Computational mental health screening using multimodal speech and text has shown great promise. However, existing models often assume all clinical speech protocols carry equivalent evidentiary validity. In reality, heterogeneous protocols, from free interviews to fixed reading tasks, support fundamentally different evidence. Forcing uniform reasoning flattens these boundaries, causing models to hallucinate symptoms from irrelevant text or overclaim support. Even advanced long chain-of-thought LLMs fail to resolve this issue, as free-form reasoning can exacerbate boundary violations. To address this, we reformulate multimodal screening as an evidence-bounded reasoning problem. We introduce the Evidence Package Benchmark, integrating 1,870 packages across six heterogeneous sources with explicit modality masks and evidence permissions. We further propose EviBound, a protocol-aware evidence control framework. Unlike direct LLM prompting, EviBound uses a profile-aware planner to restrict reasoning scope, orchestrates evidence tools via five-way acoustic consensus, and enforces a boundary critic to suppress unsupported claims. Empirical results show EviBound achieves a held-out test Depression AUROC of 0.8658, exceeding the strongest direct omni-modal baseline by +0.0811 AUROC while maintaining zero claim violations. Our work moves beyond unconstrained accuracy toward evidence-consistent, protocol-aware systems for safer clinical NLP research.

Language-model-based agents operating over extended interaction horizons face persistent challenges in preserving temporally grounded information and maintaining behavioral consistency across sessions, a failure mode we term “soul erosion.” We present BMAM (Brain-inspired Multi-Agent Memory), a general-purpose memory architecture that models agent memory as a set of functionally specialized subsystems rather than a single unstructured store. Inspired by cognitive memory systems, BMAM decomposes memory into episodic, semantic, salience-aware, and control-oriented components that operate at complementary time scales, organised as a six-phase memory lifecycle. To support long-horizon reasoning, BMAM organises episodic memories along explicit timelines and retrieves evidence by fusing multiple complementary signals. Experiments on the LoCoMo benchmark show that BMAM achieves 78.45% accuracy, outperforming seven memory-augmented baselines. Pairwise ablations reveal super-additive synergy between brain-region components rather than redundant stacking, and a Soul Portability Test demonstrates 87.5% identity-integrity across full memory export, clear, and restore. A targeted refinement of the temporal-trigger heuristics raises LongMemEval multi-session accuracy from 45.2% to 56.4%, validating the architectural decomposition behind BMAM.Code is available at https://github.com/innovation64/BMAM.

Lifelong model editing seeks to continuously incorporate new knowledge into Large Language Models (LLMs) without costly retraining. Memory-augmented editors preserve pretrained weights to protect unrelated capabilities, but struggle to balance Generalization to rephrased edit queries and Locality on non-edited knowledge. This challenge becomes particularly severe under dense editing settings, where semantically related edits induce substantial memory interference. To tackle this problem, we propose MH-HoReN, a multi-head Hopfield retrieval framework that performs retrieval across complementary subspaces and combines head-wise matching signals for more reliable memory selection. We further extend Hopfield-based editing into a clustered batched scheme that combines shared codebook memories for related edits with token-level pooling for efficient knowledge storage and query routing. Extensive experiments across diverse LLMs and editing benchmarks demonstrate that MH-HoReN achieves better locality-generalization trade-offs as editing scales increase, improving the sum of Locality and Generalization by more than 12% over existing methods while preserving almost perfect Reliability in sequential editing, alongside 7.7% gains in Overall Performance under batched editing. Our code is available here.

A situated query like "where is Lin Wei?" often encodes more than its literal content: the user may also want to know whether Lin Wei is free, in a good mood, or worth interrupting now. Standard tool-use agents answer the literal question and stop. AURA inserts an inference step between scene perception and tool use that produces an IntentFrame: a structured estimate of the implicit need with a scalar gap score that controls per-query probe budget and tool selection. On a 100-query four-scene implicit-intent benchmark, AURA improves implicit-need coverage over ReAct-style probing (Delta = +0.07, p < 10^-6); three of four scenes are individually significant, the gain reproduces on a second backbone, and a prompt ablation attributes the lift to gap calibration rather than answer memorisation. A component ablation localises the gain further: the gap-derived budget ceiling alone outperforms the full architecture (p = 0.02), and bin-wise calibration analysis (r = -0.33) identifies the gap score as a probe trigger rather than a calibrated value estimator. On factual lookup the controller trades raw accuracy for 82% fewer probes and zero forbidden-tool violations on a privacy-sensitive slice; scope conditions are detailed in Limitations. Code, simulator, and benchmark are released at https://github.com/innovation64/AURA, with the implicit-intent benchmark also on the HuggingFace Hub (innovation64/aura-implicit-intent).

Brain tumor MRI diagnosis requires models not only to identify lesions but also to perform fine-grained medical reasoning based on imaging modality, tumor location, and key imaging findings. However, existing medical multimodal datasets often lack structured diagnostic semantics, and closed-ended evaluations may overestimate model ability due to option cues or elimination shortcuts. To address this issue, we introduce MM-NeuroOnco, a large-scale multimodal instruction dataset and benchmark for brain tumor MRI diagnostic reasoning, containing 24,726 MRI slices from 20 public data sources and approximately 200,000 multimodal instructions grounded in imaging evidence. We develop a multi-model collaborative pipeline for semantic completion and quality control, which supplements original tumor labels and segmentation masks with key diagnostic attributes, and further construct MM-NeuroOnco-Bench with a no-LLM-inference policy for ground-truth labels. The benchmark adds a “None of the above” option to closed-ended questions to reduce evaluation bias caused by option shortcuts. Experiments show that existing general-purpose and medical-specialist multimodal large models remain limited on this task, with the strongest baseline achieving only 41.88% accuracy on diagnosis-related questions. Fine-tuning on MM-NeuroOnco yields NeuroOnco-GPT, achieving a 27% absolute accuracy improvement and demonstrating the effectiveness of our dataset in enhancing fine-grained multimodal medical reasoning. Code and dataset documentation are available at https://github.com/gfnnnb/MM-NeuroOnco.

Multi-modal machine translation (MMT) can reduce ambiguity and semantic distortion compared with traditional machine translation (MT) by utilizing auxiliary information such as images. However, current MMT methods face two primary challenges. The first is their underperformance compared to MT methods based on pre-trained models. The second is the inadequate exploitation and integration of the image modality within the model, primarily due to a lack of triplet training data. A mainstream approach is to introduce large amounts of parallel and monolingual data to train the text model and the visual model separately. However, incorporating extensive external data can result in data imbalance, which may introduce biases during training. Additionally, the collection and cleaning of such large datasets is labor-intensive. To overcome these challenges, we introduce a novel, low-cost, large language model-based data augmentation method called LAMBDA, which can enrich the original samples and expand the dataset without requiring external images and text. We propose a fine-grained image captioning module with a noise filter to hierarchically and accurately extract unexploited information from images. Additionally, we design two specific prompts to guide the GPT-3.5 model in generating enriched texts and the corresponding translations. The enriched samples contain diverse text and strong connections between text and images, leading to significant improvements for MMT baselines, with the highest being an increase of up to 3.83 BLEU score and 3.61 METEOR score.