CTS Professionalisation Talk 2021/22 #4
On Friday 22 October, Nick Lambson gave the current cohort of translation and interpreting students a professionalisation talk on machine translation (MT) services in the language service industry.
Nick Lambson is a senior localisation instructor at Beijing Language and Culture University and a localisation engineer for MediaLocate.
The highlight of the talk was the application of machine translation to the business world. Lambson pointed out that a big problem in the translation industry is that content is growing dramatically while traditional translation can hardly meet the industrial requirements. However, MT and post-editing can be a remedy for this, as it strikes a balance between efficiency and quality. To illustrate the reliability of MT, Lambson gave us an insight into the development of MT technologies. He demonstrated the advantages of the popular neural MT over the traditional statistical MT and explained the training process of a neural model vividly, using the game Mastermind as an analogy.
Lambson also introduced two kinds of evaluation metrics for MT: human evaluation and automatic evaluation such as BLUE, Edit Distance and COMET. He warned that, although automatic evaluation is economical and convenient, its results can be difficult to interpret without the guidance of human evaluation. In particular, he noted the discrepancy between BLUE scores and human assessments, and recommended alternatives to BLUE.
Abundant graphs and statistics were used to prove that neural MT outperforms statistical MT in various aspects. That said, Lambson also reminded translators to be alert to neural MT’s tendency to generate misleading ‘hallucinations’ which can easily escape the post-editors’ notice.
Based on statistics, Lambson showed us that not only can customised MT boost our productivity, but that the post-edited translation can also be better quality than purely human translation.
Finally, Lambson discussed whether human translators will be replaced by MT, which was encouraging to students who plan to step into the translation industry. After analyzing comments from professionals and customers, he came to the conclusion that human translators will not be phased out, but they should acquire new skills and adapt to new roles.
At the end of the talk, a Q&A session took place, where Lambson answered the following questions from our host, Faruk Mardan, and the students:
How is data collected for Edit Distance?
The distance can be calculated either by CAT tools or a Python script.
Should we use MT when it is not discouraged by commissioners?
It’s fine to use MT as long as you are sure it can improve the quality of your translation because the aim is to generate the best translation.
What skills should translators develop to adapt to the new industrial environment with MT as a major player?
They should be familiar with CAT tools and integrate MT into the CAT tools. To stand out, they need to be experts in fields that they are passionate about.
For those interested in localisation, Lambson also has an online course called Python for Localisation where you can learn basic programming for localisation work.
Author: Ran Ji.