Best Practices on Using Machine Translation
Machine Translation has been on a continuous development cycle for over a decade. When Google Translate launched 10 years ago, it dramatically increased public awareness of Machine Translation (MT). Global users marveled at how snippets of text could be translated into a range of languages with the simple click of a button.
Other major search engines, such as Yahoo, Microsoft, Baidu, and Yandex, responded by quickly releasing their own MT services. Throughout the last decade, TransPerfect has been driving and influencing the innovation for MT for the best optimization for our customers benefit based on industry focus. The MT evolution started as a rule-based process tied closely to the target language, then shifted to more organizations focused primarily based on statistical models, using previously translated texts (corpora) to translate the new sentences, and finally using a hybrid model to drive efficiency. Now it appears that we are on to the next phase. Recently the industry has published research suggesting that Neural Machine Translation and predictive analytics will be enhancing product performance.
While it’s exciting to see these technological advancements in the MT space, it’s important for global businesses (who are understandably excited at the prospect of high-quality, instant translation) to fully understand the implications of this progress.
First off, even the most accurate machine translation technology can’t automatically account for branded content, terminology preferences, formatting considerations, subject-matter, cultural issues, and other important factors. Professional translation services (which typically involve a multi-step process of preflight, review, and quality control) are better equipped to account for all of these factors in a company and audience specific-fashion. Even in this moment of incredible MT innovation, companies who want professional-quality results need to partner with a language services provider.
The good news is, most language services providers offer a wide variety of options that can be customized to the needs of each client and project, ranging from a human translation and review process supported by Translation Memory (TM) technology to a mixed process involving various grades of machine translation combined with human proofing and QA processes.
TransPerfect is investing more than ever in technology to increase productivity, reduce turnaround times, and cut costs—including Neural Machine Translation technology. Similar to the software industry that provides several Service Level Agreements (SLAs), we offer multiple Translation Quality Levels to cater to specific projects, clients, and industries.
While human translation has grown increasingly cost effective with the advent of TM technology and globalization management system (GMS) solutions, there are some situations where an expertly tuned MT approach better suits the client’s needs. Our MT solutions can include various degrees of software training inputs combined with light or heavy post-editing to refine final proofs.
Also of note, TransPerfect’s proprietary production technology utilizes a hybrid process; a statistical core with layers of linguistic processing both at the source and at the target level, which offers a higher-quality result than many competing technologies out there. Over the years, our skilled engineers have created cutting-edge solutions to further enhance these services, such as client-trained engines that provide on-the-fly machine translation for internal documents and text, so that employees don’t translate confidential information through public engines like Google Translate. These solutions are essentially bulletproof and can be installed within clients’ data centers or in a private cloud if required.
What’s more, TransPerfect is playing an active role in MT and Natural Language Processing Technology by developing systems based on deep learning in two scenarios: the automatic post-processing of linguist rules and the deployment of advanced MT technology for languages where the gain with respect to Statistical Machine Translation are bigger, such as Chinese or Japanese. Furthermore, we are integrating predictive modeling and artificial intelligence into our MT solutions so that we help translators be even more efficient, improving ROI for our clients.
The key takeaway is that while Google is doing incredible work advancing MT technology, a one-size-fits-all approach does not exist. While MT is one way to reduce costs, it’s not the only way, nor is it always the best way. Companies who wish to get the most bang for their buck (while also ensuring the best quality result) have a whole host of options at their fingertips. The important thing is to find a partner who gives you access to a full range of solutions and then effectively guides you to the one that best suits your needs.
TransPerfect will continue to play an integral role in pushing innovation within this technology advancement. We look forward to exploring the value that MT may bring your company. Contact us for more details.
