This is an exciting step for Lingo24, putting us at the heart of the future of…
Global enterprises are under pressure to deliver more and more content to more people, faster. Lingo24’s Neil Campbell explores how a Neural approach offers data gathering opportunities to help them prioritise their content for translation.
It’s well known that people prefer to engage with content in their own language. But, for global enterprises with high volumes of content, prioritising what should be translated can be challenging. Key product and services pages are clearly candidates for professional translation, to convert and engage customers. But what about internal news feeds or lighter content like lifestyle blogs?
All too often, limited budgets and time constraints mean that lower-tier content is left untouched. But advances in Neural Machine Translation are making this type of content more viable for translation: a customised approach can accommodate large volumes of content, and produce quality output at speed, unlocking more content than ever before.
A data-driven approach with Neural MT
Now, some global enterprises are gathering data and reviewing how people interact and engage with raw Neural Machine Translation content (MT with no human editing). They use this data to gauge how valuable their content actually is for their audience and whether the content is viable for professional translation. Businesses that adopt a data-driven approach with Neural MT replace the guesswork in their localization strategy with analytical, informed decisions, saving them time and money.
This data-gathering process is unlocked by the power of a customised Neural MT engine that can produce content quickly and at scale with a decent level of quality with greater fluency.
Prioritising content for translation with the feedback loop
Data gathering isn’t new, some have already taken it a step further by seeking feedback on the quality of the content. Indeed, many of the technology giants (Symantec, Microsoft, etc) have led the way here. By adding disclaimers to warn users if content is from raw MT output and asking them for feedback, enterprises can capture even more data.
By combining the use of Machine Translation with a feedback mechanism from users and the usage analytics from their site, they are able to make informed decisions and prioritise for investment. For example, if a page attracts heavy traffic but receives many one star ratings, it’s a clear candidate for professional translation.
Collecting feedback from users isn’t innovative in itself, but combining it with a raw MT approach is unlocking vast amounts of content that previously would have remained untranslated.
Collecting feedback from users isn’t innovative in itself, but combining it with a raw MT approach is unlocking vast amounts of content that previously would have remained untranslated because of budget constraints or a lack of understanding of where to target the localization budget. All in all, it’s a win both for users and for the business.
No need for disclaimers
Some of our enterprise customers have begun deploying raw MT content without disclaimers, due to the high performance levels of the output.
Ooni Pizza Ovens specialise in high-end outdoor pizza ovens that offer restaurant quality at home. Ooni’s data tell them that customer conversions are higher when their content reflects their strong brand tone of voice. But it was a challenge to localize the huge volumes of content the company had already built in English as their business exploded internationally. How to localize their content fast enough was a key issue.
Based on their content type and data available, Lingo24 built Ooni a customised NMT engine. The goal of the project was always to post-edit this content with the aim of building an asset for the future.
Without this customised approach the content wouldn’t have been considered for translation, resulting in a mix of local language and English content.
Data is King
Using data in this manner is an innovative way to help enterprises learn what content their audience values, and where they should invest their translation budget and resources. Using high-quality Raw MT output from a customised Neural MT engine is a key enabler.
Neural MT is unlocking other opportunities, including localization at scale, user generated content and internal knowledge sharing.