seanghay/khmer-spelling-corrections

Khmer Spelling Corrections

Naturally occurring Khmer misspellings paired with the word the writer meant.

The labels are not annotated, they are observed. Search sessions record the whole typing trajectory toward a single word, so when a user types something, fails, adjusts and lands on a real dictionary headword, the failed attempt and the headword form a correction pair produced by a real person under no instruction to make mistakes.

Only pairs within two edits of the target are kept, and only where at least 5 distinct people made the same mistake — so every row is an error the writing system reliably induces, not a one-off slip.

Columns

column	description
misspelling	what was typed
correction	the headword the session resolved to
sessions	sessions exhibiting this pair
clients	distinct anonymous clients contributing (always >= 5)

Uses

Training and evaluating Khmer spell checkers, search query correction, OCR and ASR post-correction, and keyboard design. Khmer error-correction resources are scarce; this is drawn from real usage rather than synthetic corruption, so the error distribution reflects what the script actually causes people to get wrong — register consonant pairs, vowel height, and a spurious final រ above all.

Limitations

The target is what the session resolved to, which is strong evidence of intent but not a human annotation. A user who gives up and searches a different word can contribute a wrong pair. The 5-client threshold suppresses most of this but the data has not been manually verified.

Derived from anonymous search logs of khmerdict.com, collected December 2024 - August 2026 across roughly 3.2M searches by ~153,000 anonymous clients.

Privacy

This dataset contains no logs and no identifiers. It was produced by an aggregation pipeline that:

• requires every row to be attributable to at least 5 distinct clients, so no row can describe a single person's search;
• discards client identifiers, IP addresses, user agents and timestamps entirely — they are never read into the output;
• restricts all Khmer text to single words matching [\u1780-\u17DD]+, excluding digits, Latin characters, punctuation and whitespace, so no row can carry a phone number, email address, URL or freeform phrase;
• constrains words to dictionary headwords or strings within two edits of one.

At k=5, roughly 9% of distinct queries survive while retaining about 79% of total search volume: the discarded tail is where re-identification risk lives, not where the linguistic value is.

The source logs are not published and will not be.