A morphological analyzer and generator for Tigre (Ge'ez script), built from Shlomo Raz, A Descriptive Study of Tigre Grammar (1973). Type a word to get its root or lemma, word class, features, its pronunciation in Raz's transcription (which shows the gemination, ə and vowel length that Ge'ez spelling hides) and an English gloss, ranked by a model learned from an 11-million-word Tigre corpus.
GitHub ·
pip install git+https://github.com/beshir-a-Ibrahim/tigmorph.git ·
v2.1 · runs entirely in your browser
Score 0 = best reading; higher = less likely.
| Word | Root / lemma | Class | Features | Pronunciation | Score |
|---|
Text from many sources spells the same word in several ways (ሀለ / ህለ, ይዐለ / ኢዐለ / የዐለ,
ሠ for ሰ, ፡ between words, :: for ።). The normalizer writes every word in one standard spelling.
chars: characters only (NFC, ሠ→ሰ, ፀ→ጸ, ኀ→ሐ, punctuation, spaces).
safe: plus spelling pairs that differ only by a known free alternation.
full: every row of the learned table, including pairs that differ in other letters (review these).
About 3,300 verb stems, 9,000 nominal entries and 280 function words.
| Test | Words | Analysed | Best reading right |
|---|---|---|---|
| Raz's example sentences (running text) | 5,268 | 95.0% | 89.9% |
| Raz's verb paradigms (Ch. 5) | 2,013 | 100% | — |
| Raz's function words and clitics | 245 | 100% | 97.6% |
| Raz's deverbal nominals (§ 2.7) | 94 | 100% | 80% |
| Raz's nouns: singular, plural, possessive | 634 | 99.8% | 88% |
| Verb forms weighted by corpus frequency | 361k tokens | 100% | 96.9% |
| Corpus tokens (10.9M) analysed from the lexicon | 83.8% |
Ge'ez spelling hides gemination and vowel length, so some forms have two real readings that only context can settle, mainly verb type B vs C, and a participle vs the prefixless imperfect.
Type A–D · Deriv derivation · TAM Perf / Impf / Jus / Imp, or a verbal noun
(PartAct, PartPass, Agent, Inf, Instr, class V>N) · Subj / Obj person (1Sg … 3PlF) ·
Neg negative ʼi- · Colloq prefixless imperfect · Poss pronoun ending ·
Art=Def article · Rel relative la- · Cl attached ወ-/ከ- · Prep
attached preposition · Encl -መ/-ዲ. Full list:
docs/TAGS.md.
Root-independent templates following Raz Ch. 5 and Ch. 2 are compiled with pynini into finite-state transducers
(root + tags ⇄ Raz transcription ⇄ Ge'ez spelling). A small pure-Python runtime reads them. On this page
that runtime runs in your browser through Pyodide, so nothing you type is sent
to a server.
Fill in a sheet in contrib/:
frequent words the lexicon doesn't know yet, Wiktionary verbs to check, or 300 corpus words to annotate.
Issues and pull requests are welcome.
Code MIT; data CC0 / CC BY 4.0 / CC BY-SA 4.0 by source (compiled FSTs: CC BY-SA 4.0). If you use tigmorph, please cite it and Raz (1973); see CITATION.cff. Corpus: BeitTigreAI/tigre-data-monolingual-text. Lexical data: the Tigre Wiktionary contributors.