Monolingual models outperform multilingual on Dravidian languages

Researchers have developed and evaluated five GPT-2 architecture models to assess the performance of multilingual language models on Dravidian languages. Four of these models were trained monolingually for Tamil, Telugu, Kannada, and Malayalam, respectively, each with its own tokenizer. A fifth model was trained multilingually, sharing a tokenizer across all four languages. The study found that the monolingual models outperformed the multilingual model, mGPT, on tasks like sentiment classification and named entity recognition, and demonstrated greater tokenizer efficiency. AI

IMPACT This research highlights the potential limitations of current multilingual models for underrepresented languages and suggests monolingual models may offer superior performance for specific linguistic tasks.

RANK_REASON The cluster contains an academic paper detailing the evaluation of language models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• GPT-2
• Hugging Face
• Kannada
• Malayalam
• Samanantar
• Tamil
• Telugu
• Venkata Naga Sai Vishnu Rohit Pulipaka
• Wikipedia

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →