Patrick Queiroz Da Silva, Hari Sethuraman, Dheeraj Rajagopal, Hannaneh Hajishirzi, Sachin Kumar. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Training data attribution (TDA) methods aim to attribute model outputs back to specific training examples, and the application of these methods to large language model (LLM) outputs could significantly advance model transparency and data curation. However, it has been challenging to date to apply these methods to the full scale of LLM pretraining. In this paper, we refine existing gradient-based methods to work effectively at scale, allowing us to retrieve influential examples for an 8B-parameter language model from a pretraining corpus of over 160B tokens with no need for subsampling or pre-filtering. Our method combines several techniques, including optimizer state correction, a task-specific Hessian approximation, and normalized encodings, which we find to be critical for performance at scale. In quantitative evaluations on a fact tracing task, our method performs best at identifying examples that influence model predictions, but classical, model-agnostic retrieval methods such as BM25 still perform better at finding passages which explicitly contain relevant facts. These results demonstrate a misalignment between factual *attribution* and causal *influence*. With increasing model size and training tokens, we find that influence more closely aligns with factual attribution. Finally, we examine different types of examples identified as influential by our method, finding that while many directly entail a particular fact, others support the same output by reinforcing priors on relation types, common entities, and names. We release our prompt set and model outputs, along with a web-based visualization tool to explore influential examples for factual predictions, commonsense reasoning, arithmetic, and open-ended generation for an 8B-parameter LLM.

Collecting diverse human opinions is costly and challenging.This leads to a recent trend in exploiting large language models (LLMs) for generating diverse data for potential scalable and efficient solutions.However, the extent to which LLMs can generate diverse perspectives on subjective topics is still unclear.In this study, we explore LLMs' capacity of generating diverse perspectives and rationales on subjective topics such as social norms and argumentative texts.We introduce the problem of extracting maximum diversity from LLMs.Motivated by how humans form opinions based on values, we propose a criteria-based prompting technique to ground diverse opinions.To see how far we can extract diverse perspectives from LLMs, or called diversity coverage, we employ a step-by-step recall prompting to generate more outputs from the model iteratively.Our methods, applied to various tasks, show that LLMs can indeed produce diverse opinions according to the degree of task subjectivity.We also find that LLMs performance of extracting maximum diversity is on par with human. 1

Large language models (LLMs) are now available from cloud API providers in various sizes and configurations. While this diversity offers a broad spectrum of choices, effectively leveraging the options to optimize computational cost and performance remains challenging. In this work, we present AutoMix, an approach that strategically routes queries to larger LMs, based on the approximate correctness of outputs from a smaller LM. Central to AutoMix are two key technical contributions. First, it has a few-shot self-verification mechanism, which estimates the reliability of its own outputs without requiring extensive training. Second, given that self-verification can be noisy, it employs a POMDP based router that can effectively select an appropriately sized model, based on answer confidence. Experiments across five language models and five challenging datasets show that Automix consistently surpasses strong baselines, reducing computational cost by over 50\% for comparable performance.

Large pre-trained language models have achieved impressive results on various style classification tasks, but they often learn spurious domain-specific words to make predictions (Hayati et al., 2021).While human explanation highlights stylistic tokens as important features for this task, we observe that model explanations often do not align with them.To tackle this issue, we introduce StyLEx, a model that learns from human annotated explanations of stylistic features and jointly learns to perform the task and predict these features as model explanations.Our experiments show that StyLEx can provide human-like stylistic lexical explanations without sacrificing the performance of sentence-level style prediction on both indomain and out-of-domain datasets.Explanations from StyLEx show significant improvements in explanation metrics (sufficiency, plausibility) and when evaluated with human annotations.They are also more understandable by human judges compared to the widely-used saliency-based explanation baseline.1 * currently at Google

Conditional set generation learns a mapping from an input sequence of tokens to a set.Several NLP tasks, such as entity typing and dialogue emotion tagging, are instances of set generation.SEQ2SEQ models, a popular choice for set generation, treat a set as a sequence and do not fully leverage its key properties, namely order-invariance and cardinality.We propose a novel algorithm for effectively sampling informative orders over the combinatorial space of label orders.We jointly model the set cardinality and output by prepending the set size and taking advantage of the autoregressive factorization used by SEQ2SEQ models.Our method is a model-independent data augmentation approach that endows any SEQ2SEQ model with the signals of order-invariance and cardinality.Training a SEQ2SEQ model on this augmented data (without any additional annotations) gets an average relative improvement of 20% on four benchmark datasets across various models: BART-base, T5-11B, and GPT3-175B. 1

Document authoring involves a lengthy revision process, marked by individual edits that are frequently linked to comments. Modeling the relationship between edits and comments leads to a better understanding of document evolution, potentially benefiting applications such as content summarization, and task triaging. Prior work on understanding revisions has primarily focused on classifying edit intents, but falling short of a deeper understanding of the nature of these edits. In this paper, we present explore the challenge of describing an edit at two levels: identifying the edit intent, and describing the edit using free-form text. We begin by defining a taxonomy of general edit intents and introduce a new dataset of full revision histories of Wikipedia pages, annotated with each revision’s edit intent. Using this dataset, we train a classifier that achieves a 90% accuracy in identifying edit intent. We use this classifier to train a distantly-supervised model that generates a high-level description of a revision in free-form text. Our experimental results show that incorporating edit intent information aids in generating better edit descriptions. We establish a set of baselines for the edit description task, achieving a best score of 28 ROUGE, thus demonstrating the effectiveness of our layered approach to edit understanding.

Vidhisha Balachandran, Artidoro Pagnoni, Jay Yoon Lee, Dheeraj Rajagopal, Jaime Carbonell, Yulia Tsvetkov. Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume. 2021.

Defeasible reasoning is the mode of reasoning where conclusions can be overturned by taking into account new evidence.Existing cognitive science literature on defeasible reasoning suggests that a person forms a mental model of the problem scenario before answering questions.Our research goal asks whether neural models can similarly benefit from envisioning the question scenario before answering a defeasible query.Our approach is, given a question, to have a model first create a graph of relevant influences, and then leverage that graph as an additional input when answering the question.Our system, CURIOUS, achieves a new stateof-the-art on three different defeasible reasoning datasets.This result is significant as it illustrates that performance can be improved by guiding a system to "think about" a question and explicitly model the scenario, rather than answering reflexively. 1

We introduce SELFEXPLAIN, a novel selfexplaining model that explains a text classifier's predictions using phrase-based concepts.SELFEXPLAIN augments existing neural classifiers by adding (1) a globally interpretable layer that identifies the most influential concepts in the training set for a given sample and (2) a locally interpretable layer that quantifies the contribution of each local input concept by computing a relevance score relative to the predicted label.Experiments across five text-classification datasets show that SELFEX-PLAIN facilitates interpretability without sacrificing performance.Most importantly, explanations from SELFEXPLAIN show sufficiency for model predictions and are perceived as adequate, trustworthy and understandable by human judges compared to existing widely-used baselines.1

Niket Tandon, Keisuke Sakaguchi, Bhavana Dalvi, Dheeraj Rajagopal, Peter Clark, Michal Guerquin, Kyle Richardson, Eduard Hovy. Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP). 2020.

Xuchao Zhang, Dheeraj Rajagopal, Michael Gamon, Sujay Kumar Jauhar, ChangTien Lu. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

To perform tasks specified by natural language instructions, autonomous agents need to extract semantically meaningful representations of language and map it to visual elements and actions in the environment. This problem is called task-oriented language grounding. We propose an end-to-end trainable neural architecture for task-oriented language grounding in 3D environments which assumes no prior linguistic or perceptual knowledge and requires only raw pixels from the environment and the natural language instruction as input. The proposed model combines the image and text representations using a Gated-Attention mechanism and learns a policy to execute the natural language instruction using standard reinforcement and imitation learning methods. We show the effectiveness of the proposed model on unseen instructions as well as unseen maps, both quantitatively and qualitatively. We also introduce a novel environment based on a 3D game engine to simulate the challenges of task-oriented language grounding over a rich set of instructions and environment states.

We present a novel approach to automated question generation that improves upon prior work both from a technology perspective and from an assessment perspective. Our system is aimed at engaging language learners by generating multiple-choice questions which utilize specific inference steps over multiple sentences, namely coreference resolution and paraphrase detection. The system also generates correct answers and semantically-motivated phrase-level distractors as answer choices. Evaluation by human annotators indicates that our approach requires a larger number of inference steps, which necessitate deeper semantic understanding of texts than a traditional single-sentence approach.

SenticNet is a publicly available semantic and affective resource for concept-level sentiment analysis. Rather than using graph-mining and dimensionality-reduction techniques, SenticNet 3 makes use of "energy flows" to connect various parts of extended common and common-sense knowledge representations to one another. SenticNet 3 models nuanced semantics and sentics (that is, the conceptual and affective information associated with multi-word natural language expressions), representing information with a symbolic opacity of an intermediate nature between that of neural networks and typical symbolic systems.