Explore the vast range of titles available.

Automated fault diagnosis can facilitate diagnostics assistance, speedier troubleshooting, and better-organised logistics. Currently, most AI-based prognostics and health management in the automotive industry ignore textual descriptions of the experienced problems or symptoms. With this study, however, we propose an ML-assisted workflow for automotive fault nowcasting that improves on current industry standards. We show that a multilingual pre-trained Transformer model can effectively classify the textual symptom claims from a large company with vehicle fleets, despite the task’s challenging nature due to the 38 languages and 1357 classes involved. Overall, we report an accuracy of more than 80% for high-frequency classes and above 60% for classes with reasonable minimum support, bringing novel evidence that automotive troubleshooting management can benefit from multilingual symptom text classification.

Amid ongoing efforts to develop extremely large, multimodal models, there is increasing interest in efficient Small Language Models (SLMs) that can operate without reliance on large data-centre infrastructure. However, recent SLMs (e.g., LLaMA or Phi) with up to three billion parameters are predominantly trained in high-resource languages, such as English, which limits their applicability to industries that require robust NLP solutions for less-represented languages and low-resource settings, particularly those requiring low latency and adaptability to evolving label spaces. This paper examines a retrieval-based approach to multi-label text classification (MLC) for a media monitoring dataset, with a particular focus on less-represented languages, such as Slovene. This dataset presents an extreme MLC challenge, with instances labelled using up to twelve thousand categories. The proposed method, which combines retrieval with computationally efficient prediction, effectively addresses challenges related to multilinguality, resource constraints, and frequent label changes. We adopt a model-agnostic approach that does not rely on a specific model architecture or language selection. Our results demonstrate that techniques from the extreme multi-label text classification (XMC) domain outperform traditional Transformer-based encoder models, particularly in handling dynamic label spaces without requiring continuous fine-tuning. Additionally, we highlight the effectiveness of this approach in scenarios involving rare labels, where baseline models struggle with generalisation.

Established approaches to analyze multilingual text corpora require either a duplication of analysts’ efforts or high-quality machine translation (MT). In this paper, I argue that multilingual sentence embedding (MSE) is an attractive alternative approach to language-independent text representation. To support this argument, I evaluate MSE for cross-lingual supervised text classification. Specifically, I assess how reliably MSE-based classifiers detect manifesto sentences’ topics and positions compared to classifiers trained using bag-of-words representations of machine-translated texts, and how this depends on the amount of training data. These analyses show that when training data are relatively scarce (e.g., 20K or less-labeled sentences), MSE-based classifiers can be more reliable and are at least no less reliable than their MT-based counterparts. Furthermore, I examine how reliable MSE-based classifiers label sentences written in languages not in the training data, focusing on the task of discriminating sentences that discuss the issue of immigration from those that do not. This analysis shows that compared to the within-language classification benchmark, such “cross-lingual transfer” tends to result in fewer reliability losses when relying on the MSE instead of the MT approach. This study thus presents an important addition to the cross-lingual text analysis toolkit.

Text categorization and sentiment analysis are two of the most typical natural language processing tasks with various emerging applications implemented and utilized in different domains, such as health care and policy making. At the same time, the tremendous growth in the popularity and usage of social media, such as Twitter, has resulted on an immense increase in user-generated data, as mainly represented by the corresponding texts in users’ posts. However, the analysis of these specific data and the extraction of actionable knowledge and added value out of them is a challenging task due to the domain diversity and the high multilingualism that characterizes these data. The latter highlights the emerging need for the implementation and utilization of domain-agnostic and multilingual solutions. To investigate a portion of these challenges this research work performs a comparative analysis of multilingual approaches for classifying both the sentiment and the text of an examined multilingual corpus. In this context, four multilingual BERT-based classifiers and a zero-shot classification approach are utilized and compared in terms of their accuracy and applicability in the classification of multilingual data. Their comparison has unveiled insightful outcomes and has a twofold interpretation. Multilingual BERT-based classifiers achieve high performances and transfer inference when trained and fine-tuned on multilingual data. While also the zero-shot approach presents a novel technique for creating multilingual solutions in a faster, more efficient, and scalable way. It can easily be fitted to new languages and new tasks while achieving relatively good results across many languages. However, when efficiency and scalability are less important than accuracy, it seems that this model, and zero-shot models in general, can not be compared to fine-tuned and trained multilingual BERT-based classifiers.

Urdu and English are widely used for visual text communications worldwide in public spaces such as signboards and navigation boards. Text in such natural scenes contains useful information for modern-era applications such as language translation for foreign visitors, robot navigation, and autonomous vehicles, highlighting the importance of extracting these texts. Previous studies focused on Urdu alone or printed text pasted manually on images and lacked sufficiently large datasets for effective model training. Herein, a pipeline for Urdu and English (bilingual) text detection and recognition in complex natural scene images is proposed. Additionally, a unilingual dataset is converted into a bilingual dataset and augmented using various techniques. For implementations, a customized convolutional neural network is used for feature extraction, a recurrent neural network (RNN) is used for feature learning, and connectionist temporal classification (CTC) is employed for text recognition. Experiments are conducted using different RNNs and hidden units, which yield satisfactory results. Ablation studies are performed on the two best models by eliminating model components. The proposed pipeline is also compared to existing text detection and recognition methods. The proposed models achieved average accuracies of 98.5% for Urdu character recognition, 97.2% for Urdu word recognition, and 99.2% for English character recognition.

Current student-centred, multilingual, active teaching methodologies require that teachers have continuous access to texts that are adequate in terms of topic and language competence. However, the task of finding appropriate materials is arduous and time consuming for teachers. To build on automatic readability assessment research that could help to assist teachers, we explore the performance of natural language processing approaches when dealing with educational science documents for secondary education. Currently, readability assessment is mainly explored in English. In this work we extend our research to Basque and Spanish together with English by compiling context-specific corpora and then testing the performance of feature-based machine-learning and deep learning models. Based on the evaluation of our results, we find that our models do not generalize well although deep learning models obtain better accuracy and F1 in all configurations. Further research in this area is still necessary to determine reliable characteristics of training corpora and model parameters to ensure generalizability.

Sentiment analysis is pivotal in facilitating informed decision-making for businesses, governments, and organizations by comprehending public opinion. However, the task becomes challenging when dealing with linguistic diversity and limited resources for specific languages. This paper presents a novel method, KNetwork, for conducting cross-lingual sentiment analysis of Hindi and English text. The KNetwork leverages the feature vectors generated from translated and transliterated text, aiming to enhance the accuracy of sentiment analysis in cross-lingual settings. Specifically, this paper addresses the challenges associated with sentiment analysis in countries like India, which possess a rich linguistic heritage. The KNetwork model is rigorously evaluated on multiple review datasets, showcasing its performance against state-of-the-art models. Moreover, KNetwork achieves superior results in terms of accuracy of 92.5% and an F1-score of 0.922, outperforming existing models. With an AUC-ROC value of 0.934, it excels in cross-lingual sentiment analysis. This study advances the sentiment analysis for languages with limited resources and underscores the KNetwork’s efficacy in enhancing accuracy, with far-reaching implications for informed decision-making.

Using synthetic data sets to train medicine-focused machine learning models has been shown to enhance their performance; however, most research focuses on English texts. In this paper, we explore generating non-English synthetic medical texts. We propose a methodology for generating medical synthetic data, showcasing it by generating medical texts written in a non-English mixed language. We evaluate our approach with thirteen different language models that are open-source and proprietary, and assess the quality of the data sets in two ways: performing a statistical comparison between the original data set and the generated data sets, and training a classifier to distinguish between original and synthetic examples. The Llama-3.2-3B model achieves the best F1 score of 0.821 ± 0.007 and accuracy of 0.816 ± 0.016, making it most suitable for generating indistinguishable medical synthetic data. In contrast, models like Aya-23, Phi-3, and SmoLLM variants achieve high F1 scores (0.945–0.948), indicating their synthetic data is easily distinguishable from original data. These findings highlight the importance of model selection when generating synthetic medical data sets in non-English languages.

Recent advances in vision-language models such as BLIP-2 have made AI-generated image descriptions increasingly fluent and difficult to distinguish from human-authored texts. This paper investigates whether such differences can still be reliably detected by introducing a novel bilingual dataset of English and Romanian captions. The English subset was derived from the T4SA dataset, while AI-generated captions were produced with BLIP-2 and translated into Romanian using MarianMT; human-written Romanian captions were collected via manual annotation. We analyze the problem from two perspectives: (i) semantic alignment, using CLIP similarity, and (ii) supervised classification with both traditional and transformer-based models. Our results show that BERT achieves over 95% cross-validation accuracy (F1 = 0.95, ROC AUC = 0.99) in distinguishing AI from human texts, while simpler classifiers such as Logistic Regression also reach competitive scores (F1 ≈ 0.88). Beyond classification, semantic and linguistic analyses reveal systematic cross-lingual differences: English captions are significantly longer and more verbose, whereas Romanian texts—often more concise—exhibit higher alignment with visual content. Romanian was chosen as a representative low-resource language, where studying such differences provides insights into multilingual AI detection and challenges in vision-language modeling. These findings emphasize the novelty of our contribution: a publicly available bilingual dataset and the first systematic comparison of human vs. AI-generated captions in both high- and low-resource languages.

This study investigates the application of a state-of-the-art zero-shot and few-shot natural language processing (NLP) technique for text classification tasks in Catalan, a moderately under-resourced language. The approach involves reformulating the downstream task as textual entailment, which is then solved by an entailment model. However, unlike English, where entailment models can be trained on huge Natural Language Inference (NLI) datasets, the lack of such large resources in Catalan poses a challenge. In this context, we comparatively explore training on monolingual and (larger) multilingual resources, and identify the strengths and weaknesses of monolingual and multilingual individual components of entailment models: pre-trained language model and NLI training dataset. Furthermore, we propose and implement a simple task transfer strategy using open Wikipedia resources that demonstrates significant performance improvements, providing a practical and effective alternative for languages with limited or no NLI datasets