Generative Large Language Models (LLMs) have achieved remarkable advances in various NLP tasks. In this work, our aim is to explore the multilingual capabilities of large language models by using machine translation as a task involving English and 22 Indian languages. We first investigate the translation capabilities of raw large-language models, followed by exploring the in-context learning capabilities of the same raw models. We fine-tune these large language models using parameter-efficient fine-tuning methods such as LoRA and additionally with full fine-tuning. Through our study, we have identified the model that performs best among the large language models available for the translation task.Our results demonstrate significant progress, with average BLEU scores of 13.42, 15.93, 12.13, 12.30, and 12.07, as well as chrF scores of 43.98, 46.99, 42.55, 42.42, and 45.39, respectively, using two-stage fine-tuned LLaMA-13b for English to Indian languages on IN22 (conversational), IN22 (general), flores200-dev, flores200-devtest, and newstest2019 testsets. Similarly, for Indian languages to English, we achieved average BLEU scores of 14.03, 16.65, 16.17, 15.35 and 12.55 along with chrF scores of 36.71, 40.44, 40.26, 39.51, and 36.20, respectively, using fine-tuned LLaMA-13b on IN22 (conversational), IN22 (general), flores200-dev, flores200-devtest and newstest2019 testsets. Overall, our findings highlight the potential and strength of large language models for machine translation capabilities, including languages that are currently underrepresented in LLMs.

Named Entity Recognition (NER) is a use-ful component in Natural Language Process-ing (NLP) applications. It is used in varioustasks such as Machine Translation, Summa-rization, Information Retrieval, and Question-Answering systems. The research on NER iscentered around English and some other ma-jor languages, whereas limited attention hasbeen given to Indian languages. We analyze thechallenges and propose techniques that can betailored for Multilingual Named Entity Recog-nition for Indian Languages. We present a hu-man annotated named entity corpora of ∼40Ksentences for 4 Indian languages from two ofthe major Indian language families. Addition-ally, we show the transfer learning capabilitiesof pre-trained transformer models from a highresource language to multiple low resource lan-guages through a series of experiments. Wealso present a multilingual model fine-tunedon our dataset, which achieves an F1 score of∼0.80 on our dataset on average. We achievecomparable performance on completely unseenbenchmark datasets for Indian languages whichaffirms the usability of our model.

Machine Translation for low-resource languages presents significant challenges, primarily due to limited data availability. We have a baseline model and a primary model. For the baseline model, we first fine-tune the mBART model (mbart-large-50-many-to-many-mmt) for the language pairs English-Khasi, Khasi-English, English-Manipuri, and Manipuri-English. We then augment the dataset by back-translating from Indic languages to English. To enhance data quality, we fine-tune the LaBSE model specifically for Khasi and Manipuri, generating sentence embeddings and applying a cosine similarity threshold of 0.84 to filter out low-quality back-translations. The filtered data is combined with the original training data and used to further fine-tune the mBART model, creating our primary model. The results show that the primary model slightly outperforms the baseline model, with the best performance achieved by the English-to-Khasi (en-kh) primary model, which recorded a BLEU score of 0.0492, a chrF score of 0.3316, and a METEOR score of 0.2589 (on a scale of 0 to 1), with similar results for other language pairs.

This paper presents the systems submitted by the Yes-MT team for the Low-Resource Indic Language Translation Shared Task at WMT 2024, focusing on translating between English and the Assamese, Mizo, Khasi, and Manipuri languages. The experiments explored various approaches, including fine-tuning pre-trained models like mT5 and IndicBart in both Multilingual and Monolingual settings, LoRA finetune IndicTrans2, zero-shot and few-shot prompting with large language models (LLMs) like Llama 3 and Mixtral 8x7b, LoRA Supervised Fine Tuning Llama 3, and training Transformers from scratch. The results were evaluated on the WMT23 Low-Resource Indic Language Translation Shared Task’s test data using SacreBLEU and CHRF highlighting the challenges of low-resource translation and show the potential of LLMs for these tasks, particularly with fine-tuning.

In this paper, we present our work in the EHRSQL 2024 shared task which tackles reliable text-to-SQL modeling on Electronic Health Records. Our proposed system tackles the task with three modules - abstention module, text-to-SQL generation module, and reliability module. The abstention module identifies whether the question is answerable given the database schema. If the question is answerable, the text-to-SQL generation module generates the SQL query and associated confidence score. The reliability module has two key components - confidence score thresholding, which rejects generations with confidence below a pre-defined level, and error filtering, which identifies and excludes SQL queries that result in execution errors. In the official leaderboard for the task, our system ranks 6th. We have also made the source code public.

In this paper, we present our work to the MEDIQA-M3G 2024 shared task, which tackles multilingual and multimodal medical answer generation. Our system consists of a lightweight Vision-and-Language Transformer (ViLT) model which is fine-tuned for the clinical dermatology visual question-answering task. In the official leaderboard for the task, our system ranks 6th. After the challenge, we experiment with training the ViLT model on more data. We also explore the capabilities of large Vision-Language Models (VLMs) such as Gemini and LLaVA.

Obtaining sufficient information in one’s mother tongue is crucial for satisfying the information needs of the users. While high-resource languages have abundant online resources, the situation is less than ideal for very low-resource languages. Moreover, the insufficient reporting of vital national and international events continues to be a worry, especially in languages with scarce resources, like Mizo. In this paper, we conduct a study to investigate the effectiveness of a simple methodology designed to generate a holistic summary for Mizo news articles, which leverages English-language news to supplement and enhance the information related to the corresponding news events. Furthermore, we make available 500 Mizo news articles and corresponding enriched holistic summaries. Human evaluation confirms that our approach significantly enhances the information coverage of Mizo news articles.

Natural language processing (NLP) modelshave achieved remarkable progress in recentyears, particularly in tasks related to semanticanalysis. However, many existing benchmarksprimarily focus on lexical and syntactic un-derstanding, often overlooking the importanceof numerical reasoning abilities. In this pa-per, we argue for the necessity of incorporatingnumeral-awareness into NLP evaluations andpropose two distinct tasks to assess this capabil-ity: Numerical Reasoning and Headline Gener-ation. We present datasets curated for each taskand evaluate various approaches using both au-tomatic and human evaluation metrics. Ourresults demonstrate the diverse strategies em-ployed by participating teams and highlight thepromising performance of emerging modelslike Mixtral 8x7b instruct. We discuss the im-plications of our findings and suggest avenuesfor future research in advancing numeral-awarelanguage understanding and generation.

In this paper, we introduce a machine-generated text detection system designed totackle the challenges posed by the prolifera-tion of large language models (LLMs). Withthe rise of LLMs such as ChatGPT and GPT-4,there is a growing concern regarding the po-tential misuse of machine-generated content,including misinformation dissemination. Oursystem addresses this issue by automating theidentification of machine-generated text acrossmultiple subtasks: binary human-written vs.machine-generated text classification, multi-way machine-generated text classification, andhuman-machine mixed text detection. We em-ploy the RoBERTa Base model and fine-tuneit on a diverse dataset encompassing variousdomains, languages, and sources. Throughrigorous evaluation, we demonstrate the effec-tiveness of our system in accurately detectingmachine-generated text, contributing to effortsaimed at mitigating its potential misuse.

Semantic hallucinations in neural language gen-eration systems pose a significant challenge tothe reliability and accuracy of natural languageprocessing applications. Current neural mod-els often produce fluent but incorrect outputs,undermining the usefulness of generated text.In this study, we address the task of detectingsemantic hallucinations through the SHROOM(Semantic Hallucinations Real Or Mistakes)dataset, encompassing data from diverse NLGtasks such as definition modeling, machinetranslation, and paraphrase generation. We in-vestigate three methodologies: fine-tuning onlabelled training data, fine-tuning on labelledvalidation data, and a zero-shot approach usingthe Mixtral 8x7b instruct model. Our resultsdemonstrate the effectiveness of these method-ologies in identifying semantic hallucinations,with the zero-shot approach showing compet-itive performance without additional training.Our findings highlight the importance of robustdetection mechanisms for ensuring the accu-racy and reliability of neural language genera-tion systems.

Detecting abusive language in multimodal videos has become a pressing need in ensuring a safe and inclusive online environment. This paper focuses on addressing this challenge through the development of a novel approach for multimodal abusive language detection in Tamil videos and sentiment analysis for Tamil/Malayalam videos. By leveraging state-of-the-art models such as Multiscale Vision Transformers (MViT) for video analysis, OpenL3 for audio analysis, and the bert-base-multilingual-cased model for textual analysis, our proposed framework integrates visual, auditory, and textual features. Through extensive experiments and evaluations, we demonstrate the effectiveness of our model in accurately detecting abusive content and predicting sentiment categories. The limited availability of effective tools for performing these tasks in Dravidian Languages has prompted a new avenue of research in these domains.

Abusive comments in online platforms have become a significant concern, necessitating the development of effective detection systems. However, limited work has been done in low resource languages, including Dravidian languages. This paper addresses this gap by focusing on abusive comment detection in a dataset containing Tamil, Tamil-English and Telugu-English code-mixed comments. Our methodology involves logistic regression and explores suitable embeddings to enhance the performance of the detection model. Through rigorous experimentation, we identify the most effective combination of logistic regression and embeddings. The results demonstrate the performance of our proposed model, which contributes to the development of robust abusive comment detection systems in low resource language settings. Keywords: Abusive comment detection, Dravidian languages, logistic regression, embeddings, low resource languages, code-mixed dataset.

This study addresses the challenge of detecting fake news in Dravidian languages by leveraging Google’s MuRIL (Multilingual Representations for Indian Languages) model. Drawing upon previous research, we investigate the intricacies involved in identifying fake news and explore the potential of transformer-based models for linguistic analysis and contextual understanding. Through supervised learning, we fine-tune the “muril-base-cased” variant of MuRIL using a carefully curated dataset of labeled comments and posts in Dravidian languages, enabling the model to discern between original and fake news. During the inference phase, the fine-tuned MuRIL model analyzes new textual content, extracting contextual and semantic features to predict the content’s classification. We evaluate the model’s performance using standard metrics, highlighting the effectiveness of MuRIL in detecting fake news in Dravidian languages and contributing to the establishment of a safer digital ecosystem. Keywords: fake news detection, Dravidian languages, MuRIL, transformer-based models, linguistic analysis, contextual understanding.

This paper presents the overview of the shared task on emotional analysis in Tamil. The result of the shared task is presented at the workshop. This paper presents the dataset used in the shared task, task description, and the methodology used by the participants and the evaluation results of the submission. This task is organized as two Tasks. Task A is carried with 11 emotions annotated data for social media comments in Tamil and Task B is organized with 31 fine-grained emotion annotated data for social media comments in Tamil. For conducting experiments, training and development datasets were provided to the participants and results are evaluated for the unseen data. Totally we have received around 24 submissions from 13 teams. For evaluating the models, Precision, Recall, micro average metrics are used.

We present our findings from the first shared task on Multi-task Learning in Dravidian Languages at the second Workshop on Speech and Language Technologies for Dravidian Languages. In this task, a sentence in any of three Dravidian Languages is required to be classified into two closely related tasks namely Sentiment Analyis (SA) and Offensive Language Identification (OLI). The task spans over three Dravidian Languages, namely, Kannada, Malayalam, and Tamil. It is one of the first shared tasks that focuses on Multi-task Learning for closely related tasks, especially for a very low-resourced language family such as the Dravidian language family. In total, 55 people signed up to participate in the task, and due to the intricate nature of the task, especially in its first iteration, 3 submissions have been received.

Parsing has been gaining popularity in recent years and attracted the interest of NLP researchers around the world. It is challenging when the language under study is a free-word order language that allows ellipsis like Telugu. In this paper, an attempt is made to parse subordinate clauses especially, non-finite verb clauses and relative clauses in Telugu which are highly productive and constitute a large chunk in parsing tasks. This study adopts a knowledge-driven approach to parse subordinate structures using linguistic cues as rules. Challenges faced in parsing ambiguous structures are elaborated alongside providing enhanced tags to handle them. Results are encouraging and this parser proves to be efficient for Telugu.

In this work, two Neural Machine Translation (NMT) systems have been developed and evaluated as part of the bidirectional Tamil-Telugu similar languages translation subtask in WMT21. The OpenNMT-py toolkit has been used to create quick prototypes of the systems, following which models have been trained on the training datasets containing the parallel corpus and finally the models have been evaluated on the dev datasets provided as part of the task. Both the systems have been trained on a DGX station with 4 -V100 GPUs. The first NMT system in this work is a Transformer based 6 layer encoder-decoder model, trained for 100000 training steps, whose configuration is similar to the one provided by OpenNMT-py and this is used to create a model for bidirectional translation. The second NMT system contains two unidirectional translation models with the same configuration as the first system, with the addition of utilizing Byte Pair Encoding (BPE) for subword tokenization through the pre-trained MultiBPEmb model. Based on the dev dataset evaluation metrics for both the systems, the first system i.e. the vanilla Transformer model has been submitted as the Primary system. Since there were no improvements in the metrics during training of the second system with BPE, it has been submitted as a contrastive system.

This paper presents an overview of the shared task on machine translation of Dravidian languages. We presented the shared task results at the EACL 2021 workshop on Speech and Language Technologies for Dravidian Languages. This paper describes the datasets used, the methodology used for the evaluation of participants, and the experiments’ overall results. As a part of this shared task, we organized four sub-tasks corresponding to machine translation of the following language pairs: English to Tamil, English to Malayalam, English to Telugu and Tamil to Telugu which are available at https://competitions.codalab.org/competitions/27650. We provided the participants with training and development datasets to perform experiments, and the results were evaluated on unseen test data. In total, 46 research groups participated in the shared task and 7 experimental runs were submitted for evaluation. We used BLEU scores for assessment of the translations.

This paper describes the IIITK team’s submissions to the offensive language identification, and troll memes classification shared tasks for Dravidian languages at DravidianLangTech 2021 workshop@EACL 2021. Our best configuration for Tamil troll meme classification achieved 0.55 weighted average F1 score, and for offensive language identification, our system achieved weighted F1 scores of 0.75 for Tamil, 0.95 for Malayalam, and 0.71 for Kannada. Our rank on Tamil troll meme classification is 2, and offensive language identification in Tamil, Malayalam and Kannada are 3, 3 and 4 respectively.