) in a few-shot setting to perform critique-to-preference transformation, and we test two methods for retrieving recommendations: one that matches embeddings, and another that fine-tunes an LM for the task. We instantiate this approach in the restaurant domain and evaluate it using a new dataset of restaurant critiques. In an ablation study, we show that utilizing critique-to-preference transformation improves recommendations, and that there are at least three general cases that explain this improved performance.

Neural Network Language Models (NNLMs) generate probability distributions by applying a softmax function to a distance metric formed by taking the dot product of a prediction vector with all word vectors in a high-dimensional embedding space. The dot-product distance metric forms part of the inductive bias of NNLMs. Although NNLMs optimize well with this inductive bias, we show that this results in a sub-optimal ordering of the embedding space that structurally impoverishes some words at the expense of others when assigning probability. We present numerical, theoretical and empirical analyses which show that words on the interior of the convex hull in the embedding space have their probability bounded by the probabilities of the words on the hull.

Representation learning is a critical ingredient for natural language processing systems. Recent Transformer language models like BERT learn powerful textual representations, but these models are targeted towards token- and sentence-level training objectives and do not leverage information on inter-document relatedness, which limits their document-level representation power. For applications on scientific documents, such as classification and recommendation, accurate embeddings of documents are a necessity. We propose SPECTER, a new method to generate document-level embedding of scientific papers based on pretraining a Transformer language model on a powerful signal of document-level relatedness: the citation graph. Unlike existing pretrained language models, Specter can be easily applied to downstream applications without task-specific fine-tuning. Additionally, to encourage further research on document-level models, we introduce SciDocs, a new evaluation benchmark consisting of seven document-level tasks ranging from citation prediction, to document classification and recommendation. We show that Specter outperforms a variety of competitive baselines on the benchmark.

Language models pretrained on text from a wide variety of sources form the foundation of today’s NLP. In light of the success of these broad-coverage models, we investigate whether it is still helpful to tailor a pretrained model to the domain of a target task. We present a study across four domains (biomedical and computer science publications, news, and reviews) and eight classification tasks, showing that a second phase of pretraining in-domain (domain-adaptive pretraining) leads to performance gains, under both high- and low-resource settings. Moreover, adapting to the task’s unlabeled data (task-adaptive pretraining) improves performance even after domain-adaptive pretraining. Finally, we show that adapting to a task corpus augmented using simple data selection strategies is an effective alternative, especially when resources for domain-adaptive pretraining might be unavailable. Overall, we consistently find that multi-phase adaptive pretraining offers large gains in task performance.

Recent advances in commonsense reasoning depend on large-scale human-annotated training sets to achieve peak performance. However, manual curation of training sets is expensive and has been shown to introduce annotation artifacts that neural models can readily exploit and overfit to. We propose a novel generative data augmentation technique, G-DAUGˆC, that aims to achieve more accurate and robust learning in a low-resource setting. Our approach generates synthetic examples using pretrained language models and selects the most informative and diverse set of examples for data augmentation. On experiments with multiple commonsense reasoning benchmarks, G-DAUGˆC consistently outperforms existing data augmentation methods based on back-translation, establishing a new state-of-the-art on WinoGrande, CODAH, and CommonsenseQA, as well as enhances out-of-distribution generalization, proving to be robust against adversaries or perturbations. Our analysis demonstrates that G-DAUGˆC produces a diverse set of fluent training examples, and that its selection and training approaches are important for performance.

Recurrent neural network language models (RNNLM) form a valuable foundation for many NLP systems, but training the models can be computationally expensive, and may take days to train on a large corpus. We explore a technique that uses large corpus n-gram statistics as a regularizer for training a neural network LM on a smaller corpus. In experiments with the Billion-Word and Wikitext corpora, we show that the technique is effective, and more time-efficient than simply training on a larger sequential corpus. We also introduce new strategies for selecting the most informative n-grams, and show that these boost efficiency.

We introduce a novel topic modeling approach based on constructing a semantic set cover for clusters of similar documents. Specifically, our approach first clusters documents using their Tf-Idf representation, and then covers each cluster with a set of topic words based on semantic similarity, defined in terms of a word embedding. Computing a topic cover amounts to solving a minimum set cover problem. Our evaluation compares our topic modeling approach to Latent Dirichlet Allocation (LDA) on three metrics: 1) qualitative topic match, measured using evaluations by Amazon Mechanical Turk (MTurk) workers, 2) performance on classification tasks using each topic model as a sparse feature representation, and 3) topic coherence. We find that qualitative judgments significantly favor our approach, the method outperforms LDA on topic coherence, and is comparable to LDA on document classification tasks.

Commonsense reasoning is a critical AI capability, but it is difficult to construct challenging datasets that test common sense. Recent neural question answering systems, based on large pre-trained models of language, have already achieved near-human-level performance on commonsense knowledge benchmarks. These systems do not possess human-level common sense, but are able to exploit limitations of the datasets to achieve human-level scores. We introduce the CODAH dataset, an adversarially-constructed evaluation dataset for testing common sense. CODAH forms a challenging extension to the recently-proposed SWAG dataset, which tests commonsense knowledge using sentence-completion questions that describe situations observed in video. To produce a more difficult dataset, we introduce a novel procedure for question acquisition in which workers author questions designed to target weaknesses of state-of-the-art neural question answering systems. Workers are rewarded for submissions that models fail to answer correctly both before and after fine-tuning (in cross-validation). We create 2.8k questions via this procedure and evaluate the performance of multiple state-of-the-art question answering systems on our dataset. We observe a significant gap between human performance, which is 95.3%, and the performance of the best baseline accuracy of 65.3% by the OpenAI GPT model.

Recurrent neural network language models (RNNLMs) are the current standard-bearer for statistical language modeling. However, RNNLMs only estimate probabilities for complete sequences of text, whereas some applications require context-independent phrase probabilities instead. In this paper, we study how to compute an RNNLM’s em marginal probability: the probability that the model assigns to a short sequence of text when the preceding context is not known. We introduce a simple method of altering the RNNLM training to make the model more accurate at marginal estimation. Our experiments demonstrate that the technique is effective compared to baselines including the traditional RNNLM probability and an importance sampling approach. Finally, we show how we can use the marginal estimation to improve an RNNLM by training the marginals to match n-gram probabilities from a larger corpus.

We describe a deployed scalable system for organizing published scientific literature into a heterogeneous graph to facilitate algorithmic manipulation and discovery. The resulting literature graph consists of more than 280M nodes, representing papers, authors, entities and various interactions between them (e.g., authorships, citations, entity mentions). We reduce literature graph construction into familiar NLP tasks (e.g., entity extraction and linking), point out research challenges due to differences from standard formulations of these tasks, and report empirical results for each task. The methods described in this paper are used to enable semantic features in www.semanticscholar.org.

Intelligent systems require common sense, but automatically extracting this knowledge from text can be difficult. We propose and assess methods for extracting one type of commonsense knowledge, object-property comparisons, from pre-trained embeddings. In experiments, we show that our approach exceeds the accuracy of previous work but requires substantially less hand-annotated knowledge. Further, we show that an active learning approach that synthesizes common-sense queries can boost accuracy.

We propose an unsupervised importance sampling approach to selecting training data for recurrent neural network (RNNs) language models. To increase the information content of the training set, our approach preferentially samples high perplexity sentences, as determined by an easily queryable n-gram language model. We experimentally evaluate the heldout perplexity of models trained with our various importance sampling distributions. We show that language models trained on data sampled using our proposed approach outperform models trained over randomly sampled subsets of both the Billion Word (Chelba et al., 2014 Wikitext-103 benchmark corpora (Merity et al., 2016).

Many Natural Language Processing (NLP) models rely on distributed vector representations of words. Because the process of training word vectors can require large amounts of data and computation, NLP researchers and practitioners often utilize pre-trained embeddings downloaded from the Web. However, finding the best embeddings for a given task is difficult, and can be computationally prohibitive. We present a framework, called VecShare, that makes it easy to share and retrieve word embeddings on the Web. The framework leverages a public data-sharing infrastructure to host embedding sets, and provides automated mechanisms for retrieving the embeddings most similar to a given corpus. We perform an experimental evaluation of VecShare’s similarity strategies, and show that they are effective at efficiently retrieving embeddings that boost accuracy in a document classification task. Finally, we provide an open-source Python library for using the VecShare framework.