The complete [1 to 5]-gram Gumar Corpus in the style of Google n-grams.
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

The Gumar Corpus N-grams

Copyright © 2017-2018 New York University Abu Dhabi

Computational Approaches to Modeling Language (CAMeL) Lab


We present the Gumar Corpus n-grams. The n-grams are Generated from the Gumar Corpus, a large-scale corpus of Gulf Arabic containing more than 100 million words [1,2]. The n-grams are in order of 5, that is 5, 4, 3, 2 and 1 grams with their respective frequency counts and number of documents they appear in. The n-grams are counted across the entire corpus and also across each dialect category individually. The format of the n-gram files follows a similar format of Google n-grams with the exception of the year column which we don't produce.


  • All documents of the corpus are converted into plain text.
  • Basic UTF-8 character cleaning.
  • Punctuation separation.

Dialect Categorization

Below are categorizations of the dialects and their respective document counts. For specific information per document please refer to the spreadsheet attached with this package.

Tag Dialect Document Count
SA Saudi 770
AE Emirati 115
KW Kuwaiti 87
OM Omani 14
QA Qatari 10
BA Bahraini 8
MSA Modern Standard Arabic 82
EGY Egyptian 3
LEV Levantine 5
MOR Moroccan 1
IRQ Iraqi 5
YEM Yemeni 1
UNID_GA Unidentified Gulf Arabic 116
MIXED_GA Mixed Gulf Arabic 11
MIXED Gulf Arabic mixed with other Arabic dialects 4


You can download the GUMAR n-grams here.

The n-grams are split by dialect into seperate compressed folders of the form <TAG>.tar.xz where <TAG> is one of the dialect tags listed above. There is an additional file ALL.tar.xz that contains n-grams of all the dialects combined.

Once downloaded, you can extract the files by running the following:

tar -xJf <TAG>.tar.xz

This will generate a folder <TAG>/ in the current working directory.

Directory Structure

Each folder contains the following n-gram files:

  • 1-grams_<TAG>.tsv
  • 2-grams_<TAG>.tsv
  • 3-grams_<TAG>.tsv
  • 4-grams_<TAG>.tsv
  • 5-grams_<TAG>.tsv


Each n-gram file consists of three tab separated columns as follows:

<n-gram> TAB <frequency> TAB <# of documents> NEWLINE

Each <n-gram> larger than one is single space separated.

Example of a 2-grams row:

انتظر منك	85	69

* Note that the example above is displayed right-to-left but the columns are in the order described.

Each n-gram file is sorted by <frequency> in descending order.

Data Sources

If you would like more details on the data used to generate the n-grams, take a look at the Gumar_Info.tsv file. It is a Tab Separated Values file containing author and title information for each document, as well as its dialect and the link it was downloaded from. Duplicate entries for title-author pairs indicate that a document was split into multiple files.

* Please note that some entries in Gumar_Info.tsv containing double-quotes have been escaped. We recommend using a TSV reader (eg. Microsoft Excel, Apple Numbers, Google Docs, etc.) to parse these properly.


Please use the following citation when referencing or using this resource:

Khalifa, Salam, Nizar Habash, Dana Abdulrahim, and Sara Hassan. "A Large Scale Corpus of Gulf Arabic." In Language Resources and Evaluation Conference. 2016. Portorož, Slovenia


The Gumar Corpus n-grams are licensed under a Creative Commons Attribution 3.0 Unported License.


[1] Khalifa, Salam, Nizar Habash, Dana Abdulrahim, and Sara Hassan. "A Large Scale Corpus of Gulf Arabic." In Language Resources and Evaluation Conference. 2016. Portorož, Slovenia

[2] Khalifa, Salam, Nizar Habash, Fadhl Eryani, Ossama Obeid, Dana Abdulrahim, and Meera Al Kaabi. "A Morphologically Annotated Corpus of Emirati Arabic". In Language Resources and Evaluation Conference. 2018. Miyazaki, Japan