The paper proposes a content‑based addressing scheme for long‑context models that replaces the growing token counter in Rotary Position Embedding (RoPE) with unit‑level addresses derived from the content of each unit. By dividing the token stream into units, the method preserves local RoPE behavior while allowing new units to be addressed via learned content maps, avoiding positional mismatches when extending context length. Experiments on character‑level Tiny Shakespeare show that a model trained on 256‑character contexts achieves lower perplexity at 4096 characters using this scheme, and a second diagnostic demonstrates retrieval of multiple serialized facts.
By Mahesh Godavarti
The paper introduces GAPE (Gated Adaptive Positional Encoding), a drop‑in augmentation for rotary positional encodings that adds content‑aware bias to attention logits while preserving rotary geometry. GAPE uses a query‑dependent gate to contract irrelevant context and a key‑dependent gate to keep salient distant tokens, leading to exponential attenuation of weakly protected context. Experiments show that GAPE improves long‑context robustness, extrapolating up to eight times the training length, and can be retrofitted into a pretrained 7B model without losing benchmark performance.
By Riccardo Ali, Alessio Borgi, Mario Severino, Pietro Li\`o, Christopher Irwin
The paper introduces STAR, a metric that measures sentence-level alignment between source and target documents in document-to-document machine translation. Using STAR, the authors develop StarPO, a preference‑optimization framework that ranks translation hypotheses by structural quality and applies a dynamic alignment mask to focus learning on misaligned segments. Experiments on news and literary data show that StarPO improves both translation quality and structural integrity, enabling small models to outperform large proprietary systems such as GPT‑4o while remaining more token‑efficient.
By Yichen Dong, Hao Wang, Junhui Li, Linlong Xu, Longyue Wang, Weihua Luo
The paper introduces LLM-Microscope, a toolkit for measuring how Large Language Models encode contextual information at the token level. It shows that seemingly minor tokens—such as determiners, stopwords, and punctuation—carry surprisingly high contextual weight, and removing them degrades performance on benchmarks like MMLU and BABILong-4k. The study also finds a strong link between contextualization and linearity, indicating that the transformation between layers can be approximated by a single linear mapping when tokens are well contextualized.
By Anton Razzhigaev, Matvey Mikhalchuk, Temurbek Rahmatullaev, Elizaveta Goncharova, Polina Druzhinina, Ivan Oseledets, Andrey Kuznetsov
arXiv:2602.13194v3 Announce Type: replace-cross
Abstract: Humans and large language models can predict next letter or word from its prior context much better than random guessing, indicating strong r...
By Weishun Zhong, Doron Sivan, Tankut Can, Mikhail Katkov, Misha Tsodyks
The paper introduces SCSP, a training‑free framework that improves long‑context embeddings by selectively pooling informative tokens. SCSP partitions documents into sentence‑aware chunks, adds a semantic compression prompt to each chunk, and uses prompt‑isolated attention masks to estimate token importance. The selected tokens’ intermediate‑layer representations are aggregated to form the final embedding, yielding consistent performance gains across zero‑shot and fine‑tuned models on long‑context benchmarks.
By Zifeng Cheng, Jie Zheng, Zhiwei Jiang, Shuwen Wang, Fei Shen, Shiping Ge, Qing Gu
arXiv:2601. 04098v2 Announce Type: replace-cross Abstract: Transformer language models systematically prefer tokens at specific input positions regardless of semantic relevance---a phenomenon known as positional bias.
By Maryam Rahimi, Mahdi Nouri, Yadollah Yaghoobzadeh
MGAL is a new multilingual benchmark for evaluating long‑context large language models, built from United Nations reports in six official UN languages and covering 8K to 128K tokens. It tests four linguistic granularities—word, sentence, paragraph, and document—while also stratifying examples by their position within the document (begin, middle, end). Experiments show that models excel at word‑level tasks but struggle with coarser granularity, and that closed‑source models outperform others in lower‑resource languages, revealing challenges such as local semantic crowding and a fluency‑consistency gap.
By Chunhan Li, Chenglin Xu, Zongyang Zhang, Jiale Liu, Zhuoxi Rao, Xudong Jia, Junxiu He, Menglin Yang, Wenjuan Gong, Zhengzhe Liu, Chengwei Qin
arXiv:2609.34240v2 Announce Type: replace-cross
Abstract: Existing open-ended generation metrics measure likelihood, lexical diversity, or distributional similarity in generic representation space, y...
By Jinnuo Liu, Junhao Zhu, Weifeng Jiang, Haoming Liu, Hongyi Wen
The paper proposes an encoder Transformer that explicitly separates semantic, absolute positional (AP), and relative positional (RP) information, restricting the masked‑language‑modeling objective to the semantic stream. This disentanglement reveals that the AP subspace collapses into a low‑frequency two‑dimensional manifold reflecting document structure, that attention heads specialize into structure‑ and semantic‑oriented groups with RP supporting only the latter, and that standard positional encodings fail to robustly encode macroscopic structure. The approach preserves positional encoding and improves performance on 49 out of 65 linguistic phenomena in the Flash‑Holmes probing benchmark.
By Pierre-Antoine Lequeu, Camille Barboule, Benjamin Piwowarski
The paper investigates why supervised AI‑text detectors, specifically a RoBERTa‑based model, can be fooled by subtle changes in language. By applying semantic, structural, and tokenizer‑level perturbations to a large dataset and controlled Mistral‑7B‑Instruct outputs, the authors show that increasing verb diversity makes machine text easier to detect and that detection scores correlate with statistical complexity, leading to a high false‑positive rate on formal human writing. They also evaluate an event‑based latent space detector, finding that paraphrasing and homoglyphs significantly alter extracted event sequences and verbs, yet the detector’s performance remains modest (AUC 0.577).
By Claudiu Creanga, Liviu Dinu
arXiv:2512. 14391v3 Announce Type: replace-cross Abstract: In-context learning is fundamental to modern Large Language Models (LLMs); however, prevailing architectures impose a rigid and fixed contextual structure by assigning linear or constant positional indices.
By Huayang Li, Tianyu Zhao, Deng Cai, Richard Sproat