As social-media platforms emerge and evolve faster than the regulations meant to oversee them, automated detoxification might serve as a timely tool for moderators to enforce safe discourse at scale. We here describe our submission to the PAN 2025 Multilingual Text Detoxification Challenge, which rewrites toxic single-sentence inputs into neutral paraphrases across 15 typologically diverse languages. Building on a 12B-parameter Gemma-3 multilingual transformer, we apply parameter-efficient LoRA SFT fine-tuning and prompting techniques like few-shot and Chain-of-Thought. Our multilingual training corpus combines 3,600 human-authored parallel pairs, 21,600 machine-translated synthetic pairs, and model-generated pairs filtered by Jaccard thresholds. At inference, inputs are enriched with three LaBSE-retrieved neighbors and explicit toxic-span annotations. Evaluated via Style Transfer Accuracy, LaBSE-based semantic preservation, and xCOMET fluency, our system ranks first on high-resource and low-resource languages. Ablations show +0.081 joint score increase from few-shot examples and +0.088 from basic CoT prompting. ANOVA analysis identifies language resource status as the strongest predictor of performance ($η^2$ = 0.667, p < 0.01).

Fact-checking in code-mixed, low-resource languages such as Hinglish remains an underexplored challenge in natural language processing. Existing fact-verification systems largely focus on high-resource, monolingual settings and fail to generalize to real-world political discourse in linguis - tically diverse regions like India. Given the widespread use of Hinglish by public figures, particularly political figures, and the growing influence of social media on public opin - ion, there's a critical need for robust, multilingual and con - text-aware fact-checking tools. To address this gap a novel benchmark HiFACT dataset is introduced with 1,500 real-world factual claims made by 28 Indian state Chief Minis - ters in Hinglish, under a highly code-mixed low-resource setting. Each claim is annotated with textual evidence and veracity labels. To evaluate this benchmark, a novel graph-aware, retrieval-augmented fact-checking model is proposed that combines multilingual contextual encoding, claim-evi - dence semantic alignment, evidence graph construction, graph neural reasoning, and natural language explanation generation. Experimental results show that HiFACTMix outperformed accuracy in comparison to state of art multi - lingual baselines models and provides faithful justifications for its verdicts. This work opens a new direction for multi - lingual, code-mixed, and politically grounded fact verifica - tion research..

In this work, we introduce our solution for the Multilingual Text Detoxification Task in the PAN-2025 competition for the ylmmcl team: a robust multilingual text detoxification pipeline that integrates lexicon-guided tagging, a fine-tuned sequence-to-sequence model (s-nlp/mt0-xl-detox-orpo) and an iterative classifier-based gatekeeping mechanism. Our approach departs from prior unsupervised or monolingual pipelines by leveraging explicit toxic word annotation via the multilingual_toxic_lexicon to guide detoxification with greater precision and cross-lingual generalization. Our final model achieves the highest STA (0.922) from our previous attempts, and an average official J score of 0.612 for toxic inputs in both the development and test sets. It also achieved xCOMET scores of 0.793 (dev) and 0.787 (test). This performance outperforms baseline and backtranslation methods across multiple languages, and shows strong generalization in high-resource settings (English, Russian, French). Despite some trade-offs in SIM, the model demonstrates consistent improvements in detoxification strength. In the competition, our team achieved ninth place with a score of 0.612.

Large Language Models (LLMs) have achieved remarkable success in code generation tasks, powering various applications like code completion, debugging, and programming assistance. However, existing benchmarks such as HumanEval, MBPP, and BigCodeBench primarily evaluate LLMs on English-only prompts, overlooking the real-world scenario where multilingual developers often use code-mixed language while interacting with LLMs. To address this gap, we introduce CodeMixBench, a novel benchmark designed to evaluate the robustness of LLMs on code generation from code-mixed prompts. Built upon BigCodeBench, CodeMixBench introduces controlled code-mixing (CMD) into the natural language parts of prompts across three language pairs: Hinglish (Hindi-English), Spanish-English, and Chinese Pinyin-English. We comprehensively evaluate a diverse set of open-source code generation models ranging from 1.5B to 15B parameters. Our results show that code-mixed prompts consistently degrade Pass@1 performance compared to their English-only counterparts, with performance drops increasing under higher CMD levels for smaller models. CodeMixBench provides a realistic evaluation framework for studying multilingual code generation and highlights new challenges and directions for building robust code generation models that generalize well across diverse linguistic settings.

This paper presents our process for developing a sample-efficient language model for a conversational Hinglish chatbot. Hinglish, a code-mixed language that combines Hindi and English, presents a unique computational challenge due to inconsistent spelling, lack of standardization, and limited quality of conversational data. This work evaluates multiple pre-trained cross-lingual language models, including Gemma3-4B and Qwen2.5-7B, and employs fine-tuning techniques to improve performance on Hinglish conversational tasks. The proposed approach integrates synthetically generated dialogues with insights from existing Hinglish datasets to address data scarcity. Experimental results demonstrate that models with fewer parameters, when appropriately fine-tuned on high-quality code-mixed data, can achieve competitive performance for Hinglish conversation generation while maintaining computational efficiency.

The problems of online hate speech and cyberbullying have significantly worsened since the increase in popularity of social media platforms such as YouTube and Twitter (X). Natural Language Processing (NLP) techniques have proven to provide a great advantage in automatic filtering such toxic content. Women are disproportionately more likely to be victims of online abuse. However, there appears to be a lack of studies that tackle misogyny detection in under-resourced languages. In this short paper, we present a novel dataset of YouTube comments in mix-code Hinglish collected from YouTube videos which have been weak labelled as `Misogynistic' and `Non-misogynistic'. Pre-processing and Exploratory Data Analysis (EDA) techniques have been applied on the dataset to gain insights on its characteristics. The process has provided a better understanding of the dataset through sentiment scores, word clouds, etc.

The advent of Large Language Models (LLMs) has advanced the benchmark in various Natural Language Processing (NLP) tasks. However, large amounts of labelled training data are required to train LLMs. Furthermore, data annotation and training are computationally expensive and time-consuming. Zero and few-shot learning have recently emerged as viable options for labelling data using large pre-trained models. Hate speech detection in mix-code low-resource languages is an active problem area where the use of LLMs has proven beneficial. In this study, we have compiled a dataset of 100 YouTube comments, and weakly labelled them for coarse and fine-grained misogyny classification in mix-code Hinglish. Weak annotation was applied due to the labor-intensive annotation process. Zero-shot learning, one-shot learning, and few-shot learning and prompting approaches have then been applied to assign labels to the comments and compare them to human-assigned labels. Out of all the approaches, zero-shot classification using the Bidirectional Auto-Regressive Transformers (BART) large model and few-shot prompting using Generative Pre-trained Transformer- 3 (ChatGPT-3) achieve the best results

Large language models (LLMs) have exerted a considerable impact on diverse language-related tasks in recent years. Their demonstrated state-of-the-art performance is achieved through methodologies such as zero-shot or few-shot prompting. These models undergo training on extensive datasets that encompass segments of the Internet and subsequently undergo fine-tuning tailored to specific tasks. Notably, they exhibit proficiency in tasks such as translation, summarization, question answering, and creative writing, even in the absence of explicit training for those particular tasks. While they have shown substantial improvement in the multilingual tasks their performance in the code switching, especially for machine translation remains relatively uncharted. In this paper, we present an extensive study on the code switching task specifically for the machine translation task comparing multiple LLMs. Our results indicate that despite the LLMs having promising results in the certain tasks, the models with relatively lesser complexity outperform the multilingual large language models in the machine translation task. We posit that the efficacy of multilingual large language models in contextual code switching is constrained by their training methodologies. In contrast, relatively smaller models, when trained and fine-tuned on bespoke datasets, may yield superior results in comparison to the majority of multilingual models.

In the past few years due to the Covid19 pandemic the adoption of e-learning platforms has increased significantly. The widespread restrictions have forced students to continue their education via online means which causes them to spend a significant amount of their time watching videos and attending classes. This sudden change from offline to online learning has affected a lot of students therefore making an attempt to build systems that can accurately simulate the experience of offline learning can help in smoothing out this drastic transition. Live classes is one such way that gives the students a chance to escape the monotony of watching recorded videos on a daily basis. The interaction aspect of such classes allow the students to clarify small scale doubts instantaneously and at the same time gives teachers the opportunity to compliment the students on good behaviour. All these tiny bits significantly affect the learning outcome for a student by making the course content more interesting and thus improving their overall engagement on the platform. In order to mimic this offline style of interaction there can be a multitude of implementations like live polls or quizzes to check whether the student is paying attention, dynamic interactive diagrams that fuel the curiosity of students by giving them a chance to tinker with it, in-session feedback to understand the student's opinions or the in-class chats mechanism between the participants of a given session. Unlike all the other mechanisms, chats are the most open medium of communication and provide the maximum opportunity to interact with each other.

This paper describes the Stevens Institute of Technology's submission for the WMT 2022 Shared Task: Code-mixed Machine Translation (MixMT). The task consisted of two subtasks, subtask $1$ Hindi/English to Hinglish and subtask $2$ Hinglish to English translation. Our findings lie in the improvements made through the use of large pre-trained multilingual NMT models and in-domain datasets, as well as back-translation and ensemble techniques. The translation output is automatically evaluated against the reference translations using ROUGE-L and WER. Our system achieves the $1^{st}$ position on subtask $2$ according to ROUGE-L, WER, and human evaluation, $1^{st}$ position on subtask $1$ according to WER and human evaluation, and $3^{rd}$ position on subtask $1$ with respect to ROUGE-L metric.