Metraipolis › archive › dc_24381eae8698
Evaluating Dedicated Monolingual and Joint Multilingual Causal Models for Dravidian Languages — arXiv:2608.07727
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.07727
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-19T08:49:47.192Z
- SHA-256 of the archived text
81d5325fd39bded18d41a8114a457228cc3100d685d4d6edc9b39c716e74e91e- Stable URL of this page
https://metraipolis.com/documento/dc_24381eae8698
How to cite
Evaluating Dedicated Monolingual and Joint Multilingual Causal Models for Dravidian Languages — arXiv:2608.07727 — arXiv (en), https://arxiv.org/abs/2608.07727 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_24381eae8698 (Metraipolis archive dc_24381eae8698, sha256 81d5325fd39bded18d41a8114a457228cc3100d685d4d6edc9b39c716e74e91e)
BibTeX
@misc{dc24381eae8698,
title = {Evaluating Dedicated Monolingual and Joint Multilingual Causal Models for Dravidian Languages — arXiv:2608.07727},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.07727},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_24381eae8698; sha256 81d5325fd39bded18d41a8114a457228cc3100d685d4d6edc9b39c716e74e91e},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.07727. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Dravidian languages, mainly Tamil, Telugu, Kannada, and Malayalam make up only a small part of the data used to train multilingual language models, so it's not clear how much per-language ability these models actually keep. I have trained five GPT-2 architecture models from scratch to compare four monolingual models (one each for Tamil, Telugu, Kannada, and Malayalam, each with its own 32K-vocabulary subword tokenizer) against one multilingual model sharing a 64K-vocabulary subword tokenizer across all four languages. All the 5 models are trained on cleaned CC-100, Wikipedia, and Samanantar data. I have tested the models on perplexity, bits-per-byte, tokenizer efficiency, and fine-tuning results which are compared against mGPT. The monolingual models outperform mGPT on sentiment classification and named entity recognition, and their tokenizers proved more efficient than the shared multilingual model across all the languages tested. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-11 in cs.CL). Categorie arXiv: Computation and Language (cs.CL). Pagina del lavoro: https://arxiv.org/abs/2608.07727 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-19T08:49:41.829Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-08-11?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).