Systematic Analysis of Large Language Models and Transformer-Based Machine Translation for English-Tamil and Tamil-English Across Diverse Datasets

2026-07-27 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Advanced, quick

Summary

This research systematically evaluates several multilingual translation models for English-Tamil and Tamil-English across diverse datasets, including NTREX, EnTamV2, WikiMatrix, and PMIndia. The study assesses supervised Neural Machine Translation (NMT) systems, NLLB and mBART, using both BLEU and chrF metrics, examining their performance across varying data quality and domains. It also incorporates an attention-based analysis to visualize token alignments between English source texts and Tamil translations, enhancing model interpretability. Furthermore, the study demonstrates that in-context prompting with the Tamil-capable TamilLaMA model can achieve effective few-shot English-to-Tamil and Tamil-to-English translation, qualitatively comparing it to supervised methods. Key findings indicate that dataset quality and domain alignment significantly impact model performance, attention mechanisms improve explainability, and few-shot large language models can produce structurally coherent Tamil translations.

Key takeaway

For NLP engineers developing machine translation systems for low-resource languages like Tamil, prioritize curating high-quality, domain-aligned datasets, as this significantly impacts model performance. You should also integrate attention-based analysis to diagnose translation errors and enhance model interpretability. Consider exploring in-context prompting with large language models such as TamilLaMA for few-shot translation, as it can produce structurally coherent outputs, offering a viable alternative or complement to supervised approaches.

Key insights

Dataset quality and domain alignment are critical for English-Tamil machine translation, with few-shot LLMs showing promise for coherent output.

Principles

• Dataset quality and domain alignment are crucial.
• Attention mechanisms enhance model interpretability.
• Few-shot LLMs can yield coherent low-resource translations.

Method

The study evaluates supervised NMT (NLLB, mBART) with BLEU/chrF, performs attention analysis, and qualitatively compares few-shot LLM (TamilLaMA) prompting for English-Tamil translation.

In practice

• Prioritize high-quality, domain-aligned datasets.
• Use attention analysis for translation error diagnosis.
• Explore in-context prompting for low-resource languages.

Topics

• Machine Translation
• Low-Resource Languages
• English-Tamil Translation
• Large Language Models
• Attention Mechanisms
• Few-Shot Learning

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• Disentangling meaning from language in LLM-based machine translation — LLM machine translation functions are disentangled into distinct, specialized attention head sets for language and meaning.
• BERT-based Models vs. Large Language Models for Low-Resource Named Entity Recognition: A Comparative Study on Marathi — Task-specific, language-focused models outperform general-purpose LLMs for low-resource language NER.
• Translation as Augmentation: Effect of Translated Data on Assessment of Difficulty — Machine translation can augment scarce native data to improve text difficulty assessment in low-resource languages.
• The Evolving Landscape of LLM Evaluation — Recent AI advancements include top open-weights models, zero-shot MT, and evolving job market trends.
• MELLA: Bridging Linguistic Capability and Cultural Groundedness for Low-Resource Language MLLMs — Effective low-resource MLLMs require both linguistic fluency and deep cultural understanding, beyond mere translation.
• Large Language Models in NLP — LLMs integrate NLP advancements with massive scale to create general-purpose, prompt-driven language systems.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.