latin stringlengths 1 132 | khmer stringlengths 1 105 | persona stringclasses 5
values | source stringclasses 6
values |
|---|---|---|---|
>>Quick Selection Tools: prae | >>Quick Selection Tools: ααααΎ | mixed | hplt |
SOMRAP TWER KA SELECT TOU LER | ααααΆααααααΎααΆα Select αα
ααΎ | careful | hplt |
rouppab dael man laknahjbah vea manlaknah | ααΌαααΆα αααααΆα αααααα α
αααΆαα ααΆααΆα αααααα | lazy | hplt |
ngeai sruol nhing japrorhosaaaa | ααΆα αααα½α αα·α ααΆαααα αα | mixed | hplt |
daoy louk nak gruan dai choch mau | ααα ααα α’ααα ααααΆαα αα α
α»α
ααα
| lazy | hplt |
khang jveng jap new tianh tow | ααΆαααααα α
αΆαα new ααΆα αα
| mixed | hplt |
LOE TOMRONGROUPPEAB LOK NEAK | ααΎ ααααα ααΌαααΆα ααα α’ααα | lazy | hplt |
bnnrh ( travte thveu bonti mdang2 kom ao huorh | ααα»ααααα(ααααααααααΎααααα·ααααααα»αα²ααα α½α | lazy | hplt |
kangkrav roppeab ) | ααΆααααα
ααΌαααΆα) | lazy | hplt |
>> Magic | >> Magic | mixed | hplt |
wand Tools prae prasaTools nih | wand Tools ααααΎ ααααΆααTools ααα | mixed | hplt |
knong ka Select | αααα»α ααΆα Select | careful | hplt |
leurouppheap dael | ααΎ ααΌαααΆα ααα | mixed | hplt |
mean mean | ααΆα ααΆα | mixed | hplt |
phtey Background por no tae muy haey men | αααα Background αααα αααα½α α αΎαααΆα | lazy | hplt |
laknah cbas vea | αααααα α
αααΆαα ααΆ | careful | hplt |
mian laknah ngii ng jabror ho so bomput douy | ααΆα αααααα ααΆα αα·α ααΆαα αα αα αααα»α ααα | lazy | hplt |
louk neakgroan'tae choch tov loe rouppap | ααα α’ααα ααααΆαααα α
α»α
αα
ααΎ ααΌαααΆα | careful | hplt |
brl nuhwa ng choen | αααααα ααΆ αα·α α
αα | mixed | hplt |
Select Vaccine douy kluon tov ler | Select Vaccine ααα αααα½α αα
ααΎ | careful | hplt |
roupiab teanglay na dail | ααΌααΆα ααΆααα‘αΆα ααΆ ααα | mixed | hplt |
mean pon sradeang kna | ααΆα ααα ααααααααααΆ | mixed | hplt |
novknong | αα
αααα»α | mixed | hplt |
krom Rectangular Marquee Tools: prae | αααα»α Rectangular Marquee Tools:ααααΎ | mixed | hplt |
somrab tvoe kar katrouppheap mean reang | ααααΆαα ααααΎααΆα ααΆαα ααΌαααΆα ααΆα ααΆα | mixed | hplt |
buon jrong nhing mean booking ringjea | αα½α αααα»α αα·α ααΆα booking ααΆα ααΆ | lazy | hplt |
Nurse rungvung via mian Tools | nurse αααααα ααΆααΆα Tools | careful | hplt |
cham'nuon 4 dl min leak brand | α
ααα½α4 ααα ααΆα αααα brand | mixed | hplt |
srodiang kna | αααααα ααααΆ | lazy | hplt |
douc khangkraomea | ααΌα
ααΆααααααα | mixed | hplt |
>> Single row | >> Single row | careful | hplt |
Marquee Tools: Praeu Samrab Kat Rubpheab Chea | Marquee Tools: ααααΎ ααααΆαα ααΆαα ααΌαααΆα ααΆ | formal | hplt |
reang trou weng del mean teung | ααΆα αααααα ααα ααΆα ααΉα | careful | hplt |
jamnuon 1 pech sail (Pixel) ge praeu | α
ααα½α 1 αα·α
ααα(Pixel)αα ααααΎ | careful | hplt |
vea | ααΆ | mixed | hplt |
somrab sop dveuja sday jnot jnot map | ααααΆαα αα»α ααααΎααΆ ααααΆα ααααΌαα map | lazy | hplt |
bandouy pi | ααααααααΈ | mixed | hplt |
jveng tea tov sdam | ααααα tea αα
ααααΆα | careful | hplt |
>> Single column Marquee Tools: praeu somrap | >> Single column Marquee Tools: ααααΎ ααααΆαα | mixed | hplt |
kat rouppap jea reang tro | ααΆαα ααΌαααΆα ααΆααΆα ααα | mixed | hplt |
veng 7am del man toeng | ααα 7am ααα ααΆα ααΉα | careful | hplt |
jom nuan 1 pic sael (Pixel) ke proe | α
ααα½α 1 αα·α
ααα(Pixel)αα ααααΎ | mixed | hplt |
va somrab sob thveujea sdai jnoud | ααΆ ααααΆαα αα»α ααααΎααΆ ααααΆα ααααΌα | careful | hplt |
Piler joh krom. | α ααΈααΎα
α»ααααααα | careful | hplt |
nouknong krom Lasso Tool | αα
αααα»ααααα»α Lasso Tool | mixed | hplt |
nih man | αααααΆα | lazy | hplt |
Obkornor prohak prohel knea | α§ααααα αααα αΆαα αααα αα ααααΆ | mixed | hplt |
chamnuon weekend bei ke | α
ααα½α weekend ααΈ αα | lazy | hplt |
Prae wia samrap kat tovtam rouppiap del ke chong | ααααΎααΆ ααααΆαα ααΆαα αα
ααΆα ααΌαααΆα ααα ααα
α | careful | hplt |
baan | ααΆα | mixed | hplt |
pol gi ge meun jong ban Back groundke teng | ααα ααΈ αααα·α α
αα ααΆα Back ground ααααα | careful | hplt |
tae proe Tools torm bi ni. | ααααααΎ Tools ααΆαααΈ αααα | careful | hplt |
Tools teang nuh mean doj | Tools ααΆαα αααααΆα ααΌα
| careful | hplt |
ja | ααΆα | mixed | hplt |
>> Lasso | >> Lasso | careful | hplt |
Tool: praeu samrab kat | Tool: ααααΎααααΆαα ααΆαα | formal | hplt |
rup phap douy prer mou joch jop | ααΌαααΆα ααα ααααΎ αα
α
α»α
ααΆαα | mixed | hplt |
chea nuch mean leakna bat baen | ααΆ αα·α
αα
ααΆα αααααα ααααα | careful | hplt |
TOV TAM JUT COMNOAM | αα
ααΆα α
α·ααα α
αααΆα | mixed | hplt |
pel katmun triv leng mau nus tey | ααα ααΆαα αα·αααααΌα ααα ααα
ααα αα | mixed | hplt |
trov choch mou o choap | ααααΌα α
α»α
αα
α²ααααΆαα | careful | hplt |
cheanech. | ααΆαα·α
αα
α | lazy | hplt |
>> Polygonal Lasso Tool: Praeu Samrab | >> Polygonal Lasso Tool: ααααΎααααΆαα | formal | hplt |
Select kat tov leu rouppeab daoy | Select ααΆαα αα
ααΎ ααΌαααΆαααα | mixed | hplt |
preumav choch tou leu | ααααΎααα
α
α»α
αα
ααΎ | lazy | hplt |
vea | ααΆ | mixed | hplt |
mdang mouy domnat2 | αααα αα½α ααααΆααα | careful | hplt |
vea mean leaknah chea bantoat roat trang trang | ααΆααΆα ααααααααΆ αααααΆαα ααΆαα αααααα | formal | hplt |
>> Magnetic Lasso Tool: | >> Magnetic Lasso Tool: | mixed | hplt |
proe samrap kat rupphap tov tam | ααααΎ ααααΆαα ααΆαα ααΌαααΆα αα
ααΆα | careful | hplt |
tum rung robas vea del yeung changban vea mean peapngeay sruol douy lok | ααααα ααααααΆ ααα ααΎα α
ααααΆα ααΆ ααΆα ααΆαααΆα αααα½α ααα ααα | careful | hplt |
neak ticket kruan tae ous tov tam | α’ααα ticket ααααΆαα αα α’αΌα αα
ααΆα | mixed | hplt |
Tumrung rubhaeuy novpel dol | αααααααΌα α αΎα αα
ααα ααα | mixed | hplt |
kanleng na man sa peabmeun jbas greanor | αααααα ααΆααΆα αααΆα αα·αα
αααΆαα ααααΆα | mixed | hplt |
te chuch | ααα
α»α
| lazy | hplt |
mow jweing te suang bnn. | αα
ααααα αα αα½α ααα»αααααα | lazy | hplt |
>> tek thla Slice Tool: prey | >> ααΉαααααΆ Slice Tool: ααααΈ | mixed | hplt |
somrab tveu kar kat roubpab | ααααΆαα ααααΎααΆα ααΆαα ααΌαααΆα | mixed | hplt |
chea | ααΆ | careful | hplt |
pnaik2 deltrov yok hot | αααααα ααα ααααΌα αα hot | careful | hplt |
tov prae daembey phchorp | αα
ααααΎ ααΎααααΈ ααααΆαα | careful | hplt |
tov neungHTML Text nov knung Website | αα
αα·α HTML Text αα
αααα»α Website | careful | hplt |
>> prer samrab kat | >> ααααΎααααΆαα ααΆαα | careful | hplt |
jam damruem kaem ney rubpheab teangmul | jam ααααΉα ααα αα ααΌαααΆα ααΆααααΌα | formal | hplt |
Posat domrum yok tae reang | αααα·αααΆαα ααααΉα αα αα ααΆα | careful | hplt |
bon jrong khang knung tae bon noh robeap brer ngiaysrol lok | αα½ααααα»αααΆα αααα»αααααα»ααααα ααααααααΎ ααΆααααα½α ααα | mixed | hplt |
nekkroan te | α’ααα ααααΆαααα | mixed | hplt |
choch mao chveng oy chop haey | α
α»α
ααα
ααααα α’ααααΆαα α αΎα | careful | hplt |
dianh tow ler ko nleng roubpiab dael louk | ααΆα αα
ααΎα αααααααΌαααΆα ααα ααα | careful | hplt |
nak jang tver ka Crop | α’ααα α
αα ααααΎααΆα Crop | careful | hplt |
novbontuabmok joch Key | αα
αααααΆαααα α
α»α
Key | careful | hplt |
Enter nov pel nus roobpeab krau | Enter αα
αααααα ααΌαααΆα αααα
| careful | hplt |
Select raboh size lok neak ningggg | Select αααα size αααα’ααα αα·α | mixed | hplt |
bat dear novsal te rouppheap nov knung Select ponnua | ααΆαα dear αα
ααααα ααΌαααΆα αα
αααα»α Select ααα»ααααα | mixed | hplt |
. training | α training | careful | hplt |
>> Spot Healing Brush Tool: prer | >> Spot Healing Brush Tool: ααααΎ | careful | hplt |
somrap dvaeka lup bambat cham nuch toch tochdel mean | ααααΆαα ααααΎααΆα αα»α ααααΆαα α
ααα»α
ααΌα
α ααα ααΆα | mixed | hplt |
NOW LEU ROUBPEAB DOU CHEA MOUN BRCHROUY NG CHAMNUC PSE | αα
ααΎ ααΌαααΆα ααΌααΆ αα»α ααααααα»α αα·α α
ααα»α
αααα | mixed | hplt |
teat jea jraen teat β¦. | α ααα ααΆ α
αααΎα αααβ¦. | mixed | hplt |
>> Patch tool: prer somrab Select | >> Patch tool: ααααΎααααΆαα Select | mixed | hplt |
Samdei Type Fine-tuning Data
These are the exact fine-tuning (stage 2) and adaptation (stage 3) pairs of the Samdei Type romanized-Khmer β Khmer-script converters, Samdei Type Mini and Samdei Type Base.
Pretraining (stage 1, samdei/samdei-type-pretraining-data) teaches Khmer words and spelling from 6.5β12.8M generator pairs. These two short stages teach how people actually type in chat:
- Stage 2: looser LLM spellings, spelling habits mined from public word lists, and chat-like abbreviations.
- Stage 3: the spoken register and stretched letters.
Each stage is 3β6k steps (10β36 minutes on a laptop).
Every Khmer target is real text, except 2,483 chat sentences written by Gemma 4. Only the romanized input varies, by source.
Paper: technical report. Inference code: github.com/Samdei-Labs/samdei-type. Test set: samdei/samdei-type-bench.
Configs
The data is organized by kind, so each source of spelling noise can be used, inspected or left out on its own. Stages 2 and 3 share one dataset because stage 3 replays the whole stage-2 mix; the stage-3 data is the casual config.
| Config | Files (rows) | What it is | Stage | Used by |
|---|---|---|---|---|
chat_mix (default) |
ft_v2 (629,878) |
a chat-heavy sample of generator pairs: more LLM-written chat sentences and ALT, fewer web pieces; English chat words and place names inserted | 2, 3 | Mini, Base |
llm_romanized |
train_gemrom2 (18,273) |
Gemma 4 romanizations of 6,382 real sentences, 4 style prompts, Γ3 copies with random casing | 2, 3 (Γ4 in the mix) | Mini, Base |
abbreviated |
ft_v2_abbr3 (139,704), train_gemrom2_abbr3 (16,302) |
chat-like abbreviated copies of chat_mix and llm_romanized pairs; only the input changes (tov β tv, bong β bg) |
2, 3 (Gemma part Γ2) | Mini, Base |
mined_tables_chh |
ft_rules_chh (289,820) |
new pairs spelled with the table mined from the Chhunneng word list (dictionary-like scheme), alone or with mixed / lazy noise habits | 2, 3 | Mini |
mined_tables_chh_wd |
ft_rules_chhwd (289,647) |
made the same way from the same sources, but spelled 70% with the Chhunneng table and 30% with the table mined from Wikidata names (official, name-style scheme) | 2, 3 | Base |
casual |
ft_casual_noakk (426,946) |
21.6k spoken-register web sentences, spelled by the generator with stretched letters (tovvv, baaan) |
3 only | Mini, Base |
synthetic_heldout |
train_v2_s3.dev (12,232) |
synthetic held-out pairs monitored during training (validation split) | β | β |
Mini and Base use the same data. The only difference is the mined-table file: Mini uses mined_tables_chh, Base mined_tables_chh_wd.
| Mix | Files, with repeats | Pairs |
|---|---|---|
| Stage 2, Base | ft_v2 + ft_v2_abbr3 + train_gemrom2 Γ4 + train_gemrom2_abbr3 Γ2 + ft_rules_chhwd |
1,164,925 |
| Stage 2, Mini | the same with ft_rules_chh |
1,165,098 |
| Stage 3 | the stage-2 mix of the model + ft_casual_noakk |
1,591,871 (Base), 1,592,044 (Mini) |
recipes.json gives the exact mix of each checkpoint (files in order, repeats, steps, learning rate, initialization), read from the training arguments stored in the checkpoints.
generator/ ft_v2.parquet
llm_romanized/ train_gemrom2.parquet
abbreviated/ ft_v2_abbr3.parquet train_gemrom2_abbr3.parquet
mined_tables/ ft_rules_chh.parquet ft_rules_chhwd.parquet
casual/ ft_casual_noakk.parquet
validation/ train_v2_s3.dev.parquet
recipes.json
Columns
| Column | Content |
|---|---|
latin |
romanized input, as the model saw it |
khmer |
Khmer-script target, normalized with khnormal; English words, digits and brand names are copied from the input |
persona |
how the input was made (below) |
source |
where the Khmer target comes from: hplt, fineweb2, wikipedia, alt (a piece of a web sentence); gemma (a piece of an LLM-written chat sentence); casual (a piece of a spoken-register HPLT or FineWeb-2 sentence); gazetteer; gemrom_<x> (a whole sentence from <x>, romanized by Gemma 4) |
persona |
Meaning |
|---|---|
formal, careful, mixed, lazy |
the generator's typing personas, from standard spelling to heavy chat shortcuts and typos |
place |
standalone place-name pair from our gazetteer |
rules_chh, rules_chh+mixed, rules_chh+lazy |
Chhunneng-mined table, alone or with the noise habits of mixed / lazy |
rules_wd, rules_wd+mixed |
Wikidata-mined table |
β¦~stretch |
the persona also stretches letters inside the sentence (stage 3) |
gemrom0βgemrom3 |
Gemma 4 romanization with one of four style prompts |
abbr |
abbreviated copy of another pair |
How to load
from datasets import load_dataset
casual = load_dataset("samdei/samdei-type-finetuning-data", "casual", split="train")
abbr = load_dataset("samdei/samdei-type-finetuning-data", "abbreviated", split="train")
To rebuild a mix exactly, read recipes.json and concatenate its files in order, with repeats. Training starts from the stage-1 checkpoints, samdei/samdei-type-base-pretrained and samdei/samdei-type-mini-pretrained.
How the pairs were made
chat_mix. The persona generator (second version) on a chat-heavy sample of sentences. The generator works in four steps:- it cuts sentences into pieces of 1β12 words;
- it pronounces each word with Google's Khmer pronunciation lexicon;
- it spells the piece in a persona (formal 8%, careful 32%, mixed 40%, lazy 20%);
- it adds noise (glued or split words, dropped vowels, swaps, typos), English chat words and place names.
llm_romanized. Gemma 4 12B, run locally, romanized real sentences "the way Cambodians type on Messenger", 10 sentences per call. The Khmer side stays the original sentence, so the target is always correct, and Gemma's looser spelling acts as human-like noise on the input. 97% of the outputs passed our filters.mined_tables_*. A rule miner learns per-slot spelling tables (onset, vowel, coda, glide, context) by Viterbi alignment and hard EM from public (Khmer word, romanization) lists: the Chhunneng word list (24.9k pairs, MIT) and 1.6k Wikidata person and place names (CC0). Words that are targets of a word-level evaluation set are removed first. The tables are insamdei/samdei-type-resources. The pairs here are new sentences (HPLT, FineWeb-2, ALT, chat sentences) spelled with them.abbreviated.- It covers 40% of
chat_mixpairs and allllm_romanizedpairs. - Operations: consonant skeleton (
bongβbg), dropped vowels, first + last letter (jongβjg), dropped digraph h (khlangβklang), dropped cluster member. - Vowels are dropped only from one-syllable words among the 1,000 most frequent.
- An abbreviation that the chat lexicon reserves for one word (
nhfor αααα»α) is never produced for another.
- It covers 40% of
casual. 21.6k HPLT and FineWeb-2 sentences that contain one of 13 spoken-register words (α αααΉα, α αΉα, α’αΈ, ααα, β¦) as a token; misordered PDF text is dropped. They are spelled with stretched letters by the generator (formal 5%, careful 25%, mixed 35%, lazy 15%, Chhunneng table with mixed habits 20%). No Akkhara-derived spelling rules are used.
Test separation.
- Every sentence of every evaluation set, including Samdei Type Bench and its chat context, is banned in the generator.
- The rule miner never sees a test row or a word-level test target.
- No Samdei Type Bench message equals a training target. The only shared strings are common single words.
Reproducibility. These files are the exact pairs the released models were trained on, so training can be repeated without regenerating them. The data generator and the training code are not public; the released code (github.com/Samdei-Labs/samdei-type) is for inference only.
Limitations
- Synthetic inputs. No row is real typing. Chat habits are covered only as far as our personas, two mined tables, one LLM and our abbreviation rules reach.
- Scheme bias of the mined tables. The Chhunneng table follows a dictionary-like scheme and the Wikidata table an official name scheme; neither is chat spelling as such.
- Pieces, not messages. Inputs are short (median 26 characters).
- Personal data. Web text may contain names, phone numbers or other personal data of the original authors. Lines with URLs or e-mail-like strings were dropped, but no other personal-data filter was run. To have a row removed, open a discussion in the Community tab.
- LLM text. The Gemma-written chat sentences and romanizations can contain mistakes.
License and attribution
CC BY-SA 4.0 (license text). The rows contain pieces of the source sentences, so the source licenses travel with them: Khmer Wikipedia's share-alike terms set the license of the whole set, and rows from the other sources also stay under their own terms below.
| Source | Side | License |
|---|---|---|
HPLT v2 (HPLT/HPLT2.0_cleaned, khm_Khmr) |
Khmer | CC0 packaging; Common Crawl terms of use apply to the text |
FineWeb-2 (HuggingFaceFW/fineweb-2, khm_Khmr) |
Khmer | ODC-By 1.0; Common Crawl terms of use apply to the text |
Khmer Wikipedia (wikimedia/wikipedia, 20231101.km) |
Khmer | CC BY-SA 4.0 (share-alike) |
| Asian Language Treebank (ALT), Khmer side | Khmer | CC BY 4.0 |
| LLM-written chat sentences (ours, Gemma 4) | Khmer | CC BY 4.0 (Gemma 4: Apache-2.0) |
| Gemma 4 romanizations (ours) | Latin | Apache-2.0 model output; Khmer side as its source |
| Google Khmer pronunciation lexicon | Latin (generator) | CC BY 4.0 |
| Chhunneng romanized Khmer dataset (Β© 2025 Chrea Chanchhunneng) | Latin (mined table) | MIT |
| Wikidata names | Latin (mined table) | CC0 |
| Chat lexicon, place names, English chat words (ours) | Latin (generator) | CC0 |
Rows whose source is wikipedia or gemrom_wikipedia carry Wikipedia's share-alike terms. They occur only in chat_mix, llm_romanized and abbreviated; the mined_tables_* and casual configs contain none.
The MIT notice of the Chhunneng dataset, which the mined_tables_* tables derive from, is included as LICENSE-chhunneng.txt.
Changes to the source text. The sentences were normalized with khnormal, filtered and deduplicated; the generator cut them into pieces of 1β12 words (the llm_romanized rows keep whole sentences). Each Khmer target is paired with a romanization in the latin column (see "How the pairs were made").
Citation
@misc{samdei-type-2026,
title = {Samdei Type: Separating What Is Written from How It Is Spelled for Romanized Khmer Transliteration},
author = {Mengsay Loem},
year = {2026},
note = {Technical report},
url = {https://github.com/Samdei-Labs/samdei-type/blob/main/paper/samdei-type.pdf}
}
Contact: open a discussion in this repository's Community tab.
- Downloads last month
- 71