Low-resource languages pose persistent challenges for Natural Language Processing tasks such as lemmatization and part-of-speech (POS) tagging. This paper investigates the capacity of recent large language models (LLMs), including GPT-4 variants and open-weight Mistral models, to address these tasks in few-shot and zero-shot settings for four historically and linguistically diverse under-resourced languages: Ancient Greek, Classical Armenian, Old Georgian, and Syriac. Using a novel benchmark comprising aligned training and out-of-domain test corpora, we evaluate the performance of foundation models across lemmatization and POS-tagging, and compare them with PIE, a task-specific RNN baseline. Our results demonstrate that LLMs, even without fine-tuning, achieve competitive or superior performance in POS-tagging and lemmatization across most languages in few-shot settings. Significant challenges persist for languages characterized by complex morphology and non-Latin scripts, but we demonstrate that LLMs are a credible and relevant option for initiating linguistic annotation tasks in the absence of data, serving as an effective aid for annotation.
In this paper, we study the structure of the complete asymptotic expansion of the probability that a large combinatorial object is connected or consists of a given number of connected components. For rapidly growing labeled families of structures, the coefficients involved in these expansions are possibly negative integers. Using species theory, we interpret these coefficients as the difference between the counting sequences of two derivative species of structures. In particular, we show that this difference can be viewed as the counting sequence of the virtual species obtained with the help of an "anti-SEQ" operator applied to the initial family of structures. Applications include P-angulated discrete surfaces, quadratic square-tiled surfaces, and non-orientable graph encoded manifolds, which were not reachable with our previous methods. Moving on to the weighted species, we establish the whole structure of the asymptotic expansion of the probability that a graph is connected in the Erdős-Rényi model G(n,p). Here, the asymptotic coefficients are polynomials in p/1-p and can be described both in terms of simple graphs and irreducible tournaments with ties. We also provide general asymptotic results for sequence and cycle decomposition, as well as the complete asymptotic expansion of the probability that a random labeled tournament with ties is irreducible.
The escalating environmental footprint of the Information and Communication Technol- ogy (ICT) sectorwith global Data Center electric- ity usage potentially reaching 3
As Large Language Models (LLMs) increasingly serve as foundational reasoning engines, their tendency to hallucinate remains a critical vulnerability. While recent internal state probes offer a promising alternative to slow external retrieval systems, they largely reduce hallucination detection to a token-wise binary classification task, failing to capture the structured, sequential boundaries of semantic drift. Here, we introduce an internal hidden state framework for fine-grained, span-level hallucination detection. By inspecting layer-wise activation patterns, we attempt to detect the exact hallucination onset and continuation tokens in an LLM generation. Our experiments show that this approach successfully isolates hallucination onsets, achieving substantial improvements in Precision-Recall AUC over random baselines despite extreme class imbalance. Ultimately, we propose a novel cross-model detection framework in which one model observes the internal representations elicited by another model's generation. We find that an external observer can match or exceed a generator's self-detection of its own hallucination onsets, including when the observer is the smaller model, suggesting that self-detection is not the ceiling for onset localisation.
As systems-on-a-chip (SoCs) that power our devices become increasingly heterogeneous to meet low-power, high-performance computational requirements, they also become more vulnerable to the possible impacts of an on-chip resource availability attack. In this article, we present SoCurity, the first network-on-chip (NoC) counter-based hardware monitoring approach for enhancing heterogeneous SoC security. With this monitoring approach, we develop a fast, lightweight anomalous activity detection and localization system at the hardware level. The presented detection system uses semi-supervised machine learning models, and requires no prior attack knowledge for detecting anomalous activity. This design choice provides protection against existing and novel future attacks impacting on-chip resource availability. Our localization engine leverages the inherent interpretability of SoCurity’s NoC counters to locate detected anomalous activity in the SoC for focused recovery. We demonstrate our techniques with a case study on a real SoC implementation for a connected autonomous vehicle system and find up to 96% detection accuracy and fast detection (∼30µs for prediction on an ASIC) in online experiments. Furthermore, we show that the flagged anomalous activity can be reliably localized for up to 99% of detected anomalous activity in our experiments.
