Gene Roddenberry, of Star Trek fame, solved the language problem by inventing something non-existent. Is it here now?

It would defy logic that the crazy alien races on the original and subsequent Star Trek shows and movies would be able to communicate freely, without interpreters. So Gene Roddenberry, the show’s creator, invented a fictional device called a Universal Translator that solved the problem. Today, with neural machine translation (NMT), artificial intelligence (AI), and machine learning (ML), we are starting to see claims of something like a universal translator. But are they legitimate? And how does this affect international market research?

Market research translation is a natural target for automation

The reality is that language is a very challenging problem for automation, not because we’re not making strides towards useful machine translation (we are), but because languages, especially when you get away from European languages (that have been priorities for NMT), are complex. You have colloquialisms, obscure cultural references, localized dialects, a wide variety of alphabets, pronunciation issues…multiplied dozens of times by the sheer variety of languages. As we’ve pointed out in a series of articles, cultural factors alone can derail market research survey results based on language use. In spite of this, for a variety of factors, automation can and is being applied to market research translation. There are a number of things that make market research translation a good candidate for automation. But first you have to come down to reality and understand what ‘automation’ means.

Ignore the man (or woman) behind the curtain!

Automation in translation and localization workflows does not mean a machine does everything, despite marketing that implies a scenario like that is coming or here. NMT can’t do localization, nor can it edit itself. There are humans behind the curtain turning that machine translation into understandable and culturally relevant language. There have to be people involved and that is for the foreseeable future. The problems are simply too complex, especially when multiple languages are involved. Are you surveying people in China? What languages are you using? Where are they living? Dialects change with regional differences. Don’t assume there is a language called Chinese. There isn’t. There is Mandarin and Cantonese and simplified Chinese and local dialects…you need human translators who know the difference and the underlying culture. This typically means native speakers who are in-country in the areas you are targeting.

What automation can do for market research translation

So, what does automation in market research translation mean? It turns out that MR, because of the platforms like Decipher and Confirmit that are used, is ready to connect with translation workflows via APIs built into these applications. The formats are standardized and the applications can export file formats like XML that can maintain formatting throughout a translation process. They are translated and reviewed by humans and then they flow back into the platform ready for delivery to respondents. The automation doesn’t remove the human element, it incorporates it for gains in speed and cost efficiency, without sacrificing accurate and sensitive localization processes.

Market research surveys are bringing information back and it is often colloquial

The wild card in the rosy MR automation story is open-ended comment field response translation. People seldom respond in nicely structured syntax that lends itself to machine translation. If a survey is not heavily reliant on open comments, MT can give some insights into meaning, but they must not be heavily weighted in compiling results. This area in particular, comment translation, is very dependent on human translators who know the nuances of meaning in common language answers.

Unlike the translation of most business content, surveys and survey results often target ‘regular’ people rather than demographics for whom you can assume a certain level of language sophistication, i.e. knowledge of business English. You have to speak in their language and they will reply in their day to day language. Localized translation is particularly critical to getting enough responses and completions. And that, ultimately, that requires human translators. The good news is everything else can and is being automated for faster turnaround and cost savings.