Phonetic transcription is vital for speech processing and linguistic documentation, particularly in languages like Tamil with complex phonology and dialectal variation. Challenges such as consonant gemination, retroflexion, vowel length, and one-to-many grapheme-phoneme mappings are compounded by limited data on Sri Lankan Tamil dialects. We present a dialect-aware, rule-based transcription tool for Tamil that supports Indian and Jaffna Tamil, with extensions underway for other dialects. Using a two-stage pipeline: Tamil script to Latin, then to IPA with context-sensitive rules, the tool handles dialect shifts. A real-time interface enables dialect selection. Evaluated on a 7,830-word corpus, it achieves 94.54% accuracy for Jaffna Tamil and is higher than other tools like eSpeak NG, advancing linguistic preservation and accessible speech technology for Tamil communities.

This paper introduces the TaLK Corpus, the first speech benchmark corpus for Sri Lankan Tamil Automatic Speech Recognition (ASR) covering speech from 22 administrative districts of Sri Lanka. The corpus contains 1 hour and 33 minutes of speech from 22 native speakers (one per district) and includes rich metadata on demographics, location history, recording conditions, and domain information, along with transcriptions in Tamil script and the International Phonetic Alphabet (IPA). Standardised preprocessing (16 kHz mono WAV format) and segmentation using Silero Voice Activity Detection (VAD) resulted in 1,214 utterances. All recordings were manually transcribed by trained linguists, and MD5-based file naming used to ensure data integrity and consistency. TaLK corpus enables district-wise benchmarking of ASR systems and supports dialect-sensitive evaluation. We establish baseline results for multilingual models (Whisper Large-V3 and Facebook’s MMS) in zero-shot settings. The evaluation reveals substantial performance disparities across districts, highlighting the impact of regional phonological variation in low-resource Sri Lankan Tamil. Although Whisper Large-V3 outperforms MMS overall, it shows considerable variability, with mean Word Error Rates ranging from 0.672 to 0.903 across districts. These findings demonstrate strong regional effects even within a single model. By releasing TaLK-Corpus under the CC-BY-NC 4.0 licence, we aim to support dialect-robust ASR research and foster inclusive speech technologies for Sri Lankan Tamil-speaking communities.

This paper presents an overview of the Shared Task on Multimodal Hate and Sentiment Understanding in Low-Resource Memes, organized as part of the Second Workshop on Challenges in Processing South Asian Languages (CHiPSAL 2026) at LREC 2026. The task addresses automated content understanding in low-resource settings by focusing on monolingual Nepali memes written in Devanagari script. Built upon the NeMeme dataset, the task comprises two subtasks: (1) binary hate speech detection and (2) three-class sentiment analysis. The competition attracted 23 teams for hate detection and 13 teams for sentiment analysis. Participating teams employed diverse strategies, including late-fusion multimodal architectures combining multilingual text encoders with vision models, caption-based approaches using large vision-language models, and ensemble techniques. The top-performing system achieved macro-F1 scores of 80.52% on hate detection and 68.81% on sentiment analysis using a late-fusion hybrid architecture with discriminative learning rates. Our analysis reveals that multimodal fusion consistently outperforms unimodal baselines, sentiment analysis poses greater challenges than hate detection due to increased semantic nuance, and the scarcity of Devanagari-centric pretrained models remains a significant bottleneck. This shared task establishes a benchmark for multimodal understanding in low-resource South Asian languages and provides insights for developing inclusive content moderation systems.

Detecting online toxicity in morphologically rich, low-resource languages like Tamil remains a major computational challenge. Standard transformer models often struggle with sub-word fragmentation, which can dilute the semantic intensity of regional insults and out-of-vocabulary slang. To mitigate this limitation, we train a multi-layer hybrid framework that fuses the deep contextual representations of L3Cube-TamilBERT with the character-level robustness of FastText embeddings. Our architecture leverages Last-4 Layers averaging and a dual pooling strategy (Mean + Max) to capture both global sentence intent and extract high-activation spikes of offensive cues typically lost in single layer representations. Experiments show that this hybrid model achieves a Macro-F1 of 0.7883, notably enhancing Hate Recall (0.7503) for detection of offensive content. Additionally, as reported by other studies, stacking ensemble achieves peak hate precision (0.9296), providing a high accuracy alternative for moderation scenarios requiring minimal false positives. By combining deep contextual hidden states with FastText embeddings, the proposed feature-fusion ensemble approach with multi-layer hybrid framework approach establishes a new benchmark for hate speech detection for Tamil.

Large Language Models (LLMs) have shown strong generalization across tasks in high-resource languages; however, their linguistic competence in low-resource and morphologically rich languages such as Tamil remains largely unexplored. Existing multilingual benchmarks often rely on translated English datasets, failing to capture the language specific linguistic and cultural nuances of the target language. To address this gap, we introduce ILAKKANAM, the first Tamil-specific linguistic evaluation benchmark manually curated using 820 questions from Sri Lankan school-level Tamil subject examination papers spanning Grades 1–13. Each question is annotated by trained linguists under five linguistic categories and a factual knowledge category. We evaluate both closed-source and open-source LLMs using a standardized evaluation pipeline. Our results show that Gemini 2.5 achieves the highest overall performance, while open-source models lag behind, highlighting the gap in linguistic grounding. Category- and grade-wise analyses reveal that all models perform well on lower-grade questions but show a clear decline as the grade level and the linguistic complexity of the questions increase. Further, no strong correlation is observed between a model’s overall performance and its ability to identify linguistic categories, suggesting that performance may be driven by exposure rather than genuine understanding. The code and dataset used in this study are publicly available in our repository, where the dataset consists only of extracted examination questions to mitigate potential data leakage. Keywords: Tamil, Linguistic Benchmark, Linguistic diagnostics, Low-resource language

This paper presents the findings of the second workshop on Challenges in Processing South Asian Languages (CHiPSAL 2026), held as part of LREC 2026. South Asia is one of the most linguistically diverse regions in the world, yet its languages remain severely underrepresented in language resources and technologies, particularly in the era of large language models (LLMs). The workshop brings together research addressing key challenges in this space, including data scarcity, morphological complexity, code-mixing, script diversity, and the lack of culturally grounded evaluation benchmarks. The workshop received 57 submissions, covering a wide range of languages, tasks, and modalities, including both widely spoken languages (e.g., Bengali, Hindi, Tamil, and Urdu) and extremely low-resource and endangered languages such as Burushaski, Limbu, and Nepal Bhasha (Newari). Several contributions introduce arguably first-of-their-kind resources and benchmarks for these languages, spanning both text and speech domains, and focusing on linguistically informed and culturally grounded data creation. In addition to the main track, the workshop hosted a shared task on Multimodal Hate and Sentiment Understanding in Low-Resource Memes for Nepali, attracting strong community participation. The results highlight the effectiveness of multimodal approaches while also revealing persistent challenges in modelling culturally nuanced and low-resource data. Across the accepted papers and shared task, key insights include the central role of high-quality data, the limitations of current multilingual models in low-resource settings, and the need for culturally aware and data-centric approaches. Overall, CHiPSAL 2026 demonstrates the growing momentum in South Asian language processing and highlights the importance of sustained, community-driven efforts to build inclusive and representative language technologies.

Tokenizers act as a bridge between human language and the latent space of language models, influencing how language is represented in these models. Despite the dominance of English-Centric (EC) Large Language Models (LLMs), tokenization methods often fail to fairly represent complex scripts like Tamil, Sinhala, and Hindi, primarily due to pre-tokenization choices. This study demonstrates that pre-tokenization has a more significant impact than tokenization algorithms on achieving egalitarian representation. To address this, we introduce an improvement to the Byte Pair Encoding (BPE) algorithm by incorporating graphemes, which we term Grapheme Pair Encoding (GPE). Our experiments show that grapheme-based character extraction outperforms byte-level tokenizers for complex scripts. We validate this approach through experiments on Tamil, Sinhala, and Hindi. The codebase and resources used in this work are publicly available at https://github.com/vmenan/tokenizers-coling2025.

The growing use of Devanagari-script languages such as Hindi, Nepali, Marathi, Sanskrit, and Bhojpuri on social media presents unique challenges for natural language understanding (NLU), particularly in language identification, hate speech detection, and target classification. To address these challenges, we organized a shared task with three subtasks: (i) identifying the language of Devanagari-script text, (ii) detecting hate speech, and (iii) classifying hate speech targets into individual, community, or organization. A curated dataset combining multiple corpora was provided, with splits for training, evaluation, and testing. The task attracted 113 participants, with 32 teams submitting models evaluated on accuracy, precision, recall, and macro F1-score. Participants applied innovative methods, including large language models, transformer models, and multilingual embeddings, to tackle the linguistic complexities of Devanagari-script languages. This paper summarizes the shared task, datasets, and results, and aims to contribute to advancing NLU for low-resource languages and fostering inclusive, culturally aware natural language processing (NLP) solutions.

In this paper, we provide a brief summary of the inaugural workshop on Challenges in Processing South Asian Languages (CHiPSAL) held as part of COLING 2025. The workshop included regular papers, invited keynotes, and shared task papers, fostering a collaborative platform for exploring challenges in processing South Asian languages. The shared task focused on Devanagari-script language understanding, encompassing subtasks on language identification, hate speech detection, and target classification. This workshop series aims to address linguistic and cultural nuances, resource constraints, and orthographic complexities in low-resource South Asian languages while advancing NLP research and promoting multilingual inclusivity.

Tamil is a relatively low-resource language in the field of Natural Language Processing (NLP). Recent years have seen a growth in Tamil NLP datasets in Natural Language Understanding (NLU) or Natural Language Generation (NLG) tasks, but high-quality linguistic resources remain scarce. In order to alleviate this gap in resources, this paper introduces Aalamaram, a treebank with rich linguistic annotations for the Tamil language. It is hitherto the largest publicly available Tamil treebank with almost 10,000 sentences from diverse sources and is annotated for the tasks of Part-of-speech (POS) tagging, Named Entity Recognition (NER), Morphological Parsing and Dependency Parsing. Close attention has also been paid to multi-word segmentation, especially in the context of Tamil clitics. Although the treebank is based largely on the Universal Dependencies (UD) specifications, significant effort has been made to adjust the annotation rules according to the idiosyncrasies and complexities of the Tamil language, thereby providing a valuable resource for linguistic research and NLP developments.

This paper reports the development of the first dependency treebank for the Sinhala language (STB). Sinhala, which is morphologically rich, is a low-resource language with few linguistic and computational resources available publicly. This treebank consists of 100 sentences taken from a large contemporary written text corpus. These sentences were annotated manually according to the Universal Dependencies framework. In this paper, apart from elaborating on the approach that has been followed to create the treebank, we have also discussed some interesting syntactic constructions found in the corpus and how we have handled them using the current Universal Dependencies specification.

Sentiment analysis in natural language processing (NLP), endeavors to computationally identify and extract subjective information from textual data. In code-mixed text, sentiment analysis presents a unique challenge due to the mixing of languages within a single textual context. For low-resourced languages such as Tamil and Tulu, predicting sentiment becomes a challenging task due to the presence of text comprising various scripts. In this research, we present the sentiment analysis of code-mixed Tamil and Tulu Youtube comments. We have developed a Bidirectional Long-Short Term Memory (BiLSTM) networks based models for both languages which further uses contextualized word embeddings at input layers of the models. For that purpose, ELMo embeddings have been trained on larger unannotated code-mixed text like corpora. Our models performed with macro average F1-scores of 0.2877 and 0.5133 on Tamil and Tulu code-mixed datasets respectively.

This paper describes an ongoing development of a grammar error checker for the Tamil language using a state-of-the-art deep neural-based approach. This proposed checker capture a vital type of grammar error called subject-predicate agreement errors. In this case, we specifically target the agreement error that occurs between nominal subject and verbal predicates. We also created the first-ever grammar error annotated corpus for Tamil. In addition, we experimented with different multi-lingual pre-trained language models to capture syntactic information and found that IndicBERT gives better performance for our tasks. We implemented this grammar checker as a multi-class classification on top of the IndicBERT pre-trained model, which we fine-tuned using our annotated data. This baseline model gives an F1 Score of 73.4. We are now in the process of improving this proposed system with the use of a dependency parser.

This paper describes how we developed a neural-based dependency parser, namely ThamizhiUDp, which provides a complete pipeline for the dependency parsing of the Tamil language text using Universal Dependency formalism. We have considered the phases of the dependency parsing pipeline and identified tools and resources in each of these phases to improve the accuracy and to tackle data scarcity. ThamizhiUDp uses Stanza for tokenisation and lemmatisation, ThamizhiPOSt and ThamizhiMorph for generating Part of Speech (POS) and Morphological annotations, and uuparser with multilingual training for dependency parsing. ThamizhiPOSt is our POS tagger, which is based on the Stanza, trained with Amrita POS-tagged corpus. It is the current state-of-the-art in Tamil POS tagging with an F1 score of 93.27. Our morphological analyzer, ThamizhiMorph is a rule-based system with a very good coverage of Tamil. Our dependency parser ThamizhiUDp was trained using multilingual data. It shows a Labelled Assigned Score (LAS) of 62.39, 4 points higher than the current best achieved for Tamil dependency parsing. Therefore, we show that breaking up the dependency parsing pipeline to accommodate existing tools and resources is a viable approach for low-resource languages.

This paper describes a new and larger coverage Finite-State Morphological Analyser (FSM) and Generator for the Dravidian language Tamil. The FSM has been developed in the context of computational grammar engineering, adhering to the standards of the ParGram effort. Tamil is a morphologically rich language and the interaction between linguistic analysis and formal implementation is complex, resulting in a challenging task. In order to allow the development of the FSM to focus more on the linguistic analysis and less on the formal details, we have developed a system of meta-morph(ology) rules along with a script which translates these rules into FSM processable representations. The introduction of meta-morph rules makes it possible for computationally naive linguists to interact with the system and to expand it in future work. We found that the meta-morph rules help to express linguistic generalisations and reduce the manual effort of writing lexical classes for morphological analysis. Our Tamil FSM currently handles mainly the inflectional morphology of 3,300 verb roots and their 260 forms. Further, it also has a lexicon of approximately 100,000 nouns along with a guesser to handle out-of-vocabulary items. Although the Tamil FSM was primarily developed to be part of a computational grammar, it can also be used as a web or stand-alone application for other NLP tasks, as per general ParGram practice.