Classical and neural NLP: translation, question answering, tokenization and the evaluation of language understanding.
arXiv:2609.16312v1 Announce Type: cross Abstract: One-to-many machine translation (MT) is computationally expensive for autoregressive (AR) systems, which suffer from linear latency scaling with both...
Lit3R is a system developed by tus-nlp for the LitTraceQA shared task, which focuses on evidence-grounded question answering over scientific literature. The system combines off-the-shelf retrieval, reranking, and large language model components without task-specific training, using an iterative retrieval process that merges BM25-based sparse and dense retrieval, cross-encoder reranking, and LLM verification, along with paper-to-paper expansion. In the official test set, Lit3R achieved a 4th place ranking on the leaderboard.
The paper investigates counterfactual self‑explanations in large language models, where a model edits an input minimally to change its own prediction. Experiments on sentiment analysis and natural language inference with ten instruction‑tuned models from the LLaMA‑3 and Qwen‑2.5 families show that larger models produce more faithful, minimal, and human‑aligned counterfactuals. While rationale‑guided prompts improve minimality and alignment, they do not consistently enhance faithfulness, indicating that explanation quality depends heavily on model capacity and requires empirical validation.
arXiv:2609.17458v1 Announce Type: cross Abstract: Table understanding is a core task in document intelligence, encompassing two key subtasks: table reconstruction and table visual question answering...
arXiv:2609.16997v1 Announce Type: new Abstract: Crisis sentiment analysis is especially challenging for low-resource languages such as Bangla, where language, context, and public reaction shift rapid...
SPEAR NeXT is a compact, pixel‑wise multimodal spectral‑temporal foundation model that learns temporal self‑supervision by predicting future latent Earth states from past observations. It encodes instantaneous states from optical, radar, and environmental data into 32‑dimensional embeddings, then models their evolution with a causally masked transformer that forecasts multiple future horizons. The model uses Rotary Position Embeddings to capture relative temporal order and month/year embeddings to encode seasonal and interannual context.
MechReason is a new benchmark for multi-image, multi-hop reasoning in mechanical engineering, featuring 12,000 question-answer pairs with explicit reasoning chains and 21,000 visual items across nine evidence types. It covers eight task types—explanation, prediction, design, and diagnosis—within four reasoning dimensions, and is constructed through a four-stage pipeline that extracts engineering claims, generates verification-masked questions, and validates multimodal quality. Even state‑of‑the‑art models score only 62.89% on this dataset, highlighting its difficulty.
SAVOR is a training framework for multimodal large language models that adds token and answer confidence to the output schema, optimises a Group Relative Policy Optimisation objective to penalise calibration error and poor abstention, and uses the learned confidence at inference to revisit visual evidence only when uncertain. Experiments on POPE, HallusionBench, AMBER, and MMHal-Bench with InternVL3-8B and Qwen3-VL-8B backbones show that SAVOR reduces hallucination while maintaining general capability on MME and MMBench, achieving lower Expected Calibration Error than DPO and decoding baselines.
The paper examines how open‑weight language models expose the control tokens used in chat templates, allowing attackers to forge turn boundaries that the model treats as legitimate. An audit of 256 deployed tokenizers shows all are vulnerable, and the commonly recommended flag fails to protect 56.6% of cases. The authors introduce nameless tokenization, which removes surface strings for control identifiers while preserving their internal representation, achieving identical token streams on clean data and significantly improving accuracy on delimiter‑bearing text.
The paper introduces ReImaGin, a method that uses image generation models as a flexible visual reasoning tool for multimodal large language models. Unlike traditional fixed-function vision tools, ReImaGin accepts natural language commands and can perform open-ended visual operations such as removing occlusions or creating floorplans from multiple views. Experiments on six diverse visual reasoning tasks show that ReImaGin outperforms both text-only reasoning and specialist vision-tool baselines, achieving up to a 25% improvement.
AsyncCouple-Flow introduces a new framework for multi‑modal spatio‑temporal forecasting that tackles three key challenges: differing sampling rates, missing modalities, and autoregressive error accumulation. It employs a Modality‑Aware Token Sparsification module to produce equal‑length sequences, an Asynchronous Cross‑Modal Coupling Graph to fuse data under arbitrary asynchrony and missingness, and a Flow‑Matching Forecasting Head that models multi‑step prediction as a conditional ODE. Experiments on weather and traffic datasets demonstrate that the method outperforms state‑of‑the‑art baselines and remains robust even when up to two modalities are missing.
The paper proposes a new approach to single-document extractive summarization by constructing a sentence hypergraph where sentences are nodes and keywords or named entities are hyperedges. A greedy algorithm is then used to find a dominating set of this hypergraph, which yields the sentences that compose the summary. The study compares this hypergraph-based method with existing graph-based summarization techniques.
arXiv:2609.15991v1 Announce Type: new Abstract: Standard subword tokenizers either treat every orthographic variation of a word (such as hello, Hello, HELLO, and H\'ello) as unrelated vocabulary entr...
SciNLP is a new benchmark dataset for full‑text entity and relation extraction in the NLP domain, comprising 60 manually annotated papers with 6,429 entities and 1,649 relations. It is the first dataset to provide full‑text annotations of entities and their relationships specifically for NLP literature. Experiments show that models trained on SciNLP outperform baselines on certain tasks, and the dataset enabled the automatic construction of a fine‑grained knowledge graph with an average node degree of 3.3.
The paper introduces VT-Transformer, a model that predicts answerability scores for Visual Question Answering by treating the task as a regression problem rather than a binary classification. It leverages visual and textual features within a Transformer architecture and demonstrates improved performance and robustness on the VizWiz 2020 dataset compared to existing baselines.
arXiv:2507.23248v2 Announce Type: replace-cross Abstract: Bengali is spoken by more than 230 million people, yet no standardized instrument evaluates large language models (LLMs) on Bengali across th...
The article surveys how large language models (LLMs) can be incorporated into networked control systems, cyber‑physical systems, and multi‑agent networks without violating stability and safety guarantees. It proposes treating the LLM as a slow supervisor that sets high‑level goals, while a fast, certified inner loop preserves physical stability. The survey maps LLM characteristics—such as inference latency, API failures, tokenization, and hallucinations—to classical control challenges and highlights the growing gap between model capability and formal safety assurances, calling for future research on stability proofs.
The paper introduces ‘DP-SPIN’, a trusted‑curator framework that generates differentially private semantic plans for aggregate insight generation. ‘DP-SPIN’ maps each record to a bounded sparse nonnegative vector over pre‑defined semantic concepts, sums these vectors into a semantic sketch, and releases a noisy plan containing admitted concepts and their masses. The framework provides user‑level privacy by clipping each user’s contribution and ensures that the final summary is differentially private through post‑processing, with guarantees established under both add/drop and replacement adjacency.
arXiv:2609.16340v1 Announce Type: cross Abstract: Machine translation systems are periodically upgraded to stronger models, but the available preference signal is human post-edits of an older system'...
The paper introduces ReMova, a pipeline for cleaning Belarusian data and fine‑tuning large language models (LLMs) for English‑to‑Belarusian translation. It uses a correction tool to handle the two orthographies of Belarusian, remove noise, filter out interference from other languages, and correct common misspellings found online. Ablation experiments on unfiltered data show that filtering benefits all fine‑tuned models, with LLM‑based models gaining about twice as much as a dedicated encoder‑decoder MT system, highlighting data quality as a key bottleneck for Belarusian MT.