IMB translation tool: Lionbridge

by Manuel Herranz The news of the month has undoubtedly been the announcement by Lionbridge to partner with IBM to develop (and probably offer) machine translation solutions. Possibly, the intention is to offer the advantages of MT to Lionbridge existing clients and maybe to control the technology. After all, whoever controls the technology, has a good chance of gaining (or consolidating) market dominance. The move must be welcome by all true believers of MT as the (new?) force of change in the translation industry. However, even though Lionbridge is the biggest language company jumping in the MT-DIY boat, it is not the first one to combine the offer of MT+PE as a substitute to the (increasingly old-fashioned?) TM or T+E+P models. With translator production reaching 850-1000+ words per hour and mounting production and price pressures, 20th-century technologies seem too cumbersome and antiquated for the demands of multilingual digital content. Who wins, who loses and what are the advantages for a MLV like Lionbridge to embrace MT? And above, why? WHY – All players in the industry are faced with the same questions and issues, buyers and vendors alike

Huge increases in the volumes of content to be translated.
A reduction is product/services-to-market times (faster production, faster reaction times are needed in digital economies). Translation “users” expect faster and faster translation deliveries.
Budgetary strains /unwillingness to increase or pay more translation costs / translation perceived as a cost, not as revenue-generating exercise

A fourth key point is the need to increase and speed human productivity in communication (just like the telephone brought immediacy compared to the letter, and TV news as compared with cinema news updates between the 20’s and 50’s and then 24h news channels in the 90’s). However, the need for immediacy and higher productivity is not new and has always been there, the digital era has only increased the speed of change from the typewriter to CAT tools to MT+PE. Translation buyers will gladly welcome any changes that provide for an increase from the 2,500-word/person mark to x3 or x4 times that figure with well-trained engines. That is the real driver for the adoption of MT: more content needs to be made available faster for global audiences. Nothing points to a decrease in content growth. Quite the opposite, and thus the pressures on cost and for higher productivity get stronger. But… all this will mean lower rates, right? How can then a language company benefit from lower rates in an ever more competitive market with an already low level of entry threshold? LOSERS Any “big” move by any of the big players in the industry will send shockwaves to smaller players. In this case, the players are manifold, although basically can be summarized into two groups:

1. LSPs caught out the loop and with no automatization procedures may not be able to produce the services at the speed and price required. It may be much for difficult to compete with the toughest cowboy in town when he also has the biggest gun. • Paradoxically, the new battle may include Lionbridge itself because its size and infrastructure costs. Projected loss in revenues from current clients may not make up for the increase in the number of words translated.

2. Computer Assisted Translation (CAT) vendors - unless they make a move for MT integration... fast! Most CAT vendors have integrated some sort of MT plug-in for a while (SDL Trados since 2007/08, Swordfish with a link to GT, etc.). However, as MT quality improves and more companies automate or customize engines, questions about the function of a CAT tool environment will not take too long to be on the table. If CAT tools become post-editing environment, there'll be no need for further [and costly] licenses, unless CAT vendors come up with some clever idea integrating the best of both worlds (see below). But can they without hurting their own interests?

3. MT technology suppliers may suffer if the combination proves too successful and wipes out open source and other MT alternatives, although the strength of the blow is not on them. They can still service corporate users and LSPs eager to fence off competition by gaining technology advantages (particularly those with a vertical structure and supplying to specific markets). WINNERS

First and foremost, translation buyers in general, because the agreement heralds the trend to come for years. As other companies adopt MT+PE as their technology and workflow offering, they will get the job done hopefully at no extra cost: the bursting in new content will be managed.
Savvy technology integrators and infrastructure (crowdsourcing) companies. All those companies offering workflow systems, CAT tools etc., will feel the need to integrate MT within their processes. Translation workflows will become part of any workflow system and those tools which have not integrated some kind of MT plug-in will feel the need to do so. * Initiatives with a vision to create clear win-win scenarios for all the stakeholders (buyers, translators, technology developers, even LSPs) will flourish.
Savvy LSPs now realizing that the old CAT tools must be put aside in their workflows. They can become nice post-editing environments. I am biased to say so, but PangeaMT is the only one LSP to have “reinvented” itself as an LSP offering SMT services + PE – and sharing the knowledge with clients and other LSPs.
For several years, data-sharing initiatives and platforms (from EU funded platforms now in development to industry leading organizations like TAUS’ TDA and smaller organizations offering web-spider services for data gathering) will become data repositories to enhance and acquire relevant data for MT training.

THE “BATTLE” TO COME - FLASHING-FORWARD

• Expect the emergence of new technology vendors. Medium-term, a wave of Merger and Acquisition (defensive moves and pre-emptive attacks). • The older generations of translators and old-fashioned LSPs (the traditional localization work) face a major re-training challenge with a whole new business paradigm. However, translators are key players in the change. We will witness an scenario similar to the adoption of TM technologies in the 90’s. • The value of the localization industry may be questioned. • A new set of tensions as a result of the new MT technologies and engine customizations being released into the market. • Price drops and price wars: As mentioned above, a signal that old business model for "fixed content" and TEP cycles are coming to an end.

We do not have all the answers at PangeaMT :) but we try to provide news and analysis and help the industry think and create favorable opinions for the advancement of the industry as whole. We do believe that the development and application of customized MT engines is the way forward (and in this respect Statistical MT always proves much more customizable or hybridizable than other technologies). Our next post will analyze the new future scenarios and above all if 1+1 will make 3 or perhaps less than 2. IBM's MT technology has been around for decades, fathering a few spin-offs and a few jokes (" Every time I fire a linguist, my system's performance improves”). This raises the question of why IBM has not become the revolutionary force behind MT systems deployment and language automation. It was down to another software company but with advertising as its major source of revenue to shake the translation industry to set the path offering free SMT for all. IBM follows the historical trend of decades of investment in rule-based MT and a non-reversible switch to Statistical MT. As the industry's omnipresent player, Lionbridge is well-known in the industry for its "captive client" philosophy and unmatched resources, but hardly as an innovator. In the meantime, here's some other perspectives

From SDL (Mark Lancaster's view on SDL blog, 4th May)
From the Automated Language Translation Group
From TAUS
From GTS Blog and review in the same blog
A view from an ex-LIOX manager, now turned LIOX freelancer (in Spanish)
Translators' point of view from PROZ to spice the moment (on TM ownership, too)

Next time you think languages, think Pangeanic

Manuel Herranz

05/10/10