The Arabic English Newswire Translation Collection was produced by the Linguistic Data Consortium (LDC). It consists of approximately 550,000 words of Arabic newswire text and its English translation from Agence France Presse (France), An Nahar (Lebanon) and Assabah (Tunisia). The source Arabic text was used in LDC's Arabic Treebank, specifically, in Part 1 (Part 1 v. 2.0; Part 1 v. 3.0), Part 3 (Part 3 v. 1.0; Part 3 v. 2.0) and Part 4 (Part 4 v. 1.0). A subset of Agence France Presse (AFP) source text from Arabic Treebank: Part 1 v. 2.0 was previously translated and released by LDC in Arabic Treebank: Part 1 - 10K-word English Translation, LDC2003T07. The English translations in this corpus were provided by translation agencies using LDC's Arabic Translation Guidelines.
The number of stories and their epochs for each source are as follows:
|AFP ||734 stories; July 2000 - November 2000 |
|An Nahar ||600 stories; January 2002 - December 2002 |
| Assabah ||397 stories; September 2004 - November 2004 |
|Total ||1731 stories
Word count of Arabic tokens by source is shown in the following table:
The original source files used different encodings for the Arabic characters, including UTF8 and ASMO. SGML tags were used for marking sentence and paragraph boundaries and for annotating other information about each story. All Arabic source data was converted to UTF and most SGML tags were removed or replaced by "plain text" markers.
Portions © 2000 Agence-France Presse, © 2002 An Nahar, © 2004 Assabah, © 2002-2005, 2009 Trustees of the University of Pennsylvania