AlcLaM is an Arabic dialectal pretrained-language model introduced in the paper.
alclam-base-v1 and alclam-base-v2 are publically available in huggingface:https://huggingface.co/rahbi/
A pipeline for DID is also available on huggingface:
from transformers import pipeline
classifier = pipeline("text-classification", model="rahbi/alclam-base-v1-DID")
result = classifier("ما ذلحين والله ما عاد اصدقك")
print(result)
Output:
[{'label': 'يمني', 'score': 0.999321460723877}]
To use our model, simply run fine_tuning.py. To achieve the results presented in the paper, use the folder Ft.
Below is a performance comparison of AlcLaM with other popular pretrained-language models on various datasets.
| Task | mBERT | LaBSE | AraBERT | ArBERT | MdBERT | CAMeL | MARBERT | AlcLaM | |
|---|---|---|---|---|---|---|---|---|---|
| DID | MADAR-2 | 72.9 ± 16.9 | 86.6 ± 0.5 | 87.1 ± 0.2 | 87.1 ± 0.2 | 86.0 ± 0.6 | 87.5 ± 1.0 | 85.3 ± 3.8 | 98.2 ± 0.1 |
| MADAR-6 | 91.3 ± 0.1 | 91.1 ± 0.2 | 91.6 ± 0.1 | 91.6 ± 0.2 | 91.6 ± 0.0 | 92.0 ± 0.1 | 92.2 ± 0.2 |
93.2 ± 0.1 |
|
| MADAR-9 | 75.5 ± 0.5 | 75.7 ± 0.2 | 76.8 ± 0.3 | 74.5 ± 4.3 | 75.9 ± 0.5 | 77.5 ± 0.4 | 78.2 ± 0.3 | 81.9 ± 0.3 | |
| MADAR-26 | 60.5 ± 0.2 | 62.0 ± 0.2 | 62.0 ± 0.1 | 61.7 ± 0.1 | 60.2 ± 0.4 | 62.9 ± 0.1 | 61.5 ± 0.4 |
66.3 ± 0.1 |
|
| NADI | 17.6 ± 0.5 | 17.6 ± 0.5 | 22.6 ± 0.5 | 22.6 ± 0.5 | 24.9 ± 0.6 | 25.9 ± 0.5 |
28.6 ± 0.8 |
25.6 ± 0.6 | |
| SA | SemEval | 51.3 ± 1.3 | 64.2 ± 0.7 | 65.4 ± 0.5 | 64.4 ± 0.9 | 65.6 ± 0.3 | 67.1 ± 0.7 | 66.4 ± 0.3 |
69.2 ± 0.4 |
| ASAD | 59.8 ± 0.0 | 62.4 ± 0.0 | 41.3 ± 0.0 | 66.9 ± 0.0 | 67.5 ± 0.0 | 65.8 ± 0.0 | 66.8 ± 0.0 | 66.7 ± 0.0 | |
| AJGT | 86.4 ± 0.3 | 92.4 ± 0.7 | 92.7 ± 0.3 | 92.6 ± 0.4 | 93.6 ± 0.0 | 93.6 ± 0.3 | 93.7 ± 0.1 |
95.0 ± 0.3 |
|
| ASTD | 46.3 ± 1.4 | 55.7 ± 0.4 | 57.5 ± 2.3 | 59.7 ± 0.1 | 61.9 ± 0.4 | 60.2 ± 0.2 | 61.0 ± 0.5 |
64.6 ± 0.1 |
|
| LABR | 81.1 ± 0.0 | 85.4 ± 0.0 | 85.9 ± 0.0 | 85.9 ± 0.0 | 84.7 ± 0.0 | 86.3 ± 0.0 | 85.0 ± 0.0 | 84.9 ± 0.0 | |
| ARSAS | 73.2 ± 0.7 | 76.2 ± 0.6 | 76.8 ± 0.3 | 76.1 ± 0.2 | 76.3 ± 0.2 | 77.1 ± 0.3 | 76.2 ± 0.2 |
77.9 ± 0.3 |
|
| HSOD | HateSpeech | 67.9 ± 1.4 | 73.7 ± 1.1 | 76.4 ± 1.2 | 76.8 ± 1.4 | 80.0 ± 0.1 | 78.8 ± 0.6 | 80.0 ± 0.8 |
81.4 ± 0.5 |
| Offense | 85.3 ± 0.5 | 87.2 ± 0.5 | 90.5 ± 0.4 | 90.5 ± 0.4 | 90.8 ± 0.2 | 89.2 ± 0.5 | 90.8 ± 0.3 |
91.3 ± 0.3 |
|
| Adult | 87.9 ± 0.1 | 87.2 ± 0.3 | 88.6 ± 0.1 | 88.4 ± 0.6 | 88.1 ± 0.0 | 88.6 ± 0.3 | 88.3 ± 0.1 |
89.3 ± 0.3 |
| Task | mBERT | LaBSE | AraBERT | ArBERT | MdBERT | CAMeL | MARBERT | AlcLaM | |
|---|---|---|---|---|---|---|---|---|---|
| DID | MADAR-2 | 97.3 ± 0.8 | 98.0 ± 0.1 | 98.1 ± 0.0 | 98.1 ± 0.0 | 98.0 ± 0.1 | 98.1 ± 0.1 | 97.2 ± 0.7 | 99.7 ± 0.0 |
| MADAR-6 | 91.3 ± 0.1 | 91.1 ± 0.2 | 91.6 ± 0.1 | 91.6 ± 0.2 | 91.6 ± 0.0 | 92.0 ± 0.1 | 92.2 ± 0.2 | 93.2 ± 0.1 * | |
| MADAR-9 | 78.5 ± 0.5 | 79.1 ± 0.1 | 80.4 ± 0.2 | 77.7 ± 3.6 | 79.1 ± 0.5 | 80.5 ± 0.2 | 81.1 ± 0.3 | 83.4 ± 0.4 | |
| MADAR-26 | 60.6 ± 0.2 | 61.9 ± 0.2 | 61.9 ± 0.1 | 61.7 ± 0.2 | 60.1 ± 0.3 | 62.9 ± 0.2 | 61.3 ± 0.3 | 66.1 ± 0.2 * | |
| NADI | 33.4 ± 0.6 | 33.4 ± 0.6 | 38.9 ± 1.7 | 38.9 ± 1.7 | 41.9 ± 1.9 | 42.7 ± 1.6 | 47.3 ± 0.1 * | 46.6 ± 1.0 | |
| SA | SemEval | 53.4 ± 1.5 | 65.0 ± 0.6 | 66.1 ± 0.5 | 65.1 ± 0.8 | 66.1 ± 0.3 | 68.0 ± 0.3 | 66.9 ± 0.3 | 69.5 ± 0.3 * |
| ASAD | 74.6 ± 0.0 | 75.2 ± 0.0 | 70.6 ± 0.0 | 78.4 ± 0.0 | 77.6 ± 0.0 | 77.0 ± 0.0 | 77.6 ± 0.0 | 79.5 ± 0.0 | |
| AJGT | 86.4 ± 0.3 | 92.4 ± 0.7 | 92.8 ± 0.3 | 92.6 ± 0.4 | 93.6 ± 0.0 | 93.6 ± 0.3 | 93.8 ± 0.1 | 95.0 ± 0.3 * | |
| ASTD | 46.7 ± 1.7 | 55.6 ± 0.6 | 57.7 ± 2.4 | 59.7 ± 0.3 | 62.0 ± 0.3 | 60.1 ± 0.2 | 61.0 ± 0.3 | 64.9 ± 0.1 * | |
| LABR | 90.4 ± 0.0 | 92.3 ± 0.0 | 92.8 ± 0.0 | 92.8 ± 0.0 | 91.9 ± 0.0 | 93.0 ± 0.0 | 92.6 ± 0.0 | 92.6 ± 0.0 | |
| ARSAS | 74.5 ± 0.8 | 77.2 ± 0.7 | 77.6 ± 0.3 | 77.0 ± 0.3 | 77.5 ± 0.3 | 78.0 ± 0.3 | 77.4 ± 0.4 | 78.6 ± 0.5 * | |
| HSOD | HateSpeech | 75.2 ± 2.2 | 80.0 ± 0.7 | 80.5 ± 1.4 | 80.8 ± 1.9 | 84.3 ± 0.3 | 83.3 ± 0.6 | 84.4 ± 0.4 | 84.6 ± 0.7 * |
| Offense | 91.7 ± 0.1 | 92.8 ± 0.4 | 94.5 ± 0.2 | 94.6 ± 0.4 | 94.6 ± 0.2 | 93.6 ± 0.2 | 94.8 ± 0.0 | 94.9 ± 0.1 * | |
| Adult | 95.0 ± 0.0 | 94.4 ± 0.2 | 95.2 ± 0.1 | 94.9 ± 0.4 | 95.1 ± 0.1 | 95.2 ± 0.2 | 95.1 ± 0.0 | 95.6 ± 0.0 |
If you use the code, please cite the paper:
@article{murtadha2024alclam,
author = {Murtadha, Ahmed and Saghir, Alfasly and Wen, Bo and Jamaal, Qasem and Mohammed, Ahmed and Liu, Yunfeng},
title = {AlcLaM: Arabic Dialectal Language Model},
journal = {Arabic NLP 2024},
year = {2024},
url = {https://arxiv.org/abs/2407.13097},
eprinttype = {arXiv},
eprint = {2407.13097}
}