Yes and no. Between languages that are quite closely related, such as Anglo-Saxon languages English and Swedish, it does work rather well and definitely helps to understand the basic subject or idea of a text. Between languages that systemically differ a great deal, such as English and Finnish, not so well. What is needed in both of these cases is a whole new line of business called post-editing.
We’ve all used Google Translate to write a short message or answer to a friend abroad in their own language, which has made them very happy and us appear more sophisticated than we actually are. GT is a great tool for that, and for getting an idea of what is the point on a certain web page or in a product description. However, by machine translation I mainly refer to other more advanced tools, such as Sunda, that are being developed by geeks and academics around the world, also in Finland, in both private companies and universities. And our small language Finnish (about 5 million speakers) is a mouthful to handle to say the least.
These more advanced machine translation systems learn during the translation process so that translator can train them to produce the kind of language that is needed for their specific field, and over time, results turn good. In this case, machine translation is estimated to save up to 30% in working time. Translation machine works together with a translation memory program (e.g. memoQ, Studio), and TMs are kept “clean” by post-editing. Within the translation industry, post-editing is now somewhat of a buzz word and something that agencies and professionals are getting interested in as a new line of work and business. Finnish Kites Cluster Project is working on the subject, arranging seminars and workshops, in which my knowledge is based. Thank you KITES and the presenters from the translation departments of Finnish universities for all the background info for this blog.
According to a survey of 75 translation agencies globally (TAUS 2010), about half of translation agencies offer post-editing services. Post-editing is estimated to amount to between 5% and 40% of all translation production (van der Meer, J. and Ruopp, A. 2014). According to a survey of 438 translators in Europe, 30% used machine translation, and out of them, 70% performed post-editing always or sometimes (Gaspari, F., Almaghout, H., and Doherty, S. 2015). According to a survey of 238 Finnish translators, knowledge of machine translation was not considered particularly important, and translators were most familiar with Google Translate (Mikhailov, M. 2015).
The translation departments of Finnish universities in Helsinki, Turku and Tampere consider that there is a market for machine translation and post-editing, and they are arranging training for students on these. Dr. Dorothy Kenny of Dublin City University has studied the translator’s role in statistical machine translation and says that the time is ripe for translator educators to engage with SMT in more profound ways than they have done to date. Indeed translators need information about this new technology, what benefits it can bring to our profession, and how to measure the benefits.
Currently machine translation can make translating simple, repetitive texts like instructions faster and more economical, providing that the machine has been trained to manage domain-specific content. Success also depends on the type of language, grammatical structure and pre-editing of the source text. However, creative texts like marketing materials, journalistic articles, and literature are a whole different ballgame that machines cannot handle and perhaps never will.
This post was posted for the first time 12 September 2015. An update from #kites2019 symposium: the latest machine translation development involves self-attentional neural networks.