Datasets:
Tamil Names
The 100 longest unique names (by character count, after lowercasing) from MeghanaKap/tamil_names_roman, transliterated into Tamil script.
Method
Transliterated word-by-word using a fine-tuned AI4Bharat IndicXlit
checkpoint (fairseq, checkpoint_1_100.pt from this project's fine-tuning
work). Single-letter tokens (name initials, e.g. the p in "vijaya p") are
left in Roman script rather than transliterated, since a lone letter has no
well-defined Tamil transliteration on its own.
Columns
customer_name_roman-- the original romanized name, lowercasedcustomer_name_tamil-- word-by-word Tamil-script transliteration, with single-letter initials left as-is
Known limitations
- Word-level model: multi-word names are transliterated word-by-word, not as a full sequence, so cross-word context isn't used.
- Business/company names (a small fraction of the source data) are transliterated the same way as personal names -- phonetically reasonable but not curated for brand-name conventions.
- Downloads last month
- 20