Lisa Globalisation Conference Berkeley

The Role of Translation Software

Gudrun Magnusdóttir
CEO, ESTeam AB

NOTE: All summaries for the Berkeley Globalization Conference were done by volunteers. They are designed to be used in conjunction with the slides, i.e., the volunteers generally recorded only the information not already included in the slides. We thank Tim Parsons for this one.
Minimally revised by Gudrun Magnusdóttir


Gudrun Magnusdottir has 25 years of experience in the field of translation and 15 years experience working with high-end machine translation software for large organizations and companies.

Initially, the translation tools available were powerless because computers were unable to process large amounts of data.

However, the latest from Google news releases indicate they are still continually building and shipping many servers to handle increasing demand for their web tools, and translation is among their most popular tools.

The success of Google is on the whole positive because many more web pages in diverse languages, including small-population minority languages, are now accessible. Also, for average consumers and small organizations, the translation tool offered by Google is sufficiently useful to preclude any need to purchase expensive translation software. The current market for translation software is now highly competitive. However, Google’s usefulness is limited because corporations and large organizations point out that Google provides no assurance of confidentiality, 24/7 operational security nor enhanced clients customisation of in the information being processed. This is a problem for most large companies, as well as for the European Commission. Thus, Google is not used in the industrial translation environment for large organizations, corporations, etc.

Magnusdottir recommends the combination of TM (Translation Memory) + MT (Machine Translation) as the best solution now and for the foreseeable future because the ombination of functions and checking is more powerful than any other standalone solution available today.However, TM + MT is an expensive arrangement, and small players in the translation market cannot compete with this scenario and are getting squeezed out.

Many commercially available TMs are full sentence-based. However, full sentence-based translation software (in contrast to phrase-based TM software) is only useful for revising finished documents.

In the future, the combination of TM+MT will remain the best practice. And the trend is to use TM as an MT tool for reworking old documents.

In terms of building translation support tools, Magnusdottir was shocked at a meeting with the European Union (EU) to learn that translators are using Google widely. A vast array of older, already developed language EU/EC repositories remains untapped because the information contained therein is regarded as being too difficult to access.

Magnusdottir remarked that among trademark lawyers there is global use of MT for years. Also, they have become accustomed to reading and working with translations of poor quality but better than Google.

Looking ahead, most users of Google are satisfied, and there is a fear that Google will replace traditional translators. On the whole, Google is good for single users who are working with easy text, i.e., users should not expect high quality. In the future, all translation could be automated providing that the right software is built, except maybe poetry, but only if the TM is available and accurate.

Current TM fuzzy is problematic. It is a bottleneck on applying an automatic mode in TM. A meaning sensitive fuzzy assure better translation quality making it possible to reuse TM resources below the sentence also improve the throughput. Thus a more advanced fuzzy results in sub-sentence level output that is good. For quality purposes, Magnusdottir generally considers the combined tools of automatic translation with human translator support as the best translation method in use since 2001, and expects this to be the same for many years to come.

Q&A

1. In general, do translators not like MT output?
Answer: Yes, this is true.

2. MT for the European languages is successful, but is it also good for Asian languages?
Answer: Yes, but there are special cases. For example, the Japanese language’s inability to say “No”neces sitates the use of a human translator to create a Japanese-localized context. In this case, a rule-based system is too inflexible and renders poor quality.

3. Is there a role for additional software in the translation process?
Answer: Yes. My organization is now developing and deploying new software with metadata features, wherein redundancies are fixed, and terminology is made more flexible and accessible. Keep in mind that access to terminology is appreciated by human translators, but MT does not offer this capability.

4. Do you think minority languages will eventually be dominated by English? What is the viability of using TM with languages spoken by a smaller number of speakers?
Answer: Currently, to my knowledge, almost every available source for small languages is being collected to build repositories for these languages. So yes, the smaller languages will have more important roles than previously in the field of translation.

5. Assuming open standard and sentence-based TM is the trend, can your approach be modularized into something that can be adopted by corporations?
Answer: Yes, the open source scenario enables corporations to force the task of integrating their needs and the tools they’ve developed onto translation software suppliers. Suppliers will increasingly need to accommodate the particular needs and the homegrown translation of tools of their customers.

Berkeley Globalization Conference 2009
© 2009 ● The Localization Industry Standards Association ● All rights reserved

back to conference list

Downloads
Presentation