Latin Macronizer - Help

← Back to the macronizer ↑ All Languages

What it does

Latin was written without marking vowel length, but length is what makes the language scan and, often, what tells two words apart: malum (an evil) and mālum (an apple), liber (free) and līber (a book). This tool reads your Latin and adds the macrons — the bars over long vowels.

It is not a dictionary lookup. The same spelling can be several different words, so the tool has to work out which one you meant from the sentence around it: rosa is nominative rosa or ablative rosā depending on the grammar. That is why it needs a part-of-speech tagger, and why it sometimes gets it wrong — see reading the output.

How it works

Everything runs inside your browser; your text is never sent to a server. Your text passes through four stages:

• Tokenizing — the text is split into words, and enclitics (-que, -ve, -ne) are separated off, so virumque is understood as virum + -que.
• POS tagging — RFTagger, trained on the Latin Dependency Treebank, decides each word's part of speech, case, number, gender, tense and so on from its context.
• Lookup — the tagged form is looked up in a wordlist of about 812,000 Latin word-forms with their vowel lengths.
• Morphological analysis — words that aren't in the wordlist go to Morpheus, the Perseus project's Latin analyser, which works the form out from stems and endings.

Reading the output

Most words come back as plain macronized text. Two kinds are highlighted, and both are invitations to check the tool's work:

• Yellow — the word has more than one possible reading, and the tool picked the one it thought fit the sentence. Click the word (or focus it with Tab and press Enter) to cycle through the alternatives; the text updates, and so do the Copy and export buttons. If the tool guessed Galliā where you wanted Gallia, this is how you fix it.
• Red — the word is not in the wordlist. Morpheus is asked to analyse it from its stem and ending; if even Morpheus doesn't recognise it, the vowel lengths are guessed from the word's ending and the part of speech the tagger predicted. That guess can be confidently wrong — invent a word like accaemem and you will still get accaēmem, because -emem looks like an ending whose e is long. Treat red words with more suspicion than the rest; proper names and rare or late words end up here.

Hover a word (or tap it on a phone) to open its analysis: the RFTagger tag decoded into plain features, the lemma, whether it was found in the wordlist, and every reading Morpheus found. The reading currently shown in the text is highlighted among the candidates.

The options

• Mark long vowels — the macrons themselves. Turn it off and the tool will still tag and analyse, but leave the spelling alone.
• Also mark māius etc. — a handful of words (maius, eius, cuius…) whose first vowel is long by position before consonantal i. Editors disagree about marking these, so it is a choice.
• u → v and i → j — classical orthography. Latin wrote VNVS and IACTVS; modern editions vary between unus/vnus and iactus/jactus. These convert consonantal u and i where the analysis says they are consonants — so cum stays cum.
• Scan as — if your text is verse, choosing its metre (dactylic hexameter, elegiac distichs, hendecasyllable, iambic trimeter + dimeter) lets the tool prefer the vowel lengths that make the line scan, which resolves many ambiguities that prose cannot. The feet it found are shown at the end of each line. Leave it on Prose for ordinary text.
• De-macronize — shows the text below with all macrons stripped out (the input box is left as it is). Useful for checking an already-macronized text.

The one-time download

Because the whole pipeline runs locally, the browser must first fetch the tagger model, the Morpheus database and the wordlist. They are served gzipped, so the download is about 10 MB (65 MB once unpacked in your browser). This happens once: the files are then stored in your browser, so later visits start in a few seconds and work offline. Clear cache at the bottom of the page deletes them; the next visit will download them again.

Accuracy

Expect a small number of mistakes in ordinary prose, and more in verse, in poetry with unusual word order, and around proper names. The highlighted words are where the tool is least certain — if a text matters, read those and cycle the ones that are wrong. It is a good corrector, not an oracle.

Credits

This is a WebAssembly port of Johan Winge's Latin macronizer, whose design and data it follows closely; the method is described in his thesis Automatic annotation of Latin vowel length (Uppsala, 2015). It builds on RFTagger (Schmid & Laws), the Morpheus analyser and the Latin Dependency Treebank. The browser port lives at latin-macronizer-wasm.

---

For technical issues or suggestions, please visit our GitHub repository.