Status: shipping

Building the Romanian NLP API that should already exist

LexicRo — open-core Romanian language intelligence infrastructure. Looking for early feedback, collaborators, and anchor users.

NLP Romanian API Open Source B1+ tools EdTech

Phase 2 is live

The morphological analyser is shipped. /analyze returns lemma, part of speech and full morphology for every token in context - 98.14% POS accuracy, 95.50% lemma accuracy (UD Romanian test split, gold tokenisation). Conjugation is live too. Free tier: 1,000 requests/day, no credit card.

Measured on the UD Romanian RRT test split with gold tokenisation: the standard Romanian benchmark, comparable to published results. End-to-end accuracy on raw text is a different and currently unmeasured quantity.

Try the demo → Documentation → API explorer → Status →

If you've ever tried to do anything programmatic with Romanian text — parse a sentence, get the correct inflected form of a noun, check whether a verb conjugation is right — you've probably hit the same wall.

There's no clean API for it. Verbecc handles conjugation, but there's nothing for morphology or lemmatisation. This is not good enough for production.

The specific gap

Given an arbitrary Romanian sentence, return for each token its lemma, part of speech, case, number, gender, person and tense. This is what spaCy does for English, French and German in a pip install. LexicRo is that as a hosted, versioned HTTP contract for Romanian: callable without shipping a model, stamping the version it answered with, and telling you per token whether the answer came from a lexicon or a prediction.

Romanian NLP tooling lags well behind English. The academic resources are there — DEXonline (313k+ lemmas), RoLEX (330k morphosyntactic entries), the Universal Dependencies Romanian Treebank — they're just not packaged in a way developers can actually use.

That's what LexicRo is for.

What we're building

A hosted REST API — with an open-source core — covering the endpoints Romanian developers actually need:

    # Morphological analysis

    POST /analyze

    → lemma, POS, case, gender, number, person, tense per token, plus where each answer came from.

    # Verb conjugation

    GET /conjugate/{verb}

    → seven moods, including perfect simplu
  and viitor I. For a verb the conjugator does not recognise,
  returns a predicted paradigm and marks it as predicted.

The core is open source (MIT). It is built on openly licensed Romanian language resources — see attribution — and licensing terms for the model weights are still being worked out. The hosted API is available now either way, with a generous free tier (1,000 req/day, no credit card).

Built on what already exists

We're not starting from scratch. The data and models are there — they just need engineering:

Data: the MULTEXT-East Romanian word-form lexicon (428k entries, CC BY-SA 4.0), and the UD Romanian RRT treebank (9.5k annotated sentences, CC BY-SA 4.0, developed at RACAI). Full citations on the attribution page.

Models: bert-base-romanian-cased-v1 fine-tuned for morphological tagging, paired with lexicon lookup — the model resolves what the dictionary cannot, the dictionary supplies what the model would only guess at. verbecc for conjugation, extended.

Infrastructure: FastAPI, Docker, full OpenAPI spec.

Phase 2 — live now

The hard part is done. /analyze handles contextual disambiguation, not just dictionary lookup — it knows that sare is the noun sare ("salt") in Pune sare în mâncare and the verb sări ("to jump") in Pisica sare pe masă. Measured on the UD Romanian RRT test split with gold tokenisation: 98.14% part-of-speech accuracy, 98.43% morphological features (F1), 95.50% lemma accuracy. Full numbers and caveats in the documentation.

What we're looking for right now

Honest feedback on the endpoint design Does this cover what you actually need? What's missing? What would make you use this over your current approach?

Early users willing to test the API If you're building something with Romanian text — edtech, content tools, document processing, language learning — we'd like to talk. Early users get priority feature input.

Academic and institutional connections If you work on Romanian language resources at a university or research institution, I'd like to hear from you — on tagsets, evaluation, or where an open Romanian NLP API would actually be useful in your work.

Anyone who's built adjacent to this problem If you've already built a Romanian spell checker, worked with the UD treebank, or tried to fine-tune anything on Romanian — we'd genuinely like to hear what you learned.

On the business model

Open core, hosted API, freemium tiers. The free tier is real and permanent — not a trial. The commercial tiers will fund continued development. We're pursuing language-technology funding partly because Romanian deserves proper NLP infrastructure regardless of whether the commercial model scales immediately.

Three ways in.

Everything described here is live. Read the docs, or get in touch if you want to talk about your use case.

Read the docs Get API access Get in touch