Distinguishing machine-generated text (MGT) from human-written text (HWT) becomes increasingly important due to potential misuse. However, most supervised detectors often degrade out of domain (OOD) and require large, diverse training sets. In this work, we analyze the linearity and quality of MGT representations and show that simple linear probes outperform a wide range of detectors while being substantially more sample-efficient. We first show that MGT and HWT latent representations are linearly separable in low-dimensional space, and provide a plausible explanation for this separability through systematic differences in their representation quality. Motivated by these insights, we train two variants of simple linear probes and evaluate them across 4 benchmarks against 16 baselines. Probes consistently improve OOD detection (+11 AUC), requiring solely samples to reach near-peak performance. We show that this transferability arises because probes recover a shared latent MGT direction that generalizes across diverse settings. Finally, we demonstrate that probing vectors capture a continuous spectrum of “machineness”, highlighting their potential for fine-grained estimation of AI-edited text. Overall, our work provides insights into latent-space differences between MGT and HWT and demonstrates the potential of linear probes as as robust and sample-efficient MGT detectors. We release our code on github.

The highly formal and negotiated language of United Nations (UN) resolutions presents unique challenges for argument mining. This paper describes our system submitted to the ArgMining 2026 Shared Task: Reconstructing the Reasoning in United Nations Resolutions. Adhering to the strict constraint of utilising open-weight models with ≤ 8 billion parameters, we propose a hybrid, compute-efficient architecture powered by Qwen3-8B. For the preambular-operative classification, we implement a set of deterministic rules related to the specificity of UN documents, supplemented by zero-shot prompting to handle edge-case paragraphs that fall outside these heuristics. For tagging, we implement a hierarchical multi-agent debate system. For relation prediction, we deploy a two-stage retrieve-and-rank pipeline, introducing an asymmetric distance-decay heuristic to model the backward-referencing nature of UN legal texts. Our approach shows that careful pipeline engineering can allow highly constrained models to perform sophisticated argumentative reasoning.

In automated fact-checking (AFC), check-worthiness detection identifies claims requiring verification based on domain-specific criteria. On Wikipedia, this task instantiates as Citation Needed Detection (CND), which flags claims lacking supporting citations. However, existing research has largely overlooked lower-resource languages, and recent AFC pipelines rely on large language models (LLMs), which are inaccessible to low-resource organizations. We introduce MCN, a multilingual CND corpus spanning 18 languages across three resource levels, on which we conduct an extensive study of small decoder-based language models (SLMs). Our experiments show that SLMs fine-tuned with an encoder-style objective substantially outperform prompted LLMs across languages. We further present one of the first studies on cross-lingual CND, demonstrating that SLMs fine-tuned solely on English claims surpass LLMs, even with little to no target-language adaptation. Our findings have important implications for lower-resource Wikipedia communities and suggest that compact, task-specific models are preferable to LLMs for CND.

Given Wikipedia’s role as a trusted source of high-quality, reliable content, there are growing concerns about the proliferation of low-quality machine-generated text (MGT) produced by large language models (LLMs) on its platform. Reliable detection of MGT is therefore essential, yet existing work primarily evaluates MGT detectors on generic generation tasks, rather than on tasks more commonly performed by Wikipedia editors. This misalignment can lead to poor generalisability when applied to real-world Wikipedia contexts.We introduce WETBench, a multilingual, multi-generator, and task-specific benchmark for MGT detection. We define three editing tasks empirically grounded in Wikipedia editors’ perceived use cases for LLM-assisted editing: Paragraph Writing, Summarisation, and Text Style Transfer, which we implement using two new datasets across three languages. For each writing task, we evaluate three prompts, produce MGT across multiple generators using the best-performing prompt, and benchmark diverse detectors.We find that, across settings, training-based detectors achieve an average accuracy of 78%, while zero-shot detectors average 58%. These results demonstrate that detectors struggle with MGT in realistic generation scenarios and underscore the importance of evaluating such models on diverse, task-specific data to assess their reliability in editor-driven contexts.

The evaluation and exchange of large lexicon databases remains a challenge in many NLP applications. Despite the existence of commonly accepted standards for the format and the features used in a lexicon, there is still a lack of precise and interoperable specification requirements about how lexical entries of a particular language should look like, both in terms of the numbers of forms and in terms of features associated with these forms. This paper presents the notion of “lexical masks”, a powerful tool used to evaluate and exchange lexicon databases in many languages.

We propose a new multilingual language model benchmark that is composed of 40+ languages spanning several scripts and linguistic families. With around 40 billion characters, we hope this new resource will accelerate the research of multilingual modeling. We train monolingual causal language models using a state-of-the-art model (Transformer-XL) establishing baselines for many languages. We also introduce the task of multilingual causal language modeling where we train our model on the combined text of 40+ languages from Wikipedia with different vocabulary sizes and evaluate on the languages individually. We released the cleaned-up text of 40+ Wikipedia language editions, the corresponding trained monolingual language models, and several multilingual language models with different fixed vocabulary sizes.