Datasets:
Tonos keyboard data
The data files of Tonos, a polytonic Greek keyboard for iPhone and iPad: its spell-check word lists, the compiled dictionary built from them, its movable nu and elision table, and its next-word language model. They are the exact files the app ships, published under the Creative Commons Attribution-ShareAlike 4.0 International license (CC BY-SA 4.0).
This revision holds data version 1.3.10-cb6b6ee, named for the newest Dilemma export among its files (release and short commit). Every data version the app ships has a tag of that name in this repository, and the app links to the tag of the data it carries, so a tag never changes once it exists. The main branch holds the newest version. manifest.json lists every file with its size and SHA-256, to check a download against.
Files
| File | Size | Contents | In the app |
|---|---|---|---|
dictionary/grc_polytonic.dic |
35.8 MB | Ancient and Medieval Greek word list, Hunspell format, with a frequency class per entry | compiled into dictionary/grc_polytonic.dict |
dictionary/grc_polytonic.aff |
6.7 kB | Affix rules of the Ancient Greek word list | compiled into dictionary/grc_polytonic.dict |
dictionary/grc_polytonic.version |
109 bytes | Version of the Ancient Greek word list | as is |
dictionary/grc_mg_polytonic.dic |
3.5 MB | Polytonic Modern Greek word list: words the Ancient Greek list lacks, ratings of shared words, and spellings Modern Greek avoids | compiled into dictionary/grc_polytonic.dict |
dictionary/grc_mg_polytonic.aff |
273 bytes | Affix file of the Modern Greek word list (no rules) | as is |
dictionary/grc_mg_polytonic.version |
1.6 kB | Version and sources of the Modern Greek word list | as is |
dictionary/grc_polytonic.dict |
76.6 MB | The two word lists compiled to the app's binary dictionary | xz-compressed, as grc_polytonic.dict.lzma |
morphology/grc_morph.json |
1.2 MB | Movable nu and elision table | compiled into morphology/grc_morph.bin |
morphology/grc_morph.bin |
1.2 MB | The movable nu and elision table compiled to the app's binary format | xz-compressed, as grc_morph.bin.lzma |
language-model/grc_ngram.bin |
61.9 MB | Trigram next-word model, GNLM format version 2 | xz-compressed, as grc_ngram.bin.lzma |
language-model/grc_ngram.version |
628 bytes | Version, settings and counts of the language model | as is |
keyboard-fallback/grc_polytonic.dic |
1.0 kB | 98 common words the keyboard checks spelling against until the app has installed the full dictionary | as is, in the keyboard extension |
keyboard-fallback/grc_polytonic.aff |
254 bytes | Affix file of the fallback word list | as is, in the keyboard extension |
The app stores three files xz-compressed; they are published here decompressed, and decompressing the app's copies gives these bytes. The word lists and grc_morph.json are not in the app themselves: the app carries them compiled, and they are published here as the form to read and change.
How the files were made
Every file except the keyboard fallback was built by Dilemma, an open-source Greek lemmatizer and data pipeline whose code is under the MIT License. These files come from its exports 1.3.8 (commit 20b8a1b) and 1.3.10 (commit cb6b6ee); each .version file names the export of its file. grc_morph.json has no version file; it is the output of Dilemma's export_morphology.py.
The files hold words, word forms, frequency classes, counts and probabilities computed from the sources below, not the source texts. Dilemma extracted forms and lemmas from the dictionaries and annotated corpora, normalized them to Unicode NFC, merged, filtered and ranked them by frequency, generated inflected forms from paradigms, and counted word sequences in tokenized sentences, keeping the most frequent continuations of each context. Texts whose license allows only NonCommercial use were left out. The app then compiled the word lists and the morphology table into its binary formats without changing their content. The keyboard fallback is a short hand-made list of common words.
License
These files are licensed under CC BY-SA 4.0. Some sources are licensed under CC BY-SA 3.0, whose terms allow an adaptation to be licensed under a later version. If you share these files or something you make from them, credit the sources below, say what you changed, and license your adaptation under CC BY-SA 4.0 or a license compatible with it.
Sources and attribution
| Source | Author or maintainer | License | Used for |
|---|---|---|---|
| English Wiktionary and Greek Wiktionary | Wiktionary contributors (Wikimedia community), obtained as JSONL through kaikki.org (Tatu Ylonen, Wiktextract) | CC BY-SA 3.0 | Ancient Greek word list, Modern Greek word list, Morphology table |
| A Greek-English Lexicon (LSJ), 9th edition, 1940 | H. G. Liddell, R. Scott, H. S. Jones, through the LSJ9 processed exports | Public domain | Ancient Greek word list |
| Greek Lexicon of the Roman and Byzantine Periods | E. A. Sophocles, TEI through Ionian University and the Internet Archive | Public domain | Ancient Greek word list |
| GLAUx | Alek Keersmaekers (2021); works whose source license is NonCommercial are left out | CC BY-SA 4.0 | Ancient Greek word list, Morphology table, Language model |
| The Ancient Greek Dependency Treebank (AGDT) | Perseus Project (G. Crane et al.) | CC BY-SA 3.0 US | Ancient Greek word list, Morphology table |
| The Diorisis Ancient Greek Corpus | A. Vatri and B. McGillivray (2018) | CC BY 4.0 | Ancient Greek word list, Morphology table, Language model |
| Pedalion Trees | Perseids publications | CC BY-SA 4.0 | Ancient Greek word list |
| Harrington Trees | Perseids publications | CC BY-SA 4.0 | Ancient Greek word list |
| UD_Ancient_Greek-PTNK (Septuagint: Genesis, Ruth) | Universal Dependencies contributors | CC BY-SA 4.0 | Ancient Greek word list |
| STEPBible TAGNT (Translators Amalgamated Greek New Testament) | Tyndale House Cambridge, STEP Bible (www.STEPBible.org) | CC BY 4.0 | Ancient Greek word list |
| Opera Graeca Adnotata (OGA) v0.2.0 | Giuseppe G. A. Celano, doi:10.5281/zenodo.14206061 | CC BY-SA 4.0 | Ancient Greek word list |
| First1KGreek | Open Greek and Latin | CC BY-SA 4.0 | Ancient Greek word list |
| Patristic Text Archive (PTA) | Patristic Text Archive; files with a NonCommercial license are left out | Creative Commons licenses recorded per file | Ancient Greek word list |
| Patrologia Graeca (Migne), corrected OCR | Public-domain text, through the Open Greek Corpus correction of the Calfa OCR | CC BY 4.0 (OCR) | Ancient Greek word list |
| Open Greek Corpus | Open Greek; source and license of every work recorded in its corpus_editions.json | CC BY-SA 4.0 | Ancient Greek word list |
| Byzantine and Early Modern Greek corpus | Francisco Riordan; license of every text recorded in its README | Public domain and CC BY-SA 3.0, per text | Language model |
| Greek Wikisource, polytonic texts | Wikisource contributors (el.wikisource.org), through the glossAPI dataset Wikisource_Greek_texts | CC BY-SA 3.0 | Modern Greek word list, Morphology table, Language model |
| FrequencyWords, Modern Greek full list | Hermit Dave, el_full.txt, counted from the OpenSubtitles2018 corpus of OPUS; only the counts are used | CC BY-SA 4.0 | Modern Greek word list |
Dilemma's NOTICE lists every source of its tables in full, including the reference lexica it checks headwords against without copying their entries.
Formats
.dicand.affare Hunspell files. The first line of a.dicis its number of entries; each entry is a word, optionally followed by/and an affix flag, then a tab and fields separated by spaces.fr:C,fr:M,fr:Randfr:Xgive the frequency class (common, medium, rare, not seen in the corpora). In the Modern Greek list,mg:generatedmarks a form generated from a paradigm rather than attested,mg:sharedrates a word the Ancient Greek list also has, andmg:avoidnames an Ancient Greek spelling that Modern Greek writes otherwise. See Word lists and Hunspell export.grc_ngram.binis a memory-mappable trigram model, GNLM format version 2: a 128-byte little-endian header, a sorted UTF-8 vocabulary, unigram counts, and sorted bigram and trigram tables of 6-byte entries (a word id and a fixed-point log probability). See Binary format.grc_morph.jsonmaps the forms that take movable nu (nu) and full forms to their elided forms (el, andel_modernfor Modern Greek). See Boundary rewrites.grc_polytonic.dictis the app's compiled dictionary (magicTONOSDIC, little-endian): every affix rule expanded, the entries sorted by their UTF-8 bytes with an offset table, an index of the entries without diacritics, and one frequency byte per entry. It holds the words of the two lists.grc_morph.binis the app's compiledgrc_morph.json(magicGMRF, little-endian): sorted UTF-8 strings with offset tables for the nu forms and for the elision pairs.
Citation
Tonos keyboard data, version 1.3.10-cb6b6ee. https://huggingface.co/datasets/ciscoriordan/tonos-data. License: CC BY-SA 4.0.
- Downloads last month
- 73