Chapimenge/amharic-gemination-lexicon

1# Running the extractor2 3`gemination.py` has no dependencies of its own. To turn new words into4gemination patterns it needs HornMorpho 5.3.6 (Michael Gasser,5https://github.com/hltdi/HornMorpho) with two small patches.6 7## One command8 9```bash10./install_hornmorpho.sh .venv-hm          # Python 3.9 or newer; set PYTHON=... to choose one11LANG=C.UTF-8 .venv-hm/bin/python gemination.py ይገኛል ክልል መለወጥ12LANG=C.UTF-8 .venv-hm/bin/python test_gemination.py13```14 15Expected output of the second line (tab-separated):16 17```18ይገኛል  ይገ፟ኛ፟ል  positions=[1, 2]  sources=['t-prefix', 'root']  pos=V  root_freq=3688  seg=----ይ</ገ/ኝ>---ኣ/ል---19ክልል   ክል፟ል   positions=[1]     sources=['lexicon']            pos=N  root_freq=1     seg=-<ክ/ልል>------20መለወጥ  መለ፟ወጥ  positions=[1]     sources=['t-prefix']           pos=N  root_freq=84    seg=-<መ/-ለወጥ>------21መለወጥ  መለወ፟ጥ  positions=[2]     sources=['root']               pos=N  root_freq=84    seg=-<መ-ለ/ወጥ>------22```23 24መለወጥ has two readings (passive mälläwäṭ, active mäläwwäṭ), so it prints two25patterns. Only sentence context can choose between them.26 27## What the script does, step by step28 291. **Install the wheel.** HornMorpho is not on PyPI. The script downloads30   `dist/hornmorpho-5.3.6-py3-none-any.whl` from the HornMorpho repository at31   commit `7e3d93af760e27ea6dbc3b7a078d2d9c3335f618` and checks its sha25632   (`e1b0e4ed...fa7`). Dependencies `conllu` and `requests` come with it.332. **Patch 1, Tk GUI stub** (`patches/gui.py`, replaces `hm/morpho/gui.py`).34   `hm/morpho/corpus.py` does `from .gui import *` at import time, so35   `import hm` fails on any Python without `_tkinter`. The stub defines the36   one name `corpus.py` refers to. Applied only when `import tkinter` fails.373. **Patch 2, lexicon loader** (`patches/morphology.patch`, applied with38   `patch -p1` inside site-packages). `words1.srf` lines are39   `ortho phon POS` with `#` comment lines, but the loader built a dict from40   the raw split, which raises on the first real entry.414. **Amharic data.** `src/hm/languages/a.tgz` from the same commit42   (sha256 `3274f71d...95c`), unpacked into `hm/languages/`.43 44Always run with a UTF-8 locale (`LANG=C.UTF-8`).45 46## The call that matters47 48```python49import hm50readings = list(hm.anal("a", "ይገኛል", degem=False))   # degem=False keeps the '/' marks51```52 53With the default `degem=True` the `/` marks are stripped from `seg`, and the54only gemination information left is the `+gemN` root feature, which misses55every geminate created by a prefix or a suffix. That was the bug in v1 of this56lexicon.57 58```python59import gemination as G60G.analyze("ይገኛል")                 # runs HornMorpho, returns patterns, most frequent root first61G.word_gemination(word, readings)  # same, from readings you already have62G.reading_gemination(word, {"seg": "----ይ</ገ/ኝ>---ኣ/ል---"})   # one seg, no HornMorpho needed63G.mark("ይገኛል", [1, 2])            # 'ይገ፟ኛ፟ል'64```65 66## Rebuilding the release files67 68`build_lexicon.py` converts the internal lexicon (one entry per word with the69stored HornMorpho segs) into `data/`. It re-derives every pattern from its segs70and stops if any disagrees, so the data cannot drift from this extractor.71