phonsobon/khmer-khmer-dictionary

Khmer-Khmer Dictionary — cleaned for model training

This is not the original dictionary. It is a cleaned version of the Khmer Dictionary 2022 (វចនានុក្រមខ្មែរ ២០២២) of the National Council of Khmer Language (NCKL), Royal Academy of Cambodia, reshaped for training models. Rows were removed and fields were changed, as listed below. For the authoritative text, use the official NCKL dictionary.

One row per sense, in the flat layout of metythorn/khmer-khmer-dictionary. Every row has both a word and a pronunciation.

Summary

	rows
Original dictionary rows	44,698
Removed (word without pronunciation)	164
This dataset	44,534 (38,511 unique words)

What was removed

164 rows that had a word but no pronunciation:

	rows	example
Idioms, proverbs and sayings (sub-entries)	161	អន្ទង់វែងកុំរកឆ្នាំងវែង, កណ្ដៀរកោរកណ្ដុរកាត់
Headwords without pronunciation	2	ភ្នែកសព្រាត, សច្ច
Second sense of one of those headwords	1	ភ្នែកសព្រាត

Apart from ភ្នែកសព្រាត, every removed word still appears in this dataset in another entry that has a pronunciation.

What was changed

change	rows	example
Additional senses filled in: the original leaves word and pronunciation blank on the 2nd, 3rd … sense of a word; they are copied from the entry	3,611	2nd sense of កង់បី → កង់បី [កង់-បី]
Sub-entries use their own word (the original puts the headword in the word column and the sub-word elsewhere)	13,464	ហ្វារ → ហ្វារសមុទ្រ [ផា-សៈ-ម៉ុត]
Homograph markup converted to superscript numbers	518 cells	កេងកង{¥¥}១{££} → កេងកង¹
Columns kept: word, pronunciation, poly, pos, definition, examples (etymology, notes and internal IDs dropped)	all

Columns

column	example	description
word	កង់ទីន	headword or sub-entry
pronunciation	[កង់-ទីន]	NCKL Khmer-script respelling, - between syllables; alternatives joined with ឬ. Exactly as in the source dictionary, not cleaned: reading notes such as [ភ័ស] (ជើង ត) and source typos are kept
poly	១)	sense number when the word has several senses
pos	ន.	part of speech (ន. noun, កិ. verb, គុ. adjective, …)
definition	(ម.ព. អាហារដ្ឋាន)។	definition
examples		usage examples

from datasets import load_dataset
ds = load_dataset("phonsobon/khmer-khmer-dictionary", split="train")

Known issues

About 20 phrase sub-entries carry only their headword's pronunciation in the original dictionary (e.g. ហាមាត់ចង្គ្រុង [ហា], រាជ្យសម្បត្តិ [រាច]). They are kept here as in the source. For pronunciation (G2P) training, use phonsobon/khmer-G2P, where these and other unusable pairs are removed and the data is split into train / validation / test.

Built with build_khmer_khmer_dictionary.py (in this repository) from the raw NCKL CSV.