The TAUS (Translation Automation User Society) Global Content Summit was once more held in Dublin chez Microsoft. Well protected from the first storm runners, I was able to spend a day exchanging inspiring thoughts with language process and technology experts from LSPs, technology providers and industry giants.
Language Service Provider? No!
In the first panel we heard the agencies’ side on the development of the industry. ‘Language service providers’ or ‘translation agencies’, according to Gráinne Maycock of Amplexor and Mark Rice of thebigword, these are obsolete company categories. Instead they call themselves Content Process Consultants or Global Content Providers. Is it all just a question of words? Not quite: It has been recognized that pure translation performance’ is no longer enough and that many working areas can be relieved by using machine learning applications. Therefore, more technical expertise (i.e AI + MT) and process knowledge (such as content consulting, process analysis) is required. Conclusion for translators: language experts who develop in these areas of expertise are on the right track!
Not without light post-editing
Then came the assessment of the customer side: Richard Turncliffe from Google, Paul Leahy from Oracle and our gracious host Vincent Gadani from Microsoft exchanged views on current developments in the use of language technologies. Interesting confession: Google does not publish raw machine translation (MT) internally, but exclusively PEMT (Post Edited Machine Translation), using ‘light post-editing’.
A challenge for all three industry giants: Assessing the quality of MT. Microsoft has even developed its own module and workflow for this purpose, in which the quality of MT is assessed with artificial intelligence and, depending on the result, either published ‘raw’ or routed towards post-editing.
Exciting: All companies have large linguistic teams to fall back on. They continue to see a need here because such processes could not be controlled and improved without linguistic expertise!
According to Oracle, the correct use of terminology is the key element for good MT quality. And some language gaps need to be filled, but above all the data is missing.
Learning: Rowing against a wild stream of information
Learning is more important today than ever – and in the language technology sector, it is still necessary to bridge large gaps in knowledge. Although technical content plays an important role, attitudes towards learning and the changing environment are even more relevant: In the past, people stockpiled their knowledge, and most of the information they needed was retrieved directly from their head, but today people learn ‘on-demand’ and how and, above all, where they can quickly find the information they need. Companies can support these processes if they not only give their employees the freedom to learn but also enable them to put what they have learned directly into practice. Only then can knowledge be used profitably in the company. We support our customers by providing relevant methods and content for digital qualification.
Marcin Kotwicki from the European Commission is, among other things, concerned with the question of how much one can ‘know’ about one’s own data and the translators and post-editors, whether one should make this knowledge available in the data stream. And if, in the end, it leads to producing better quality by knowing more about the origin and context of your own data.
Maciej Krupinksi of GET IT Language Solutions believes that the current training of language specialists does not reflect reality. He advocates specializing in translator training and integrating more technical knowledge and expertise. The challenge here is to prepare for jobs we don’t even know yet.
No AI without data
Dace Dzeguze and Jaap van der Meer from TAUS presented cool new services and offers. First things first: TAUS Services can now be used much more flexible, e.g. you can now subscribe to DQF and the Dashboard without being a TAUS member.
The ‘Human Language Project’‘ of TAUS is very exciting, in which TAUS experts produce and offer data for rare languages on the basis of their own data platform.
TAUS Matching Data: Using sample data of the customer, suitable data is automatically compiled at the segment level. The data is cleansed and provided anonymously, according to the customer’s wishes.
News from the tech corner
In the tech slot, Natasha Latysheva from Welocalize showed us the magic trick of synthetic data generation for MT training when missing language data. And Diane O’Reilly from Iconic Translation had a sophisticated MT workflow for the pharmaceutical industry. It’s exciting what you can do with language technology!
Business as usual?
Andrew Campbell from Lingo24 shared his business model with us, which relies heavily on technology but still considers the linguist to be a central figure. And Margaret Ann Dowling brainstormed with us about revolutionizing the magazine world – with content for all in all languages. All it needs is a technical platform – and an understanding of the publishing world for ‘technical’ language processes.
These are just a few impressions from a very colorful day. Not to mention the exciting coffee conversations 🙂
Now I am very much looking forward to the TAUS Asia Conference in Singapur vom 16.-18.10, where I will meet some of the participants from Dublin again and will certainly make new acquaintances. And I am particularly looking forward to receiving information about the language business in Asia. I will report on my findings and experiences in Singapore in my next blog in two weeks’ time.
So, stay tuned