We submit to both tracks of the RAG4Reports challenge with two complementary components: PREFNUGGET, which derives concise nugget banks from pairwise preference judgments between system responses, and CRUCIBLE, a nugget-first pipeline that uses such banks to assemble reports on a given topic. The shared nugget-level representation unifies our approach to report evaluation (Task A) and report generation (Task B).

Generative retrieval (GR) maps queries directly to document identifiers (docids) using parametric knowledge. However, this design makes corpus expansion costly: adding new documents requires updating model parameters to encode new document-docid associations incurs repeated training and catastrophic forgetting of previously indexed documents. In this work, we revisit incremental GR as an in-context retrieval problem, where newly added documents are supplied as inference-time document-docid evidence. We propose ICICLE, an in-context indexing framework that performs source-aware docid generation over both parametric memory and context-provided document-docid pairs. ICICLE combines a ‘[COPY]‘-based routing mechanism, preference-based calibration, and large context adaptation to distinguish context-grounded retrieval from parametric retrieval. Experiments on MS MARCO and NQ320K show that ICICLE improves retrieval of newly introduced documents while preserving seen-document retention without corpus-specific retraining. Our analysis further shows that high-shot degradation is mainly caused by routing failure, highlighting source-selection calibration as a key bottleneck for scaling in-context generative retrieval.

We introduce the task of grounded article generation with the goal of creating a Wikipedia-style article from multiple diverse videos about real-world events—from natural disasters to political elections—where all the information in the article is supported by video evidence. Videos are intuitive sources for retrieval-augmented generation (RAG), but most contemporary RAG workflows focus heavily on text while existing methods for video-based summarization focus on low-level scene understanding rather than high-level event semantics. To close this gap, we introduce , a benchmark consisting of expert-written articles and densely annotated videos that provide evidence for articles’ claims, facilitating the integration of video into RAG pipelines and enabling the creation of in-depth content that is grounded in multimodal sources. We further propose Collaborative Article Generation (CAG), a novel interactive method for article creation from multiple videos. CAG leverages an iterative interaction between an r1-style reasoning model and a VideoLLM to draw higher-level inferences about the target event than is possible with VideoLLMs alone, which fixate on low-level visual features. We benchmark state-of-the-art VideoLLMs and CAG in both oracle retrieval and RAG settings and find that CAG consistently outperforms alternative methods, while suggesting intriguing avenues for future work.

Long-form retrieval tasks such as report generation, multi-document summarization, and retrieval-augmented generation require systems to surface evidence covering multiple aspects (subtopics) of a complex query. Existing approaches to information coverage typically fall into two families: (1) methods such as Maximum Marginal Relevance (MMR) that rerank initial retrievals using document-similarity heuristics, and (2) expansion-based methods that invoke large language models at query time to produce subqueries, pseudo-answers, or expanded keywords. We propose SPLADE-COVER, a learned sparse retrieval model whose expansion mechanism is trained to be coverage-aware. SPLADE-COVER inherits the SPLADE-v3 architecture but is trained with an objective that distills multi-aspect subquery supervision into a single sparse representation, requiring no language-model invocation or reranking at query time. On benchmarks with nugget-level judgments, SPLADE-COVER often improves over its base model as measured by both information coverage metrics and relevance metrics, at a fraction of the query-time cost of generation-based methods.

We introduce MiRAGE, an evaluation framework for retrieval-augmented generation (RAG) from multimodal sources. As audiovisual media becomes a more prevalent source of information online, RAG systems must integrate such media into generation. Yet, existing evaluation methods for RAG are largely text-centric and do not readily transfer to multimodal settings. MiRAGE is a claim-centric approach to multimodal RAG evaluation, consisting of InfoF1, which assesses factuality and information coverage, and CiteF1, which assesses citation support and completeness. We show that, when applied by humans, MiRAGE strongly aligns with extrinsic judgments of output quality. We additionally introduce an automatic implementation of MiRAGE and compare it to multimodal variants of three prominent text-centric RAG metrics—ALCE, ARGUE, and RAGAS—finding that MiRAGE outperforms all three on text while being the only one to generalize to multimodal sources. We release open-source implementations and outline evaluation methods for multimodal RAG.

Despite recent advancements in neural retrieval, representing text fragments or phrases with proper contextualized embeddings is still challenging. Particularly in video retrieval, where documents are text extracted through OCR from the frames or ASR from audio tracks, the textual content is rarely complete sentences but only a bag of phrases. In this work, we propose FORTIFY, a generative model fine-tuning approach for noisy document rewriting and summarization, to improve the downstream retrieval effectiveness. By experimenting on MultiVENT 2.0, an informational video retrieval benchmark, we show Llama fine-tuned with FORTIFY provides an effective document expansion, leading to a 30% improvement over prompting an out-of-box Llama model on nDCG@10. Zero-shot transferring the model tailored for MultiVENT 2.0 to two out-of-distribution datasets still demonstrates competitive retrieval effectiveness to other document preprocessing alternatives.

Legal professionals need to write analyses that rely on citations to relevant precedents, i.e., previous case decisions. Intelligence systems assisting legal professionals in writing such documents provide great benefits but are challenging to design. Such systems need to help locate, summarize, and reason over salient precedents in order to be useful. To enable systems for such tasks, we work with legal professionals to create a colossal dataset. supporting two important backbone tasks: information retrieval (IR) and retrieval-augmented generation (RAG). This dataset CLERC (Case Law Evaluation and Retrieval Corpus), is constructed for training and evaluating models on their ability to (1) find corresponding citations for a given piece of legal analysis and to (2) compile the text of these citations (as well as previous context) into a cogent analysis that supports a reasoning goal. We benchmark state-of-the-art models on CLERC, showing that current approaches still struggle: GPT-4o generates analyses with the highest ROUGE F-scores but hallucinates the most, while zero-shot IR models only achieve 48.3% recall@1000.

Despite the recent successes of transformer-based models in terms of effectiveness on a variety of tasks, their decisions often remain opaque to humans. Explanations are particularly important for tasks like offensive language or toxicity detection on social media because a manual appeal process is often in place to dispute automatically flagged content. In this work, we propose a technique to improve the interpretability of these models, based on a simple and powerful assumption: a post is at least as toxic as its most toxic span. We incorporate this assumption into transformer models by scoring a post based on the maximum toxicity of its spans and augmenting the training process to identify correct spans. We find this approach effective and can produce explanations that exceed the quality of those provided by Logistic Regression analysis (often regarded as a highly-interpretable model), according to a human study.

Offensive language detection is an important and challenging task in natural language processing. We present our submissions to the OffensEval 2020 shared task, which includes three English sub-tasks: identifying the presence of offensive language (Sub-task A), identifying the presence of target in offensive language (Sub-task B), and identifying the categories of the target (Sub-task C). Our experiments explore using a domain-tuned contextualized language model (namely, BERT) for this task. We also experiment with different components and configurations (e.g., a multi-view SVM) stacked upon BERT models for specific sub-tasks. Our submissions achieve F1 scores of 91.7% in Sub-task A, 66.5% in Sub-task B, and 63.2% in Sub-task C. We perform an ablation study which reveals that domain tuning considerably improves the classification performance. Furthermore, error analysis shows common misclassification errors made by our model and outlines research directions for future.