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.

Transformer-based pre-trained language models have dominated the field of Natural Language Processing (NLP) for quite some time now. However, the Nepali language, spoken by approximately 32 million people worldwide, remains significantly underrepresented in this domain. This underrepresentation is primarily attributed to the scarcity of monolingual data corpora and limited available resources for the Nepali language. While existing efforts have predominantly concentrated on basic encoder-based models, there is a notable gap in the exploration of decoder-based architectures. To address this gap, we have collected 27.5 GB of Nepali text data, approximately 2.4x larger than any previously available Nepali language corpus. Leveraging this data, we pre-trained three different models i.e., BERT, RoBERTa, and GPT-2, exclusively for the Nepali Language. Furthermore, we performed instruction tuning and explored its potential for monolingual Nepali data, providing a foundation for future research. Our models outperformed the existing best model by 2 points on Nep-gLUE benchmark, scoring 95.60 and also outperformed existing models on text generation tasks, demonstrating improvements in both understanding and generating Nepali text.

Modern general-purpose speech recognition systems are more robust in languages with high resources. However, achieving state-of-the-art accuracy for low-resource languages is still challenging. To deal with this challenge, one of the popular practice is fine-tuning the pre-trained model on low-resource settings. Nevertheless, pre-trained or fine-tuned model fails to capture the complex character and word constituency in the Devanagari script transcription. We proposed a complementary loss function designed to force the model to learn the character constituency of Devanagari script. Our complementary loss function, called as Rule-Based Character Constituency Loss (RBCCL), that penalizes incorrect transcriptions and updates the overall loss during the model training phase. This loss function can be combined with CTC loss or cross-entropy loss as well which are widely used in ASR training. Our experiment shows that combining the existing cross-entropy loss with new complementary loss (RBCCL) improves the Word Error Rate (WER ), reducing it from 47.1% to 23.41% which is quite promising result.

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.

The performance of Automatic Speech Recognition (ASR) systems has improved significantly, driven by advancements in large-scale pre-trained models. However, adapting such models to low-resource languages such as Nepali is challenging due to the lack of labeled data and computational resources. Additionally, adapting the unique speech parameters of the speaker to a model is also a challenging task. Personalization helps to target the model to fit the particular speaker. This work investigates parameter-efficient fine-tuning (PEFT) methods like Low-Rank Adaptation (LoRA) and Decomposed Weight Low-Rank Adaptation (DoRA) to improve the performance of fine-tuned Whisper ASR models for Nepali ASR tasks by Personalization. These experiments demonstrate that the PEFT methods obtain competitive results while significantly reducing the number of trainable parameters compared to full fine-tuning. LoRA and DoRA show a relative WER to FT_Base increment of 34.93% and 36.79%, respectively, and a relative CER to FT_Base increment of 49.50% and 50.03%, respectively. Furthermore, the results highlight a 99.74% reduction in total training parameters.

Modern general-purpose speech recognition systems are more robust in languages with high resources. In contrast, achieving state-of-the-art accuracy for low-resource languages is still challenging. The fine-tuning of the pre-trained model is one of the highly popular practices which utilizes the existing information while efficiently learning from a small amount of data to enhance the precision and robustness of speech recognition tasks. This work attempts to diagnose the performance of a pre-trained model when transcribing the audio from the low-resource language. In this work, we apply an adapter-based iterative parameter-efficient fine-tuning strategy on a limited dataset aiming to improve the quality of the transcription of a previously fine-tuned model. For the experiment we used Whisper’s multilingual pre-trained speech model and Nepali as a test language. Using this approach we achieved Word Error Rate of 27.9%,which is more than 19% improvement over pre-trained Whisper Large − V2.

Offensive and profane content has been on the rise in Nepali Social Media, which, is very disturbing to users. This is partly due to the absence of proper tools and mechanisms for the Nepali language to deal with profanity and offensive texts. In this work, we attempt to develop a deep learning-based profanity and offensive comments detection tool. We develop a Bi-LSTM (Bidirectional Long Short Term Memory) based model for the classification of Profane and Offensive comments and study different variations of the task. Furthermore, Multilingual BERT embedding and vocab embedding were used among others for an accurate understanding of the intent and decency of the posts. While previous related studies in the Nepali language are more focused on sentiment and offensiveness detection only, our study explores profanity and offensiveness detection as two distinct tasks. Our Bi-LSTM model outputs 87.8% accuracy for

Large Language Models (LLMs) have made significant advancements in Natural Language Processing (NLP) by excelling in various NLP tasks. This study specifically focuses on evaluating the performance of LLMs for Named Entity Recognition (NER) and Part-of-Speech (POS) tagging for a low-resource language, Nepali. The aim is to study the effectiveness of these models for languages with limited resources by conducting experiments involving various parameters and fine-tuning and evaluating two datasets namely, ILPRL and EBIQUITY. In this work, we have experimented with eight LLMs for Nepali NER and POS tagging. While some prior works utilized larger datasets than ours, our contribution lies in presenting a comprehensive analysis of multiple LLMs in a unified setting. The findings indicate that NepBERTa, trained solely in the Nepali language, demonstrated the highest performance with F1-scores of 0.76 and 0.90 in ILPRL dataset. Similarly, it achieved 0.79 and 0.97 in EBIQUITY dataset for NER and POS respectively. This study not only highlights the potential of LLMs in performing classification tasks for low-resource languages but also compares their performance with that of alternative approaches deployed for the tasks.

Nepali, a low-resource language belonging to the Indo-Aryan language family and spoken in Nepal, India, Sikkim, and Burma has comparatively very little digital content and resources, more particularly in the legal domain. However, the need to translate legal documents is ever-increasing in the context of growing volumes of legal cases and a large population seeking to go abroad for higher education or employment. This underscores the need for developing an English-Nepali Machine Translation for the legal domain. We attempt to address this problem by utilizing a Neural Machine Translation (NMT) System with an encoder-decoder architecture, specifically designed for legal Nepali-English translation. Leveraging a custom-built legal corpus of 125,000 parallel sentences, our system achieves encouraging BLEU scores of 7.98 in (Nepali → English) and 6.63 (English → Nepali) direction

Nepal Script (also known as Prachalit Script) is the widely used script of Nepal Bhasa, the native language of the Kathmandu Valley in Nepal. Derived from the Brahmi Script, the Nepal Script was developed in the 9th century and was extensively used till the 20th century, before being replaced by the Devanagari script. Numerous ancient manuscripts, inscriptions, and documents written in the Nepal Script are still available containing immense knowledge on architecture, arts, astrology, ayurveda, literature, music, tantrism, etc. To preserve and revive Nepal Bhasa, digitizing such documents plays a crucial role. This paper presents our work on text recognition for the Nepal Script. The implementation includes the Nepal Script text recognizer based on CRNN CTC architecture aided by line and word segmentations. Leveraging a carefully curated dataset that encompasses handwritten and printed texts in the Nepal Script, our work has achieved CER of 6.65% and WER of 13.11%. The dataset used for this work is available as Nepal Script Text Dataset on Kaggle. The paper further explores the associated challenges due to the complex nature of the script such as conjuncts, modifiers and variations; and the current state of the script.

The Automatic Speech Recognition (ASR) has come up with significant advancements over the course of several decades, transitioning from a rule-based method to a statistical approach, and ultimately to the use of end-to-end (E2E) frameworks. This phenomenon continues with the progression of machine learning and deep learning methodologies. The E2E approach for ASR has demonstrated predominant success in the case of resourceful languages with larger annotated corpus. However, the accuracy is quite low for low-resourced languages such as Nepali. In this regard, language-specific tools such as tokenizers seem to play a vital role in improving the performance of the E2E model for low-resourced languages like Nepali. In this paper, we propose a pronunciationaware syllable tokenizer for the Nepali language which improves the results of the E2E model. Our experiment confirm that the introduction of the proposed tokenizer yields better performance with the Character Error Rate (CER) 8.09% compared to other language-independent tokenizers.

Fine tuning of the pre-trained language model is a technique which can be used to enhance the technologies of low-resourced languages. The unsupervised approach can fine-tune any pre-trained model with minimum or even no language-specific resources. It is highly advantageous, particularly for languages that possess limited computational resources. We present a novel approach for fine-tuning a pre-trained Automatic Speech Recognition (ASR) model that is suitable for low resource languages. Our methods involves iterative fine-tuning of pre-trained ASR model. mms-1b is selected as the pretrained seed model for fine-tuning. We take the Nepali language as a case study for this research work. Our approach achieved a CER of 6.77%, outperforming all previously recorded CER values for the Nepali ASR Systems.

Research on Deep learning-based Text-toSpeech (TTS) systems has gained increasing popularity in low-resource languages as this approach is not only computationally robust but also has the capability to produce state-ofthe-art results. However, these approaches are yet to be significantly explored for the Nepali language, primarily because of the lack of adequate size datasets and secondarily because of the relatively sophisticated computing resources they demand. This paper explores the FastPitch acoustic model with HiFi-GAN vocoder for the Nepali language. We trained the acoustic model with two datasets, OpenSLR and a dataset prepared jointly by the Information and Language Processing Research Lab (ILPRL) and the Nepal Association of the Blind (NAB), to be further referred to as the ILPRLNAB dataset. We achieved a Mean Opinion Score (MOS) of 3.70 and 3.40 respectively for the same model with different datasets. The synthesized speech produced by the model was found to be quite natural and of good quality.

Many image captioning tasks have been carried out in recent years, the majority of the work being for the English language. A few research works have also been carried out for Hindi and Bengali languages in the domain. Unfortunately, not much research emphasis seems to be given to the Nepali language in this direction. Furthermore, the datasets are also not publicly available in the Nepali language. The aim of this research is to prepare a dataset with Nepali captions and develop a deep learning model based on the Convolutional Neural Network (CNN) and Transformer combined model to automatically generate image captions in the Nepali language. The dataset for this work is prepared by applying different data preprocessing techniques on the Flickr8k dataset. The preprocessed data is then passed to the CNN-Transformer model to generate image captions. ResNet-101 and EfficientNetB0 are the two pre-trained CNN models employed for this work. We have achieved some promising results which can be further improved in the future.

Language model pre-training has significantly impacted NLP and resulted in performance gains on many NLP-related tasks, but comparative study of different approaches on many low-resource languages seems to be missing. This paper attempts to investigate appropriate methods for pretraining a Transformer-based model for the Nepali language. We focus on the language-specific aspects that need to be considered for modeling. Although some language models have been trained for Nepali, the study is far from sufficient. We train three distinct Transformer-based masked language models for Nepali text sequences: distilbert-base (Sanh et al., 2019) for its efficiency and minuteness, deberta-base (P. He et al., 2020) for its capability of modeling the dependency of nearby token pairs and XLM-ROBERTa (Conneau et al., 2020) for its capabilities to handle multilingual downstream tasks. We evaluate and compare these models with other Transformer-based models on a downstream classification task with an aim to suggest an effective strategy for training low-resource language models and their fine-tuning.

In this era of AI and Deep Learning, Speech Recognition has achieved fairly good levels of accuracy and is bound to change the way humans interact with computers, which happens mostly through texts today. Most of the speech recognition systems for the Nepali language to date use conventional approaches which involve separately trained acoustic, pronunciation and language model components. Creating a pronunciation lexicon from scratch and defining phoneme sets for the language requires expert knowledge, and at the same time is time-consuming. In this work, we present an End-to-End ASR approach, which uses a joint CTC- attention-based encoder-decoder and a Recurrent Neural Network based language modeling which eliminates the need of creating a pronunciation lexicon from scratch. ESPnet toolkit which uses Kaldi Style of data preparation is the framework used for this work. The speech and transcription data used for this research is freely available on the Open Speech and Language Resources (OpenSLR). We use about 159k transcribed speech data to train the speech recognition model which currently recognizes speech input with the CER of 10.3%.

The Tamang language is spoken mainly in Nepal, Sikkim, West Bengal, some parts of Assam, and the North East region of India. As per the 2011 census conducted by the Nepal Government, there are about 1.35 million Tamang speakers in Nepal itself. In this regard, a Machine Translation System for Tamang-Nepali language pair is significant both from research and practical outcomes in terms of enabling communication between the Tamang and the Nepali communities. In this work, we train the Transformer Neural Machine Translation (NMT) architecture with attention using a small hand-labeled or aligned Tamang-Nepali corpus (15K sentence pairs). Our preliminary results show BLEU scores of 27.74 for the Nepali→Tamang direction and 23.74 in the Tamang→Nepali direction. We are currently working on increasing the datasets as well as improving the model to obtain better BLEU scores. We also plan to extend the work to add the English language to the model, thus making it a trilingual Machine Translation System for Tamang-Nepali-English languages.

Due to the general availability, relative abundance and wide diversity of opinions, news Media texts are very good sources for sentiment analysis. However, the major challenge with such texts is the difficulty in aligning the expressed opinions to the concerned political leaders as this entails a non-trivial task of named-entity recognition and anaphora resolution. In this work, our primary focus is on developing a Natural Language Processing (NLP) pipeline involving a robust Named-Entity Recognition followed by Anaphora Resolution and then after alignment of the recognized and resolved named-entities, in this case, political leaders to the correct class of opinions as expressed in the texts. We visualize the popularity of the politicians via the time series graph of positive and negative sentiments as an outcome of the pipeline. We have achieved the performance metrics of the individual components of the pipeline as follows: Part of speech tagging – 93.06% (F1-score), Named-Entity Recognition – 86% (F1-score), Anaphora Resolution – 87.45% (Accuracy), Sentiment Analysis – 80.2% (F1-score).

This paper describes an annotation scheme for argumentation in opinionated texts such as newspaper editorials, developed from a corpus of approximately 500 English texts from Nepali and international newspaper sources. We present the results of analysis and evaluation of the corpus annotation ― currently, the inter-annotator agreement kappa value being 0.80 which indicates substantial agreement between the annotators. We also discuss some of linguistic resources (key factors for distinguishing facts from opinions, opinion lexicon, intensifier lexicon, pre-modifier lexicon, modal verb lexicon, reporting verb lexicon, general opinion patterns from the corpus etc.) developed as a result of our corpus analysis, which can be used to identify an opinion or a controversial issue, arguments supporting an opinion, orientation of the supporting arguments and their strength (intrinsic, relative and in terms of persuasion). These resources form the backbone of our work especially for performing the opinion analysis in the lower levels, i.e., in the lexical and sentence levels. Finally, we shed light on the perspectives of the given work clearly outlining the challenges.