One English Spelling, Thirteen Different Thai Words
TL;DR: Measured across 16,559 Thai dictionary headwords, 3,334 of them share their official English spelling with at least one other word. That is one word in five. The worst strings, chan and sai, each carry thirteen unrelated words. Writing the tones and vowel lengths, as Paiboon does, cuts it to one word in twenty-three but does not close it.
Romanisation is a one-way street, and the return trip is the problem
Everybody knows romanisation loses the tones. It is the first thing any guide to reading Thai tells you, usually with the same five words that all come out khao. What the anecdote does not tell you is how often it happens, and that turns out to be the number that matters, because it decides whether a romanised word can be turned back into a Thai one at all.
Going one way is fine. You have ข้าว in front of you, you want to say it out loud, and khao gets you most of the way there. Going the other way is where it falls over. Somebody texts you the name of a dish, or a guidebook prints a street, and you are standing in front of a sign trying to work out which of the Thai words on it is the one you were told about. There is no rule that gets you back, because the spelling that was thrown away is the only thing that distinguished them.
What was measured
The dictionary frame published with the romanisation study holds 17,274 Thai headwords, each carrying two romanisations: one under the Royal Institute system, which is what road signs, passports and maps use, and one under Paiboon, the dictionary system that writes tone and vowel length. Removing the 715 entries whose whole definition is that they are an obsolete, archaic or misspelled form of another word leaves 16,559 distinct Thai spellings, and those are what the counts below are taken over.
Then the question is simply asked backwards. Group the words by their romanisation instead of by their spelling, and count how many groups hold more than one word. Anyone can repeat it: the dataset is published as a CSV beside the study, under the licences named on its page, and the two columns are ri_wiktionary and paiboon_wiktionary.
Phayan is an app for learning to read Thai script, letter by letter from zero, with audio on every word and progress kept in the browser before there is any account to keep it in. This measurement is the argument for its existence stated as a number, which is worth declaring up front rather than pretending to have stumbled on it.
One word in five loses its identity on the official system
The 16,559 words produce 14,317 distinct Royal Institute spellings. Most of those are unique, and the collisions concentrate: 1,092 strings, 7.6 per cent of them, are shared, but because a shared string usually holds three or four words rather than two, they account for 3,334 of the words themselves.
| System | Distinct spellings | Shared spellings | Words affected | Worst case |
|---|---|---|---|---|
| Royal Institute (signs, passports) | 14,317 | 1,092 (7.6%) | 3,334 (20.1%) | chan, sai: 13 words each |
| Paiboon (tone and vowel length) | 16,167 | 326 (2.0%) | 718 (4.3%) | gaan: 6 words |
| Thai script | 16,559 | none | none | every word its own spelling |
The third row is there because it is easy to miss what it means. The Thai spelling is what the whole dataset is keyed on, so of course it does not collide. That is the point rather than a trick: the script is the only representation of a Thai word that is guaranteed to name exactly one word, and every romanisation is a compression of it that throws some of that away.
chan, thirteen different words
The largest collision is worth seeing in full, because a percentage does not convey what it is like to meet one. Every word in this table is written chan on a sign.
| Thai | Paiboon | Means |
|---|---|---|
| ฉัน | chǎn | I, me |
| ชั้น | chán | floor, class, grade |
| จาน | jaan | plate, dish |
| จันทร์ | jan | the moon, and so Monday |
| ชัน | chan | steep |
| ชาญ | chaan | skilled, proficient |
| ชาน | chaan | an outer platform or terrace |
| ฉาน | chǎan | Shan, the people and the state |
Five more sit under the same string: จัน and จันทน์, two trees; จาร, to inscribe; ฉันท์, a kind of contentment; and ฌาน, the meditative absorption. Notice what the first column is doing. ฉัน and จาน and ชั้น are not near each other in sound, in tone or in meaning. They are near each other only in the romanisation, which is an artefact of the writing system and nothing to do with Thai.
The words you meet first are the worst affected
One word in five sounds survivable. It is not the number a beginner experiences, because collisions are a property of short words, and short words are the common ones. The dataset marks which of its headwords appear in the Phayan teaching corpus, which is the vocabulary a learner meets in the first levels: 2,149 of them. Of those, 1,596 share their Royal Institute spelling with another word. That is 74.3 per cent.
The same 2,149 words in Paiboon: 327 collide, or 15.2 per cent. So tone marks and vowel length do most of the work, and the residue is still one word in seven. For the vocabulary you are actually learning, the official romanisation is ambiguous about three times out of four, and the learner-friendly one about once in seven.
What writing the tone actually buys you
Paiboon exists to fix precisely this, and mostly it does: 20.1 per cent falls to 4.3 per cent, which is a four-fold improvement from adding two pieces of information. But the residue is instructive. The worst Paiboon collision is gaan, six words: การ (work, and the prefix that makes nouns of verbs), กาล (time), กานต์ and กานท์ (beloved, and verse), การณ์ (a matter or circumstance), and กาฬ (black). They are homophones in speech too. No romanisation can separate them, because the thing separating them is not pronounced.
That is the honest ceiling on the whole idea. A romanisation is a transcription of how a word sounds, so the most it can ever do is tell apart words that sound different. Thai has a large number of words that sound identical and are spelled differently, mostly from Pali and Sanskrit, and those are invisible to any system that writes sounds. It is the same phenomenon as 42 letters spelling 21 sounds, seen from the other end.
What this does not say
It does not say romanisation is bad. It is doing a job it is good at. The Royal Institute system exists so that someone who cannot read Thai can pronounce a street name well enough to be understood by a taxi driver, and for that a lost tone is an acceptable price. Two different words arriving at the same sign is a feature of the job, not a defect in the system.
What the numbers say is narrower: a romanisation cannot be used as an index. You cannot look a word up by it, you cannot match it against a sign reliably, and you cannot learn vocabulary through it without learning ambiguity along with the words. Twenty per cent of the dictionary and seventy-four per cent of a beginner vocabulary is the size of the ambiguity, and it is not reducible by choosing a better system, only by reading the letters. Forty-two of those, in everyday use, is the whole of what it takes; the beginner path through them is the shortest route, and the comparison of the systems themselves covers what each one is for.
Frequently Asked Questions
Why do different Thai words have the same spelling in English?
Because English letters cannot carry the tone, the vowel length or the choice of Thai letter, and in Thai all three change the word. Measured over 16,559 dictionary headwords, 3,334 of them share their Royal Institute spelling with at least one other word, so one word in five is ambiguous before you start. Phayan sidesteps the problem by teaching the script itself rather than a romanisation, which is the same reason Anki decks built on Thai characters outlast ones built on romanised fields.
How do I find the Thai spelling of a word I only know romanised?
Look it up by meaning rather than by the romanisation, and confirm it against the tone, because the romanisation alone cannot get you back to one word. A dictionary that shows the Paiboon form narrows it a long way, since only 4.3 per cent of words still collide once the tone and vowel length are written. Phayan shows the Thai spelling beside every word it teaches for that reason, and ThaiPod101 lessons that print both forms are usable the same way.
Is it safe to learn Thai vocabulary from romanisation only?
It is safe for a phrase or two and unsafe as a method, because the ambiguity is concentrated exactly where a beginner lives. Of the 2,149 dictionary words that appear in the Phayan teaching corpus, 1,596 share their official romanisation with another word, so three quarters of an early vocabulary is learned without the thing that identifies it. Pocket Thai Speaking is honest about this trade and is a reasonable choice if speaking a fixed set of phrases is genuinely all that is wanted.
What is the difference between RTGS and Paiboon romanisation?
RTGS is the Royal Institute system used on road signs, passports and maps, and it writes neither tone nor vowel length; Paiboon is the dictionary system that writes both. The difference is measurable: RTGS leaves 20.1 per cent of dictionary words sharing a spelling and Paiboon leaves 4.3 per cent. Phayan displays Paiboon by default with an RTGS toggle, because the two answer different questions and a learner needs the first one.
So how long before I can just read the sign instead?
Sooner than the size of the alphabet suggests, because signs reuse a small set of words and the letters are learned in an order that front-loads the common ones. The honest answer is that it depends on how often you practise, and anyone quoting a fixed number of days is selling something. Phayan runs letter by letter from zero and keeps your progress without an account, so the first session costs nothing to try.