From Inspiration to Commercialization

On October 9 and 10, the Université du Québec en Outaouais hosted the TAMA 2008 Conference. TAMA stands for Terminology in Advanced Management Applications.

One hundred and forty-eight participants from 20 countries took part in this conference. According to Gabriel Huard, Director of the Terminology Standardization Directorate at the Translation Bureau, Public Works and Government Services Canada: "This conference met participant expectations not just in terms of the presentations and the quality of the location, but also in terms of UQO's reception and services."

TAMA is a conference regularly organized by TermNet, the International Network for Terminology, to allow international players in the field of terminology to convene. TermNet President Frieda Steurs and Executive Secretary Gabrielle Sauberer gave the opening address.

UQO Rector Jean Vaillancourt highlighted the leading research role the university plays in this field.

TAMA is a top-rated international event where representatives from institutions, organizations and companies from around the world converge.

TAMA participants identify and discuss world trends as well as new technologies in the vast field of applied terminology.

TAMA is recognized as a business-oriented conference where particular attention is paid to developing systems, methods and services related to terminology management within the content and language industries.

Another meeting took place: the Language Industry Association showcase was held November 17. To visit the event website, please click on the following link:

http://ailia.4poyntzdezign.com/content.php?sec=4

The Translation Bureau's professional magazine, Language Update, is celebrating its 40th anniversary this year! The first issue of Terminology Update, as the journal was originally called, was published in January 1968: this is almost a longevity record for a linguistic periodical.

Language Update is published four times a year. Along with TERMIUM^®, this prestigious journal is one of the terminological standardization tools used throughout the public service. It provides us with news of developments in the language industry and features a highly practical section dealing with the difficulties that language professionals, communicators and non specialists are faced with on a daily basis. There is something in it for all language lovers!

Since March 2008, the Translation Bureau has been disseminating Language Update free of charge on its Web site.

Read the special 40th anniversary issue of Language Update, discover its story and watch as your contentious language problems fade away!

The next edition of the MT Summit international conference, an event that is organized every other year jointly by IAMT (International Association for Machine Translation) and AMTA (Association for Machine Translation in the Americas), will be held from August 26 to August 30, 2009, at the Château Laurier in Ottawa (Ontario). The last edition, MT Summit XI, was held in Copenhagen (Denmark) in September, 2007.

MT Summit XII will bring together scientists, developers, and users of machine translation and other tools for multilingual processing. The conference will survey the international state of the art and practice in machine translation with:

• A full research program track
• Government and Commercial user program tracks
• Special sessions on technology in translator training and tools for translators
• A technology showcase with live demonstrations of MT systems and applications incorporating machine translation.

Before the main conference, one full day (Wednesday, August 26) will be devoted to tutorials, and after the conference another day (Sunday, August 30) will be devoted to specialized workshops. These wide-ranging activities will allow participants from research, business, and government to get together to explore how to use machine translation technology to conquer language barriers and overcome the digital divide.

As this was written, we learned that NIST (National Institute of Standards and Technology, U.S.A.) has committed to collocating their OpenMT evaluation workshop with MT Summit XII. They will hold their workshop on August 31 and September 1^st, just following the Summit. This is an international workshop by invitation-only that could gather 125 MT specialists; its objective is to help advance the state of the art of technologies that translate text between human languages, with the goal of producing outputs of an adequate and fluent translation of the original.

The call for research papers has been announced: submissions are due by the end of April, 2009. The Research Program Committee for MT Summit XII is chaired by Pierre Isabelle and co-chaired by Roland Kuhn, both from the NRC Interactive Language Technologies Group and well-known in the machine translation community. For further information, see the conference Web site (http://summitxii.amtaweb.org/).

Separate calls will be made in the coming weeks for Commercial and Government user Presentations and Case Studies, as well as for Workshop and Tutorial Proposals.

On May 14, 2008, NRC-IIT ILTG Group Leader Pierre Isabelle presented at the Université du Québec en Outaouais the latest research results obtained from the application of machine translation. The results were extremely encouraging in terms of the quality of the translations produced. The session was a great success as it attracted over fifty people from the public and private sectors, some of which were from outside Quebec.

Following a review of the evolution of machine translation (MT), from the rule-based (RBMT) approach to the most recent machine-learning-based statistical approaches, Pierre Isabelle provided some details on the NRC's statistical machine translation (SMT) system – PORTAGE – to bring the audience up to speed on the current context of MT. He then presented the results obtained by using the approach known as "Statistical Post-Editing," in which an SMT system "learns" to perform automatic post-editing of results produced by an RBMT system. This "learning" is done via a "bilingual" corpus (in fact, in one language only!) composed, on the one hand, of raw RBMT output and, on the other hand, either of the manually post-edited texts from the RBMT output or of independently produced human translations. The experiments involving the first approach were carried out with Job Bank (Service Canada), while experiments using the second approach were conducted with SYSTRAN, a company well known in the translation industry. In the latter case, it was possible to measure the SYSTRAN-PORTAGE hybrid system against its competitors through evaluation of the results conducted by humans: the SYSTRAN-PORTAGE combination was considered best and second in quality, respectively, in the two competitions in which the system participated, thereby showing that hybrid MT systems perform very well in terms of the quality of resulting translations.

A collaboration between two researchers from the Xerox Research Centre Europe (XRCE) in Grenoble (France) – Nicola Cancedda and Marc Dymetman – and two researchers from the NRC Interactive Language Technologies Group in Gatineau (QC, Canada) – Cyril Goutte and George Foster – resulted in the publication of a new textbook which will become the reference in statistical machine translation, and to which a number of specialists have contributed.

Entitled 'Learning Machine Translation' and published by MIT Press, this book covers the state-of-the-art contribution of machine learning methods to statistical machine translation. For further details, see: http://mitpress.mit.edu/catalog/item/default.asp?ttype=2&tid=11753.

The research team of the European project SMART (Statistical Multilingual Analysis for Retrieval and Translation, http://www.smart-project.eu/node/1) is ready to take an important step forward in its work plan during tests which will integrate statistical machine translation (SMT; in this case, the NRC PORTAGEshared system) into a translation memory environment (Trados of SDL International). The methods of SMT can, in principle, use feedback obtained from translators when the latter post-edit the SMT output. However, one had not yet been able to produce a system that would be capable of taking such feedback into account 'on-line', i.e. immediately after feedback is given by the translator: to-date, feedback was indeed exploited only following an 'off-line' re-training of the system. The methods to be tested by the SMART project researchers aim at producing systems that will have a real on-line learning capacity and the objective of the tests now in progress is to verify the feasibility of this new concept within a broader environment for translation support.

Since the on-line launch of the TerminoWeb 1.0 prototype in December, 2006, research and studies related to computational terminology have progressed well in the NRC Interactive Language Technologies Group and with their collaborators, taking advantage of the flexibility offered by this research platform. In particular:

• more than 30 users and user groups have accessed TerminoWeb 1.0: terminologists, translators, terminology researchers, linguists, and others;
• structured tests have been completed by three terminologists from the Terminology Standardization Directorate of the Translation Bureau of Canada;
• TerminoWeb has been used by researchers from the Observatoire linguistique sens-texte (OLST) laboratory of Université de Montréal, in the SACOT project (Semi-automatic Construction of Ontologies from Texts) of Defence R&D Canada (Valcartier, QC);
• TerminoWeb has been used during graduate studies seminars in terminology and terminotics in several universities: the University of Ottawa, the Université du Québec en Outaouais, and Glendon College (affiliated to York University, Toronto);
• a tutorial for TerminoWeb is being developed at the University of Ottawa to be included in their CERTT initiative (http://aix1.uottawa.ca/~certt/CERTT-main-EN.htm), which aims at giving access to a variety of technologies, tutorials, and exercises for teaching purposes, and also for the broader benefit of translators, terminologists, writers, and other language professionals who have an increasing need for technology tools.

Caroline Barrière currently puts the final touch to version 2.0 of TerminoWeb, which is planned for on-line launch by the end of 2008. This new version of TerminoWeb will bring important improvements, such as:

• the creation and analysis of text corpora in French (only English was covered to-date);
• the exploration of term lists in a more targeted fashion taking into account previously selected terms;
• the definition of sets of semantic relations in order to better answer the needs of a variety of types of analyses (e.g., a user may define a set of 'medical relations' which would be comprised of a number of semantic relations specific to that domain);
• the creation of corpora relevant to a set of terms (rather than a specific domain);
• the possibility to carry out more flexible searches in the corpus under study, e.g. by using as a linguistic pattern "is a % part of", which can identify sentences containing either "is a specific part of" or "is a large part of".

See the TerminoWeb prototype Web site: http://termino.iit.nrc.ca/termino_e.html

The NRC Institute for Information Technology's Interactive Language Technologies Group has been working since 2004 on a major Statistical Machine Translation (SMT) project: the PORTAGE project, the expertise of which is already recognized internationally. One of the preliminary processing tasks of the PORTAGE system on a bilingual corpus in which sentences are aligned is to carry out more precise alignments, at the phrase (sentence segment) level and also at the word level. This word alignment task can now be carried out independently, thanks to the PORTalign software offered by NRC under a commercial licence.

Word alignment, although developed for the purpose of SMT, opens the door to potential applications outside of the SMT field itself, for example in multilingual terminology applications and in the improvement of the functionalities of translation memory tools. The PORTalign software allows such exploration and implementations independently of a full implementation of SMT.

For further information, please contact: Michel.Mellinger@cnrc-nrc.gc.ca

The LTRC thanks the readers of Cybermag-TL for their interest and for helping to disseminate the newsletter. This online information magazine, which is also posted on the LTRC’s website, is designed to keep language industry and information technology partners and stakeholders up to date with the latest news from the LTRC. Please continue to spread the word among your colleagues, and contact us with your feedback, suggestions and comments!

Language Technologies Research Centre
283 Alexandre-Taché Boulevard
P.O. Box 1250, Hull Branch
Gatineau QC J8X 3X7
Tel. : (819) 595-3999
info@ltrc.ca

Go to the website