README.md

---
language:
- kab
license: cc-by-4.0
configs:
- config_name: conjugation-tables
  data_files:
  - split: train
    path: conjugation-tables/train-*
- config_name: lemmatizer
  data_files:
  - split: train
    path: lemmatizer/train-*
  - split: validation
    path: lemmatizer/validation-*
  - split: test
    path: lemmatizer/test-*
- config_name: seq2seq
  data_files:
  - split: train
    path: seq2seq/train-*
  - split: validation
    path: seq2seq/validation-*
  - split: test
    path: seq2seq/test-*
dataset_info:
- config_name: conjugation-tables
  features:
  - name: id
    dtype: string
  - name: name
    dtype: string
  - name: translation
    dtype: string
  - name: hasDirectionalParticle
    dtype: bool
  - name: isIrregular
    dtype: bool
  - name: isDerived
    dtype: bool
  - name: pattern_id
    dtype: string
  - name: pattern_verb
    dtype: string
  - name: pattern_number
    dtype: string
  - name: imperative
    dtype: string
  - name: aorist
    dtype: string
  - name: preterite
    dtype: string
  - name: negativePreterite
    dtype: string
  - name: aoristParticiple
    dtype: string
  - name: preteriteParticiple
    dtype: string
  - name: negativePreteriteParticiple
    dtype: string
  - name: intensiveForms
    dtype: string
  - name: hasIntensiveForms
    dtype: bool
  splits:
  - name: train
    num_bytes: 12414868
    num_examples: 6198
  download_size: 2508441
  dataset_size: 12414868
- config_name: lemmatizer
  features:
  - name: form
    dtype: string
  - name: target
    dtype: string
  - name: infinitif
    dtype: string
  - name: tense
    dtype: string
  - name: person
    dtype: string
  splits:
  - name: train
    num_bytes: 27658845
    num_examples: 310270
  - name: validation
    num_bytes: 1533254
    num_examples: 17237
  - name: test
    num_bytes: 1536410
    num_examples: 17238
  download_size: 10130858
  dataset_size: 30728509
- config_name: seq2seq
  features:
  - name: infinitif
    dtype: string
  - name: tense
    dtype: string
  - name: person
    dtype: string
  - name: input
    dtype: string
  - name: target
    dtype: string
  splits:
  - name: train
    num_bytes: 27658845
    num_examples: 310270
  - name: validation
    num_bytes: 1533254
    num_examples: 17237
  - name: test
    num_bytes: 1536410
    num_examples: 17238
  download_size: 10130831
  dataset_size: 30728509
---
# Kabyle Verbs — Kabyle Verb Conjugation

Kabyle verb conjugation dataset — 6,198 verbs, ~344,000 conjugated forms, covering aorist, preterite, imperative, participles, and intensive forms.

Data source: [amyag.com](https://amyag.com), work by Kamal Nait Zerrad.

---

## Summary

| Property | Value |
|-----------|--------|
| Language | Kabyle (taqbaylit) |
| Verbs | 6,198 |
| Total conjugated forms | 344,745 |
| Unique forms | 214,276 |
| Grammatical tenses | 11 (aorist, preterite, negative preterite, imperative, intensive aorist, intensive imperative, participles...) |
| Persons | 1s, 2s, 3s m, 3s f, 1p, 2p m, 2p f, 3p m, 3p f + participle |
| Format | Structured JSON / HuggingFace Datasets |
| License | CC-BY-SA 4.0 |

---

## Repository Structure

This repository contains **3 configurations**:

### 1. `conjugation-tables` — Raw Tables
Raw dataset with complete conjugation tables for each verb.

```python
from datasets import load_dataset
ds = load_dataset("boffire/kabyle-verbs", "conjugation-tables")
```

**Fields:**
- `id` — unique verb identifier
- `name` — infinitive (e.g., `yeɣra`, `addi`)
- `translation` — French translation
- `hasDirectionalParticle` — directional particle present
- `isIrregular` — irregular verb
- `isDerived` — derived verb
- `imperative`, `aorist`, `preterite`, `negativePreterite` — forms by person (JSON)
- `aoristParticiple`, `preteriteParticiple`, `negativePreteriteParticiple` — participles
- `intensiveForms` — intensive forms with their own tenses
- `pattern` — morphological pattern (id, model verb, number)

### 2. `seq2seq` — Pairs for Automatic Conjugator
`(input, target)` format for training a seq2seq model (T5, mT5, etc.) to conjugate.

```python
ds = load_dataset("boffire/kabyle-verbs", "seq2seq")
```

**Example format:**
```
input  : "yeɣra | aorist | 1s"
target : "ɣraɣ"

input  : "addi | imperative | 2s"
target : "addi"

input  : "addi | preterite participle | participle"
target : "yuddin"
```

**Splits:**
- train: 310,270 examples
- validation: 17,237 examples
- test: 17,238 examples

### 3. `lemmatizer` — Pairs for Lemmatization
Inverse format: `(form, context)` for training a lemmatizer / morphological analyzer.

```python
ds = load_dataset("boffire/kabyle-verbs", "lemmatizer")
```

**Example format:**
```
form   : "ɣraɣ"
target : "yeɣra | aorist | 1s"
```

**Splits:**
- train: 310,270 examples
- validation: 17,237 examples
- test: 17,238 examples

---

## Tense Distribution

| Tense | Number of forms |
|-------|-----------------|
| Intensive aorist | 76,004 |
| Preterite | 60,183 |
| Negative preterite | 60,150 |
| Aorist | 59,156 |
| Intensive imperative | 23,134 |
| Imperative | 18,245 |
| Intensive aorist participle | 14,374 |
| Preterite participle | 9,854 |
| Aorist participle | 9,849 |
| Negative intensive aorist participle | 7,720 |
| Negative preterite participle | 6,076 |

---

## Use Cases

### Automatic Conjugator
Train a T5/mT5 model to generate the conjugated form from the verb, tense, and person.

```python
from transformers import T5Tokenizer, T5ForConditionalGeneration

tokenizer = T5Tokenizer.from_pretrained("google/mt5-small")
model = T5ForConditionalGeneration.from_pretrained("google/mt5-small")

# Fine-tune on the seq2seq dataset
```

### Lemmatizer / Morphological Analyzer
Train a model to recover the infinitive, tense, and person from an inflected form.

### Orthographic Correction
Use reference forms to automatically correct conjugation errors in Kabyle text.

### MT/ASR Corpus Enrichment
Generate paraphrases by varying verb tenses in parallel corpora (Tatoeba, Common Voice, etc.).

### Linguistic Resource
Reference for linguists, Kabyle learners, and language learning tools.

---

## Notes on Overlapping Forms

Approximately **22.7%** of forms appear in multiple contexts. This is expected because the dataset lists conjugation tables for both the preterite and the negative preterite, and Kabyle forms the negative using preverbal particles (`ur... ara`) rather than by altering the verb stem. Consequently, the verb form itself remains identical across these two tenses. Some participle forms also overlap with finite forms. This is not a defect in the data, but a reflection of the Kabyle morphological system.

For the **conjugator** (forward task), this is not a problem: the model generates the correct form given the explicit tense and person. For the **lemmatizer** (inverse task), a contextual model is needed, or ambiguity must be accepted.

---

## Related Resources

- [boffire/kabyle-pos](https://huggingface.co/datasets/boffire/kabyle-pos) — Morpho-syntactic tagging
- [boffire/kabyle-english-TM](https://huggingface.co/datasets/boffire/kabyle-english-TM) — English-Kabyle translation corpus
- [boffire/kabyle-tokenizer-T5](https://huggingface.co/datasets/boffire/kabyle-tokenizer-T5) — SentencePiece tokenizer adapted for Kabyle
- [boffire/mT5-kabyle-model](https://huggingface.co/boffire/mT5-kabyle-model) — mT5 model fine-tuned on Kabyle

---

## Citation

If you use this dataset, please cite:

```bibtex
@dataset{kabyle-verbs-2026,
  author = {MOKRAOUI, Athmane},
  title = {Kabyle Verbs: Kabyle Verb Conjugation},
  year = {2026},
  publisher = {HuggingFace},
  url = {https://huggingface.co/datasets/boffire/kabyle-verbs}
}
```

---

## Contact

- **Author**: Athmane MOKRAOUI (boffire)
- **HF Profile**: https://huggingface.co/boffire
- **Language**: Kabyle (taqbaylit) — Amazigh language spoken in Algeria
- **Data source**: [amyag.com](https://amyag.com), work by Kamal Nait Zerrad