Zain Muhammad Mujahid, Dustin Wright, Isabelle Augenstein
Published in arXiv, 2025
We demonstrate that factuality metrics are brittle for evaluating query focused long document summarization.
Download here
Dustin Wright, Zain Muhammad Mujahid, Lu Wang, Isabelle Augenstein, David Jurgens
Published in EMNLP, 2025
We propose the task of unstructured evidence attribution for long context query focused summarization and generate a synthetic dataset (SUnsET) to improve model performance on it.
Download here
Dustin Wright, Christian Igel, Gabrielle Samuel, and Raghavendra Selvan
Published in Communications of the ACM, 2025
We present a perspective on why efficiency will not make AI sustainable and propose systems thinking as a paradigm for the AI community to adopt.
Download here
Jiaxin Pei, Dustin Wright, Isabelle Augenstin, David Jurgens
Published in arXiv, 2025
We survey readers in the US and UK about their perceptions of science news and develop models to predict science engagement based on this.
Download here
Dustin Wright and Isabelle Augenstein
Published in PLOS One, 2025
We perform an empirical analysis and propose new methods for aggregating labels from crowd annotations
Download here
Kevin Roitero, Dustin Wright, Michael Soprano, Isabelle Augenstein, Stefano Mizzaro
Published in SIGIR, 2025
We demonstrate that summaries of webpages shown to crowdworkers can be used to effecient and effective fact-checking.
Download here
Dustin Wright, Sarah Masud, Jared Moore, Srishti Yadav, Maria Antoniak, Peter Ebert Christensen, Chan Young Park, Isabelle Augenstein
Published in arXiv, 2025
We assess the risk of knowledge collapse with LLMs by measuring epistemic diversity across 27 models in multiple settings.
Download here
Jonathan Ivey, Shivani Kumar, Jiayu Liu, Hua Shen, Sushrita Rakshit, Rohan Raju, Haotian Zhang, Aparna Ananthasubramaniam, Junghwan Kim, Bowen Yi, **Dustin Wright**, Abraham Israeli, Anders Giovanni Møller, Lechen Zhang, David Jurgens
Published in arxiv, 2024
We show that simulations of real people in dialogue settings are not faithful
Download here
Dustin Wright*, Arnav Arora*, Nadav Borenstein, Shrishti Yadav, Serge Belongie, and Isabelle Augenstein
Published in EMNLP Findings, 2024
We generate 156,000 responses to 62 political propositions across 6 language models and demonstrate systematic biases in their stances and plain-text responses.
Download here
Dustin Wright, Christian Igel, and Raghavendra Selvan
Published in NeurIPS (Spotlight), 2024
We derive thresholdless pruning rules for structured pruning and empirically demonstrate their automatic pruning capability.
Download here
Amelie Wührl, Dustin Wright, Roman Klinger, and Isabelle Augenstein
Published in Findings of ACL, 2024
We collect a dataset of fine-grained distortions in scientific reporting of findings, and run several analysis and benchmarks on this data.
Download here
Andreas Nugaard Holm, Dustin Wright, and Isabelle Augenstein
Published in Information, 2023
We analyze the tradeoffs between softmax and MC dropout for uncertainty estimation
Download here
Dustin Wright*, Jiaxin Pei*, David Jurgens, and Isabelle Augenstein
Published in EMNLP, 2022
We develop a new dataset and models for measuring information change in science communication, providing improved performance on scientific evidence retrieval and several large scale analyses of science communication.
Download here
Dustin Wright, David Wadden, Kyle Lo, Bailey Kuehl, Arman Cohan, Isabelle Augenstein, and Lucy Lu Wang
Published in ACL, 2022
We develop methods for generating and evaluating atomic, valid scientific claims from citation texts.
Download here
Andreas Nugaard Holm, Barbara Plank, Dustin Wright and Isabelle Augenstein
Published in AAAI 2022 Workshop on Scientific Document Understanding (SDU 2022), 2022
We present a method and dataset for the novel task of predicting the trajectory of citations a paper will receive over time.
Download here
Dustin Wright and Isabelle Augenstein
Published in EMNLP, 2021
We formalize and introduce a test set for exaggeration detection of health science, and propose MT-PET, an extension of Pattern Exploiting Training, to perform the task in a few-shot setting.
Download here
Dustin Wright and Isabelle Augenstein
Published in Findings of ACL, 2021
We introduce the CiteWorth dataset for cite-worthiness detection, provide several strong baselines for the task, and demontrate downstream usefulness of pre-training on cite-worthiness detection.
Download here
Dustin Wright and Isabelle Augenstein
Published in EMNLP, 2020
We demonstrate that when using large pretrained transformer models, mixture of experts methods can lead to significant gains in domain adaptation settings while domain adversarial training does not. We provide evidence that such models are relatively robust across domains, making homogenous predictions despite being fine-tuned on different domains.
Download here
Pepa Atanasova* and Dustin Wright* and Isabelle Augenstein
Published in EMNLP, 2020
We propose a novel method using universal adversarial triggers and GPT-2 to generate difficult adversarial claims for fact checking models which preserve label direction and are semantically coherent, showing that such generated claims easily fool fact checking models.
Dustin Wright and Isabelle Augenstein
Published in Findings of EMNLP, 2020
We show that when performing the task of claim check-worthiness detection, positive-unlabelled learning helps across multiple domains. Additionally, we highlight key similarities and differences in check-worthiness detection datasets.
Download here
Varsha Dave Badal, Dustin Wright, Yannis Katsis, Ho-Cheol Kim, Austin D Swafford, Rob Knight, Chun-Nan Hsu
Published in Microbiome, 2019
We describe and highlight challenges in the construction of knowledge bases for human microbiome-disease associations, surveying relevant literature and providing recommendations for KB construction in this domain going forward.
Download here
Dustin Wright, Yannis Katsis, Raghav Mehta, Chun-Nan Hsu
Published in Automated Knowledge Base Construction, 2019
We develop a lightweight model for performing disease name normalization utilizing pretrained word-embeddings, distant supervision, and a dictionary of disease terms to outperform state of the art on disease name normalization on two datasets. AKBC 2019 Best Application Paper
Download here
Eunjeong Stella Koh, Shlomo Dubnov, Dustin Wright
Published in IEEE 20th International Workshop on Multimedia Signal Processing (MMSP), 2018
We present promising results with symbolic music generation using a variational autoencoder with CNN encoder.
Download here