Hey Google, How Do I Translate for Voice Search?
In the past decade, voice search has exploded. 20% of all Google mobile searches are voice queries. 41% of adults and 55% of teens use voice search daily, according to Google. 65% of people who own a Google Home or Amazon Echo “can’t imagine going back.” It’s no surprise then that experts agree that 50% of all searches will be voice searches by 2020.
Marketing and SEO specialists have already started to adjust their content strategies to optimize for voice. But what effects and opportunities will this new technology bring to the translation industry?
Translating content to train voice search machines comes with its own unique challenges and interesting points. The best voice search engines have put a lot of effort into generating the right responses, so much so that Google now might respond with an “uh-huh” instead of a “yes.” So the algorithms involved need to be meticulously trained to recognize context and produce natural-sounding responses. This principle applies to every language and puts specific constraints on translators and engineers.
Voice search localization for translators
Voice search requires a different kind of translation. Translators will get text files that are taken directly from someone’s speech. In order to create the translations that train the engine, a translator will need to think like an interpreter might, by translating only the parts that have meaning rather than every stutter and “um…” Once a translator has general contextual bits added, they’ll still need to build up all the subject matter expert vocabulary and phrases.
When producing a translation, a translator will need to record the audio of their translation to be fed to the machine, in addition to the text. While audio of a certain quality is necessary, it’s actually better to have “normal” people doing the recording rather than voice artists. This allows the machine to learn how to pick up the rhythms of natural speech. Voice search companies need this variety in their translations as well: translators of all ages, genders, and accents will give them the broadest set of data for machine learning.
Voice search localization for engineers
These new challenges for machine learning require more and more precise engineering, so the text can match up with the machine learning process.
For example, when an engineer gets an audio file, they need to be able to provide a very precise start and end time for the sound wave in order for the machine to sync up correctly. At other times, the file or timing formats might be different than translation industry standards. This means engineers may have to develop scripts to automate timestamp format conversion (or in the worst case scenario, manually adjust their workflows) to accommodate those changes.
There is no doubt that voice search will be getting more prevalent and keep providing interesting new avenues for the translation and localization industry. Venga’s specialists can help you translate your content for voice search across the globe—contact us to find out more.