GALE Program: Arabic to English Translation Guidelines

Version 2.7

September 7, 2010

Linguistic Data Consortium

http://www.ldc.upenn.edu/Projects/GALE

2

1 Introduction .............................................................................................................. 3

2 Translation Teams ................................................................................................... 3

3 File Formats ............................................................................................................. 4

3.1 Source file......................................................................................................... 4

3.2 Completed translation file ................................................................................. 5

5.3 Numbers ........................................................................................................... 8

5.4 Capitalization .................................................................................................... 9

5.5 Punctuation ....................................................................................................... 9

5.6 Idioms ............................................................................................................. 10

5.7 English or other language content .................................................................. 10

5.8 Factual Errors in Source Text ......................................................................... 11

5.9 Typographical errors ....................................................................................... 11

5.10 Difficult to Translate Source Text .................................................................... 11

5.11 Headlines and Titles ....................................................................................... 12

5.12 Special Issues for Translation of Speech Sources.......................................... 12

5.13 Special Issues for Translation of Weblogs and Newsgroups .......................... 16

6 Quality Control at LDC ........................................................................................... 16

7 Guidelines .............................................................................................................. 19

Appendix: Transcription and translation mark-ups ........................................................ 20

3

These guidelines provide an overview of requirements for translation of Arabic text into

English to support the GALE Program evaluation of machine translation technology.

Genres covered by these guidelines include the following:

•

•

•

This document describes the format of the source text and its translation, and

addresses specific issues when translating text from different genres.

2 Translation Teams

Initial translation is performed by translation services employing professional translators.

Services must assign files to be translated to a team consisting of at least two

A translation service may have multiple teams working simultaneously. Proofreaders

may be shared among teams, unless LDC provides instructions to the contrary for a

particular project.

Translation teams may use an automatic machine translation system and/or a

translation memory system to assist them during translation.

Translation teams must be documented as follows:

• Translator and proofreader profiles consisting of name or pseudonym, native

4

such metadata as speaker labels, timestamps, section and turn boundaries or other

information. LDC converts all source text into a standard translation format before

sending data out to translators in order to 1) make the source files easy for translators

to read; 2) to avoid translator’s tampering with the metadata; and 3) to aid automatic

processing of the data after the translation is returned to LDC.

3.1 Source file

Each source file delivered from LDC is divided into segments that roughly correspond to

sentences or sentence-like units. Each Arabic segment in the source file consists of 3

components.

be translated.

A sample source file appears below:

ب ثدا اه   ن آ   أ و رآد-ن أ ؟

يا ر ا  ر ن ترأو  أ أ.

5

3.2 Completed translation file

The completed translation should be formatted exactly the same as the source file; the

difficult for LDC to automatically post-process the translation files. Similarly, in cases

where a single Arabic sentence is translated into multiple English sentences, no blank

lines should be inserted between the English sentences.

Speaker IDs in square brackets, e.g. [speaker1], are provided to facilitate clear

understanding of conversational speech. They should not be translated or copied over

into the translation. Occasionally speaker names do not appear inside square brackets

but instead appear at the start of the source text. In such cases, the names should be

translated as usual, using the standard guidelines for translating proper names (see

Section 5.2).

6

Completed translation files should be submitted electronically, as zipped email

attachments, via FTP or by web upload as specified by LDC. Paper transmission is not

acceptable. All completed translation files must be in plain text, not in some proprietary

software format (like Microsoft Word).

Using a descriptive header helps LDC direct your message or delivery to the right place,

ensuring timely response.

7

Whenever an Arabic proper name has an existing conventional translation in English,

that translation should be used. For example, Gamal Abdel Nasser ا  ل( ), the

late former president of Egypt, should be translated as Gamal Abdel Nasser, not Jamal

Abdel Nasser as Modern Arabic would have suggested.

8

The order of last name, first name presentation for the name in the source file should be

changed to Bin Laden Osama in the translation.

Recall that Speaker IDs that appear inside square brackets are provided as an aid to

9

In American English, commas are generally used for numbers with more than three

digits unless they are the names of years.

Use a combination of numerals and words for very large numbers.

 و ن ا اه رود أ نرأو و.

Translators should follow standard written English rules for capitalization unless there is

strong evidence in the source text that suggests a different treatment. Proper names

16

Affairs of the Hour

Weblogs and Newsgroups frequently contain emoticons and URLs. Please follow the

instructions below when translators come across these cases.

online posting. Emoticons should be copied over to English translation.

The following is an incomplete list of popular emoticons you may see in weblogs and

English, but mostly in its accuracy. Our definition of “quality” first and foremost requires

17

the translation delivery to be faithful to the source, preserving its original meaning and

style. This should be accomplished with the consideration that the translation should

also be comprehensible and in fluent English.

Each translation delivery received by LDC is reviewed for completeness, accuracy and

overall quality. Payment for completed translation is contingent upon successful

The grading system used by all translation reviewers is outlined below:

18

If more than 40 points are deducted from a 1200-word sample, the translation will be

considered unacceptable and the whole delivery will be sent back to the translation

team for improvement.

If a delivery is sent back to the translation team for further proofreading, the improved

Information about the data under review (volume, genre, etc.) and an overall rating for

this delivery as excellent, very good, good, fair, poor.

For each significant deduction above, at least one example is provided, along with the

following information:

19

In case these guidelines prove to be unclear, LDC reserves the right to modify them.

Agencies will always use the latest version.

20

Appendix: Transcription and translation mark-ups