Teaching computers to translate
An ambitious project to develop software capable of automatically translating between the European Union’s 23 official languages promises to help governments, businesses and citizens communicate more easily and cheaply.
Called Moses, the open source software is being developed by a team of researchers in the EUROMATRIX project. They aim to provide the EU’s private and public sectors with the means to translate documents quickly and accurately between any of the 27-nation bloc’s language pairs.
To achieve that goal, the researchers are relying on innovative technologies to take machine translation beyond the current state of the art.
Billions of euros spent on translations
Currently, European governments, European institutions, companies and citizens spend billions of euros a year employing an army of translators to translate legislation, business documents and legal papers between European languages.
The workload has more than doubled in recent years as the EU has gone from 15 Member States to 27 and from 11 official languages to 23.
Machine translation, in which computers rather than people do the work, has been an option for years, although its poor quality has generally made it unviable for anything but the most rudimentary uses.
As anyone who has ever attempted to translate text on the internet knows, the output of many automated systems is often littered with punctuation errors, misplaced words and grammatical mistakes that can make the text almost unintelligible.
Developing a viable alternative to human translators
To make machine translation a viable alternative to human translation, the EUROMATRIX researchers saw the need for a new approach. Instead of simply following the path of earlier systems, which translate texts using a set of predefined linguistic rules and a constrained lexicon, the EUROMATRIX team developed software to allow the computer to learn to translate from past work and experience.
In essence, the approach relies on the computer program referring to an existing body of translated text and using statistical analysis to determine how words are used.
The internet makes finding the necessarily vast body of existing translations easy, allowing the system to draw on texts in different languages to greatly improve the accuracy of translations over uniquely rule-based approaches.
The researchers have also investigated a hybrid approach, merging both statistical and rule-based systems.
Translation workshops and commercial interest
In order to demonstrate Moses, the EUROMATRIX team has organised a series of workshops involving universities across Europe, while the open source software itself has already elicited commercial interest.
Moses is currently being used by several big European organisations as well as some small and medium businesses, according to the project partners.
Call for proposal
See other projects for this call
Funding SchemeSTREP - Specific Targeted Research Project
38123 Povo - Trento