A few weeks ago I wrote about building the Romanian NLP API that should already exist. Today Phase 1 is live.
What shipped in Phase 1:
GET /conjugate/{verb}
→ conjugation table — 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.
GET /lookup/{word} → definitions from DEXonline (DEX '09, MDA2, DLRLC), HTML stripped, source attributed
GET /inflect/{word} → basic inflection info extracted from dictionary headers
POST /difficulty → word validation against standard Romanian dictionaries
Free tier: 1,000 requests/day, no credit card, no account required for anonymous use (10 req/day without a key).
Try it: https://api.lexicro.com/docs
The interactive Swagger UI is live — you can call every endpoint directly from the browser without writing any code.
What's next (Phase 2): Fine-tuning bert-base-romanian-cased-v1 for morphological analysis — the POST /analyze endpoint that returns lemma, POS, case, gender, number, person, and tense per token. That's the hard part. Building in public — feedback still very welcome.
Update, 2026-08-30: /lookup, /inflect and /difficulty were withdrawn on 2026-08-17 and are not currently served. /analyze shipped and is live. Demo, no key required: https://demo.lexicro.com
→ Original announcement: https://dev.to/peterabolins/building-the-romanian-nlp-api-that-should-already-exist-2gg7
→ Demo: https://demo.lexicro.com
→ GitHub: https://github.com/LexicRo/lexicro
Top comments (0)