arXiv subjects

Serge Sharoff

Publications and source records attributed to Serge Sharoff.

At least 19 recordsLinked to original sources

arXiv2026

UOL@IDEM at BEA 2026 Shared Task 1: Neural Fusion and Feature-Rich Modeling for L1-Aware Vocabulary Difficulty Prediction

This paper describes UOL@IDEM's closed-track submission to the BEA 2026 shared task on L1-aware vocabulary difficulty prediction. We model the task as regression and train separate systems for Spanish, German, and Mandarin Chinese\footnote{Below we use \emph{Chinese} for brevity.}. Our system combines multilingual contextual representations with engineered features capturing frequency, surface form, retrieval evidence, semantic alignment, cognate similarity, and masked-language-model predictability. Development results show consistent gains over the official closed-track baselines, with sentence-embedding encoders such as BGE-M3, multilingual E5, and LaBSE performing best. Official submissions achieve RMSE scores of 1.132, 1.037, and 0.891 for Spanish, German, and Chinese, respectively. Feature analysis identifies frequency as the most stable predictor, while contextual predictability, form similarity, retrieval, and semantic features provide complementary L1-sensitive signals. Error analysis shows strong ranking performance but weaker calibration for the easiest items, which are often overpredicted. See https://github.com/Nouran-Khallaf/UoL-IDEM-BEA2026-Vocabulary-Difficulty-Prediction

cs.CL↗

arXiv2026

The Almost Intelligent Revolution: Options for Scaling Up Deliberation and Empowering People with AI

The increasing prominence of Large Language Models (LLMs) in public discourse presents both opportunities and challenges for democratic deliberation. While red teaming strategies help mitigate specific risks, broader concerns persist regarding linguistic constraints, biases, and the sycophantic tendencies of LLMs. This chapter explores how LLMs can be used to significantly scale up and democratise deliberation, particularly in fostering inclusivity and empowering traditionally marginalised groups. Drawing on concepts from Systemic-Functional Linguistics, the chapter examines how variations across language users (for example, with respect to socio-demographic groups) and across language use (for example, with respect to communicative functions) shape participation in AI-supported deliberation. The chapter presents AI-driven deliberation studies and assesses their potential to scaffold argumentation, enhance access, and reduce the influence of exclusionary linguistic norms and biases which are embedded in prestigious registers. At the same time, the chapter cautions against both overclaiming, which leads to unrealistic expectations, and underclaiming, which risks missed opportunities for AI-assisted engagement. The chapter concludes by identifying future research directions to maximise the democratic potential of AI-assisted participation while embedding ethical safeguards to counteract the reproduction of linguistic inequalities.

cs.CL↗

arXiv2026

Align and Shine: Building High-Quality Sentence-Aligned Corpora for Multilingual Text Simplification

Text simplification plays a crucial role in improving the accessibility and comprehensibility of written information for diverse audiences, including language learners and readers with limited literacy. Despite its importance, large-scale, high-quality datasets for training and evaluating text simplification models remain scarce for languages other than English. This paper reports an experimental study on the collection and processing of crowd-sourced simplification data from comparable corpora to construct a corpus suitable for both training and testing text simplification systems across multiple languages (Catalan, English, French, Italian and Spanish). We report mechanisms for sentence-level alignment from document-level data. The resulting dataset of the aligned sentence pairs is publicly available.

cs.CL↗

arXiv2026

To Predict or Not to Predict? Towards reliable uncertainty estimation in the presence of noise

This study examines the role of uncertainty estimation (UE) methods in multilingual text classification under noisy and non-topical conditions. Using a complex-vs-simple sentence classification task across several languages, we evaluate a range of UE techniques against a range of metrics to assess their contribution to making more robust predictions. Results indicate that while methods relying on softmax outputs remain competitive in high-resource in-domain settings, their reliability declines in low-resource or domain-shift scenarios. In contrast, Monte Carlo dropout approaches demonstrate consistently strong performance across all languages, offering more robust calibration, stable decision thresholds, and greater discriminative power even under adverse conditions. We further demonstrate the positive impact of UE on non-topical classification: abstaining from predicting the 10\% most uncertain instances increases the macro F1 score from 0.81 to 0.85 in the Readme task. By integrating UE with trustworthiness metrics, this study provides actionable insights for developing more reliable NLP systems in real-world multilingual environments. See https://github.com/Nouran-Khallaf/To-Predict-or-Not-to-Predict

cs.CL↗

arXiv2026

How Much Noise Can BERT Handle? Insights from Multilingual Sentence Difficulty Detection

Noisy training data can significantly degrade the performance of language-model-based classifiers, particularly in non-topical classification tasks. In this study we designed a methodological framework to assess the impact of denoising. More specifically, we explored a range of denoising strategies for sentence-level difficulty detection, using training data derived from document-level difficulty annotations obtained through noisy crowdsourcing. Beyond monolingual settings, we also address cross-lingual transfer, where a multilingual language model is trained in one language and tested in another. We evaluate several noise reduction techniques, including Gaussian Mixture Models (GMM), Co-Teaching, Noise Transition Matrices, and Label Smoothing. Our results indicate that while BERT-based models exhibit inherent robustness to noise, incorporating explicit noise detection can further enhance performance. For our smaller dataset, GMM-based noise filtering proves particularly effective in improving prediction quality by raising the Area-Under-the-Curve score from 0.52 to 0.92, or to 0.93 when de-noising methods are combined. However, for our larger dataset, the intrinsic regularisation of pre-trained language models provides a strong baseline, with denoising methods yielding only marginal gains (from 0.92 to 0.94, while a combination of two denoising methods made no contribution). Nonetheless, removing noisy sentences (about 20\% of the dataset) helps in producing a cleaner corpus with fewer infelicities. As a result we have released the largest multilingual corpus for sentence difficulty prediction: see https://github.com/Nouran-Khallaf/denoising-difficulty

cs.CL↗

arXiv2026

Can LLM Reasoning Be Trusted? A Comparative Study: Using Human Benchmarking on Statistical Tasks

This paper investigates the ability of large language models (LLMs) to solve statistical tasks, as well as their capacity to assess the quality of reasoning. While state-of-the-art LLMs have demonstrated remarkable performance in a range of NLP tasks, their competence in addressing even moderately complex statistical challenges is not well understood. We have fine-tuned selected open-source LLMs on a specially developed dataset to enhance their statistical reasoning capabilities, and compared their performance with the human scores used as a benchmark. Our results show that the fine-tuned models achieve better performance on advanced statistical tasks on the level comparable to a statistics student. Fine-tuning demonstrates architecture-dependent improvements, with some models showing significant performance gains, indicating clear potential for deployment in educational technology and statistical analysis assistance systems. We also show that LLMs themselves can be far better judges of the answers quality (including explanation and reasoning assessment) in comparison to traditional metrics, such as BLEU or BertScore. This self-evaluation capability enables scalable automated assessment for statistical education platforms and quality assurance in automated analysis tools. Potential applications also include validation tools for research methodology in academic and industry settings, and quality control mechanisms for data analysis workflows.

cs.CL↗

arXiv2026

Almost Clinical: Linguistic properties of synthetic electronic health records

This study evaluates the linguistic and clinical suitability of synthetic electronic health records in mental health. First, we describe the rationale and the methodology for creating the synthetic corpus. Second, we examine expressions of agency, modality, and information flow across four clinical genres (Assessments, Correspondence, Referrals and Care plans) with the aim to understand how LLMs grammatically construct medical authority and patient agency through linguistic choices. While LLMs produce coherent, terminology-appropriate texts that approximate clinical practice, systematic divergences remain, including registerial shifts, insufficient clinical specificity, and inaccuracies in medication use and diagnostic procedures. The results show both the potential and limitations of synthetic corpora for enabling large-scale linguistic research otherwise impossible with genuine patient records.

cs.CL↗

arXiv2025

Reading Between the Lines: A dataset and a study on why some texts are tougher than others

Our research aims at better understanding what makes a text difficult to read for specific audiences with intellectual disabilities, more specifically, people who have limitations in cognitive functioning, such as reading and understanding skills, an IQ below 70, and challenges in conceptual domains. We introduce a scheme for the annotation of difficulties which is based on empirical research in psychology as well as on research in translation studies. The paper describes the annotated dataset, primarily derived from the parallel texts (standard English and Easy to Read English translations) made available online. we fine-tuned four different pre-trained transformer models to perform the task of multiclass classification to predict the strategies required for simplification. We also investigate the possibility to interpret the decisions of this language model when it is aimed at predicting the difficulty of sentences. The resources are available from https://github.com/Nouran-Khallaf/why-tough

cs.CL↗

arXiv2024

Controlling Out-of-Domain Gaps in LLMs for Genre Classification and Generated Text Detection

This study demonstrates that the modern generation of Large Language Models (LLMs, such as GPT-4) suffers from the same out-of-domain (OOD) performance gap observed in prior research on pre-trained Language Models (PLMs, such as BERT). We demonstrate this across two non-topical classification tasks: 1) genre classification and 2) generated text detection. Our results show that when demonstration examples for In-Context Learning (ICL) come from one domain (e.g., travel) and the system is tested on another domain (e.g., history), classification performance declines significantly. To address this, we introduce a method that controls which predictive indicators are used and which are excluded during classification. For the two tasks studied here, this ensures that topical features are omitted, while the model is guided to focus on stylistic rather than content-based attributes. This approach reduces the OOD gap by up to 20 percentage points in a few-shot setup. Straightforward Chain-of-Thought (CoT) methods, used as the baseline, prove insufficient, while our approach consistently enhances domain transfer performance.

cs.CL↗

arXiv2023

BERT Goes Off-Topic: Investigating the Domain Transfer Challenge using Genre Classification

While performance of many text classification tasks has been recently improved due to Pre-trained Language Models (PLMs), in this paper we show that they still suffer from a performance gap when the underlying distribution of topics changes. For example, a genre classifier trained on \textit{political} topics often fails when tested on documents about \textit{sport} or \textit{medicine}. In this work, we quantify this phenomenon empirically with a large corpus and a large set of topics. Consequently, we verify that domain transfer remains challenging both for classic PLMs, such as BERT, and for modern large models, such as GPT-3. We also suggest and successfully test a possible remedy: after augmenting the training dataset with topically-controlled synthetic texts, the F1 score improves by up to 50\% for some topics, nearing on-topic training results, while others show little to no improvement. While our empirical results focus on genre classification, our methodology is applicable to other classification tasks such as gender, authorship, or sentiment classification. The code and data to replicate the experiments are available at https://github.com/dminus1/genre

cs.CL↗

arXiv2023

Beyond Images: An Integrative Multi-modal Approach to Chest X-Ray Report Generation

Image-to-text radiology report generation aims to automatically produce radiology reports that describe the findings in medical images. Most existing methods focus solely on the image data, disregarding the other patient information accessible to radiologists. In this paper, we present a novel multi-modal deep neural network framework for generating chest X-rays reports by integrating structured patient data, such as vital signs and symptoms, alongside unstructured clinical notes.We introduce a conditioned cross-multi-head attention module to fuse these heterogeneous data modalities, bridging the semantic gap between visual and textual data. Experiments demonstrate substantial improvements from using additional modalities compared to relying on images alone. Notably, our model achieves the highest reported performance on the ROUGE-L metric compared to relevant state-of-the-art models in the literature. Furthermore, we employed both human evaluation and clinical semantic similarity measurement alongside word-overlap metrics to improve the depth of quantitative analysis. A human evaluation, conducted by a board-certified radiologist, confirms the model's accuracy in identifying high-level findings, however, it also highlights that more improvement is needed to capture nuanced details and clinical context.

cs.CV↗

arXiv2023

GATology for Linguistics: What Syntactic Dependencies It Knows

Graph Attention Network (GAT) is a graph neural network which is one of the strategies for modeling and representing explicit syntactic knowledge and can work with pre-trained models, such as BERT, in downstream tasks. Currently, there is still a lack of investigation into how GAT learns syntactic knowledge from the perspective of model structure. As one of the strategies for modeling explicit syntactic knowledge, GAT and BERT have never been applied and discussed in Machine Translation (MT) scenarios. We design a dependency relation prediction task to study how GAT learns syntactic knowledge of three languages as a function of the number of attention heads and layers. We also use a paired t-test and F1-score to clarify the differences in syntactic dependency prediction between GAT and BERT fine-tuned by the MT task (MT-B). The experiments show that better performance can be achieved by appropriately increasing the number of attention heads with two GAT layers. With more than two layers, learning suffers. Moreover, GAT is more competitive in training speed and syntactic dependency prediction than MT-B, which may reveal a better incorporation of modeling explicit syntactic knowledge and the possibility of combining GAT and BERT in the MT tasks.

cs.CL↗

arXiv2023

Syntactic Knowledge via Graph Attention with BERT in Machine Translation

Although the Transformer model can effectively acquire context features via a self-attention mechanism, deeper syntactic knowledge is still not effectively modeled. To alleviate the above problem, we propose Syntactic knowledge via Graph attention with BERT (SGB) in Machine Translation (MT) scenarios. Graph Attention Network (GAT) and BERT jointly represent syntactic dependency feature as explicit knowledge of the source language to enrich source language representations and guide target language generation. Our experiments use gold syntax-annotation sentences and Quality Estimation (QE) model to obtain interpretability of translation quality improvement regarding syntactic knowledge without being limited to a BLEU score. Experiments show that the proposed SGB engines improve translation quality across the three MT tasks without sacrificing BLEU scores. We investigate what length of source sentences benefits the most and what dependencies are better identified by the SGB engines. We also find that learning of specific dependency relations by GAT can be reflected in the translation quality containing such relations and that syntax on the graph leads to new modeling of syntactic aspects of source sentences in the middle and bottom layers of BERT.

cs.CL↗

arXiv2022

Estimating Confidence of Predictions of Individual Classifiers and Their Ensembles for the Genre Classification Task

Genre identification is a subclass of non-topical text classification. The main difference between this task and topical classification is that genres, unlike topics, usually do not correspond to simple keywords, and thus they need to be defined in terms of their functions in communication. Neural models based on pre-trained transformers, such as BERT or XLM-RoBERTa, demonstrate SOTA results in many NLP tasks, including non-topical classification. However, in many cases, their downstream application to very large corpora, such as those extracted from social media, can lead to unreliable results because of dataset shifts, when some raw texts do not match the profile of the training set. To mitigate this problem, we experiment with individual models as well as with their ensembles. To evaluate the robustness of all models we use a prediction confidence metric, which estimates the reliability of a prediction in the absence of a gold standard label. We can evaluate robustness via the confidence gap between the correctly classified texts and the misclassified ones on a labeled test corpus, higher gaps make it easier to improve our confidence that our classifier made the right decision. Our results show that for all of the classifiers tested in this study, there is a confidence gap, but for the ensembles, the gap is bigger, meaning that ensembles are more robust than their individual models.

cs.CL↗

arXiv2022

Multimodal Pipeline for Collection of Misinformation Data from Telegram

The paper presents the outcomes of AI-COVID19, our project aimed at better understanding of misinformation flow about COVID-19 across social media platforms. The specific focus of the study reported in this paper is on collecting data from Telegram groups which are active in promotion of COVID-related misinformation. Our corpus collected so far contains around 28 million words, from almost one million messages. Given that a substantial portion of misinformation flow in social media is spread via multimodal means, such as images and video, we have also developed a mechanism for utilising such channels via producing automatic transcripts for videos and automatic classification for images into such categories as memes, screenshots of posts and other kinds of images. The accuracy of the image classification pipeline is around 87%.

cs.SI↗

arXiv2022

Towards Arabic Sentence Simplification via Classification and Generative Approaches

This paper presents an attempt to build a Modern Standard Arabic (MSA) sentence-level simplification system. We experimented with sentence simplification using two approaches: (i) a classification approach leading to lexical simplification pipelines which use Arabic-BERT, a pre-trained contextualised model, as well as a model of fastText word embeddings; and (ii) a generative approach, a Seq2Seq technique by applying a multilingual Text-to-Text Transfer Transformer mT5. We developed our training corpus by aligning the original and simplified sentences from the internationally acclaimed Arabic novel "Saaq al-Bambuu". We evaluate effectiveness of these methods by comparing the generated simple sentences to the target simple sentences using the BERTScore evaluation metric. The simple sentences produced by the mT5 model achieve P 0.72, R 0.68 and F-1 0.70 via BERTScore, while, combining Arabic-BERT and fastText achieves P 0.97, R 0.97 and F-1 0.97. In addition, we report a manual error analysis for these experiments. \url{https://github.com/Nouran-Khallaf/Lexical_Simplification}

cs.CL↗

arXiv2021

Experiments with adversarial attacks on text genres

Neural models based on pre-trained transformers, such as BERT or XLM-RoBERTa, demonstrate SOTA results in many NLP tasks, including non-topical classification, such as genre identification. However, often these approaches exhibit low reliability to minor alterations of the test texts. A related probelm concerns topical biases in the training corpus, for example, the prevalence of words on a specific topic in a specific genre can trick the genre classifier to recognise any text on this topic in this genre. In order to mitigate the reliability problem, this paper investigates techniques for attacking genre classifiers to understand the limitations of the transformer models and to improve their performance. While simple text attacks, such as those based on word replacement using keywords extracted by tf-idf, are not capable of deceiving powerful models like XLM-RoBERTa, we show that embedding-based algorithms which can replace some of the most ``significant'' words with words similar to them, for example, TextFooler, have the ability to influence model predictions in a significant proportion of cases.

cs.CL↗

arXiv2021

Automatic Difficulty Classification of Arabic Sentences

In this paper, we present a Modern Standard Arabic (MSA) Sentence difficulty classifier, which predicts the difficulty of sentences for language learners using either the CEFR proficiency levels or the binary classification as simple or complex. We compare the use of sentence embeddings of different kinds (fastText, mBERT , XLM-R and Arabic-BERT), as well as traditional language features such as POS tags, dependency trees, readability scores and frequency lists for language learners. Our best results have been achieved using fined-tuned Arabic-BERT. The accuracy of our 3-way CEFR classification is F-1 of 0.80 and 0.75 for Arabic-Bert and XLM-R classification respectively and 0.71 Spearman correlation for regression. Our binary difficulty classifier reaches F-1 0.94 and F-1 0.98 for sentence-pair semantic similarity classifier.

cs.CL↗