Machine translation is an evolving technology. We looked at the history of it last week: how a lack of technology stopped machine translation from delivering on what was promised, and how that ‘failure’ left the field as a taboo subject within the translation services. The stigma that followed its failure to deliver Fully Automatic High Quality Translation is not the only reason that machine translation is considered sub-par by some. Automation, in any industry, is terrifying.
Customer Partnerships
A computer can translate what is there, but how do you factor in intention? As any translator will tell you, language is much more than simply the written word. Understanding the goal of the translation, what needs to be achieved and who it is aimed at, is far more important than a direct translation. Language service providers will ensure that a project team is at hand to understand the customers’ goals. Craig Lovatt, Director of Business Development at Global Voices, comments when asked about effective translations, “We believe in forming partnerships with clients in order to understand their translation needs and to create a knowledge base of stylistic preferences.” However, with the growing demand for translation, the industry must now find a way of achieving this level of customer engagement while implementing effective machine translation technology.
Communication is changing, with every person now connected in a way the original machine translation pioneers could only dream of. This means that the language barriers have become even more of an impediment to human development. Our need for translation has exploded with the advent of the internet, with Google reporting in 2015 that its services translated over 100 billion words per day. There is simply too much for just human translators to keep up with. Machine translation seems to be the answer.
Machine Translations – the next steps
However, machine translation as we know it, is not sufficient on its own to address this growing need. Two significant changes within the language service provider industry will need to materialise to reap the full benefits of machine translation, for both the customers as well as the (linguist) providers. The first is improving the quality of the machine translation, while the second is adapting the industry to better accommodate and utilise the new tool.
Developing the technology may appear obvious but the idea that machines can imitate intention still seems to be far off – at least with our current system of Statistical Phrase-Based Machine Translations. PBSMT is the most common form of machine translation (including Google Translate up to 2016) and is ingenious in its own right. It employs a series of complicated algorithms and statistical equations to break up languages into key components and cross references them with a database of words to predict the most appropriate meaning. This is a huge leap from the Single Word Based approach that was employed before and is the reason for the sudden revitalisation of machine translation research. However, its shortcomings still require human intervention. In order to achieve Fully Automatic High Quality Translation, we need to develop a system that learns. We need artificial intelligence through applications like Neural Machine Translation.
Neural Machine Translation
The power of Neural Machine Translation is its ability to adapt to a specific situation or environment. A translator can help develop the NMT by correcting any initial mistakes, but the longer the NMT is used, the more accurate each subsequent translation will be. This hones its ability to accurately deduce the customers intention along with translating the language. It was long suspected by the machine translation pioneers that the only way for perfect machine translation was with the addition of a learning computer. This is why Artificial Intelligence research was founded alongside machine translation research: the two complement each other.
In an experiment comparing the effectiveness of NMT against PBSMT in translating 12 literary novels, it was shown that NMT was superior in every regard, with its translation rated as being as effective as a professional’s around one-third of the time.
Evolving role of the Linguist
So our technology is catching up to our translation needs, but we must also address the role of the translator in using this powerful tool to its full potential.
The translator has complete autonomy when it comes to the translations. They, in conjunction with the customer, deduce the appropriate meaning and are solely responsible for its quality and effectiveness. With a world wide implementation of NMT, there is a fear that human translators will be seen as obsolete, and eventually forced into extinction. Craig Lovatt believes the industry needs to adapt: “There is a big opportunity for the linguists to increase the quality of the output. In working together with the project team at Global Voices, the linguists are providing a very valuable element to the process for us as we aim to improve the quality of the translations but also the speed to market for an increasing amount of projects.”
Neural Machine Translation needs to be directed if it is to be used effectively and must be ‘taught’ what the correct translation are. This is not possible without a professional translator at the helm. The role of the translator will evolve to accommodate this change in the industry’s landscape, and move from a ‘worker’ role to a ‘director’ role. This will allow the linguist to focus less on ‘translating’ the texts, and more on adding value to the translation. The linguist will take enhanced ownership of ensuring the combined effort of machine and man, meets the client’s requirements. In the same way that master painters would employ students to help paint their masterpieces, so too must translators direct NMT to accomplish their goals.
It will still be their translation, their art. The difference is that the tools for completing it, are becoming more advanced.