/conjugate — Romanian verb conjugation

Given a Romanian verb, returns its full conjugation table: seven moods, every tense in each, every person and number that mood takes.

Quick start

curl https://api.lexicro.com/conjugate/merge \
  -H 'X-API-Key: lxr_your_key_here'

The a prefix is optional — /conjugate/merge and /conjugate/a%20merge return identical moods, differing only in the input field that echoes what you sent.

The response is long (nine indicative tenses alone), so here is a trimmed excerpt showing the shape:

{
  "input": "merge",
  "notes": [
    {
      "scope": "all",
      "code": "upstream_unverified",
      "message": "Forms come from verbecc 2.0.3 and are not exhaustively verified. ..."
    },
    {
      "scope": "imperativ",
      "code": "imperative_known_errors",
      "message": "The imperative has known residual errors in a small set of verbs ..."
    }
  ],
  "verb": {
    "infinitive": "merge",
    "provenance": "template",
    "template": "concu:rge"
  },
  "moods": {
    "indicativ": {
      "prezent": [
        {"form": "merg", "pronoun": "eu", "feats": {"Person": "1", "Number": "Sing"}, "source": "verbecc"},
        {"form": "mergi", "pronoun": "tu", "feats": {"Person": "2", "Number": "Sing"}, "source": "verbecc"}
      ]
    },
    "condițional": {
      "prezent": [
        {"form": "aș merge", "pronoun": "eu", "feats": {"Person": "1", "Number": "Sing"}, "source": "derived"}
      ]
    }
  }
}

Authentication

Pass your key in the X-API-Key header. Without one you get an anonymous allowance of 10 requests/day per IP address. See the /analyze guide for how to get a key — both endpoints share one key and one daily quota.

Response

Each entry in a tense's array has:

The seven moods

/conjugate is built on verbecc, an open-source Romanian conjugation library. Most forms pass through from it unchanged. Two things don't, and both matter enough to call out on their own.

condițional is composed by LexicRo, not verbecc. The library declares this mood but ships no data for it. Romanian's conditional auxiliary (aș, ai, ar, am, ați, ar) is invariant — it doesn't change with the verb — so LexicRo builds prezent as auxiliary + infinitive and perfect as auxiliary + fi + participle, using verbecc's own infinitive and participle as the base. Every form in this mood carries source: "derived".

The negative imperative's 2sg form is also composed, for the same reason. Romanian's negative imperative singular is invariantly nu + the infinitive — there's no verb where it legitimately differs from that pattern — so LexicRo builds it the same way it builds the conditional: apply the paradigm rather than trust a per-verb library entry. That form carries source: "derived" too. The 2pl negative imperative is a different pattern (nu + the plural imperative form, not nu + infinitive) and comes from verbecc as source: "verbecc" like everything else in the mood — so within one imperativ.negativ array you'll see both sources side by side:

"negativ": [
  {"form": "nu merge",   "pronoun": "tu",  "source": "derived"},
  {"form": "nu mergeți", "pronoun": "voi", "source": "verbecc"}
]

That's not an inconsistency to debug — it's two different rules for two different persons, disclosed via source so you don't have to guess which applied.

Two kinds of provenance

verb.provenance and form.source answer different questions:

• verb.provenance — did the library recognise this verb at all? "template" means it matched a known conjugation pattern. "predicted" means it didn't, and guessed a pattern from the verb's ending instead.
• form.source — which system produced this particular form, "verbecc" or "derived" (see above).

An unrecognised verb still returns 200, not 404 — a made-up or unfamiliar verb usually still conjugates the way Romanian verbs of its shape do, and that guess is often useful. verb.provenance is how you know it was a guess:

curl https://api.lexicro.com/conjugate/xyzzyti \
  -H 'X-API-Key: lxr_your_key_here'

{
  "infinitive": "xyzzyti",
  "provenance": "predicted",
  "template": "fer:i"
}

When provenance is "predicted", every form in the response is a guess — not just the ones marked derived. Only genuinely unrecognisable input (not a plausible Romanian verb at all) returns 404; see Errors.

Known limitations

Forms come from verbecc and are not exhaustively verified, and this applies across the whole response rather than to one mood. Two defects outside the imperative — the indicative present of a minți, which returned eu mit for eu mint, and the infinitive of a face, which returned a corrupted Spanish string — were reported upstream and corrected in verbecc 2.0.3. Both were found by spot check rather than by an audit, and the data has not been audited as a whole, so treat the absence of a current example as telling you nothing.

The imperative has its own, narrower set of known issues on top of that. Three verbs — a merge, a trece and a tăcea — are served their third person singular where a second person imperative belongs, so merge returns merge where correct Romanian is mergi. Separately, for some verbs the correct imperative form depends on whether the verb is used transitively — treci! versus trece-mă! — and a bare verb name doesn't tell you which sense was meant, so the returned form may not be the one your context needs.

The three verbs are also listed in that note's verbs field, so you can mark the affected forms in your own interface rather than relying on a reader noticing a caveat elsewhere on the page. Do not try to detect them from the shape of the data instead — the defect's signature, a second-person singular imperative identical to the third-person singular present, is equally true of a cânta (cântă) and a găsi (găsește), which are perfectly correct. The set is enumerated because it cannot be computed.

When the upstream fix lands, the list shrinks and this note eventually goes.

When a verb's paradigm contradicts itself

Some verbs mark persons as having no form in the present tense while other tenses supply forms for those same persons. A ninge ("to snow") gives - for every person but the third singular in indicativ prezent, then returns eu am nins in indicativ perfect-compus.

When that happens the response carries a paradigm_contradiction note, listing the affected mood/tense pairs in its tenses field. It is computed per response, so it appears only on verbs that actually exhibit it.

The note does not tell you which side is wrong, because we do not know. Both readings occur among the affected verbs: a curge really is third-person-only, so its eu am curs is the error — while a aporta is an ordinary transitive verb wrongly marked third-person-only, so there the present is the error and eu am aportat is correct. A caller who needs to choose must do so per verb; a caller who only needs to warn a user can render the note.

These are reported upstream to the library maintainers. LexicRo relays the forms verbecc produces rather than substituting its own corrections, so every value in a response is traceable to one source — the alternative, silently fixing what looks wrong, would leave you unable to tell which forms were verified and which were guessed at by whoever wrote the fix.

The notes field in every response carries this same disclosure in machine-readable form — code: "upstream_unverified" for the general case, code: "imperative_known_errors" for the imperative — so you can surface it to your own users without hardcoding this page's wording.

A tense is not one form per person

Most tenses return exactly one entry per person/number combination — eight entries for a fully personal tense (eu, tu, el, ea, noi, voi, ei, ele). Some don't. Where Romanian genuinely has two valid forms for one person, both come back as separate entries with identical feats.

The clearest example is a avea, present tense, third person singular:

[
  {"form": "a",   "pronoun": "el", "feats": {"Person": "3", "Number": "Sing", "Gender": "Masc"}, "source": "verbecc"},
  {"form": "are", "pronoun": "el", "feats": {"Person": "3", "Number": "Sing", "Gender": "Masc"}, "source": "verbecc"}
]

a (the short auxiliary form, as in el a mâncat) and are (the full main verb, as in el are un câine) are both correct third-person-singular present forms of a avea. Neither is more correct than the other, so both are returned rather than picking one. If your code indexes into a tense's array by position expecting one entry per person, it will silently pick up whichever of these it happens to land on — check the array length, or match on pronoun and feats and expect more than one hit.

Diacritics

Every field in the response uses Romanian's comma-below diacritics — ș (U+0219) and ț (U+021B) — matching /analyze. The one exception is verb.template: it is verbecc's own internal identifier, passed through exactly as the library spells it (including its legacy cedilla ş/ţ characters where present), because it's what you'd quote back to us — or to verbecc — in a bug report about a specific form.

Errors

Attribution

Conjugation is generated by verbecc. Full citations in /attribution.