Overcoming the barriers to MT adoption

After recently watching a TED talk on hackers given in June 2014 by cyber security expert Keren Elazari, I realized there were some interesting similarities between our perception of hackers and our perception of machine translation (MT). Elazari refers to hackers as the internet’s immune system and makes reference to “a love-hate relationship” between hackers and large organizations, particularly governments, since they use and need hackers but at the same time often persecute them.

The perception of MT could in some way resemble that of hackers; many in the industry fiercely object to using MT for translation, yet they may use and need it on a daily basis. Throughout the course of a day a single user browsing websites and social media will be exposed to a variety of different languages. As an example, nearly every day, posts in languages I don’t speak pop up in my social media newsfeeds — the only practical way to understand those languages is to use MT.

Over the last couple of years in organizations of all shapes and sizes we have seen the adoption rate of MT soar, and in particular MT is being used by most enterprises and language service providers (LSPs) in at least some capacity. We have heard the arguments for and against adopting MT time and time again, but regardless of our thoughts about using it, this increase in MT usage is happening for a reason.

The motivation to use MT is not coming from any one single factor, and instead its increased adoption is being driven by a matrix of contributing factors that begin with content production. In the era of “big data” and “intelligent content,” success for any business depends on capturing market share and building strong customer relationships, regardless of industry. The demand for multilingual content is also fueled by customers’ expectations for readily available information in their native languages; an increase in the use of online social communities with people who speak different languages; and the competition within many industries that creates the need to provide more value to customers and stakeholders. Considering that only around 6% of the world’s population use English as a primary language, and that over the next few years 70% of global economic growth will come from emerging markets, it is no surprise that we need MT to bridge the gap between multilingual content supply and demand. Like hackers, MT becomes the necessary evil that we need for global communication — and this is not a bad thing.

MT has already proven its worth; it offers businesses and organizations more efficient ways in which to communicate with their customers and stakeholders. It removes language barriers enabling people throughout the world to communicate with each other, in real-time and regardless of their language ability, and it brings customers closer to brands by providing more support in a greater number of languages. In light of all these benefits of using MT, the question remains: if MT is so important in this day and age of real-time multilingual content production, why are there still barriers preventing it from reaching its full potential, particularly for LSPs?

Where larger organizations can dedicate whole departments to focus on deploying MT, smaller language vendors often see it as an add-on job for their localization engineers, linguists, project managers and sales and marketing teams. As an “add-on” job, it never fully receives the attention required to be successfully integrated into the translation workflow. This often leads to confusion, uncertainty and a lack of motivation, all of which negatively affect translation quality and the rate of MT adoption.

Attitudes toward using MT have undoubtedly improved over the last few years as LSPs and their clients become more comfortable with the idea of using it. However, MT has not yet reached the familiarity level of translation memory (TM). Companies that are serious about implementing MT as a means to enhance productivity and attract new clients need to be able to communicate, learn and innovate. A working environment that nurtures and promotes these three principles will be in a better position to see a complete immersion of MT across all departments within the translation supply chain.


Communicate, learn, innovate

Communication is the critical success factor when looking to implement or change any process in a supply chain and MT is no exception. If there is an open and transparent flow of communication between everyone involved in the MT process — project managers, engineers, linguists, post-editors and the marketing and sales teams — it becomes possible to easily spot any unforeseen bumps in the road and deal with them accordingly.

Continuous learning is often just as important as clear communication. Technology is constantly changing, and when dealing with MT, you are continuously looking for ways to improve the quality and mature the engine. Engineers and linguists tasked with building and maintaining MT engines need to be willing to learn and keep abreast of new developments in the technology.

In the same way people brainstorm to come up with new ideas, innovation will naturally be borne from combining clear communication and continuous learning. The more people are trained and up-to-date with the technology behind MT, the more opportunities for innovative ideas to improve quality, time to market or even to combat inefficiencies across the supply chain.

Each department has a unique relationship with MT, and by identifying all the relevant teams or departments that touch upon the translation supply chain both directly and indirectly, it becomes easier to plan clear communication and education strategies.

Marketing: Often overlooked for MT training, marketing teams are usually the first to engage with potential MT clients. They are tasked with creating promotional materials or sales tools that will attract and interest MT buyers. This often includes brochures, articles, case-studies, blog posts, slide-decks and infographics, to name but a few. Marketing personnel might also be responsible for developing and presenting case studies or discussions about MT at language-related conferences. While these activities are well within most marketing professionals’ skillsets, the development of such content becomes more difficult with technical subject matter such as MT, and marketing teams need to be well educated on MT subject matter before diving in and creating content to engage with potential clients. They have to know the lower end and upper end of potential cost savings, the timeframe, the process of developing an MT system, the different types of MT providers out there and of course the many different use cases.

In larger organizations with in-house engineers and linguistic teams, expertise can be easily transferred through MT training sessions and workshops, which are organized to educate the sales and marketing teams. In smaller organizations, it might be more cost effective to designate someone to attend an MT training course and then have them share what they learned with the rest of the team. In any case, it is useful to build an MT resource library that can be accessed and used by anyone in the organization.

Sales: Sales teams and account managers play a fundamental role in building and maintaining relationships with clients, and so when they are introducing a service or tool such as MT, it is vital that they are confident in explaining not only the benefits but also the steps involved in developing the service. Sales personnel should be up-to-date with some of the more technical aspects of MT so they can answer questions without having to ask a more technical person — this shows the client that the company is serious about MT and will speed up the communication process.

Managing expectations is particularly important when dealing with new clients of MT. Clients need to be made aware that a one-size-fits-all approach does not work, and that a variety of factors must be considered when measuring MT quality. They also need to be educated about how to decide on an acceptable quality level for a particular job. This step is usually carried out during a pilot. If clients’ expectations are met and the project is successful, it’s likely that they will become repeat MT customers, and this will drive better-quality future MT output because their engine can be matured over time using the post-edited output.

Sales teams well educated in MT that can use clear and transparent communication are in a much better position to spot new trends or potential MT opportunities and be able to explain to current and future clients how the translation process would work.

Engineers: Whether using an online platform or installing a local version of Moses, engineers will need some training to successfully develop and maintain an MT system. Installing a local version of Moses can be time-consuming and difficult to maintain, whereas commercially available cloud-based platforms have automated many of the less appealing tasks. Getting the appropriate training speeds up the MT development process and enables companies to take on more MT projects. 

Linguists: Linguists are instrumental in building and optimizing MT systems as they work closely with engineers to develop the systems. Their knowledge of language models puts them in a unique position to help improve the quality of MT systems and mature the engines over time. It can often take some time to mature a system, meaning it has reached its maximum level of quality. Language vendors should be putting processes in place to reach this maximum level faster and more efficiently, so it runs like clockwork each time a new project comes in. Linguists play a vital role in this. They help with pre-editing the MT engine’s training data, recognizing missing words from the system, post-editing MT output and retraining the system using post-edited content, as well as helping with the system’s general language maintenance. All of which will affect the final system’s capability and development speed. The better the communication between linguists, post-editors and engineers, the faster systems can be matured. As they become more familiar with how the systems work, they will be in a better position to identify innovative opportunities to improve MT quality.

Post-editors: Post-editing MT to some degree is standard practice and should always be done with a comprehensive understanding of what the client is expecting. Since post-editors need to be able to confidently judge how much editing is required to meet those expectations, open communication is needed with the sales teams to manage client expectations, and also with the engineering teams that will apply any changes or automatic post-editing settings to avoid unnecessary repetitive errors. The fact that certified post-editor training courses and best practice guides are now becoming more widely available is further evidence that, while still a controversial topic in some circles, MT adoption is most definitely increasing.

Similar to the hacker culture, it will continue to take time to fully adapt to and embrace MT, and like hackers being considered the internet’s immune system, MT could be considered the immune system for the language services industry, as it continues to energize the demand for translation services.