YAML Metadata Warning:The task_categories "text2text-generation" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, audio-text-to-text, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-ranking, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, image-text-to-image, image-text-to-video, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, video-text-to-text, keypoint-detection, visual-document-retrieval, any-to-any, video-to-video, other

Albanian Orthographic Error Dataset - AlbHybridNet

English | Shqip

Overview

A large-scale dataset for orthographic error detection and correction in Albanian, built to train and evaluate the AlbHybridNet model - a hybrid system combining XLM-RoBERTa token classification with rule-based and embedding-based correction mechanisms.

The dataset aggregates text from six diverse sources and injects seven types of realistic spelling errors commonly observed in Albanian digital writing (social media, SMS, informal communication).

Statistics

Split	Examples
Train	133,295
Validation	16,661
Test	16,663
Total	166,619

Error samples: 43,075 (25.8%)
Clean samples: 123,544 (74.2%)

By source:

Source	Examples	%
Wikipedia	135,264	81.2%
Dibra	16,407	9.8%
Kapibara	4,779	2.9%
Synthetic	5,720	3.4%
UniMorph	1,589	1.0%
Newmark	3,057	1.8%

Error types:

Type	Description	Count
`diacritic_remove`	Diacritic removal (ë→e, ç→c)	21,317
`missing_char`	Missing character	12,882
`drop_final_vowel`	Final vowel dropped	8,698
`double_char`	Doubled character	7,690
`digraph`	Digraph substitution (gj→g, xh→x)	4,683
`diacritic_partial`	Partial diacritization	620
`morphological`	Morphological error	24

Data Format

Each line is a JSON object:

{
  "text": "Original or perturbed sentence",
  "words": ["list", "of", "tokens"],
  "labels": [0, 1, 0],
  "source": "wikipedia",
  "error_types": ["diacritic_remove"],
  "has_error": true
}

labels: 0 = correct word, 1 = erroneous word
error_types: list of injected error type(s) for this sentence

Repository Structure

data/
  train.jsonl          # 133,295 training examples
  validation.jsonl     # 16,661 validation examples
  test.jsonl           # 16,663 test examples
  augmented.jsonl      # human-verified samples
  stats.json           # dataset statistics
sources/
  newmark_clean.csv         # Newmark dictionary source
  error_pairs_all.csv       # word-level (correct, error, type) pairs
  albanian_whitelist.txt    # 686K normalised Albanian word forms
  informal_map.json         # informal→formal word mapping

Usage

from datasets import load_dataset

ds = load_dataset("aldamyzeqari/albhybridnet-albanian-orthographic-corpus")
print(ds["train"][0])

Citation

@dataset{albhybridnet2026,
  title   = {Albanian Orthographic Error Dataset},
  author  = {Myzeqari, Alda},
  year    = {2026},
  publisher = {Hugging Face},
  url     = {https://huggingface.co/datasets/aldamyzeqari/albhybridnet-albanian-orthographic-corpus}
}

Shqip

Dataset i shkallës së madhe për detektimin dhe korrigjimin e gabimeve drejtshkrimore në gjuhën shqipe, i ndërtuar për trajnimin e modelit AlbHybridNet - sistem hibrid që kombinon klasifikimin me shenjë të XLM-RoBERTa me mekanizma rregull-bazuar dhe të bazuar në embedding.

Dataset-i grumbullon tekste nga gjashtë burime të ndryshme dhe injekton shtatë lloje gabimesh drejtshkrimore realiste të vëzhguara shpesh në shkrimin dixhital shqip (media sociale, SMS, komunikim informal).

Llojet e gabimeve:

diacritic_remove - heqja e shenjave diakritike (ë→e, ç→c)
missing_char - shkronjë e munguar
double_char - shkronjë e dyfishuar
drop_final_vowel - heqja e zanores fundore
digraph - zëvendësim i digrafeve (gj→g, xh→x)
diacritic_partial - diakritikim i pjesshëm
morphological - gabim morfologjik

Downloads last month: 16

Total file size:

98.4 MB