Abstract Natural language processing (NLP) technologies increasingly shape public life, yet their deployment for social good remains unevenly distributed across domains, languages, and geographies. This piece inaugurates the NLP for Social Good column in this journal. In this piece, I map the current state of NLP for Social Good (NLP4SG) across nine application domains. The picture that emerges is one of striking imbalance: AI harms, inclusion, and digital violence attract the bulk of research attention, while poverty, peacebuilding, and environmental protection remain critically underexplored. I argue that the field must address three structural gaps, domain coverage, linguistic diversity, and evaluation methodology, if NLP is to fulfil its potential as a force for equitable social progress. The piece concludes with five directions that I believe will define the next chapter of NLP4SG research.

The increasing prevalence of mental disorders globally highlights the urgent need for effective digital screening methods that can be used in multilingual contexts. Most existing studies, however, focus on English data, overlooking critical mental health signals that may be present in non-English texts. To address this gap, we present a survey of the detection of mental disorders using social media data beyond the English language. We compile a comprehensive list of 108 datasets spanning 25 languages that can be used for developing NLP models for mental health screening. In addition, we discuss the cultural nuances that influence online language patterns and self-disclosure behaviors, and how these factors can impact the performance of NLP tools. Our survey highlights major challenges, including the scarcity of resources for low- and mid-resource languages and the dominance of depression-focused data over other disorders. By identifying these gaps, we advocate for interdisciplinary collaborations and the development of multilingual benchmarks to enhance mental health screening worldwide.

The goal of news image captioning is to generate captions by integrating news article content with corresponding images, highlighting the relationship between textual context and visual elements. The majority of research on news image captioning focuses on English, primarily because datasets in other languages are scarce. To address this limitation, we create the first multilingual news image captioning benchmark, MUNIChus, comprising 9 languages, including several low-resource languages such as Sinhala and Urdu. We evaluate various state-of-the-art neural news image captioning models on MUNIChus and find that news image captioning remains challenging. We also make MUNIChus publicly available with over 20 models already benchmarked. MUNIChus opens new avenues for further advancements in developing and evaluating multilingual news image captioning models.

Legal judgment prediction (LJP) systems support legal professionals by forecasting case outcomes, aiding in legal preparation and strategy. The development of such systems has been driven largely by benchmark datasets that support the training and evaluation of machine learning (ML) models. However, most existing datasets only contain final judgment texts that include facts written after the court hearing, which are unavailable to practitioners during the pre-hearing phase, limiting the real-world applicability of models trained on them. Additionally, benchmark datasets for Supreme Courts remain scarce, with few notable exceptions. To address both issues, we introduce the first LJP benchmark dataset for the United Kingdom Supreme Court: UKSC-JP, covering 821 cases and supporting two subtasks: (i) legal judgment classification, and (ii) court view generation. Each UKSC case includes a press summary that provides background and does not contain all adjudicated facts, offering a more realistic input for LJP systems. We evaluate several ML models, including large language models (LLMs), across both subtasks. Our results show that, despite recent advances, LJP remains a challenging task, especially when deprived of post-hearing content. We release our code and data resources publicly available at: https://github.com/DamithDR/uksc .

Identifying relevant text spans is important for several downstream tasks in NLP, as it contributes to model explainability. While most span identification approaches rely on relatively smaller pre-trained language models like BERT, a few recent approaches have leveraged the latest generation of Large Language Models (LLMs) for the task. Current work has focused on explicit span identification like Named Entity Recognition (NER), while more subjective span identification with LLMs in tasks like Aspect-based Sentiment Analysis (ABSA) has been underexplored. In this paper, we fill this important gap by presenting an evaluation of the performance of various LLMs on text span identification in three popular tasks, namely sentiment analysis, offensive language identification, and claim verification. We explore several LLM strategies like instruction tuning, in-context learning, and chain of thought. Our results indicate underlying relationships within text aid LLMs in identifying precise text spans.

We present judgeWEL, a dataset for named entity recognition (NER) in Luxembourgish, automatically labelled and subsequently verified using large language models (LLM) in a novel pipeline. Building datasets for under-represented languages remains one of the major bottlenecks in natural language processing, where the scarcity of resources and linguistic particularities make large-scale annotation costly and potentially inconsistent. To address these challenges, we propose and evaluate a novel approach that leverages Wikipedia and Wikidata as structured sources of weak supervision. By exploiting internal links within Wikipedia articles, we infer entity types based on their corresponding Wikidata entries, thereby generating initial annotations with minimal human intervention. Because such links are not uniformly reliable, we mitigate noise by employing and comparing several LLMs to identify and retain only high-quality labelled sentences. The resulting corpus is approximately five times larger than the currently available Luxembourgish NER dataset and offers broader and more balanced coverage across entity categories, providing a substantial new resource for multilingual and low-resource NER research.

Indic quality estimation (QE) and automatic post-editing (APE) data is spread across separate releases, so no single resource supports training and evaluation across tasks and language pairs on one footing. We consolidate the WMT 2020–2024 shared-task lineage with an extended English–Malayalam resource into : 126,754 instances over nine directional pairs, with up to four label types aligned on the same segment, a direct assessment, a human post-edit, word-level OK/BAD tags and an error explanation, and a test set stratified over four difficulty axes. On it we benchmark six prompted LLMs and three COMET metrics on segment-level QE, and three systems on APE. Two of the axes are defined partly on the direct assessment and select a compressed slice of it, so each axis is compared against a control drawn from the same language pair with the same score distribution. Only one survives that control: segments whose holistic and token-level quality signals conflict are ranked worse than equally-scored segments of the same language, for all nine systems and all seven pairs that carry the axis. Annotator disagreement, which looks second-hardest without the control, has no effect with it. Few-shot prompting costs every model ≤ 3.4B both correlation and output-format compliance. Within-language accuracy does not make scores comparable across pairs: of the three trained metrics, the one with the best within-language correlation loses most when the pairs are pooled. The benchmark (https://huggingface.co/datasets/surrey-nlp/IndicQE-APE) and code (https://github.com/surrey-nlp/IndicQE-APE) are released.

Post-Editing (PE) of Machine Translation (MT) output often involves repeating the same lexical and terminological corrections across many segments, especially in specialised and highly repetitive documents. Despite substantial work on Automatic Post-Editing (APE), most available corpora operate at the sentence level, others are synthetic, and overall not designed to study how corrections propagate in realistic Computer-Assisted Translation (CAT) workflows. This paper presents the APEX-VW (Automatic Post-Editing eXperiments on Virtual Wards) Corpus, a new open English-Spanish (EN-ES) dataset built from recent NHS virtual-ward documents and professional PE in Trados Studio, with controlled MT, terminology, and quality assurance settings. The corpus contains seven document-coherent source texts totalling 42k words, translated with four MT systems representing different paradigms and then post-edited by professional translators. Unlike prior resources such as WMT APE corpora, eSCAPE, MLQE-PE, or LangMark, the dataset preserves document order and CAT-tool context, making it suitable for research on terminology normalisation, correction propagation, and human-in-the-loop translation support. The paper describes the corpus design, data preparation, PE setup, and initial corpus statistics, and positions the resource as a benchmark for document-level APE and propagation-aware assistive tools.

This paper presents ltzGLUE, the first Natural Language Understanding (NLU) benchmark for Luxembourgish (LTZ) based on the popular GLUE benchmark for English. Although NLU tasks are available for many European languages nowadays, LTZ is one of the official national languages that is often overlooked. We construct new tasks and reuse existing ones to introduce the first official NLU benchmark and accompanying evaluation of encoder models for the language. Our tasks include common natural language processing tasks in binary and multi-class classification settings, including named entity recognition, topic classification, and intent classification. We evaluate various pre-trained language models for LTZ to present an overview of the current capabilities of these models on the LTZ language.

In common law systems, legal professionals such as lawyers and judges rely on precedents to build their arguments. As the volume of cases has grown massively over time, effectively retrieving prior cases has become essential. Prior case retrieval (PCR) is an information retrieval (IR) task that aims to automatically identify the most relevant court cases for a specific query from a large pool of potential candidates. While IR methods have seen several paradigm shifts over the last few years, the vast majority of PCR methods continue to rely on traditional IR methods, such as BM25. The state-of-the-art deep learning IR methods have not been successful in PCR due to two key challenges: i. Lengthy legal text limitation; when using the powerful BERT-based transformer models, there is a limit of input text lengths, which inevitably requires to shorten the input via truncation or division with a loss of legal context information. ii. Lack of legal training data; due to data privacy concerns, available PCR datasets are often limited in size, making it difficult to train deep learning-based models effectively. In this research, we address these challenges by leveraging LLM-based text embedders in PCR. LLM-based embedders support longer input lengths, and since we use them in an unsupervised manner, they do not require training data, addressing both challenges simultaneously. In this paper, we evaluate state-of-the-art LLM-based text embedders in four PCR benchmark datasets and show that they outperform BM25 and supervised transformer-based models.

The first Workshop on Language Models for Low-Resource Languages (LoResLM 2025) was held in conjunction with the 31st International Conference on Computational Linguistics (COLING 2025) in Abu Dhabi, United Arab Emirates. This workshop mainly aimed to provide a forum for researchers to share and discuss their ongoing work on language models (LMs) focusing on low-resource languages, following the recent advancements in neural language models and their linguistic biases towards high-resource languages. LoResLM 2025 attracted notable interest from the natural language processing (NLP) community, resulting in 35 accepted papers from 52 submissions. These contributions cover a broad range of low-resource languages from eight language families and 13 diverse research areas, paving the way for future possibilities and promoting linguistic inclusivity in NLP.

Recent advances in open-source vision-language models (VLMs) offer new opportunities for understanding complex and subjective multimodal phenomena such as sarcasm. In this work, we evaluate seven state-of-the-art VLMs - BLIP2, InstructBLIP, OpenFlamingo, LLaVA, PaliGemma, Gemma3, and Qwen-VL - on their ability to detect multimodal sarcasm using zero-, one-, and few-shot prompting. Furthermore, we evaluate the models' capabilities in generating explanations to sarcastic instances. We evaluate the capabilities of VLMs on three benchmark sarcasm datasets (Muse, MMSD2.0, and SarcNet). Our primary objectives are twofold: (1) to quantify each model's performance in detecting sarcastic image-caption pairs, and (2) to assess their ability to generate human-quality explanations that highlight the visual-textual incongruities driving sarcasm. Our results indicate that, while current models achieve moderate success in binary sarcasm detection, they are still not able to generate high-quality explanations without task-specific finetuning.

Predicting semantic textual similarity (STS) is a complex and ongoing challenge in natural language processing (NLP). Over the years, researchers have developed a variety of supervised and unsupervised approaches to calculate STS automatically. Additionally, various benchmarks, which include STS datasets, have been established to consistently evaluate and compare these STS methods. However, they largely focus on high-resource languages, mixed with datasets annotated focusing on relatedness instead of similarity and containing automatically translated instances. Therefore, no dedicated benchmark for multilingual STS exists. To solve this gap, we introduce the Multilingual Semantic Textual Similarity Benchmark (MUSTS), which spans 13 languages, including low-resource languages. By evaluating more than 25 models on MUSTS, we establish the most comprehensive benchmark of multilingual STS methods. Our findings confirm that STS remains a challenging task, particularly for low-resource languages.

Recently, language models (LMs) have produced excellent results in many natural language processing (NLP) tasks. However, their effectiveness is highly dependent on available pre-training resources, which is particularly challenging for low-resource languages such as Sinhala. Furthermore, the scarcity of benchmarks to evaluate LMs is also a major concern for low-resource languages. In this paper, we address these two challenges for Sinhala by (i) collecting the largest monolingual corpus for Sinhala, (ii) training multiple LMs on this corpus and (iii) compiling the first Sinhala NLP benchmark (Sinhala-GLUE) and evaluating LMs on it. We show the Sinhala LMs trained in this paper outperform the popular multilingual LMs, such as XLM-R and existing Sinhala LMs in downstream NLP tasks. All the trained LMs are publicly available. We also make Sinhala-GLUE publicly available as a public leaderboard, and we hope that it will enable further advancements in developing and evaluating LMs for Sinhala.

This article presents an approach to readability estimation that focuses on conceptual rather than linguistic complexity, using the extensive SaudiTextBooks textbooks. We introduce DARES 2.0, an enhanced concept-based readability training dataset designed to estimate the readability of Saudi educational texts. Building on DARES 1.0, DARES 2.0 extends the scope of conceptual complexity by replacing repetitive concepts and manually revising the input features with unique terms and their surrounding contexts from the SaudiTextBooks, spanning grades 1 to 12. The refined DARES 2.0 is employed to fine-tune pre-trained transformer models, including XLM-R Base, mBERT, AraELECTRA, AraBERTv2, and CAMeLBERTmix. The findings suggest that both the dataset and experimental setup require further development to ensure a larger, higher-quality dataset and to support more extensive fine-tuning experiments, in addition to exploring transfer learning from other languages and enhancing the diversity and richness of Arabic concepts. These developments pave the way for further advancements in concept-based readability estimation in educational contexts in future work.

Machine translation (MT) is widely used to translate content on social media platforms aiming to improve accessibility. A great part of the content circulated on social media is user-generated and often contains non-standard spelling, hashtags, and emojis that pose challenges to MT systems. This leads to many mistranslated instances that are presented to users of these platforms, hindering their understanding of content written in other languages. In this paper, we investigate the impact of MT on offensive language identification. We pose that MT and potential mistranslations have an important and mostly under-explored impact on social media tasks such as sentiment analysis and offensive language identification. We create MT-Offense, a novel dataset containing English originals and translations in Arabic, Hindi, Marathi, Sinhala, and Spanish produced by multiple open-access Neural Machine Translation systems. We evaluate the performance of various offensive language models on both original and MT content in different training and test set combinations. We report the F1 scores of the models. Our results show that (1) offensive language identification models perform better on original data than on MT data, and (2) the use of MT data in training helps models better identify offensive language in MT content compared to models trained exclusively on original data.