Zeineuski — Basque Dialect Identification

Fine-grained dialect identification (DID) system for Basque (Euskara). Given a text or speech sample, classifies it into one of six dialect categories: Western (Bizkaiera), Central (Gipuzkera), Navarrese, Navarrese-Labourdin, Souletin (Zuberera), or Standard Basque (Batua).

Source code: github.com/itzune/zeineuski

Architecture

Zeineuski uses a three-tier hierarchical classification architecture:

Tier 1: batua / dialectal (binary)
  └─ Tier 2: 5-class euskalkia (dialect classification)
       └─ Tier 3: 12-class azpieuskalkia (sub-dialect classification)

Classification taxonomy

The project follows Koldo Zuazo's dialect classification, which is the current linguistic consensus and the basis for Ahotsak.eus's municipality→dialect mapping.

Zuazo recognizes 6 euskalkiak (dialects):

#	Euskalkia	Our label	Notes
1	Bizkaiera / Mendebalekoa	western
2	Gipuzkera / Erdialdekoa	central
3	Goi-nafarrera	navarrese	Upper Navarrese
4	Ekialdeko nafarrera / Erronkariera	(merged into navarrese)	Extinct ~1990s; tiny data
5	Zuberera	souletin
6	Nafar-lapurtera	nav-lab
+	Euskara batua	batua	Standard unified Basque

Why 5 euskalkis + batua instead of 6 + batua?

Ekialdeko nafarrera (Salazarese/Roncalese) is linguistically a distinct dialect, but it has been functionally extinct since the 1990s (last native speaker died in 1991). Ahotsak.eus has only ~65 passages across 7 towns in the Zaraitzu and Erronkari valleys. The Klasikoak.armiarma.eus classical literature corpus — which provides most of our Tier-2 training data — maps these texts to navarrese since the dialect distinction is not present in pre-20th-century literary sources.

For Tier 3 (azpieuskalkia), we follow the Zuazo azpieuskalki taxonomy as implemented on Ahotsak.eus. The official Ahotsak municipality→ azpieuskalki mapping provides the ground truth labels for sub-dialect classification.

Data Sources

Source	Content	Dialects	Status
Klasikoak	Literary texts (pre-20th c.)	5 euskalkis	Train
Ahotsak.eus	Oral history transcriptions	12 azpieuskalkis	Train + Test
SÜ AZIA	Pastoral scripts + blog articles	Zuberera	Train + Test

See docs/data_sources/suazia_zuberotarra.md for the SÜ AZIA corpus documentation.

Models

Euskalki (Dialect) Classification — 5 euskalkis + batua (6-class)

Hierarchical 2-step classifier (binary batua/dialectal → 5-class euskalkiak):

Variant	Filename	Size	XNLI (3-class)	Test (4-class)	Batua F1
final	hier_binary_final.bin + hier_dialect_final.bin	1.5GB	92.42%	95.18%	0.962
quantized	hier_*_quantized.bin	417MB	92.38%	95.16%	0.961
compact	hier_*_compact.bin	189MB	91.78%	94.71%	0.957
tiny	hier_*_tiny.bin	112MB	91.90%	94.88%	0.961
web	hier_binary_web.bin + hier_dialect_web.bin	32MB	91.06%	94.33%	0.952

Per-class F1 (final): Western 0.953, Central 0.933, Nav-Lab 0.949, Batua 0.962.

Azpieuskalki (Sub-Dialect) Classification — 12-class (v2, 2026-06-11)

Fine-grained sub-dialect classifier trained on Ahotsak.eus oral history transcriptions and augmented with the SÜ AZIA Zuberotarra corpus (6,676 pastoral + blog sentences).

Training data (42,229 sentences):

Azpieuskalki	Sentences	%	Source
mendebal-sortaldea	13,059	30.9%	Ahotsak
erdialde-sartaldea	9,804	23.2%	Ahotsak
zuberera	6,050	14.3%	Ahotsak (441) + SÜ AZIA (6,676)
erdialde-sortaldea	4,966	11.8%	Ahotsak
nafar-ipar-sartaldea	1,966	4.7%	Ahotsak
nafar-sortaldea	1,516	3.6%	Ahotsak
naflap-sortaldea	1,395	3.3%	Ahotsak
nafar-hego-sartaldea	1,101	2.6%	Ahotsak
naflap-sartaldea	726	1.7%	Ahotsak
ekialde-nafarra	710	1.7%	Ahotsak
nafar-erdigunea	497	1.2%	Ahotsak
mendebal-sartaldea	439	1.0%	Ahotsak

Results (84.06% overall on 7,445 test samples):

Azpieuskalki	Test	Accuracy
zuberera	1,067	94.19%
mendebal-sortaldea	2,304	90.58%
nafar-ipar-sartaldea	346	88.15%
erdialde-sartaldea	1,729	83.40%
erdialde-sortaldea	876	79.11%
nafar-sortaldea	267	75.66%
naflap-sortaldea	246	71.95%
naflap-sartaldea	127	69.29%
ekialde-nafarra	125	68.00%
nafar-erdigunea	87	49.43%
nafar-hego-sartaldea	194	48.45%
mendebal-sartaldea	77	48.05%

Model variants:

Variant	Filename	Accuracy	Size	vs original
original	azpieuskalki.bin	84.06%	243MB	baseline
quantized	azpieuskalki_q.bin	82.15%	22MB	-1.91pp, 11× smaller
bucket=50K	azpieuskalki_b50000.bin	83.47%	129MB	-0.59pp, 1.9× smaller
bucket=50K Q	azpieuskalki_b50000_q.bin	81.96%	5.5MB	-2.10pp, 44× smaller

Usage

import fasttext

# Load a model
model = fasttext.load_model("azpieuskalki.bin")

# Predict
text = "Neská jin düzü, Zuñ néska?"
labels, probs = model.predict(text, k=3)
print(labels[0].replace("__label__", ""), probs[0])
# Output: zuberera 0.978

Or use the zeineuski CLI from the source repo:

uv run zeineuski predict --text "Gaur goizean goiz jaiki naiz"

Web Demo

Try it in your browser — no server, no install:

itzune.eus/euskalkid (source)

34MB of fastText models running via WebAssembly. Works offline after first load.

Training

Optimal hyperparameters (discovered via pi-autoresearch, 37 experiments over 3 sessions):

Azpieuskalki 12-class:

fasttext supervised -input train_azpieuskalki.txt -output azpieuskalki \
  -dim 200 -lr 0.2 -epoch 75 -wordNgrams 2 -minn 2 -maxn 6 -loss ns

Key insight: NO autotune — aggressive LR decay overfits to dominant classes. Character n-grams (minn=2,maxn=6) capture Basque morphological patterns (case endings, verb suffixes) that are dialect-specific (+9.4pp improvement).

Changelog

v2 (2026-06-11)

• Added 6,676 SÜ AZIA Zuberotarra sentences (pastoral scripts + blog articles)
• Zuberera training data: 750 → 7,117 sentences (1.9% → 14.3%)
• Zuberera per-class accuracy: 94.19%
• Overall 12-class accuracy: 84.06% (from 82.08% in v1)
• New model variants: q (22MB), b50000 (129MB), b50000_q (5.5MB)

v1 (2026-06-08)

• Initial 12-class azpieuskalki model: 82.08% accuracy
• 9-class variant (min_samples=600): 83.55%

License

MIT