jamalinu/tarifit-catalan-public-services

---
language:
- rif
- es
- ca
license: cc-by-4.0
task_categories:
- translation
- text-generation
- morphology-prediction
tags:
- tarifit
- amazigh
- berber
- low-resource
- morphology
- verb-conjugation
pretty_name: Tarifit Morphological Verb Dataset
size_categories:
- n<1K
---

# 🏔️ Tarifit Morphological Verb Dataset (TMVD)

## 📌 Overview
The **Tarifit Morphological Verb Dataset (TMVD)** is a high-quality, human-curated resource designed to bridge the gap in Digital Language Support for **Tarifit** (Riffian Berber). Despite being spoken by millions in the Rif region and the diaspora, Tarifit remains a severely underrepresented Afro-Asiatic language in the NLP landscape.

Unlike many large-scale datasets for low-resource languages that rely on noisy web-scraping or inaccurate machine translation, this dataset has been **manually authored and validated by a native speaker** with a professional background in linguistics and multilingual proficiency (Berber, Arabic, Catalan, Spanish, French, English).

## 🧪 Research Value
This dataset provides a granular look at Tarifit verbal morphology, characterized by complex initial and internal changes. It is particularly valuable for:
* **Instruction Tuning:** Training LLMs to understand and generate grammatically correct Tarifit.
* **Morphological Analysis:** Evaluating the ability of models like Gemini, Llama, or Claude to handle non-concatenative morphology.
* **Low-Resource Benchmarking:** Serving as a "Gold Standard" for zero-shot or few-shot translation tasks involving Catalan and Spanish.

## 📊 Dataset Structure
The dataset contains structured conjugations for core verbs (including *iri, aqqa, ttuga, gar, day*), covering various tenses and moods:
* **Preterite & Imperfective:** Capturing the aspectual nuances of the Berber verb system.
* **Future & Imperative:** Including prohibitive and intensive forms.
* **Deictic & Existential Verbs:** Essential for building dialogue systems and spatial reasoning.

### Column Metadata
| Column | Description | Example |
| :--- | :--- | :--- |
| `verb_id` | Citation form (Lemma) | `iri` |
| `translation` | Semantic gloss in Spanish/Catalan | `ser/existir` |
| `pronoun` | Subject pronoun (Independent form) | `Nec` |
| `person` | Grammatical person (1, 2, 3) | `1` |
| `number` | Grammatical number (sg, pl) | `sg` |
| `gender` | Grammatical gender (masc, fem, neuter) | `neuter` |
| `tense` | Morphosyntactic category | `preterito` |
| `form` | The target conjugated form (Latin-based) | `edjig` |

## 🛠️ Applications
This project is a foundational component of an ongoing effort to create a **Tarifit-Catalan Public Service Dialogue System**, aimed at improving accessibility and linguistic inclusion for the Berber-speaking community.

## 📜 Citation & License
If you use this dataset in your research or for model training, please cite it as follows:

```bibtex
@dataset{tarifit_verb_conjugations_2025,
author = {Jamal I. (jamalinu)},
title = {Tarifit Morphological Verb Dataset},
year = {2025},
publisher = {Hugging Face},
url = {[https://huggingface.co/datasets/jamalinu/tarifit-verb-conjugations](https://huggingface.co/datasets/jamalinu/tarifit-verb-conjugations)}
}

---
语言：
- 塔里菲特语（rif，又称里夫柏柏尔语）
- 西班牙语（es）
- 加泰罗尼亚语（ca）
许可证：知识共享署名4.0协议（CC BY 4.0）
任务类别：
- 机器翻译
- 文本生成
- 形态学预测
标签：
- 塔里菲特语（tarifit）
- 柏柏尔语（amazigh）
- 柏柏尔语（berber）
- 低资源语言
- 形态学
- 动词变位
展示名称：塔里菲特语动词形态数据集（Tarifit Morphological Verb Dataset）
规模分类：样本量少于1000（n<1K）
---

# 🏔️ 塔里菲特语动词形态数据集（TMVD）

## 📌 数据集概览
**塔里菲特语动词形态数据集（Tarifit Morphological Verb Dataset, TMVD）** 是一份高质量的人工编纂并审核的资源，旨在填补塔里菲特语（里夫柏柏尔语，Tarifit/Riffian Berber）在数字语言支持领域的空白。尽管该语言在里夫地区及海外侨民群体中拥有数百万使用者，但在自然语言处理（Natural Language Processing, NLP）领域，塔里菲特语仍是代表性严重不足的亚非语系语言。

与许多依赖嘈杂网页抓取或不准确机器翻译构建的低资源语言大规模数据集不同，本数据集**由具备语言学专业背景及多语言能力（精通柏柏尔语、阿拉伯语、加泰罗尼亚语、西班牙语、法语及英语）的母语使用者手动撰写并验证**。

## 🧪 研究价值
本数据集详细展示了塔里菲特语动词形态学特征——该语言的动词系统以复杂的词首与词内变化为典型特征。其核心应用价值包括：
* **指令微调**：用于训练大语言模型（Large Language Model, LLM）以理解并生成符合语法规范的塔里菲特语文本。
* **形态学分析**：用于评估Gemini、Llama、Claude等模型处理非拼接式形态学任务的能力。
* **低资源语言基准测试**：作为涉及加泰罗尼亚语与西班牙语的零样本（zero-shot）或少样本（few-shot）翻译任务的“金标准”基准数据集。

## 📊 数据集结构
本数据集包含核心动词的结构化变位内容（涵盖`iri`、`aqqa`、`ttuga`、`gar`、`day`等动词），覆盖多种时态与语气：
* **一般过去时与未完成体**：展现柏柏尔语动词系统的体貌细微差别。
* **将来时与祈使式**：包含禁止式与强化式变体。
* **指示与存在动词**：是构建对话系统与空间推理模型的核心组件。

### 列元数据
| 列名 | 描述 | 示例 |
| :--- | :--- | :--- |
| `verb_id` | 引语形式（词元，Lemma） | `iri` |
| `translation` | 西班牙语/加泰罗尼亚语语义释文 | `ser/existir` |
| `pronoun` | 主语代词（独立形式） | `Nec` |
| `person` | 语法人称（1、2、3） | `1` |
| `number` | 语法数（单数sg、复数pl） | `sg` |
| `gender` | 语法性别（阳性masc、阴性fem、中性neuter） | `neuter` |
| `tense` | 形态句法范畴 | `preterito` |
| `form` | 目标变位形式（基于拉丁字母书写） | `edjig` |

## 🛠️ 应用场景
本项目是正在推进的**塔里菲特语-加泰罗尼亚语公共服务对话系统**开发工作的核心组成部分，旨在提升柏柏尔语使用者的语言可及性与社会包容性。

## 📜 引用与许可
若您在研究或模型训练中使用本数据集，请按以下格式引用：
bibtex
@dataset{tarifit_verb_conjugations_2025,
author = {Jamal I. (jamalinu)},
title = {Tarifit Morphological Verb Dataset},
year = {2025},
publisher = {Hugging Face},
url = {[https://huggingface.co/datasets/jamalinu/tarifit-verb-conjugations](https://huggingface.co/datasets/jamalinu/tarifit-verb-conjugations)}
}