How Machine Translation of Market Research Saves You Money and Time
Google Translate is a gamechanger for the translation and localization business. And market research translation probably reaps the most benefits from it, though you may not see it in action. Let’s take a look at how we use it in survey translation and how it impacts turnaround time and cost.
First, a quick history lesson: The evolution of Google Translate
From the dawn of computer time (the 1950s), machine translation has been a holy grail. After all, couldn’t these miracle ‘thinking machines’ give us a universal translator? Those computer scientists back then did tackle the problem, but it proved to be a lot more complicated than they thought. And it required a lot more memory and processing power, thousands of times more of each.
The early approach was to try and fill the computer with all the rules of grammar and syntax and the entire vocabulary for each language. Then it could match these up and create a translation. That was the theory, and it was pursued up until a few years ago. But even with the massive computing power we can throw at it these days, you still ended up with those often humorous or garbled translations. Then there was a huge breakthrough.
Neural machine translation, arguably the best example of artificial intelligence in actual use
Google had a large team invested in the old way of teaching machines to translate. It was labor intensive and required constant fine-tuning by humans to see any improvement. It was really an extension of manual human translation on a very primitive basis. But Google had a policy allowing people with outlying ideas to pursue them with a small budget and team, a ‘skunkworks’. The concept was a small team trying something off the map. The term skunkworks was inherited from the aerospace industry. And one of these skunkworks teams was pursuing using artificial intelligence (AI) to do translation.
AI, more specifically machine learning (ML), could learn from existing translations and the corrections required to make them usable. Google had a huge database of this experience to draw from and the team used it. They built a demo and showed it to the core team and it showed a big improvement over the old way, that had been worked on for years with huge investments driving it. The skunkworks team approach blew the old process away, using far fewer resources. And Google made a visionary decision. Virtually overnight they changed directions and went all in with the new direction. And neural machine translation (NMT) was born.
Google Translate: A major upheaval for the market research translation business
As you might imagine, this created a lot of fear and uncertainty in the translation business. Would Google Translate eat us alive? It happened to encyclopedias, phone books, and libraries, to name a few disrupted models. The question was whether to embrace it or discredit it and both approaches were tried. But reality was upon us and forward-looking translation companies chose to see it as a powerful new tool that could benefit ourselves and our customers. The challenge was to find useful applications for it and market research survey translation proved to be a good match.
Speed baby, speed!
Market research providers depend on getting results fast. Information is volatile and public opinion even more so. But accuracy in translations is also critical to understanding the intent of those being surveyed. A poorly translated survey can return inaccurate or false results and a low completion rate. Speed and quality can butt heads on translation projects. So language service providers who do a lot of survey translation are in constant search of ways to streamline workflows while maintaining quality. Machine translation helps.
Machine translation and market research translation
Machine translation can do a baseline translation virtually instantly. This baseline can be reviewed and edited for accuracy and cultural relevance by human translators. Adding MT into our workflows does speed things up. But that is the survey side of the process. Translating comments from open comment fields can be a beast, especially if the survey is large and in multiple languages. Here’s where MT shines.
Comment translation: Getting to the core intent, quickly
People responding to surveys write comments in everyday language, conveying meaning through colloquialisms, slang, and other non-standard language. The primary goal in assessing this feedback is to get the basic meaning or to zero in on specific complaints, confusion, emotion, etc. Machine Translation can extract enough meaning from the responses to categorize them. Any MT that isn’t clear can be touched by a human translator for clarification. It is a lot faster.
This impacts cost- is your LSP passing the savings on to you?
Obviously, this makes a big difference in turnaround time. But how does it help cut costs? The real cost in translation is the human time spent translating, reviewing, and editing the work. Cut back on the time spent there and the project budget goes down. But are these savings passed on to the market research customer? The process is not transparent and it would be easy for the LSP to increase profits by using MT. But MR survey translation is incredibly competitive and competition tends to uncover inefficiencies as companies look for ways to compete pricewise. If MT isn’t actually costing them anything and they’re in a price war, it becomes a tempting target for cuts. Yes, we could profit off its use but the market really won’t let us. That is the reality.
Machine Translation is a net benefit for the Market Research business
- You get faster survey returns at a lower overall cost.
- Labor intensive comment translation is much faster.
- As AI and machine learning acquires more data on intent, the day will come (sooner rather than later) when it can actually interpret basic meaning without requiring human input, bringing costs down even more.
Your translation partner should be able to clearly explain how they use machine translation and why. The use of MT by translation professionals helps the algorithms learn faster and provide better results. And we think this is a net gain for all of us.
Visit our booth at The Market Research Event (TMRE) in November. We’ll have translation project managers there (not salespeople) who work in MR translation every day. Stop by and ask them anything!