VerChol -- Grammar-First Tokenization for Agglutinative Languages
About
Tokenization is the foundational step in all large language model (LLM) pipelines, yet the dominant approach Byte Pair Encoding (BPE) and its variants is inherently script agnostic and optimized for English like morphology. For agglutinative languages a typological class encompassing the Dravidian family (Tamil, Kannada, Telugu, Malayalam), Turkic languages (Turkish, Azerbaijani, Uzbek), Uralic languages (Finnish, Hungarian, Estonian), Korean, Japanese, Swahili, Basque, and others, a single word may encode root, tense, aspect, person, number, gender agreement, case, and postpositions into one orthographic unit. Statistical tokenizers fragment these words into byte pair chunks that sever morpheme boundaries and inflate token counts.
Prabhu Raja• 2026
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Tokenization | Tamil Wikipedia freq >= 3 (Full Evaluation) | Total Tokens9.15e+5 | 4 | |
| Token Fertility | English | -- | 4 | |
| Token Fertility | Hindi | -- | 4 | |
| Token Fertility | Tamil | -- | 4 | |
| Token Fertility | Kannada | -- | 4 | |
| Token Fertility | Telugu | -- | 4 | |
| Token Fertility | Malayalam | -- | 4 | |
| Token Fertility | Turkish | -- | 4 | |
| Token Fertility | Finnish | -- | 4 | |
| Token Fertility | Korean | -- | 4 |
Showing 10 of 11 rows