imlooh/open-tangut

Open Tangut is a project that gathers and organizes information about the Tangut script, a writing system used by the Western Xia dynasty in medieval northwestern China. The project collects data from English, Japanese, and Chinese Wiktionary, plus a source called CCAMC (a Chinese character database). It also experiments with recognizing Tangut characters from images using OCR, and includes a browser app for searching and browsing the collected data. The data covers the full Tangut range in the Unicode standard, which assigns numbers to characters so computers can display them. The main output feeds a free online dictionary called Tangutpedia. The maintainer, who goes by Looh, is still verifying the collected vocabulary and plans to add a 2025 reconstruction of Tangut pronunciation. Contributors are welcome, but since there is a single maintainer, pull requests should not go directly to the main branch. The browser app lets users search by the character itself, its Unicode codepoint, its English or Chinese meaning, its pronunciation, or its structure (how it is built from simpler components). You can navigate by dictionary index across several reference systems, step through characters by codepoint, search by component parts, draw a character for recognition, or upload an image. The app uses three Tangut fonts that ship with the repository. The data pipeline runs in stages: scrapers pull from Wiktionary and CCAMC, a merger script combines and cleans the results, and a final step builds vocabulary files. All data is stored in JSONL format, meaning one JSON object per line, which is convenient for large datasets. Each character record follows a defined schema with fields for the character, its codepoint, stroke counts, radical, structural composition, definitions in three languages plus synonyms, pronunciations from multiple scholarly reconstruction systems, example vocabulary, metadata like dictionary indexes, font variants, and source URLs. Missing values are handled gracefully by the app. The project is written in Python 3.9 or later and relies mostly on the standard library, with optional dependencies for image processing and OCR. The full README is longer than what was shown.