Speech Technology Magazine

 

Cutting the Cost of Translation

Plan for translation early in the design process.
By Sue Ellen Reager - Posted Nov 1, 2008
Page1 of 1
Bookmark and Share

The dream: To sell your product in dozens of languages. To 6 billion users worldwide. Tomorrow. 
Achieving this dream follows a prescribed path that starts with translating your application and then localizing the marketing materials. Translating the application, though, is often so difficult and expensive that many companies shy away from localization entirely.

Cost containment is essential to turning multilanguage application sales into reality. To get there, consider these five items:

  1. The cost of the engine. Many engines are available, from multilanguage ones to ones that function in only one language. If your system code is designed to handle grammars created from database entries and not manually hard-coded, then it is easier and cheaper to switch from one engine to another by language.
  2. The cost to rework system code for a new language. Cost rises dynamically with system complexity if your code is not designed for localization. It is cheap to prepare a new system for later translation, but extremely expensive to retrofit an unprepared system. It’s not just a matter of writing code but of long, extensive testing that can destroy an application’s go-to-market schedule.
  3. The cost of new code for the new language (grammars). New language grammars should be created via automation based on database content.  Using this approach, much linguistic content can be translated with automated language translation software (with its associated savings) and later refined by a human. The caveats are that automated grammar code, while more versatile, is less attractive and possibly a little slower. Note that translated grammar content will swell in size. For example, the English word "of," when translated to Spanish, becomes de, de la, del, de las, or de los. For a system to function optimally, the proper words must be included somewhere in the grammars, but it is vitally important to pare down the explosion of foreign text during later testing phases.
  4. The cost of script translation and voice prompt recording. The cost of script translation varies wildly by vendor and approach. Voice prompts require human translators. For voice recording, text-to-speech (TTS) is becoming a viable alternative to expensive voice talent and studios. The weakness of any TTS, of course, comes when words fall outside the prescribed grammar, but for many applications, TTS may be worth a serious second look, especially in foreign languages or systems requiring constant updates.
  5. The cost of testing and refining. It is cheaper and faster to test with text than with audio. Text testing involves translators or people typing a long list of sentences and phrases that they imagine someone may say, then running those sentences through the grammars to verify that the content is covered. The text testing script can be reused with later audio testing.

Many engines have a tool to test grammars. When a person speaks into a computer microphone, the words run through a grammar and the results are displayed on screen. This approach is the fastest possible for audio, but often is limited to a single grammar at a time and does not reflect a live system.

Full audio testing is the slowest and costliest testing process because it involves crawling through the entire system, with constant restarts. Savings can occur if the system is prepared for testing by code address, with the ability to jump directly to a particular place in the system.
Using the first two testing methods will significantly decrease time consumption for full audio testing because most grammars have already been perfected, and the person testing the system can concentrate on tweaking the final grammar and the application’s ability to process voice and context.

Savings can be accrued by altering the traditional path of application management and moving some manual elements to automation. But the single wisest approach is to plan for translation during the initial design stage and let the savings ripple throughout the life of the product. 


Sue Ellen Reager is CEO and founder of @International Services, a global translation services company and developer of localization software. She can be reached at sueellen@internationalservices.com.


Page1 of 1