GET /v1/conjugate/:verb
curl -s 'localhost:3000/v1/conjugate/merge' | jq{
"verb": "merge",
"classRoman": "III",
"forms": {
"inf": "merge",
"part.past": "mers",
"gerund": "mergând",
"imperative.2sg": "mergi",
"ind.prez.1sg": "merg",
"ind.prez.3sg": "merge"
},
"source": "rules"
}Both citation forms work: a fi and fi give identical results. The a
particle is stripped.
curl -s 'localhost:3000/v1/conjugate/a%20fi' | jq '.forms["ind.prez.1sg"]' # "sunt"| Key | Meaning |
|---|---|
inf |
infinitive |
inf.long |
long infinitive (mergere) |
part.past |
past participle |
gerund |
gerund |
imperative.2sg / .2pl |
imperative |
ind.prez.<1|2|3><sg|pl> |
present indicative |
ind.imperf.… |
imperfect |
ind.perfsimp.… |
simple perfect |
conj.prez.3sg / .3pl |
present subjunctive |
seeded: from the DEXonline database dump. Real data.rules: computed here.
Two mechanisms:
An irregular table, for verbs no rule derives, covering:
fi, avea, vrea, putea, vedea, face, merge, da, sta, lua, bea, ști,
veni, spune, zice, duce, trece, scrie, pune, ține, rămâne, mânca
These are the ones where the stem itself changes (face → făcut,
vedea → văzut) or the paradigm is suppletive (fi → sunt, fost). The
table lists only the cells that differ; the rest come from the rules.
A four-class rule engine for everything else, keyed on the infinitive
ending: -a (I), -ea (II), -e (III), -i/-î (IV).
A regular verb is reliable. An irregular verb not in the table above will be
conjugated by rule and may be wrong. Romanian has stem alternations
(a↔ă, e↔ea, d↔z) that the rules do not fully model.
If a word matters, cross-check against /v2/word/<verb>, where DOOM and
DEXonline publish attested forms:
curl -s 'localhost:3000/v2/word/merge?sources=doom' \
| jq '.entries[].inflections[] | { form, tags }'Those carry origin: "attested". See Accuracy.