---
language:
- en
- sq
size_categories:
- 100K<n<1M
task_categories:
- translation
- question-answering
- conversational
pretty_name: Lilium Albanicum Eng-Alb
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
dataset_info:
  features:
  - name: albanian
    dtype: string
  - name: english
    dtype: string
  - name: question
    dtype: string
  - name: response
    dtype: string
  - name: swapped
    dtype: int64
  - name: system_prompt
    dtype: string
  - name: __index_level_0__
    dtype: int64
  splits:
  - name: train
    num_bytes: 411612817
    num_examples: 502537
  - name: test
    num_bytes: 4130480
    num_examples: 5000
  download_size: 252168126
  dataset_size: 415743297
---

# Lilium Albanicum Eng-Alb


![Lilium Albanicum Dataset of QA Translation Pairs curated for LLM finetuning.](https://huggingface.co/datasets/noxneural/lilium_albanicum_eng_alb/resolve/main/lilium_albanicum.png)

**Task Categories**:
- Translation
- Question-Answering
- Conversational

**Languages**: English (en), Albanian (sq)

**Size Categories**: 100K < n < 1M

---

# Dataset Card for "Lilium Albanicum"

## Dataset Summary

The Lilium Albanicum dataset is a comprehensive English-Albanian and Albanian-English parallel corpus. The dataset includes original translations and extended synthetic Q&A pairs, which are designed to support and optimize LLM translation tasks. The synthetic pairs are generated to mimic realistic conversational scenarios, aiding in the development of more effective translation models.

## Dataset Attribution

### Translation Process:

The dataset comprises expert-generated translations, ensuring high-quality language pairs. The Q&A pairs are machine-generated, followed by rigorous human review and refinement to guarantee natural and coherent translations.

## Supported Tasks and Leaderboards

This dataset is primarily tailored for translation, question-answering, and conversational tasks, aiming to improve bilingual models' performance with a focus on contextual understanding.

## Languages

The dataset includes bilingual data in English (en) and Albanian (sq).

## Dataset Structure

### Data Instances

A typical data instance includes a text pair in English and Albanian, reflecting a conversational exchange or a Q&A format suited for translation tasks.

### Data Fields

- albanian: The corresponding Albanian translation of the text.
- english: The English version of the text.
- question: The question part of the conversational or Q&A context.
- response: The response part of the conversational or Q&A context.
- swapped: An integer (int64) indicating whether the roles in the conversation have been swapped.
- system_prompt: A string containing system prompts or instructions related to the text entry.

### Data Splits

The dataset is structured into appropriate splits for training, validation, and testing to facilitate effective machine learning practices.

## Dataset Creation

### Curation Rationale

The creation of Lilium Albanicum aims to fill the gap in high-quality, conversational-context-focused datasets for English-Albanian translation tasks, thereby enhancing the capabilities of translation models.

### Source Data

The source data originates from a well-established Albanian-English parallel corpus, enriched with synthetic yet realistic Q&A scenarios.

## Dataset Use

### Use Cases

The dataset can be employed for various NLP tasks such as bilingual translation, conversational understanding, and question-answering systems development, both in academic research and practical applications.

### Usage Caveats

The synthetic nature of some parts of the dataset may not encompass all nuances of natural language. Users should consider complementing it with naturally occurring text data for tasks requiring high levels of linguistic subtlety.

### Getting Started

The dataset is accessible through the Hugging Face datasets library, with support for streaming to handle large datasets efficiently.

________________________________________

**Dataset contributors**:

- Marlind Maksuti (contact: marlind.maksuti@gmail.com)
- StochastX team

**Acknowledgments**:

Special thanks to the creators of the original Albanian-English parallel corpus MaCoCu-sq-en 1.0 and to all contributors who participated in the generation and refinement of the Q&A pairs.

**License**:

This work is licensed under the MIT license.