Tingyu Song, Yanzhao Zhang, Mingxin Li, Zhuoning Guo, Dingkun Long, Pengjun Xie, Siyue Zhang, Yilun Zhao, Shu Wu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Reasoning-Intensive Retrieval (RIR) targets retrieval settings where relevance is mediated by latent inferential links between a query and supporting evidence, rather than semantic similarity.Motivated by the emergent reasoning abilities of Large Language Models (LLMs), recent work integrates these capabilities into the IR field, spanning the entire pipeline from benchmarks to retrievers and rerankers.Despite this progress, the field lacks a systematic framework to organize current efforts and articulate a clear path forward.To provide a clear roadmap for this rapidly growing yet fragmented area, this survey (1) systematizes existing RIR benchmarks by knowledge domains and modalities, providing a detailed analysis of the current landscape; (2) introduces a structured taxonomy that categorizes methods based on where and how reasoning is integrated into the retrieval pipeline, alongside an analysis of their tradeoffs and practical applications; and (3) summarizes challenges and future directions to guide research in this evolving field.

Reasoning-intensive retrieval aims to surface evidence that supports downstream reasoning rather than merely matching topical similarity.This capability is increasingly important for agentic search systems, where retrievers must provide complementary evidence across iterative search and synthesis.However, existing work remains limited on both evaluation and training: benchmarks such as BRIGHT provide narrow gold sets and evaluate retrievers in isolation, while synthetic training corpora often optimize single-passage relevance rather than evidence portfolio construction.We introduce BRIGHT-PRO, an expert-annotated benchmark that expands each query with multi-aspect gold evidence and evaluates retrievers under both static and agentic search protocols.We further construct RTriever-Synth, an aspect-decomposed synthetic corpus that generates complementary positives and positive-conditioned hard negatives, and use it to LoRA fine-tune RTriever-4B from Qwen3-Embedding-4B.Experiments across lexical, general-purpose, and reasoningintensive retrievers show that aspect-aware and agentic evaluation expose behaviors hidden by standard metrics, while RTriever-4B substantially improves over its base model.

Multimodal retrieval enables effective access to information across diverse modalities such as text, images, audio, and video. Audio–Text Retrieval (ATR), which retrieves relevant text given an audio query, is a key component in real-world applications including voice-enabled AI assistants and audio-integrated retrieval systems. Existing ATR methods primarily focus on aligning audio and textual semantics in a shared embedding space through contrastive learning or multimodal large language model backbones, while their robustness under realistic acoustic conditions remains largely unexplored. To bridge this gap, we introduce RoATR, a diagnostic benchmark for evaluating the Robustness of ATR systems, inspired by prior robustness studies in automatic speech recognition. RoATR defines five perturbation categories that cover eleven realistic perturbation types, including background noise, speaker variability, and device distortion. Extensive experiments on state-of-the-art ATR models reveal substantial and heterogeneous performance degradation across perturbation types. For example, LCO-Embedding suffers up to a 28.0% absolute drop in Recall@1 while no degradation under city noise. To improve the robustness of ATR models, we further propose perturbation-augmented training based on perturbed data curated following the same pipeline as RoATR, which achieves an average improvement of 16.0 percentage points in Recall@1.

We present SPORTREASON, a benchmark for retrieval-augmented reasoning on numerical sports questions.Unlike existing benchmarks limited to one or two evidence units, SPORTREASON requires combining and reasoning across free-text, structured tables, and semi-structured infoboxes.We provide 3,000 human-verified QA pairs by repurposing existing QA and table generation datasets, and by prompting large language models (LLMs).Each pair is grounded in multiple evidence from a multi-modal Wikipedia corpus containing 200K knowledge contexts.We evaluate existing retrievers and rerankers, along with agentic Retrieval-Augmented Generation (RAG) systems.The experimental results show that multi-evidence retrieval remains a challenge.Agentic RAG systems (e.g., Search-o1), despite iterative retrieval and reasoning capabilities, fail to improve performance due to imprecise query generation and distracting retrieval information.

Existing approaches typically rely on largescale fine-tuning to adapt LLMs for information reranking tasks, which is computationally expensive.In this work, we demonstrate that modern LLMs can be effectively adapted using only minimal, high-quality supervision.To enable this, we design LIMRANK-SYNTHESIZER, a reusable and open-source pipeline for generating diverse, challenging, and realistic reranking examples.Using this synthetic data, we finetune our reranker model, LIMRANK.We evaluate LIMRANK on two challenging benchmarks, i.e., BRIGHT for reasoning-intensive retrieval and FOLLOWIR for instruction-following retrieval.Our experiments demonstrate that LIMRANK achieves competitive performance, while being trained on less than 5% of the data typically used in prior work.Further ablation studies demonstrate the effectiveness of LIMRANK-SYNTHESIZER and the strong generalization capabilities of LIMRANK across downstream tasks, including scientific literature search and retrieval-augmented generation for knowledge-intensive problem solving. yale-nlp/LimRank

Large language model (LLM)-based embedding models, benefiting from large scale pretraining and post-training, have begun to surpass BERT and T5-based models on generalpurpose text embedding tasks such as document retrieval.However, a fundamental limitation of LLM embeddings lies in the unidirectional attention used during autoregressive pre-training, which misaligns with the bidirectional nature of text embedding tasks.To this end, we propose adopting diffusion language models for text embeddings, motivated by their inherent bidirectional architecture and recent success in matching or surpassing LLMs especially on reasoning tasks.We present the first systematic study of the diffusion language embedding model, which outperforms the LLM-based embedding model by 20% on long-document retrieval, 8% on reasoning-intensive retrieval, 2% on instruction-following retrieval, and achieve competitive performance on traditional text embedding benchmarks.Our analysis verifies that bidirectional attention is crucial for encoding global context in long and complex text.

Long-tail question answering presents significant challenges for large language models (LLMs) due to their limited ability to acquire and accurately recall less common knowledge.Retrieval-augmented generation (RAG) systems have shown great promise in mitigating this limitation by integrating external retrieval mechanisms.However, dense retrieval models often face the same difficulties when generalizing to rare or niche knowledge.In this study, we introduce RPDR, a novel data augmentation framework that selects high-quality easy-to-learn training data, to enhance dense retrievers.Our approach is built around three core components: synthetic data generation, data selection with Round-Trip prediction to identify easy-to-learn instances, and retriever training with these instances.We evaluate RPDR on two longtail retrieval benchmarks, POPQA and ENTI-TYQUESTIONS, demonstrating substantial improvements over existing retrievers like BM25 and Contriver, especially on extremely longtail categories.We identify the strengths and limitations of RPDR through detailed human analysis and propose a dynamic routing mechanism to dynamically route queries to specialized retrieval modules to further improve retrieval performance.

Recent breakthroughs in generative artificial intelligence have triggered a surge in demand for machine learning training, which poses significant cost burdens and environmental challenges due to its substantial energy consumption. Scheduling training jobs among geographically distributed cloud data centers unveils the opportunity to optimize the usage of computing capacity powered by inexpensive and low-carbon energy and address the issue of workload imbalance. To tackle the challenge of multi-objective scheduling, i.e., maximizing GPU utilization while reducing operational costs, we propose an algorithm based on multi-agent reinforcement learning and actor-critic methods to learn the optimal collaborative scheduling strategy through interacting with a cloud system built with real-life workload patterns, energy prices, and carbon intensities. Compared with other algorithms, our proposed method improves the system utility by up to 28.6% attributable to higher GPU utilization, lower energy cost, and less carbon emission.