Lontara

The problem

Lontara is the script of the Bugis language of South Sulawesi. It has 23 consonants and five vowel signs, but no way to write a consonant on its own. So it never records consonants at the end of a syllable, doubled consonants, a nasal before a consonant, or the glottal stop. The words mata, matta and manta are all written the same way. A reader picks the word from context.

Existing converters hide this. They return one confident Latin spelling, which is a guess presented as an answer. A signmaker can put a misspelled name on a permanent sign. A learner can be taught a rule that doesn't exist. Someone reading family manuscripts gets one reading where six are valid, with no sign that a choice was made.

What I built

The tool lists every reading the script allows

Paste Lontara into the reader and it returns a tree of possible Latin readings. Each branch point is labelled with the kind of ambiguity that caused it: a final consonant, a doubled consonant, a nasal or a glottal stop. Each choice the tool makes must carry a reason, so it can't pick a reading silently. The writer goes the other way. It lays Latin onto a palm-leaf band and marks lost information on the stroke where it was dropped.

The spelling rules live where a reviewer can read them

The eleven spelling rules sit in a data file, each marked as cited, provisional or derived, with its citation. The code only interprets that file. A Bugis reader who doesn't program has to be able to audit the orthography, and one page in the app shows every rule with its source.

Readings come from a word list, never from a guessed rule

The obvious reader would try every consonant that could close each syllable. That needs a list of Bugis final consonants, and I couldn't find one to cite. Guessing it would produce confident readings that aren't Bugis. So a reading counts only if it is a known word that the writer turns back into exactly the same Lontara. The cost is that results are only as complete as the word list, and every result says so.

Real Bugis writing caught a 56% error rate

Bugis Wikipedia has pairs of Lontara and Latin written by people who use the script. I extracted 97 of them and scored my rules against them. They agreed 56% of the time. The test found three mistakes. Two vowel letters were the reverse of what their Unicode names suggested, é was missing entirely, and people write the glottal stop as q. Fixing them raised agreement to 94%. Two other oddities, with two cases each, may be errors in the articles, so I left them for a reviewer.

The word list claims only what it can support

The word list has 1,323 entries, and each one records where it came from. 85% of the source dump turned out to be bot-generated stubs, so the list makes no claim about how common a word is, and the build fails if an entry does. Some entries are Indonesian, English or Vietnamese place names the tool can't tell apart from Bugis, so the reader badges those readings and ranks them low.

Result

Lontara is live and public, in Indonesian and English, and works offline after the first visit. It includes an on-screen Lontara keyboard, because most phones have none. By its own rules it is deployed but not launched. No Bugis reviewer has signed off yet, the code enforces that gate, and every page says so. The spelling page lists seven open questions I recorded instead of guessing. It says plainly that it is a personal project, not an authority, and can be wrong. It transliterates and never translates. I built it alone in six days, from late July to early August 2026.