Over 3 years ago at a LISA conference in Crystal City, Virginia (US) we took localization software vendors to task for slow or no innovation. Since then the biggest changes in language-centric software have largely been reactions to general phenomena such as the web and XML. While both have changed how these tools work and caused companies to revisit their information architectures, neither has particularly changed the practice of localization.

There are some bright spots. Software vendors such as Google, Language Weaver, and MultiCorpora have been more innovative and aggressive in melding computer cycles and vast storehouses of digitized data with their statistics-based machine translation and repository work. Their efforts benefit from the massive amount of digitized content in multiple languages, driving compute-intensive pattern matching and other sophisticated data discovery, mining, and harvesting techniques.

Dublin-based Alchemy took a similar statistical approach to software localization, applying mathematical formulae to the tedious tasks of adapting user interfaces (UI). The repeated dialog-by-dialog stretching and tweaking of UIs in other languages constitutes a huge sinkhole for software companies, on average taking a developer 5 to 10 minutes to stretch each dialog, menu, or control, resize boxes, and fix aspect ratios.

Alchemy’s CEO Tony O’Dowd says that Layout Manager’s probability-based artificial intelligence (AI) algorithms calculate possible transformations until it finds one that yields a good layout (the major criteria are that nothing is truncated and all the elements fit). The company claims that its tests yielded a 97% accuracy rate at high processing speeds. For example, Symantec’s Veritas unit changed Asian language fonts in 1,000 dialog boxes. It took them 408 seconds using Layout Manager, a task that would have taken 5 weeks with the previous version of Catalyst. Do the math — that’s less than 7 minutes versus 50,400 minutes, or just about 0.000138% of the time.

The history of language services might teach us a lesson here: As translation became commoditized and prices stabilized, larger LSPs made up for the flat tariff per word with profitable engineering and DTP services. Alchemy’s new tool threatens to be a disruptive technology, cutting deeply into those engineering margins at LSPs such as Lionbridge and SDL — think about reducing 3,024,000 seconds to 408 seconds. But until Layout Manager proves itself in the field, you can be sure that LSPs will use it in the back room while charging full price to the unaware — following the path they took in the early days of translation memory.

Localization managers regularly intone a litany of cost, time, and quality. If Alchemy’s claims of high accuracy and high performance play out, software companies can re-distribute developers and testers to more localization efforts — or to other functions where they see more benefit.