David Wadden, Kyle Lo, Bailey Kuehl, Arman Cohan, Iz Beltagy, Lucy Lu Wang, Hannaneh Hajishirzi
While research on scientific claim verification has led to the development of powerful systems that appear to approach human performance, these approaches have yet to be tested in a realistic setting against large corpora of scientific literature. Moving to this open-domain evaluation setting, however, poses unique challenges; in particular, it is infeasible to exhaustively annotate all evidence documents. In this work, we present SciFact-Open, a new test collection designed to evaluate the performance of scientific claim verification systems on a corpus of 500K research abstracts. Drawing upon pooling techniques from information retrieval, we collect evidence for scientific claims by pooling and annotating the top predictions of four state-of-the-art scientific claim verification models. We find that systems developed on smaller corpora struggle to generalize to SciFact-Open, exhibiting performance drops of at least 15 F1. In addition, analysis of the evidence in SciFact-Open reveals interesting phenomena likely to appear when claim verification systems are deployed in practice, e.g., cases where the evidence supports only a special case of the claim. Our dataset is available at https://github.com/dwadden/scifact-open.
David Wadden, Shanchuan Lin, Kyle Lo, Lucy Lu Wang, Madeleine van Zuylen, Arman Cohan, Hannaneh Hajishirzi
We introduce scientific claim verification, a new task to select abstracts from the research literature containing evidence that SUPPORTS or REFUTES a given scientific claim, and to identify rationales justifying each decision. To study this task, we construct SciFact, a dataset of 1.4K expert-written scientific claims paired with evidence-containing abstracts annotated with labels and rationales. We develop baseline models for SciFact, and demonstrate that simple domain adaptation techniques substantially improve performance compared to models trained on Wikipedia or political news. We show that our system is able to verify claims related to COVID-19 by identifying evidence from the CORD-19 corpus. Our experiments indicate that SciFact will provide a challenging testbed for the development of new systems designed to retrieve and reason over corpora containing specialized domain knowledge. Data and code for this new task are publicly available at https://github.com/allenai/scifact. A leaderboard and COVID-19 fact-checking demo are available at https://scifact.apps.allenai.org.
David Wadden, Ulme Wennberg, Yi Luan, Hannaneh Hajishirzi
We examine the capabilities of a unified, multi-task framework for three information extraction tasks: named entity recognition, relation extraction, and event extraction. Our framework (called DyGIE++) accomplishes all tasks by enumerating, refining, and scoring text spans designed to capture local (within-sentence) and global (cross-sentence) context. Our framework achieves state-of-the-art results across all tasks, on four datasets from a variety of domains. We perform experiments comparing different techniques to construct span representations. Contextualized embeddings like BERT perform well at capturing relationships among entities in the same or adjacent sentences, while dynamic span graph updates model long-range cross-sentence relationships. For instance, propagating span representations via predicted coreference links can enable the model to disambiguate challenging entity mentions. Our code is publicly available at https://github.com/dwadden/dygiepp and can be easily adapted for new tasks or datasets.
David Wadden, Kyle Lo, Lucy Lu Wang, Arman Cohan, Iz Beltagy, Hannaneh Hajishirzi
The scientific claim verification task requires an NLP system to label scientific documents which Support or Refute an input claim, and to select evidentiary sentences (or rationales) justifying each predicted label. In this work, we present MultiVerS, which predicts a fact-checking label and identifies rationales in a multitask fashion based on a shared encoding of the claim and full document context. This approach accomplishes two key modeling goals. First, it ensures that all relevant contextual information is incorporated into each labeling decision. Second, it enables the model to learn from instances annotated with a document-level fact-checking label, but lacking sentence-level rationales. This allows MultiVerS to perform weakly-supervised domain adaptation by training on scientific documents labeled using high-precision heuristics. Our approach outperforms two competitive baselines on three scientific claim verification datasets, with particularly strong performance in zero / few-shot domain adaptation experiments. Our code and data are available at https://github.com/dwadden/multivers.
David Wadden, Nikita Gupta, Kenton Lee, Kristina Toutanova
We introduce the task of entity-centric query refinement. Given an input query whose answer is a (potentially large) collection of entities, the task output is a small set of query refinements meant to assist the user in efficient domain exploration and entity discovery. We propose a method to create a training dataset for this task. For a given input query, we use an existing knowledge base taxonomy as a source of candidate query refinements, and choose a final set of refinements from among these candidates using a search procedure designed to partition the set of entities answering the input query. We demonstrate that our approach identifies refinement sets which human annotators judge to be interesting, comprehensive, and non-redundant. In addition, we find that a text generation model trained on our newly-constructed dataset is able to offer refinements for novel queries not covered by an existing taxonomy. Our code and data are available at https://github.com/google-research/language/tree/master/language/qresp.
David Wadden, Kejian Shi, Jacob Morrison, Alan Li, Aakanksha Naik, Shruti Singh, Nitzan Barzilay, Kyle Lo, Tom Hope, Luca Soldaini, Shannon Zejiang Shen, Doug Downey, Hannaneh Hajishirzi, Arman Cohan
We present SciRIFF (Scientific Resource for Instruction-Following and Finetuning), a dataset of 137K instruction-following instances for training and evaluation, covering 54 tasks. These tasks span five core scientific literature understanding capabilities: information extraction, summarization, question answering, claim verification, and classification. SciRIFF is unique in being entirely expert-written, high-quality instruction-following dataset for extracting and synthesizing information from research literature across diverse scientific fields. It features complex instructions with long input contexts, detailed task descriptions, and structured outputs. To demonstrate its utility, we finetune a series of large language models (LLMs) using a mix of general-domain and SciRIFF instructions. On nine out-of-distribution held-out tasks (referred to as SciRIFF-Eval), LLMs finetuned on SciRIFF achieve 70.6% average improvement over baselines trained only on general-domain instructions. SciRIFF facilitates the development and evaluation of LLMs to help researchers navigate the rapidly growing body of scientific literature.
David Wadden, Tal August, Qisheng Li, Tim Althoff
Many people struggling with mental health issues are unable to access adequate care due to high costs and a shortage of mental health professionals, leading to a global mental health crisis. Online mental health communities can help mitigate this crisis by offering a scalable, easily accessible alternative to in-person sessions with therapists or support groups. However, people seeking emotional or psychological support online may be especially vulnerable to the kinds of antisocial behavior that sometimes occur in online discussions. Moderation can improve online discourse quality, but we lack an understanding of its effects on online mental health conversations. In this work, we leveraged a natural experiment, occurring across 200,000 messages from 7,000 online mental health conversations, to evaluate the effects of moderation on online mental health discussions. We found that participation in group mental health discussions led to improvements in psychological perspective, and that these improvements were larger in moderated conversations. The presence of a moderator increased user engagement, encouraged users to discuss negative emotions more candidly, and dramatically reduced bad behavior among chat participants. Moderation also encouraged stronger linguistic coordination, which is indicative of trust building. In addition, moderators who remained active in conversations were especially successful in keeping conversations on topic. Our findings suggest that moderation can serve as a valuable tool to improve the efficacy and safety of online mental health conversations. Based on these findings, we discuss implications and trade-offs involved in designing effective online spaces for mental health support.
David Wadden, Kyle Lo
We present an overview of the SciVer shared task, presented at the 2nd Scholarly Document Processing (SDP) workshop at NAACL 2021. In this shared task, systems were provided a scientific claim and a corpus of research abstracts, and asked to identify which articles SUPPORT or REFUTE the claim as well as provide evidentiary sentences justifying those labels. 11 teams made a total of 14 submissions to the shared task leaderboard, leading to an improvement of more than +23 F1 on the primary task evaluation metric. In addition to surveying the participating systems, we provide several insights into modeling approaches to support continued progress and future research on the important and challenging task of scientific claim verification.
Yanai Elazar, Jiayao Zhang, David Wadden, Bo Zhang, Noah A. Smith
What is the effect of releasing a preprint of a paper before it is submitted for peer review? No randomized controlled trial has been conducted, so we turn to observational data to answer this question. We use data from the ICLR conference (2018--2022) and apply methods from causal inference to estimate the effect of arXiving a paper before the reviewing period (early arXiving) on its acceptance to the conference. Adjusting for confounders such as topic, authors, and quality, we may estimate the causal effect. However, since quality is a challenging construct to estimate, we use the negative outcome control method, using paper citation count as a control variable to debias the quality confounding effect. Our results suggest that early arXiving may have a small effect on a paper's chances of acceptance. However, this effect (when existing) does not differ significantly across different groups of authors, as grouped by author citation count and institute rank. This suggests that early arXiving does not provide an advantage to any particular group.
Yizhong Wang, Hamish Ivison, Pradeep Dasigi, Jack Hessel, Tushar Khot, Khyathi Raghavi Chandu, David Wadden, Kelsey MacMillan, Noah A. Smith, Iz Beltagy, Hannaneh Hajishirzi
In this work we explore recent advances in instruction-tuning language models on a range of open instruction-following datasets. Despite recent claims that open models can be on par with state-of-the-art proprietary models, these claims are often accompanied by limited evaluation, making it difficult to compare models across the board and determine the utility of various resources. We provide a large set of instruction-tuned models from 6.7B to 65B parameters in size, trained on 12 instruction datasets ranging from manually curated (e.g., OpenAssistant) to synthetic and distilled (e.g., Alpaca) and systematically evaluate them on their factual knowledge, reasoning, multilinguality, coding, and open-ended instruction following abilities through a collection of automatic, model-based, and human-based metrics. We further introduce Tülu, our best performing instruction-tuned model suite finetuned on a combination of high-quality open resources. Our experiments show that different instruction-tuning datasets can uncover or enhance specific skills, while no single dataset (or combination) provides the best performance across all evaluations. Interestingly, we find that model and human preference-based evaluations fail to reflect differences in model capabilities exposed by benchmark-based evaluations, suggesting the need for the type of systemic evaluation performed in this work. Our evaluations show that the best model in any given evaluation reaches on average 87% of ChatGPT performance, and 73% of GPT-4 performance, suggesting that further investment in building better base models and instruction-tuning data is required to close the gap. We release our instruction-tuned models, including a fully finetuned 65B Tülu, along with our code, data, and evaluation framework at https://github.com/allenai/open-instruct to facilitate future research.
Rahul Nadkarni, David Wadden, Iz Beltagy, Noah A. Smith, Hannaneh Hajishirzi, Tom Hope
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.
Fangyuan Xu, Kyle Lo, Luca Soldaini, Bailey Kuehl, Eunsol Choi, David Wadden
Large language models (LLMs) adapted to follow user instructions are now widely deployed as conversational agents. In this work, we examine one increasingly common instruction-following task: providing writing assistance to compose a long-form answer. To evaluate the capabilities of current LLMs on this task, we construct KIWI, a dataset of knowledge-intensive writing instructions in the scientific domain. Given a research question, an initial model-generated answer and a set of relevant papers, an expert annotator iteratively issues instructions for the model to revise and improve its answer. We collect 1,260 interaction turns from 234 interaction sessions with three state-of-the-art LLMs. Each turn includes a user instruction, a model response, and a human evaluation of the model response. Through a detailed analysis of the collected responses, we find that all models struggle to incorporate new information into an existing answer, and to perform precise and unambiguous edits. Further, we find that models struggle to judge whether their outputs successfully followed user instructions, with accuracy at least 10 points short of human agreement. Our findings indicate that KIWI will be a valuable resource to measure progress and improve LLMs' instruction-following capabilities for knowledge intensive writing tasks.
Team OLMo, Pete Walsh, Luca Soldaini, Dirk Groeneveld, Kyle Lo, Shane Arora, Akshita Bhagia, Yuling Gu, Shengyi Huang, Matt Jordan, Nathan Lambert, Dustin Schwenk, Oyvind Tafjord, Taira Anderson, David Atkinson, Faeze Brahman, Christopher Clark, Pradeep Dasigi, Nouha Dziri, Allyson Ettinger, Michal Guerquin, David Heineman, Hamish Ivison, Pang Wei Koh, Jiacheng Liu, Saumya Malik, William Merrill, Lester James V. Miranda, Jacob Morrison, Tyler Murray, Crystal Nam, Jake Poznanski, Valentina Pyatkin, Aman Rangapur, Michael Schmitz, Sam Skjonsberg, David Wadden, Christopher Wilhelm, Michael Wilson, Luke Zettlemoyer, Ali Farhadi, Noah A. Smith, Hannaneh Hajishirzi
We present OLMo 2, the next generation of our fully open language models. OLMo 2 includes a family of dense autoregressive language models at 7B, 13B and 32B scales with fully released artifacts -- model weights, full training data, training code and recipes, training logs and thousands of intermediate checkpoints. In this work, we describe our modified model architecture and training recipe, focusing on techniques for achieving better training stability and improved per-token efficiency. Our updated pretraining data mixture introduces a new, specialized data mix called Dolmino Mix 1124, which significantly improves model capabilities across many downstream task benchmarks when introduced via late-stage curriculum training (i.e. specialized data during the annealing phase of pretraining). Finally, we incorporate best practices from Tülu 3 to develop OLMo 2-Instruct, focusing on permissive data and extending our final-stage reinforcement learning with verifiable rewards (RLVR). Our OLMo 2 base models sit at the Pareto frontier of performance to training compute, often matching or outperforming open-weight only models like Llama 3.1, Qwen 2.5, and Gemma 2 while using fewer FLOPs and with fully transparent training data, code, and recipe. Our fully open OLMo 2-Instruct models are competitive with open-weight only models of comparable size and even some proprietary models like GPT-3.5 Turbo and GPT 4o Mini.
Ashish Sharma, Kevin Rushton, Inna Wanyin Lin, David Wadden, Khendra G. Lucas, Adam S. Miner, Theresa Nguyen, Tim Althoff
A proven therapeutic technique to overcome negative thoughts is to replace them with a more hopeful "reframed thought." Although therapy can help people practice and learn this Cognitive Reframing of Negative Thoughts, clinician shortages and mental health stigma commonly limit people's access to therapy. In this paper, we conduct a human-centered study of how language models may assist people in reframing negative thoughts. Based on psychology literature, we define a framework of seven linguistic attributes that can be used to reframe a thought. We develop automated metrics to measure these attributes and validate them with expert judgements from mental health practitioners. We collect a dataset of 600 situations, thoughts and reframes from practitioners and use it to train a retrieval-enhanced in-context learning model that effectively generates reframed thoughts and controls their linguistic attributes. To investigate what constitutes a "high-quality" reframe, we conduct an IRB-approved randomized field study on a large mental health website with over 2,000 participants. Amongst other findings, we show that people prefer highly empathic or specific reframes, as opposed to reframes that are overly positive. Our findings provide key implications for the use of LMs to assist people in overcoming negative thoughts.
Hamish Ivison, Yizhong Wang, Valentina Pyatkin, Nathan Lambert, Matthew Peters, Pradeep Dasigi, Joel Jang, David Wadden, Noah A. Smith, Iz Beltagy, Hannaneh Hajishirzi
Since the release of TÜLU [Wang et al., 2023b], open resources for instruction tuning have developed quickly, from better base models to new finetuning techniques. We test and incorporate a number of these advances into TÜLU, resulting in TÜLU 2, a suite of improved TÜLU models for advancing the understanding and best practices of adapting pretrained language models to downstream tasks and user preferences. Concretely, we release: (1) TÜLU-V2-mix, an improved collection of high-quality instruction datasets; (2) TÜLU 2, LLAMA-2 models finetuned on the V2 mixture; (3) TÜLU 2+DPO, TÜLU 2 models trained with direct preference optimization (DPO), including the largest DPO-trained model to date (TÜLU 2+DPO 70B); (4) CODE TÜLU 2, CODE LLAMA models finetuned on our V2 mix that outperform CODE LLAMA and its instruction-tuned variant, CODE LLAMA-Instruct. Our evaluation from multiple perspectives shows that the TÜLU 2 suite achieves state-of-the-art performance among open models and matches or exceeds the performance of GPT-3.5-turbo-0301 on several benchmarks. We release all the checkpoints, data, training and evaluation code to facilitate future open efforts on adapting large language models.
Muhammad Khalifa, David Wadden, Emma Strubell, Honglak Lee, Lu Wang, Iz Beltagy, Hao Peng
Large language models (LLMs) learn a vast amount of knowledge during pretraining, but they are often oblivious to the source(s) of such knowledge. We investigate the problem of intrinsic source citation, where LLMs are required to cite the pretraining source supporting a generated response. Intrinsic source citation can enhance LLM transparency, interpretability, and verifiability. To give LLMs such ability, we explore source-aware training -- a recipe that involves (i) training the LLM to associate unique source document identifiers with the knowledge in each document, followed by (ii) an instruction-tuning stage to teach the LLM to cite a supporting pretraining source when prompted. Source-aware training borrows from existing pretraining/fine-tuning frameworks and requires minimal changes to the model architecture or implementation. Through experiments on synthetic data, we demonstrate that our training recipe can enable faithful attribution to the pretraining data without a substantial impact on the model's perplexity compared to standard pretraining. Our findings also highlight the importance of pretraining data augmentation in achieving attribution. Code and data available here: \url{https://github.com/mukhal/intrinsic-source-citation}
Jian Guan, Jesse Dodge, David Wadden, Minlie Huang, Hao Peng
Recent progress in natural language processing (NLP) owes much to remarkable advances in large language models (LLMs). Nevertheless, LLMs frequently "hallucinate," resulting in non-factual outputs. Our carefully-designed human evaluation substantiates the serious hallucination issue, revealing that even GPT-3.5 produces factual outputs less than 25% of the time. This underscores the importance of fact verifiers in order to measure and incentivize progress. Our systematic investigation affirms that LLMs can be repurposed as effective fact verifiers with strong correlations with human judgments. Surprisingly, FLAN-T5-11B, the least factual generator in our study, performs the best as a fact verifier, even outperforming more capable LLMs like GPT3.5 and ChatGPT. Delving deeper, we analyze the reliance of these LLMs on high-quality evidence, as well as their deficiencies in robustness and generalization ability. Our study presents insights for developing trustworthy generation models.
Dustin Wright, David Wadden, Kyle Lo, Bailey Kuehl, Arman Cohan, Isabelle Augenstein, Lucy Lu Wang
Automated scientific fact checking is difficult due to the complexity of scientific language and a lack of significant amounts of training data, as annotation requires domain expertise. To address this challenge, we propose scientific claim generation, the task of generating one or more atomic and verifiable claims from scientific sentences, and demonstrate its usefulness in zero-shot fact checking for biomedical claims. We propose CLAIMGEN-BART, a new supervised method for generating claims supported by the literature, as well as KBIN, a novel method for generating claim negations. Additionally, we adapt an existing unsupervised entity-centric method of claim generation to biomedical claims, which we call CLAIMGEN-ENTITY. Experiments on zero-shot fact checking demonstrate that both CLAIMGEN-ENTITY and CLAIMGEN-BART, coupled with KBIN, achieve up to 90% performance of fully supervised models trained on manually annotated claims and evidence. A rigorous evaluation study demonstrates significant improvement in generated claim and negation quality over existing baselines
Chao-Chun Hsu, Erin Bransom, Jenna Sparks, Bailey Kuehl, Chenhao Tan, David Wadden, Lucy Lu Wang, Aakanksha Naik
Literature review requires researchers to synthesize a large amount of information and is increasingly challenging as the scientific literature expands. In this work, we investigate the potential of LLMs for producing hierarchical organizations of scientific studies to assist researchers with literature review. We define hierarchical organizations as tree structures where nodes refer to topical categories and every node is linked to the studies assigned to that category. Our naive LLM-based pipeline for hierarchy generation from a set of studies produces promising yet imperfect hierarchies, motivating us to collect CHIME, an expert-curated dataset for this task focused on biomedicine. Given the challenging and time-consuming nature of building hierarchies from scratch, we use a human-in-the-loop process in which experts correct errors (both links between categories and study assignment) in LLM-generated hierarchies. CHIME contains 2,174 LLM-generated hierarchies covering 472 topics, and expert-corrected hierarchies for a subset of 100 topics. Expert corrections allow us to quantify LLM performance, and we find that while they are quite good at generating and organizing categories, their assignment of studies to categories could be improved. We attempt to train a corrector model with human feedback which improves study assignment by 12.6 F1 points. We release our dataset and models to encourage research on developing better assistive tools for literature review.
Abhilasha Ravichander, Shrusti Ghela, David Wadden, Yejin Choi
Despite their impressive ability to generate high-quality and fluent text, generative large language models (LLMs) also produce hallucinations: statements that are misaligned with established world knowledge or provided input context. However, measuring hallucination can be challenging, as having humans verify model generations on-the-fly is both expensive and time-consuming. In this work, we release HALoGEN, a comprehensive hallucination benchmark consisting of: (1) 10,923 prompts for generative models spanning nine domains including programming, scientific attribution, and summarization, and (2) automatic high-precision verifiers for each use case that decompose LLM generations into atomic units, and verify each unit against a high-quality knowledge source. We use this framework to evaluate ~150,000 generations from 14 language models, finding that even the best-performing models are riddled with hallucinations (sometimes up to 86% of generated atomic facts depending on the domain). We further define a novel error classification for LLM hallucinations based on whether they likely stem from incorrect recollection of training data (Type A errors), or incorrect knowledge in training data (Type B errors), or are fabrication (Type C errors). We hope our framework provides a foundation to enable the principled study of why generative models hallucinate, and advances the development of trustworthy large language models.