Survey of Arabic/English Machine Translation

This is an overview, not a review. Nothing has been done yet on evaluating quality of output, or suitability for a particular project. Thanks to Robin Burk, Myriam Siftan, Le Blevennec, and 'Marcus Cicero' for suggestions and pointers, both here and at Winds of Change. Also the AAMT/EAMT compendium for raw data and company information, a good starting point for any language pair. Finally, the DARPA TIDES project and NIST for funding much of the development and evaluation in this area. The following appear to be the major commercial vendors in the English/Arabic market. I've omitted a few smaller, desktop-only vendors, as well as internal projects at integrators and academic-only projects. Comments and additions solicited.

In (English) alphabetical order:

AppTek Inc. (aka Applications Technology) Private, American company. MacLean, VA. Offers several bidirectional English/other language pairs, including Arabic, Farsi, Turkish, and Pashto. Multiple overseas offices including UAE, Jordan, Turkey. Standalone desktop product (TranSphere) as well as standalone or server based product for translating websites (Webtrans - PDF, as well as related speech, OCR and translation memory products. Participant in TIDES/NIST projects, likely a lot of USG agency business.

ATA Soft (Al Misbar) Private, UK company. Exclusive focus on Arabic/English. Largely desktop-based products. Formerly offered a free online translator as Al-Misbar, but this now seems to be for-pay only.

Cimos Private, French company. Offers multiple language pairs, but seems focused on English, French, and Arabic. "Semantic analyzer" based technology. Offers desktop MT products as well as automatic and manual translation as a service. Once/day trial or online subscription Arabic translator available. Cimos' products are also distributed/relabeled by MTM Linguasoft in the US.

Ciyasoft Private, American company. Specialist in English & ME languages: Arabic, Farsi, Urdu, Pashto. Offers CiyaTran MT in desktop and server versions, as well as related OCR, speech, database products. Based on MT technology acquired from "TR, Inc.". USG contractor, as well as integration services.

Language Weaver Private, venture funded, American company. Arabic/English bidirectional and other language pairs. Server and Desktop versions. A spinoff of USC/ISI natural language research. Corpus based. Backed by DARPA and In-Q-Tel. USG supplier and participant in NIST evaluations.

Sakhr Software Company (Ajeeb) Private, UAE+Egypt company. Arabic language specialists, with broad product line including OCR, voice reco, categorizers. English/Arabic MT, "enterprise class". Many ME offices as well as US. Sells to USG and prime contractors. Maintains the Ajeeb portal, including the formerly-free Tarjim online translator. Interesting profile of founder.

Systran Public, French company. Many language pairs using long-standing rule based engine. Well known as technology behind Google and Babelfish translators. Vendor to USG and EU. Arabic language pairs are new, apparently partially developed, under EU Arabic/French/English MT research program. Restricted domains of translation in area of sustainable development. The English menu is a good example of MT output in itself. I did try this one. While it's fascinating to type a French query into an English interface and retrieve documents in three languages, a clip from the English translation of an Arabic document called New destinations of the Arab tourists makes for 'interesting' reading:

Hosts that in Europe generally and until in America building do not feel complete changes in manner of the living live and the general atmospheres for life there, therefore believed that the traveling follows far countries in ['asy] of identicals adventurer of the traveling for discovery deserves this countries which does not tie us in her any common factors and consequently this trip will be if what rich trip and useful very fulfilled, and she follows Alan her postponed dream for de follows when the suitable circumstances be available. The singer Mohammed bruise dreams in visit of India for that he, just as said, her admirer in civilization and her art and her people his children curiosity for acquaintance this his rich country in culture just as appears that through the film. Hosts that what the Indian film watch in does not mean necessarily that he the truth idealizes, therefore so the rising in trip follows India will discover for him [aalkthyr] [mn] the secrets and number of the questions answers about. Says that he his visit will exploit India for visit follows...

Got that? It does make it clear why, in previous experiments with MT in social settings, the community members spent a lot of time making fun of the translation program!

Comments

[chap]

Thanks for this. I needed a good overview because I was looking for something that the new Afghan blogger could use--looks like a broke Afghani has no real resources right now!

[Adriana]

I know some resources related to Arabic/English Machine Translation. Lingo24 offers a free online Arabic-English and English-Arabic dictionary and an Arabic language paraphrasing tool.