Datasets:
Formats:
csv
Languages:
Arabic
Size:
< 1K
ArXiv:
Tags:
MMLU
reading-comprehension
commonsense-reasoning
capabilities
cultural-understanding
world-knowledge
License:
Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
@@ -3,17 +3,21 @@ license: cc-by-nc-sa-4.0
|
|
3 |
task_categories:
|
4 |
- text-classification
|
5 |
- question-answering
|
|
|
6 |
language:
|
7 |
- ar
|
8 |
tags:
|
9 |
- MMLU
|
10 |
-
-
|
11 |
-
-
|
|
|
|
|
|
|
12 |
pretty_name: 'AraDiCE -- Arabic Dialect and Cultural Evaluation'
|
13 |
size_categories:
|
14 |
- 10K<n<100K
|
15 |
dataset_info:
|
16 |
-
- config_name: ArabicMMLU-
|
17 |
splits:
|
18 |
- name: test
|
19 |
num_examples: 14455
|
@@ -21,15 +25,167 @@ dataset_info:
|
|
21 |
splits:
|
22 |
- name: test
|
23 |
num_examples: 14455
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
configs:
|
25 |
- config_name: ArabicMMLU-egy
|
26 |
data_files:
|
27 |
- split: test
|
28 |
path: ArabicMMLU_egy/test.json
|
29 |
-
- config_name: ArabicMMLU-
|
30 |
data_files:
|
31 |
- split: test
|
32 |
path: ArabicMMLU_lev/test.json
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
---
|
34 |
|
35 |
# AraDiCE: Benchmarks for Dialectal and Cultural Capabilities in LLMs
|
@@ -40,12 +196,19 @@ The **AraDiCE** dataset is designed to evaluate dialectal and cultural capabilit
|
|
40 |
|
41 |
As part of the supplemental materials, we have selected a few datasets (see below) for the reader to review. We will make the full AraDiCE benchmarking suite publicly available to the community.
|
42 |
|
43 |
-
## File/Directory
|
44 |
|
45 |
TO DO:
|
46 |
|
47 |
- **licenses_by-nc-sa_4.0_legalcode.txt** License information.
|
48 |
-
- **README.md** This file.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
49 |
|
50 |
|
51 |
## Dataset Usage
|
@@ -63,6 +226,7 @@ The dataset is distributed under the **Creative Commons Attribution-NonCommercia
|
|
63 |
|
64 |
|
65 |
## Citation
|
|
|
66 |
|
67 |
```
|
68 |
@article{mousi2024aradicebenchmarksdialectalcultural,
|
|
|
3 |
task_categories:
|
4 |
- text-classification
|
5 |
- question-answering
|
6 |
+
- multiple-choice-question
|
7 |
language:
|
8 |
- ar
|
9 |
tags:
|
10 |
- MMLU
|
11 |
+
- reading-comprehension
|
12 |
+
- commonsense-reasoning
|
13 |
+
- capabilities
|
14 |
+
- cultural-understanding
|
15 |
+
- world-knowledge
|
16 |
pretty_name: 'AraDiCE -- Arabic Dialect and Cultural Evaluation'
|
17 |
size_categories:
|
18 |
- 10K<n<100K
|
19 |
dataset_info:
|
20 |
+
- config_name: ArabicMMLU-lev
|
21 |
splits:
|
22 |
- name: test
|
23 |
num_examples: 14455
|
|
|
25 |
splits:
|
26 |
- name: test
|
27 |
num_examples: 14455
|
28 |
+
- config_name: PIQA-msa
|
29 |
+
splits:
|
30 |
+
- name: test
|
31 |
+
num_examples: 1838
|
32 |
+
- config_name: PIQA-lev
|
33 |
+
splits:
|
34 |
+
- name: test
|
35 |
+
num_examples: 1838
|
36 |
+
- config_name: PIQA-egy
|
37 |
+
splits:
|
38 |
+
- name: test
|
39 |
+
num_examples: 1838
|
40 |
+
- config_name: OBQA-msa
|
41 |
+
splits:
|
42 |
+
- name: test
|
43 |
+
num_examples: 497
|
44 |
+
- config_name: OBQA-lev
|
45 |
+
splits:
|
46 |
+
- name: test
|
47 |
+
num_examples: 497
|
48 |
+
- config_name: OBQA-egy
|
49 |
+
splits:
|
50 |
+
- name: test
|
51 |
+
num_examples: 497
|
52 |
+
- config_name: Winogrande-msa
|
53 |
+
splits:
|
54 |
+
- name: test
|
55 |
+
num_examples: 1267
|
56 |
+
- config_name: Winogrande-lev
|
57 |
+
splits:
|
58 |
+
- name: test
|
59 |
+
num_examples: 1267
|
60 |
+
- config_name: Winogrande-egy
|
61 |
+
splits:
|
62 |
+
- name: test
|
63 |
+
num_examples: 1267
|
64 |
+
- config_name: TruthfulQA-msa
|
65 |
+
splits:
|
66 |
+
- name: test
|
67 |
+
num_examples: 780
|
68 |
+
- config_name: TruthfulQA-lev
|
69 |
+
splits:
|
70 |
+
- name: test
|
71 |
+
num_examples: 780
|
72 |
+
- config_name: TruthfulQA-egy
|
73 |
+
splits:
|
74 |
+
- name: test
|
75 |
+
num_examples: 780
|
76 |
+
- config_name: BoolQ-msa
|
77 |
+
splits:
|
78 |
+
- name: test
|
79 |
+
num_examples: 892
|
80 |
+
- config_name: BoolQ-lev
|
81 |
+
splits:
|
82 |
+
- name: test
|
83 |
+
num_examples: 892
|
84 |
+
- config_name: BoolQ-egy
|
85 |
+
splits:
|
86 |
+
- name: test
|
87 |
+
num_examples: 892
|
88 |
+
- config_name: BoolQ-eng
|
89 |
+
splits:
|
90 |
+
- name: test
|
91 |
+
num_examples: 892
|
92 |
+
- config_name: AraDiCE-Culture-glf
|
93 |
+
splits:
|
94 |
+
- name: test
|
95 |
+
num_examples: 30
|
96 |
+
- config_name: AraDiCE-Culture-lev
|
97 |
+
splits:
|
98 |
+
- name: test
|
99 |
+
num_examples: 120
|
100 |
+
- config_name: AraDiCE-Culture-egy
|
101 |
+
splits:
|
102 |
+
- name: test
|
103 |
+
num_examples: 30
|
104 |
configs:
|
105 |
- config_name: ArabicMMLU-egy
|
106 |
data_files:
|
107 |
- split: test
|
108 |
path: ArabicMMLU_egy/test.json
|
109 |
+
- config_name: ArabicMMLU-egy
|
110 |
data_files:
|
111 |
- split: test
|
112 |
path: ArabicMMLU_lev/test.json
|
113 |
+
- config_name: PIQA-msa
|
114 |
+
data_files:
|
115 |
+
- split: test
|
116 |
+
path: PIQA_msa/test.json
|
117 |
+
- config_name: PIQA-lev
|
118 |
+
data_files:
|
119 |
+
- split: test
|
120 |
+
path: PIQA_lev/test.json
|
121 |
+
- config_name: PIQA-egy
|
122 |
+
data_files:
|
123 |
+
- split: test
|
124 |
+
path: PIQA_egy/test.json
|
125 |
+
- config_name: OBQA-msa
|
126 |
+
data_files:
|
127 |
+
- split: test
|
128 |
+
path: OBQA_msa/test.json
|
129 |
+
- config_name: PIQA-lev
|
130 |
+
data_files:
|
131 |
+
- split: test
|
132 |
+
path: OBQA_lev/test.json
|
133 |
+
- config_name: PIQA-egy
|
134 |
+
data_files:
|
135 |
+
- split: test
|
136 |
+
path: OBQA_egy/test.json
|
137 |
+
- config_name: Winogrande-msa
|
138 |
+
data_files:
|
139 |
+
- split: test
|
140 |
+
path: Winogrande_msa/test.json
|
141 |
+
- config_name: Winogrande-lev
|
142 |
+
data_files:
|
143 |
+
- split: test
|
144 |
+
path: Winogrande_lev/test.json
|
145 |
+
- config_name: Winogrande-egy
|
146 |
+
data_files:
|
147 |
+
- split: test
|
148 |
+
path: Winogrande_egy/test.json
|
149 |
+
- config_name: TruthfulQA-msa
|
150 |
+
data_files:
|
151 |
+
- split: test
|
152 |
+
path: TruthfulQA_msa/test.json
|
153 |
+
- config_name: TruthfulQA-lev
|
154 |
+
data_files:
|
155 |
+
- split: test
|
156 |
+
path: TruthfulQA_lev/test.json
|
157 |
+
- config_name: TruthfulQA-egy
|
158 |
+
data_files:
|
159 |
+
- split: test
|
160 |
+
path: TruthfulQA_egy/test.json
|
161 |
+
- config_name: BoolQ-msa
|
162 |
+
data_files:
|
163 |
+
- split: test
|
164 |
+
path: BoolQ_msa/test.json
|
165 |
+
- config_name: BoolQ-lev
|
166 |
+
data_files:
|
167 |
+
- split: test
|
168 |
+
path: BoolQ_lev/test.json
|
169 |
+
- config_name: BoolQ-egy
|
170 |
+
data_files:
|
171 |
+
- split: test
|
172 |
+
path: BoolQ_egy/test.json
|
173 |
+
- config_name: BoolQ-eng
|
174 |
+
data_files:
|
175 |
+
- split: test
|
176 |
+
path: BoolQ_eng/test.json
|
177 |
+
- config_name: AraDiCE-Culture-glf
|
178 |
+
data_files:
|
179 |
+
- split: test
|
180 |
+
path: AraDiCE-Culture_glf/test.json
|
181 |
+
- config_name: AraDiCE-Culture-lev
|
182 |
+
data_files:
|
183 |
+
- split: test
|
184 |
+
path: AraDiCE-Culture_lev/test.json
|
185 |
+
- config_name: AraDiCE-Culture-egy
|
186 |
+
data_files:
|
187 |
+
- split: test
|
188 |
+
path: AraDiCE-Culture_egy/test.json
|
189 |
---
|
190 |
|
191 |
# AraDiCE: Benchmarks for Dialectal and Cultural Capabilities in LLMs
|
|
|
196 |
|
197 |
As part of the supplemental materials, we have selected a few datasets (see below) for the reader to review. We will make the full AraDiCE benchmarking suite publicly available to the community.
|
198 |
|
199 |
+
<!-- ## File/Directory
|
200 |
|
201 |
TO DO:
|
202 |
|
203 |
- **licenses_by-nc-sa_4.0_legalcode.txt** License information.
|
204 |
+
- **README.md** This file. -->
|
205 |
+
## Dataset Statistics
|
206 |
+
|
207 |
+
The datasets used in this study include: *i)* four existing Arabic datasets for understanding and generation: *Arabic Dialects Dataset (ADD)*, *ADI*, *QADI*, along with a dialectal response generation dataset, and *MADAR*; *ii)* seven datasets translated and post-edited into MSA and dialects (Levantine and Egyptian), which include *ArabicMMLU*, *BoolQ*, *PIQA*, *OBQA*, *Winogrande*, *Belebele*, and *TruthfulQA*; and *iii)* *AraDiCE-Culture*, an in-house developed regional Arabic cultural understanding dataset. Please find below the types of dataset and their statistics benchmarked in **AraDiCE**.
|
208 |
+
|
209 |
+
<p align="left"> <img src="./benchmarking_tasks_datasets.png" style="width: 40%;" id="title-icon"> </p>
|
210 |
+
|
211 |
+
<p align="left"> <img src="./data_stat_table.png" style="width: 40%;" id="title-icon"> </p>
|
212 |
|
213 |
|
214 |
## Dataset Usage
|
|
|
226 |
|
227 |
|
228 |
## Citation
|
229 |
+
Please find the paper <a href="https://arxiv.org/pdf/2409.11404/" target="_blank" style="margin-right: 15px; margin-left: 10px">here.</a>
|
230 |
|
231 |
```
|
232 |
@article{mousi2024aradicebenchmarksdialectalcultural,
|