Over the past decade Machine Translation has evolved at an enormous pace and the introduction of (Deep) Neural Machine Translation as well as context aware systems present us with a wide choice of suppliers and options for almost any major language. The “right” engine applied to a specific domain and language pair typically provides such good results that humans often find it impossible separating machine translation from human translation. Moreover, localization productivity is between 30-60% higher compared to a process without machine translation. Due to these developments, our approach to Machine Translation in localization (which has not changed substantially the past two decades) requires a rethink.
Given the breadth of suppliers and quality of services available in the market, the first and primary challenge for anyone aiming to apply machine translation in localization projects effectively is not to use just any engine or supplier, but the best suited one for the project. The post-editing efforts between different engines can vary substantially; leading to differences in efforts ranging from 100-300% for the same language pair.
Machine Translation Gateway
Finding the best fit is of essential importance – not a simple feat given the wide variety of suppliers and good quality engines in the market these days. That is where tools such as the CrossLang Machine Translation Gateway with its access to all major suppliers comes into play. Not only does such a tool offer access to thousands of engines in the market through a single interface, but it also offers a simple way to filter down your choice.
One of the most important features of the Gateway, however, is the possibility to generate an advisory of the best engine for your project – helping you to find the best suited engine in a haystack of hundreds of engines and quantifying the results in clear numbers.
Rising quality level of Machine Translation
Industry leaders such as ESTEAM are even starting to question the technology stack and processes as we know them, suggesting that quality machine translation systems can replace translation management systems (TMS) altogether and provide a range of additional benefits by “understanding” the structure of your content.
While such propositions may sound outlandish to many industry veterans, there is a clear logic and merit to the approach – if the engine domain is a very close fit to your source or if the engine is trained specifically for it. What their successful implementation in a wide range of production environments does prove, is that machine translation quality has reached levels that in many domains and applications are achieving human parity.
Bottom line: it is time to rethink our approach to machine translation. While the machine translation first approach may be a bridge too far for most customers, it goes to prove that if you are not using machine translation effectively today you are likely overspending and keeping an eye on new approaches is the wise thing to do.
