About Glottosphere

Glottosphere is a free dictionary built around one idea: every word in a definition is a link. Click a word to see its own definition, and keep going, within a language or across languages.

Humanity speaks more than 7,000 languages. Most dictionaries cover a few dozen of them. The goal here is to give every language a place: a dictionary where one exists, and at least a profile with its family, where it is spoken, how many people speak it and how endangered it is.

What is here today

Dictionaries for 1,177 languages, about 4,366,919 entries in total.
Profiles of 4,000 languages.
Language family trees.

Where the data comes from

Most dictionary entries are derived from Wiktionary, extracted by Kaikki.org, and are available under the CC BY-SA 4.0 licence.
In the translator, a language's words are defined in that language where a dictionary of it does so: the Wiktionary written in it (Chinese, Japanese, Korean, German, Russian, French and others, the same licence), and for Chinese the Ministry of Education of Taiwan's 重編國語辭典修訂本, reproduced unchanged under CC BY-ND 3.0 TW. Swedish also comes from Språkbanken's Svensk ordbok (SO 2009) (CC BY 4.0), Danish from DanNet (DSL and CST, CC BY-SA 4.0), Korean from the National Institute of Korean Language's 표준국어대사전 (CC BY-SA 2.0 KR), Estonian from the Institute of the Estonian Language's EKI ühendsõnastik (CC BY 4.0), Latvian from Tēzaurs.lv (CC BY-SA 4.0) and Basque from Ibon Sarasola's Egungo Euskararen Hiztegia (UPV/EHU, CC BY 4.0). Assamese comes from Hemchandra Barua's Hemkosh (1900, public domain), which the dictionary's Assamese pages show too, with Barua's English glosses, transcribed from the Digital Library of India's scan with an OCR model trained on the book's own type (its training lines written out with AI), which may still misread a letter. Georgian comes from Sulkhan-Saba Orbeliani's სიტყჳს კონა (1685–1716, public domain), transcribed the same way from the 1949 Tbilisi edition, in Saba's spelling and his senses of the words, which are not always today's. Hebrew comes from Eliezer Ben-Yehuda's מלון הלשון העברית (volumes I–VII, which he wrote, public domain), transcribed the same way and matched to each word's sense; its senses are sometimes older than today's. Urdu comes in part from Syed Ahmed Dehlvi's فرہنگ آصفیہ (1888–1901, public domain) and Persian from Ghiyath al-Din Rampuri's غیاث اللغات (1826, public domain, in Mohammad Dabir-Siyaghi's edition, without his notes), whose pages were read with AI from the scans, page by page, and matched to each word's sense. Of the Urdu, only the pages read so far are in; the rest of the Persian was transcribed with an OCR model trained on the edition's type, which may misread a letter. Sanskrit comes from Rādhākāntadeva's Śabdakalpadruma and Tārānātha Tarkavācaspati's Vācaspatyam as the Cologne Digital Sanskrit Dictionaries (The Sanskrit Library and Thomas Malten) typed them, under CC BY-SA 4.0. Dzongkha comes from the Dzongkha Development Commission's dictionary as jamyang2002/DzongkhaDictionary keeps it, Tumbuka from Anthony Munthali's Tumbuka dictionary, Tamasheq from a Tamajaq dictionary of Niger and Luba-Kasai from the Lexique biblique français–tshiluba. Arabic also comes from Wikidata's lexemes (CC0) and, sense by sense, from Salah Abdo's Arabic WordNet 4 (CC BY 4.0), whose definitions were written with AI from Open English WordNet and Princeton WordNet, and are marked as such. The Urdu words the Farhang-e-Asifiya does not define here were defined with AI, sense by sense, from their English senses, and are marked as such too. Where none does, the English definitions are machine-translated with NLLB-200 and marked as such.
Words of smaller languages also come from PanLex, only from the sources in it whose own licences allow reuse; each such entry names its source.
For some languages without a dictionary, words were worked out from openly licensed Bible translations of the eBible corpus. Their pages say so and name the translation.
Example sentences come from Wikipedia (CC BY-SA 4.0).
Language classification, areas and countries follow Glottolog (CC BY 4.0).
Language codes follow ISO 639-3.
Endangerment status comes from the Endangered Languages Project.

Dictionaries for small languages are incomplete, and speaker numbers are estimates that differ between sources. Corrections are welcome.

Contact

Write to info@glottosphere.com.