- Machine translation can be effective when used properly
- Human oversight is required in order to make machine translation useful
- A cost-effective and high-quality translation strategy uses both machine and human translation
In a previous job, where I managed the localization process for a global enterprise, I was taken by the head of Brazilian marketing to see a company banner in Sao Paolo. In English, the banner would have said “brilliant HD,” as in “brilliant high-definition.” Unfortunately, in Portuguese, this one said “shiny hard drive.” This error illustrates the core challenge of translation – getting the context and intent correct. “Shiny hard drive” isn’t technically wrong, but it sure isn’t right, either.
This challenge is especially problematic with machine translation, which relies on software and algorithms to translate without direct human involvement. Machine translation systems don’t always catch the subtleties and nuance that a human can. Does this mean machine translation – and its promise of lower costs and faster turnarounds than with human-involved translation – is not a useful tool? No. It does mean, however, that there are a number of key considerations to keep in mind:
- Review, review, review. For machine translation to be effective, there should be a human reviewing the output to avoid the example discussed above. The percentage of content that needs to be reviewed by humans depends on the nature of the content translated and the maturity of the overall process, but some amount of review always will be needed. Resources with the appropriate language skills and time to review should be made available for this task.
- Consider your content. As a simple rule, the more subjective your content is, the less suitable it is for machine translation. Marketing and branding content – in which the meaning of every word is scrutinized and every word choice matters – is best handled by people, because subtle connotations and variations of word choices can have significant impact. On the other hand, if your content is objective, highly structured and repetitive, then it is a good fit for machine translation with the appropriate amount of human validation. Evaluate the content you need translated and use this rule to determine if machine translation makes sense.
- Create your own translation memory. Machine translation relies heavily on the translation memory – i.e. the preexisting database that maps words to their translations. As a result, the quality of the resulting translation depends directly on the quality of this translation memory. To ensure the highest level of quality in translations, companies should invest in creating their own customized translation memories that map their preferred or ideal translations. Relying on generic translation memories may be faster and cheaper, but you will get generic translations, which will reduce the effectiveness of your marketing.
Especially for companies just beginning their translation efforts – along with those looking to reduce translation costs or timelines – machine translation and its promise of immediate, cost-effective and hands-free translation is undeniably intriguing and should be explored. Keep in mind that even machine translation will require a commitment of human resources and time. It should be part of – not the entirety of – the translation strategy.