The current standard approach to scaling transformer language models trains each model size from a different random initialization. As an alternative, we consider a staged training setup that begins with a small model and incrementally increases the amount of compute used for training by applying a "growth operator" to increase the model depth and width. By initializing each stage with the output of the previous one, the training process effectively re-uses the compute from prior stages and becomes more efficient. Our growth operators each take as input the entire training state (including model parameters, optimizer state, learning rate schedule, etc.) and output a new training state from which training continues. We identify two important properties of these growth operators, namely that they preserve both the loss and the "training dynamics" after applying the operator. While the loss-preserving property has been discussed previously, to the best of our knowledge this work is the first to identify the importance of preserving the training dynamics (the rate of decrease of the loss during training). To find the optimal schedule for stages, we use the scaling laws from (Kaplan et al., 2020) to find a precise schedule that gives the most compute saving by starting a new stage when training efficiency starts decreasing. We empirically validate our growth operators and staged training for autoregressive language models, showing up to 22% compute savings compared to a strong baseline trained from scratch. Our code is available at https://github.com/allenai/staged-training.
We introduce the LongChecker system for scientific claim verification. Given a scientific claim and an evidence-containing research abstract, LongChecker predicts a veracity label and identifies supporting rationales in a multitask fashion based on a shared encoding of the claim and abstract. We perform experiments on the SciFact dataset, and find that LongChecker achieves state-of-the-art performance. We conduct analysis to understand the source of this improvement, and find that identifying the relationship between a claim and a rationale reporting a scientific finding often requires understanding the context in which the rationale appears. By making labeling decisions based on all available context, LongChecker achieves better performance on cases requiring this type of understanding. In addition, we show that LongChecker is able to leverage weakly-supervised in-domain data to facilitate few-shot domain adaptation for scientific claim verification.
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%, while plausibility of human explanations is 76%. We hope that FEB together with our proposed approach will spur the community to take on the few-shot self-rationalization challenge.
Recently proposed pre-trained generation models achieve strong performance on single-document summarization benchmarks. However, most of them are pre-trained with general-purpose objectives and mainly aim to process single document inputs. In this paper, we propose PRIMER, a pre-trained model for multi-document representation with focus on summarization that reduces the need for dataset-specific architectures and large amounts of fine-tuning labeled data. Specifically, we adopt the Longformer architecture with proper input transformation and global attention to fit for multi-document inputs, and we use Gap Sentence Generation objective with a new strategy to select salient sentences for the whole cluster, called Entity Pyramid, to teach the model to select and aggregate information across a cluster of related documents. With extensive experiments on 6 multi-document summarization datasets from 3 different domains on the zero-shot, few-shot, and full-supervised settings, our model, PRIMER, outperforms current state-of-the-art models on most of these settings with large margins. Code and pre-trained models are released at https://github.com/allenai/PRIMER
Few-shot NLP research is highly active, yet conducted in disjoint research threads with evaluation suites that lack challenging-yet-realistic testing setups and fail to employ careful experimental design. Consequently, the community does not know which techniques perform best or even if they outperform simple baselines. We formulate desiderata for an ideal few-shot NLP benchmark and present FLEX, the first benchmark, public leaderboard, and framework that provides unified, comprehensive measurement for few-shot NLP techniques. FLEX incorporates and introduces new best practices for few-shot evaluation, including measurement of four transfer settings, textual labels for zero-shot evaluation, and a principled approach to benchmark design that optimizes statistical accuracy while keeping evaluation costs accessible to researchers without large compute resources. In addition, we present UniFew, a simple yet strong prompt-based model for few-shot learning which unifies the pretraining and finetuning prompt formats, eschewing complex machinery of recent prompt-based approaches in adapting downstream task formats to language model pretraining objectives. We demonstrate that despite simplicity UniFew achieves results competitive with both popular meta-learning and prompt-based approaches.
Biomedical knowledge graphs (KGs) hold rich information on entities such as diseases, drugs, and genes. Predicting missing links in these graphs can boost many important applications, such as drug design and repurposing. Recent work has shown that general-domain language models (LMs) can serve as "soft" KGs, and that they can be fine-tuned for the task of KG completion. In this work, we study scientific LMs for KG completion, exploring whether we can tap into their latent knowledge to enhance biomedical link prediction. We evaluate several domain-specific LMs, fine-tuning them on datasets centered on drugs and diseases that we represent as KGs and enrich with textual entity descriptions. We integrate the LM-based models with KG embedding models, using a router method that learns to assign each input example to either type of model and provides a substantial boost in performance. Finally, we demonstrate the advantage of LM models in the inductive setting with novel scientific entities. Our datasets and code are made publicly available.
Readers of academic research papers often read with the goal of answering specific questions. Question Answering systems that can answer those questions can make consumption of the content much more efficient. However, building such tools requires data that reflect the difficulty of the task arising from complex reasoning about claims made in multiple parts of a paper. In contrast, existing information-seeking question answering datasets usually contain questions about generic factoid-type information. We therefore present QASPER, a dataset of 5,049 questions over 1,585 Natural Language Processing papers. Each question is written by an NLP practitioner who read only the title and abstract of the corresponding paper, and the question seeks information present in the full text. The questions are then answered by a separate set of NLP practitioners who also provide supporting evidence to answers. We find that existing models that do well on other QA tasks do not perform well on answering these questions, underperforming humans by at least 27 F1 points when answering them from entire papers, motivating further research in document-grounded, information-seeking QA, which our dataset is designed to facilitate.
Determining coreference of concept mentions across multiple documents is fundamental for natural language understanding. Work on cross-document coreference resolution (CDCR) typically considers mentions of events in the news, which do not often involve abstract technical concepts that are prevalent in science and technology. These complex concepts take diverse or ambiguous forms and have many hierarchical levels of granularity (e.g., tasks and subtasks), posing challenges for CDCR. We present a new task of hierarchical CDCR for concepts in scientific papers, with the goal of jointly inferring coreference clusters and hierarchy between them. We create SciCo, an expert-annotated dataset for this task, which is 3X larger than the prominent ECB+ resource. We find that tackling both coreference and hierarchy at once outperforms disjoint models, which we hope will spur development of joint models for SciCo.
To assess the effectiveness of any medical intervention, researchers must conduct a time-intensive and highly manual literature review. NLP systems can help to automate or assist in parts of this expensive process. In support of this goal, we release MS^2 (Multi-Document Summarization of Medical Studies), a dataset of over 470k documents and 20k summaries derived from the scientific literature. This dataset facilitates the development of systems that can assess and aggregate contradictory evidence across multiple studies, and is the first large-scale, publicly available multi-document summarization dataset in the biomedical domain. We experiment with a summarization system based on BART, with promising early results. We formulate our summarization inputs and targets in both free text and structured forms and modify a recently proposed metric to assess the quality of our system's generated summaries. Data and models are available at https://github.com/allenai/ms2
We introduce a new pretraining approach for language models that are geared to support multi-document NLP tasks. Our cross-document language model (CD-LM) improves masked language modeling for these tasks with two key ideas. First, we pretrain with multiple related documents in a single input, via cross-document masking, which encourages the model to learn cross-document and long-range relationships. Second, extending the recent Longformer model, we pretrain with long contexts of several thousand tokens and introduce a new attention pattern that uses sequence-level global attention to predict masked tokens, while retaining the familiar local attention elsewhere. We show that our CD-LM sets new state-of-the-art results for several multi-text tasks, including cross-document event and entity coreference resolution, paper citation recommendation, and documents plagiarism detection, while using a significantly reduced number of training parameters relative to prior works.