Smartling now offers predictive score on translation quality
The New York City-based company says this is the first forward-looking quality score.
Marketing comes in all kinds of languages, for all kinds of local markets. But, until computer-based systems came along, translation for multiple markets was a very labor-intensive and logistically complex operation.
Now, translation tech and service provider Smartling has released another computerized tool — a new predictive quality score that it says can help measure a translation’s success in advance, by providing brands with a future-looking forecast of how good the translation is for a given market and use.
Head of Marketing Juliana Pereira told me her company previously “only looked back” at how well a given translation had performed. She added that the new Quality Confidence Score (QCS) — which has been in a beta release with selected clients and is now available in general release — is the first predictive quality score in the industry.
The QCS utilizes a proprietary machine-learning algorithm that forecasts the chances a human in the target market will consider the translation in question to be of high quality. An 82 percent confidence score, for instance, is a good one; it means there is an 82 percent chance the translation is a quality one.
The score is based on dozens of factors from the past seven years of data for similar kinds of translations, in similar contexts and with similar factors. These can include whether it was machine or human translation, who the human translator was, and whether Smartling’s style guide or visual context tool was used. The visual context tool takes into account, for instance, whether the word for “home” should be interpreted as “house” or “homepage.”
If the score is high, the client may want to publish the translation directly without review. If it’s low, an editor’s review may be recommended. As QCS is just launched, Pereira said, there is no data yet on how accurate the predictions are.
Opinions expressed in this article are those of the guest author and not necessarily MarTech Today. Staff authors are listed here.