Few technologies provoke more debate than machine translation (MT). Futurists have long predicted its widespread adoption, some experts opine that it will be ready for prime time in anywhere from two to 200 years, and casual experimenters trade silly results from online MT services. Companies consider it, but most choose either more expensive human translation (HT) or the pocket veto of no translation at all – what we label “zero translation.”

SDL sees a huge opportunity between these extremes of HT and ZT. We met with CEO Mark Lancaster and CTO Keith Mills to discuss their plans for unifying MT, workflow, and translation memory into a Knowledge-based Translation (KbT) system. They position it as “cost-effective translation services delivered through tightly integrated translation technologies.” They pit SDL’s technical prowess against language tool and service providers that have taken the Bangalore-Beijing Offshore Express to cost-cutting. We think that SDL’s product – unifying heretofore disconnected technologies – will add MT to many more companies’ short lists for meeting their translation needs.

Slated for availability next quarter but as yet unpriced, KbT innovates in its closed-loop integration of MT, translation memory (TM), workflow, and human services. While other suppliers can deliver each of these functions, SDL offers the industry’s first end-to-end MT solution that does not require major integration by the buyer.

• The knowledge base manages translation metadata. KbT builds a repository of data about translation practices, including terminology it detects, existing translation memories it incorporates, specialized dictionaries supplied by the buyer or third parties, corrections from humans, and a set of business rules telling the system how to use all this metadata and in what sequence. A new component, SDL Phrase Finder, uses algorithms from SDL’s adaptive MT engine, Translation Server, to analyze content and automatically extract words, terms, and phrases; it then employs the MT engine to add contextual meaning to what it has found. Like terminology extractors Fusion Translate and MultiTrans, Phrase Finder creates databases where there were none. Unlike those standalone tools, it works closely with MT and translation memory to continuously enhance the knowledge base.

• A workflow broker manages tasks. A new release of SDL Workflow manages and automates communication among the knowledge base, Phrase Finder, the Translation Server, customer content stores, and humans working on the project with the SDLX translation memory.

• Humans add the "secret sauce." As part of its language services business, SDL assembled editors, computational linguists, and terminologists who edit KbT output to achieve the desired level of quality. Traditionally, translators have rejected editing MT output, claiming quality was so bad that it was easier for them to translate from scratch than use the computer’s output. This poses a marketing challenge to SDL: It must marshal its field test experience with KbT to convince these subject-area editors that the quality is up to their standards.

Finally, SDL will not push MT to every prospect. Its ideal customer will have millions of words that need translation; patience to wait a couple of months while SDL trains the knowledge base; bland content like technical manuals and FAQs rather than florid marketing prose; and the right languages. Unlike other MT innovators like Ciyasoft and Language Weaver that help the Pentagon close the intelligence gap, SDL has focused on bidirectional translation between English and major European languages.

We believe that KbT advances MT acceptability considerably by integrating useful but underutilized MT and workflow with more broadly accepted translation memory.

• Companies need help translating too much information. The supranational “Real World Enterprise” (Jan04) will increase demand for translation inside businesses, within supply chains, and to meet regulatory needs. Integrated with content stores, KbT offers a way to deal with these large volumes of critical but untranslated data.

• Software suppliers take notice. Until Bowne reveals its heretofore stealthy solution, SDL remains the only vendor with a commercially available product including MT, TM, workflow, and services. Large suppliers like channel-savvy Systran have integrated MT with enterprise systems like R/3, but lack KbT’s closed-loop capability. SDL’s more complete offering will squeeze GMS players Global Sight, Idiom, and Trados. We expect to witness long-awaited consolidation.

• Language service providers will miss the point in dismissing MT quality. LSPs will mistakenly reject MT’s benefits. They should view KbT as a productivity tool, allowing them to focus on real value-add services rather than on price. SDL must address the concerns of smaller firms that see it both as a tool provider and rival LSP. Finessing that problem could cement SDL’s leading role in the industry.

By Donald A. DePalma
March 2004

Copyright © 2004 by Common Sense Advisory, Inc. Chelmsford,  Massachusetts,  United States of America.

Published by:

Common Sense Advisory, Inc.
8 Minuteman Drive
Chelmsford, MA 01824-4646 USA
+1.866.510.6101 or +1.978.256.7621
info@commonsenseadvisory.com
www.commonsenseadvisory.com

No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, scanning or otherwise, except as permitted under Section 107 or 108 of the 1976 United States Copyright Act, without the prior written permission of the Publisher.

Permission requests should be addressed to the Permissions Department, Common Sense Advisory, Inc., 8 Minuteman Drive, Chelmsford, MA 01824-4644, +1.978.256.7621, E-Mail: info@commonsenseadvisory.com. See www.commonsenseadvisory.com/others/citation_policy.php for usage guidelines.

Trademarks: Common Sense Advisory, Global DataSet, DataPoint, Globa Vista, Quick Take, and Technical Take are trademarks of Common Sense Advisory, Inc. All other trademarks are the property of their respective owners.

Information is based on the best available resources at the time of analysis. Opinions reflect the best judgment of Common Sense Advisory’s analysts at the time, and are subject to change.