24 December 2019
LINGUISTIC INSIGHT: THE PROBLEM OF TRANSLATION QUALITY IS NOT A PROBLEM OF TECHNOLOGY
With the disruption of AI and interest in the language industry growing among private equity and venture capitalists, we offer this Linguistic Insight in order to put the role and potential of technology in translation in a wider context.
The importance of translation quality
In 2016, SDL – one of the world’s largest language service and technology providers – published a study called Translation Technology Insights Research. It surveyed over 2,700 people in the translation industry divided into three categories: LSPs (language service providers), corporates and freelancers. One of its insights was that quality is by far the most important concern yet difficult to achieve: quality is 2.5x more important than speed and 4-6x more important than cost. Yet, almost 80% of translations sourced through an LSP require rework.
The study essentially concluded that the translation quality problem is a ‘terminology’ problem, and so requires a terminology management solution. However, it is interesting in our view that the conclusion drawn was not the simplest and most obvious one, namely, the lack of people educated for the specific needs of professional translation. There is no use giving terminology to a person who does not know how to use it correctly in the given context. That is like saying that all anyone really needs to translate is a dictionary.
Since the time of the SDL study, the market for translation has only grown while developments in machine learning and AI, especially neural machine translation (NMT), have exploded, disrupting the language industry and offering potential new solutions to meet this increasing need.
The growing translation market and the potential of AI
At the SlatorCon San Francisco 2019 Investor Panel, three recent investors into the language industry, representing both venture capital and private equity and different investment approaches, agreed on one thing: the language industry is highly attractive to investors. In the most optimistic scenario for AI, one VC envisaged companies being able to offer translation solutions to customers without employing people who actually have those languages.
The language industry and its market have grown so quickly since the early days of the internet that clearly, education has not kept pace and the industry remains structurally fragmented. While into-English language combinations are among the highest in demand, foreign language learning among native speakers of English is in dramatic decline. This has contributed to the erosion of the integrity of English among its users even as it has become the world’s most dominant language. Moreover, language learning is only the basic foundation for the immensely broad knowledge and lifelong learning required for professional translation. It is understandable that the focus is on technology as the easier and faster solution over education.
Technology solutions
The problem with this approach in translation is that as long as human beings continue to produce their own texts in order to impact other human beings, they will require human translators to interpret and understand those texts in order to achieve that impact. Language is connected not only to things which can be verbalised but to things which cannot be verbalised. Language, to be understood, relies much more than we consciously realise on our understanding of words in their non-linguistic real-world context. Words on their own do not actually contain all the relevant information. One simple example is the difference between the impact on a human being and their imagination of reading a CV and when a machine reads the same CV. The true potential of the candidate will be lost on the machine.
Solving the problem of translation quality
To solve the translation quality problem with a technology solution, texts will have to be written primarily by and for machines and only secondarily by and for people. It would certainly be efficient but at the price of innovation, creativity and freedom of expression among authors of texts. And instead of translators, ‘post-editing’ specialists (as they are now called) will become adept at adjusting templates and machine output but at the price of the development of their own creativity and expertise. Editing is its own process, but fluency in translation, like fluency in language, is the result of many years of practice and exposure, eventually resulting in a high speed, intuitive understanding and a natural, spontaneous production of insightful translation decisions. In the end scenario of texts written by machines for other machines to translate, insight would not be required in any case.
These issues of course become less important the simpler and/or more mechanical a text is in terms of translation. We offer this Linguistic Insight as an additional perspective to anyone interested in the potential of machine translation and AI to offer solutions for translation needs.