Sundeep Dawadi

Author directory

2026

Subtitle translation poses challenges dis tinct from general-purpose machine trans lation model trained on web corpus. The outputs must fit strict display constraints as the text is predominantly spoken dia logue, local slangs, proper names, cultural nuances. In the "WMT26 Chinese → En glish task" all test videos belong to the historical costume drama genre, where pe riod vocabulary cannot be resolved from a single line in isolation. We address these challenges with a two-part system. We fine-tune Hy-MT2-1.8B (Zheng et al., 2026) on 500,000 Chinese-English subtitle pairs from the TVsub corpus (Wang et al., 2018) using 4-bit QLoRA (Dettmers et al., 2023), and at test time we inject show-level metadata (title, episode name, summary) into a structured system prompt to ground character names and period terms. On an in-domain 200-pair development split, QLoRA fine-tuning raises BLEU from 6.93 to 31.29 (+24.4 points) which demon strates the value of domain adaptation over the base multilingual model. Across four prompt configurations evaluated via LLM-as-judge on 20 sampled test trans lations the metadata-augmented, previous context-free variant (structured_noctx) wins 14 of 20 judged dialogues and is sent for submission. The fine-tuned model weights are released under Apache 2.0 at ShivRamSaud/hy-mt2-1.8b-wmt26.
We describe GABU, our WAT2026 English-to-Indic multimodal translation system (CITATION), which translates image grounded English captions into Hindi, Bengali, Malayalam, and Odia. The task follows the WAT English-to-Indic multimodal track setup with text only, image captioning, and multimodal tracks (CITATION). Our system builds on the pretrained, instruction-tuned vision-language model google/gemma-4-E4B-it. Each input combines a cropped visual region with the English caption, rendered through a chat template with a script constrained translation instruction. We explore two approaches which includes a zero shot baseline and a fine tuned approach of the model using QLORA (rank 64, alpha 128, 4-bit NF4, double quantization), trained for 3,000 steps due to computational resource constraints, reflecting our aim to adapt the model’s existing instruction-following ability rather than train it from scratch. We evaluate both systems using BLEU and CHRF3 under greedy decoding and report macro averaged scores across the target languages. The fine-tuned model substantially outperforms the zero shot baseline and achieves a BLEU score of 43.97 and 69.99 CHRF3 on the test split, and 40.37 BLEU and 68.25 CHRF3 on the challenge split. These results correspond to improvements of 24.39 BLEU and 18.72 CHRF3 points on the test split, and 16.70 BLEU and 11.90 CHRF3 points on the challenge split.
We describe our system for the AstroCLIMB shared task (WASP 2026), which reconstructs the citation graph of astronomy papers from scientific figures and captions alone. The task requires identifying the relationship between pairs of figures and captions as same_figure, same_paper, related_papers, or unrelated_papers. Our system finetunes google/gemma-4-E4B-it using QLoRA on 8,200 competition training rows with a shared descriptive prompt and greedy decoding. A zero shot baseline achieves 0.4408 macro F1 on a balanced 1,800 pair validation split held out from the competition training file, and the finetuned system substantially improves this citation-relation classification for the challenging related_papers class in particular. Our submitted system achieves 0.64282 F1 on all 10,000 official test pairs. These results show that multimodal finetuning can improve citation-relation inference from scientific figures and captions when document metadata is unavailable.
The Nepali language has a rich and complex morphology. Existing lemmatization research focuses on traditional rule-based or TRIE-based approaches. These methods often fail when encountering out-of-vocabulary or misspelled words. This paper investigates neural lemmatization for the under-resourced Nepali language using multilingual transformer models. We formulate lemmatization as a text-to-text generation problem and evaluate its impacts on downstream tasks by finetuning mBART-large-50, mT5-base, and mT5-small. The models were trained on a combination of publicly available and human-annotated word-lemma pair (8,000 instances) dataset. The performance is evaluated using Character Error Rate (CER), accuracy, character-level Bilingual Evaluation Understudy (BLEU), and morphological coverage. The mT5-base model achieved the highest overall performance. The model achieved 96.1% accuracy and a 1.1% CER using a learning rate of 5 × 10−4. However, it showed slightly weaker performance in handling complex morphological variations. The mBART-large-50 model followed closely with 96.0% accuracy and 0.970 morphological coverage. To assess the efficacy of these models, we applied lemmatization to downstream tasks. In Hindi-Nepali cross-lingual alignment, performance improved significantly from 12.86% to 41.61% using mBART model. In information retrieval, the Mean Average Precision (MAP)@1 using binary index increased from 0.71 to 0.90 using mBART model. These results demonstrate that multilingual transformers effectively learn morphological transformations for low-resource languages through text-to-text generation.

2025

This paper presents a comparative analysis of FastText and BERT-based approaches for Natural Language Understanding (NLU) tasks in Devanagari script languages. We evaluate these models on three critical tasks: language identification, hate speech detection, and target identification across five languages: Nepali, Marathi, Sanskrit, Bhojpuri, and Hindi. Our experiments, although with raw tweet dataset but extracting only devanagari script, demonstrate that while both models achieve exceptional performance in language identification (F1 scores > 0.99), they show varying effectiveness in hate speech detection and target identification tasks. FastText with augmented data outperforms BERT in hate speech detection (F1 score: 0.8552 vs 0.5763), while BERT shows superior performance in target identification (F1 score: 0.5785 vs 0.4898). These findings contribute to the growing body of research on NLU for low-resource languages and provide insights into model selection for specific tasks in Devanagari script processing.