# Verbacorpus

An open, source-attributed corpus of 48,787 Ukrainian proverbs and adages (1841–today) — searchable PWA plus a multi-format REST API.

[Open in the WASM portfolio](https://yemelianov.dev/?project=verbacorpus&detail=1)

Tags: typescript, pwa, api, ukrainian
Source: public
Repository: https://github.com/dmytro-yemelianov/verbacorpus
Project: https://verbacorpus.org

Verbacorpus collects Ukrainian proverbs and adages into an open, source-attributed corpus — a citable, structured dataset for researchers, language learners, and anyone building Ukrainian-language tools, rather than a scraped list. The current build holds 48,787 entries drawn from five digitized sources spanning 1841 (Ilkevich) to a modern 2009 dictionary (Mlodzynskyi), unified and deduplicated, each carrying a modern-spelling rendering, one to three categories from a fixed 27-theme taxonomy, and source attribution.

The corpus and its enrichment are kept separate and honestly labeled: `text` preserves each source's original orthography verbatim, while `modern_text`, `category`, and cleaned `explanation` fields are LLM-generated artifacts — produced by batched Claude Code agents and audited (a 40-entry quality check found modern_text about 95% acceptable and category tags about 85% acceptable). Likely dialectal variants are linked, never merged, into a `variant_group` by fuzzy string matching, with groups capped at eight members to curb over-linking. A Cloudflare Worker serves the corpus as an offline-capable Progressive Web App with lexical search on by default and an opt-in semantic-search mode backed by Workers AI embeddings in a Vectorize index. The versioned `/api/v1` REST API serves every endpoint as JSON, JSONL, XML, CSV, or TSV.

Status: public, live at verbacorpus.org. The corpus is also published as a versioned Hugging Face dataset and is citable via `CITATION.cff`. Compilation and enrichment are CC BY 4.0; the historical source texts themselves are public domain.