Decoder-only LLM rerankers struggle with long documents: inference is costly and relevance signals can be diluted by irrelevant context. Motivated by an attention analysis indicating a consistent degradation trend when non-relevant text is appended, we propose EviRerank, an evidence-based long-document reranking framework for decoder-only LLMs. EviRerank (i) scores document blocks with a lightweight selector (BM25, bi-encoder, or cross-encoder), (ii) constructs a compact reranking context under a hard token cap by dynamically budgeting evidence blocks with Adaptive Evidence Budgeting (AEB) and adding a global summary cue via Summary Augmentation (SA), and (iii) reranks with a decoder-only LLM. Across TREC DL’19, DL’22, DL’23, and MLDR-zh, EviRerank consistently outperforms full-document LLM reranking and strong block-selection baselines while substantially reducing the required input length. On TREC DL’19, EviRerank reaches up to 0.744 nDCG@10 while substantially reducing reranking latency. Additional RankZephyr-7B experiments confirm transfer to listwise reranking under matched context budgets.

Conversational legal retrieval requires identifying statutes that are better aligned with a user’s case context, even when relevant facts are revealed incrementally and incompletely across dialogue turns. Sparse and dense retrievers may return semantically related statutes that are poorly matched to these factual conditions. We propose SAGA, a training-free legal retrieval framework that injects structured applicability-oriented cues into dense retrieval. SAGA constructs a Case Applicability Frame (CAF) from user-side dialogue context to summarize dispute issues, core facts, and missing facts, and uses an offline Article Applicability Frame (AAF) to represent statute-side issue tags, prerequisites, legal effects, and evidence cues. It combines raw semantic and structural applicability streams through late fusion. Experiments on two Chinese statute retrieval benchmarks, LexRAG and STARD, show that SAGA improves over strong sparse, dense, and LLM-enhanced baselines under a unified full-corpus protocol. SAGA yields statistically significant gains on the conversational LexRAG benchmark and more modest gains on STARD over the strongest dense baseline, improving nDCG@10 from 0.3504 to 0.3696 and from 0.5862 to 0.5940, respectively. These results suggest that structured applicability-oriented cues provide useful retrieval signals beyond raw semantic matching, especially in conversational legal retrieval.

In test-time reasoning, pairing a weak local generator with a strong but expensive teacher makes deployment cost hinge on when additional teacher computation is worthwhile. We propose Cascade-SC, a containment-gated cascade that accepts the teacher’s independently solved answer when it matches the generator’s candidate set and draws additional teacher samples otherwise. The cross-model gate fires roughly 5× less often than within-model unanimity, concentrating teacher compute on harder cases. Across GSM8K, MATH500, and AIME-2024, Cascade-SC improves over self-consistency by 11–37%. On GSM8K, it matches ModelSwitch within 0.46% at roughly one-third the teacher-inference cost and beats matched-compute Teacher-only SC by 1.71%. On low-recall AIME-2024, it reaches 40.00% versus the candidate-restricted Oracle@8 ceiling of 11.11%. It also outperforms trained 7B process-reward rerankers by 4–11% without verifier training. Self-consistency and Teacher-greedy arise as special cases, while a teacher-only versus union-vote ablation shows that aggregation is a separate design choice. Source codes are available at https://github.com/SUGAR-NLP/Cascade-SC.

The semantic gap between colloquial user queries and professional legal documents presents a fundamental challenge in Legal Case Retrieval (LCR). Existing dense retrieval methods typically treat LCR as a black-box semantic matching process, neglecting the explicit juridical logic that underpins legal relevance. To address this, we propose GLIER (Generative Legal Inference and Evidence Ranking), a framework that reformulates retrieval as an inference process over latent legal variables. GLIER decomposes the task into two interpretability-driven stages: (1) A Joint Generative Inference module that translates raw queries into latent legal indicators (Charges and Legal Elements), employing a unified sequence-to-sequence strategy where charges and elements are generated jointly to enforce logical consistency; and (2) A Multi-View Evidence Fusion mechanism that aggregates generative confidence with structural and lexical signals for precise ranking. Extensive experiments on LeCaRD and LeCaRDv2 demonstrate that GLIER outperforms strong baselines like SAILER and KELLER. Notably, our framework exhibits exceptional data efficiency, maintaining robust performance even when trained with only 10% of the data.

Retrieval-Augmented Generation (RAG) grounds language models in external evidence, but multi-hop question answering remains difficult because iterative pipelines must control what to retrieve next and when the available evidence is adequate. In practice, systems may answer from incomplete evidence chains, or they may accumulate redundant or distractor-heavy text that interferes with later retrieval and reasoning. We propose S2G-RAG (Structured Sufficiency and Gap-judging RAG), an iterative framework with an explicit controller, S2G-Judge. At each turn, S2G-Judge predicts whether the current evidence memory supports answering and, if not, outputs structured gap items that describe the missing information. We map these gap items into the next retrieval query, producing stable multi-turn retrieval trajectories. To reduce noise accumulation, we maintain a sentence-level Evidence Context by extracting a compact set of relevant sentences from retrieved documents. Experiments on TriviaQA, HotpotQA, and 2WikiMultiHopQA show that S2G-RAG improves multi-hop QA performance and robustness under multi-turn retrieval. Furthermore, S2G-RAG can be integrated into existing RAG pipelines with a lightweight component, without modifying the search engine or retraining the generator.

Recent studies have demonstrated that the ability of dense retrieval models to generalize to target domains with different distributions is limited, which contrasts with the results obtained with interaction-based models. Prior attempts to mitigate this challenge involved leveraging adversarial learning and query generation approaches, but both approaches nevertheless resulted in limited improvements. In this paper, we propose to combine the query-generation approach with a self-supervision approach in which pseudo-relevance labels are automatically generated on the target domain. To accomplish this, a T5-3B model is utilized for pseudo-positive labeling, and meticulous hard negatives are chosen. We also apply this strategy on conversational dense retrieval model for conversational search. A similar pseudo-labeling approach is used, but with the addition of a query-rewriting module to rewrite conversational queries for subsequent labeling. This proposed approach enables a model’s domain adaptation with real queries and documents from the target dataset. Experiments on standard dense retrieval and conversational dense retrieval models both demonstrate improvements on baseline models when they are fine-tuned on the pseudo-relevance labeled data.

Bien que la recherche d’information neuronale ait connu des améliorations, les modèles de recherche dense ont une capacité de généralisation à de nouveaux domaines limitée, contrairement aux modèles basés sur l’interaction. Les approches d’apprentissage adversarial et de génération de requêtes n’ont pas résolu ce problème. Cet article propose une approche d’auto-supervision utilisant des étiquettes de pseudo-pertinence automatiquement générées pour le domaine cible. Le modèle T53B est utilisé pour réordonner une liste de documents fournie par BM25 afin d’obtenir une annotation des exemples positifs. L’extraction des exemples négatifs est effectuée en explorant différentes stratégies. Les expériences montrent que cette approche aide le modèle dense sur le domaine cible et améliore l’approche de génération de requêtes GPL.