Given a word and its reading, I'm interested in having it be automatically be identified as Japanese-origin (和語, but excluding 和製英語 and 和製漢語) or not. False negatives and positives aren't really a problem for my use case, but naturally they should be minimised if they can.
What would be some heuristics that can be implemented relatively easily in a computer program (say, less than 100 lines per rule (using libraries is fine) and doesn't require large amounts of training data)?
Here are some that I've thought up of. Are they correct? Can they be refined?
- Word does not contain "litte" characters (excluding ヶ) nor long vowels (not so sure about this one)
- Number of syllables (which I suppose is equivalent to mora if the above rule is true) is longer than the number of characters in the word
- Word contains at least one kanji or hiragana (to exclude most of the 外来語)