Datasets:
QCRI
/

Modalities:
Text
Formats:
parquet
Languages:
Arabic
ArXiv:
Libraries:
Datasets
pandas
License:
File size: 12,045 Bytes
cf9c2d0
 
 
 
 
463a1e6
cf9c2d0
463a1e6
b4f6dea
cf9c2d0
 
 
3eb85c3
cf9c2d0
 
 
d4f49a9
cf9c2d0
52c4aed
9520efb
1c36271
 
 
 
 
 
 
 
 
 
 
 
 
 
 
897e2ce
1c36271
897e2ce
 
 
1c36271
 
 
 
 
 
 
 
 
 
 
 
 
897e2ce
1c36271
897e2ce
 
 
1c36271
 
 
 
 
 
 
 
 
 
 
 
 
897e2ce
1c36271
897e2ce
 
1c36271
 
 
 
 
 
 
 
 
 
 
 
 
 
897e2ce
1c36271
897e2ce
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cf9c2d0
 
 
 
 
 
 
9b61a49
cf9c2d0
 
 
9b61a49
 
cf9c2d0
 
 
 
 
 
 
 
 
 
 
 
 
98e60da
cf9c2d0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
98e60da
 
 
1c36271
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
---
annotations_creators:
- expert-generated
language_creators:
- found
language:
- ar
license:
- apache-2.0
multilinguality:
- multilingual
size_categories:
- n<1K
source_datasets:
- extended
task_categories:
- token-classification
task_ids:
- part-of-speech
pretty_name: Arabic POS Dialect
dataset_info:
- config_name: egy
  features:
  - name: fold
    dtype: int32
  - name: subfold
    dtype: string
  - name: words
    sequence: string
  - name: segments
    sequence: string
  - name: pos_tags
    sequence: string
  splits:
  - name: train
    num_bytes: 269629
    num_examples: 350
  download_size: 89684
  dataset_size: 269629
- config_name: glf
  features:
  - name: fold
    dtype: int32
  - name: subfold
    dtype: string
  - name: words
    sequence: string
  - name: segments
    sequence: string
  - name: pos_tags
    sequence: string
  splits:
  - name: train
    num_bytes: 239883
    num_examples: 350
  download_size: 89178
  dataset_size: 239883
- config_name: lev
  features:
  - name: fold
    dtype: int32
  - name: subfold
    dtype: string
  - name: words
    sequence: string
  - name: segments
    sequence: string
  - name: pos_tags
    sequence: string
  splits:
  - name: train
    num_bytes: 263102
    num_examples: 350
  download_size: 97055
  dataset_size: 263102
- config_name: mgr
  features:
  - name: fold
    dtype: int32
  - name: subfold
    dtype: string
  - name: words
    sequence: string
  - name: segments
    sequence: string
  - name: pos_tags
    sequence: string
  splits:
  - name: train
    num_bytes: 245717
    num_examples: 350
  download_size: 90503
  dataset_size: 245717
configs:
- config_name: egy
  data_files:
  - split: train
    path: egy/train-*
- config_name: glf
  data_files:
  - split: train
    path: glf/train-*
- config_name: lev
  data_files:
  - split: train
    path: lev/train-*
- config_name: mgr
  data_files:
  - split: train
    path: mgr/train-*
---

# Dataset Card for Arabic POS Dialect

## Table of Contents
- [Dataset Description](#dataset-description)
  - [Dataset Summary](#dataset-summary)
  - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
  - [Languages](#languages)
- [Dataset Structure](#dataset-structure)
  - [Data Instances](#data-instances)
  - [Data Fields](#data-fields)
  - [Data Splits](#data-splits)
- [Dataset Creation](#dataset-creation)
  - [Curation Rationale](#curation-rationale)
  - [Source Data](#source-data)
  - [Annotations](#annotations)
  - [Personal and Sensitive Information](#personal-and-sensitive-information)
- [Considerations for Using the Data](#considerations-for-using-the-data)
  - [Social Impact of Dataset](#social-impact-of-dataset)
  - [Discussion of Biases](#discussion-of-biases)
  - [Other Known Limitations](#other-known-limitations)
- [Additional Information](#additional-information)
  - [Dataset Curators](#dataset-curators)
  - [Licensing Information](#licensing-information)
  - [Citation Information](#citation-information)
  - [Contributions](#contributions)

## Dataset Description

- **Homepage:** https://alt.qcri.org/resources/da_resources/
- **Repository:** https://github.com/qcri/dialectal_arabic_resources
- **Paper:** http://www.lrec-conf.org/proceedings/lrec2018/pdf/562.pdf
- **Contacts:**
- Ahmed Abdelali < aabdelali @ hbku dot edu dot qa >
- Kareem Darwish < kdarwish @ hbku dot edu dot qa >
- Hamdy Mubarak < hmubarak @ hbku dot edu dot qa >

### Dataset Summary

This dataset was created to support part of speech (POS) tagging in dialects of Arabic. It contains sets of 350 manually segmented and POS tagged tweets for each of four dialects: Egyptian, Levantine, Gulf, and Maghrebi.

### Supported Tasks and Leaderboards

The dataset can be used to train a model for Arabic token segmentation and part of speech tagging in Arabic dialects. Success on this task is typically measured by achieving a high accuracy over a held out dataset. Darwish et al. (2018) train a CRF model across all four dialects and achieve an average accuracy of 89.3%. 

### Languages

The BCP-47 code is ar-Arab. The dataset consists of four dialects of Arabic, Egyptian (EGY), Levantine (LEV), Gulf (GLF), and Maghrebi (MGR), written in Arabic script. 

## Dataset Structure

### Data Instances

Below is a partial example from the Egyptian set:
```
- `Fold`: 4
- `SubFold`: A
- `Word`: [ليه, لما, تحب, حد, من, قلبك, ...]
- `Segmentation`: [ليه, لما, تحب, حد, من, قلب+ك, ...]
- `POS`: [PART, PART, V, NOUN, PREP, NOUN+PRON, ...]
```

### Data Fields

The `fold` and the `subfold` fields refer to the crossfold validation splits used by Darwish et al., which can be generated using this [script](https://github.com/qcri/dialectal_arabic_resources/blob/master/generate_splits.sh). 

- `fold`: An int32 indicating which fold the instance was in for the crossfold validation
- `subfold`: A string, either 'A' or 'B', indicating which subfold the instance was in for the crossfold validation
- `words`: A sequence of strings of the unsegmented token
- `segments`: A sequence of strings consisting of the segments of the word separated by '+' if there is more than one segment
- `pos_tags`: A sequence of strings of the part of speech tags of the segments separated by '+' if there is more than one segment

The POS tags consist of a set developed by [Darwish et al. (2017)](https://www.aclweb.org/anthology/W17-1316.pdf) for Modern Standard Arabic (MSA) plus an additional 6 tags (2 dialect-specific tags and 4 tweet-specific tags). 

|   Tag      | Purpose | Description                          |
| -----      | ------  | -----                                |
| ADV        | MSA     | Adverb                               |
| ADJ        | MSA     | Adjective                            |
| CONJ       | MSA     | Conjunction                          |
| DET        | MSA     | Determiner                           |
| NOUN       | MSA     | Noun                                 |
| NSUFF      | MSA     | Noun suffix                          |
| NUM        | MSA     | Number                               |
| PART       | MSA     | Particle                             |
| PREP       | MSA     | Preposition                          |
| PRON       | MSA     | Pronoun                              |
| PUNC       | MSA     | Preposition                          |
| V          | MSA     | Verb                                 | 
| ABBREV     | MSA     | Abbreviation                         |
| CASE       | MSA     | Alef of tanween fatha                |
| JUS        | MSA     | Jussification attached to verbs      |
| VSUFF      | MSA     | Verb Suffix                          | 
| FOREIGN    | MSA     | Non-Arabic as well as non-MSA words  |
| FUR_PART   | MSA     | Future particle "s" prefix and "swf" |
| PROG_PART  | Dialect | Progressive particle                 |
| NEG_PART   | Dialect | Negation particle                    |
| HASH       | Tweet   | Hashtag                              |
| EMOT       | Tweet   | Emoticon/Emoji                       |
| MENTION    | Tweet   | Mention                              |
| URL        | Tweet   | URL                                  |

### Data Splits

The dataset is split by dialect. 

|   Dialect             | Tweets |  Words  |
| -----                 | ------ |  -----  |
| Egyptian (EGY)        |  350   |  7481   |
| Levantine (LEV)       |  350   |  7221   |
| Gulf (GLF)            |  350   |  6767   |
| Maghrebi (MGR)        |  350   |  6400   |

## Dataset Creation

### Curation Rationale

This dataset was created to address the lack of computational resources available for dialects of Arabic. These dialects are typically used in speech, while written forms of the language are typically in Modern Standard Arabic. Social media, however, has provided a venue for people to use dialects in written format. 

### Source Data

This dataset builds off of the work of [Eldesouki et al. (2017)](https://arxiv.org/pdf/1708.05891.pdf) and [Samih et al. (2017b)](https://www.aclweb.org/anthology/K17-1043.pdf) who originally collected the tweets.

#### Initial Data Collection and Normalization

They started with 175 million Arabic tweets returned by the Twitter API using the query "lang:ar" in March 2014. They then filtered this set using author-identified locations and tokens that are unique to each dialect. Finally, they had native speakers of each dialect select 350 tweets that were heavily accented. 

#### Who are the source language producers?

The source language producers are people who posted on Twitter in Arabic using dialectal words from countries where the dialects of interest were spoken, as identified in [Mubarak and Darwish (2014)](https://www.aclweb.org/anthology/W14-3601.pdf). 

### Annotations

#### Annotation process

The segmentation guidelines are available at https://alt.qcri.org/resources1/da_resources/seg-guidelines.pdf. The tagging guidelines are not provided, but Darwish at al. note that there were multiple rounds of quality control and revision.

#### Who are the annotators?

The POS tags were annotated by native speakers of each dialect. Further information is not known.

### Personal and Sensitive Information

[More Information Needed]

## Considerations for Using the Data

### Social Impact of Dataset

Darwish et al find that the accuracy on the Maghrebi dataset suffered the most when the training set was from another dialect, and conversely training on Maghrebi yielded the worst results for all the other dialects. They suggest that Egyptian, Levantine, and Gulf may be more similar to each other and Maghrebi the most dissimilar to all of them. They also find that training on Modern Standard Arabic (MSA) and testing on dialects yielded significantly lower results compared to training on dialects and testing on MSA. This suggests that dialectal variation should be a significant consideration for future work in Arabic NLP applications, particularly when working with social media text. 

### Discussion of Biases

[More Information Needed]

### Other Known Limitations

[More Information Needed]

## Additional Information

### Dataset Curators

This dataset was curated by Kareem Darwish, Hamdy Mubarak, Mohamed Eldesouki and Ahmed Abdelali with the Qatar Computing Research Institute (QCRI), Younes Samih and Laura Kallmeyer with the University of Dusseldorf, Randah Alharbi and Walid Magdy with the University of Edinburgh, and Mohammed Attia with Google. No funding information was included.  

### Licensing Information

This dataset is licensed under the [Apache License, Version 2.0](http://www.apache.org/licenses/LICENSE-2.0).

### Citation Information

Kareem Darwish, Hamdy Mubarak, Ahmed Abdelali, Mohamed Eldesouki, Younes Samih, Randah Alharbi, Mohammed Attia, Walid Magdy and Laura Kallmeyer (2018) Multi-Dialect Arabic POS Tagging: A CRF Approach. Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), May 7-12, 2018. Miyazaki, Japan.

```
@InProceedings{DARWISH18.562,
  author = {Kareem Darwish ,Hamdy Mubarak ,Ahmed Abdelali ,Mohamed Eldesouki ,Younes Samih ,Randah Alharbi ,Mohammed Attia ,Walid Magdy and Laura Kallmeyer},
  title = {Multi-Dialect Arabic POS Tagging: A CRF Approach},
  booktitle = {Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)},
  year = {2018},
  month = {may},
  date = {7-12},
  location = {Miyazaki, Japan},
  editor = {Nicoletta Calzolari (Conference chair) and Khalid Choukri and Christopher Cieri and Thierry Declerck and Sara Goggi and Koiti Hasida and Hitoshi Isahara and Bente Maegaard and Joseph Mariani and Hélène Mazo and Asuncion Moreno and Jan Odijk and Stelios Piperidis and Takenobu Tokunaga},
  publisher = {European Language Resources Association (ELRA)},
  address = {Paris, France},
  isbn = {979-10-95546-00-9},
  language = {english}
  }
```

### Contributions

Thanks to [@mcmillanmajora](https://github.com/mcmillanmajora) for adding this dataset.