A Look at Machine Translation
At Kent State, we’re always looking to utilize our blog to highlight key areas of the translation field for our students and prospective students. With this in mind, we recently interviewed Erik Angelone, Associate Professor of Translation Studies, to get his insights on machine translation, how the field has changed over the years, how Kent State incorporates it into courses, and more.
1. Can you tell us a little about machine translation, or MT, and how it has developed over the years?
Machine translation, in essence, automates translation based on things like the application of language rules (rule-based MT), statistical probabilities (statistical-based MT), or, more recently, neural networks based on complex deep learning (neural MT).
In the language industry, some love it, some hate it, some fear it, and some embrace it. We can probably all think of really lousy machine translation examples. That being said, in the right hands, it can enhance the translator’s productivity and consistency. Translators can use it to produce more output in a shorter period of time. Instead of translating text “from scratch” or in conjunction with their translation memories, they can leverage MT output and approach the task as one that involves post-editing.
Since the 1960s, a persistent fear among translators is that they will one day be fully replaced by machines. This hasn’t happened over the years, and who knows if it will ever happen. For the time being, translators should embrace machine translation as a tool to facilitate their performance.
2. What are some notable differences about machine translation today compared to when it was in its beginning stages?
It’s now ubiquitous, web-based, and free, meaning it’s in the hands of the masses, for better or worse. Also, users can pick and choose among multiple MT systems. Google Translate might work well for some tasks, but not others. DeepL, Systran, and a number of other MT tools can be cross-checked to see which one performs best under certain circumstances.
We are currently in a neural machine translation (NMT) era, in which machine translation is based largely on big data, massive amounts of context, and deep machine learning of patterns. This has improved the quality of MT tremendously for some language pairs and some genres. For example, instruction manuals, with their controlled language and syntactic templates. Machine translation engines can be trained to handle this type of content relatively well.
As mentioned in the previous question, machine translation has brought about a sea change in the language industry in that translators are now often called upon not only to translate content, but also to post-edit machine translation. We are also seeing a push in the direction of authoring source content in a fashion that is machine translation “friendly.” New and exciting career paths are emerging.
3. How often do professional translators utilize machine translation for translating speech? In which contexts and instances is it most beneficial?
I think the verdict is still very much out on this. Again, at least for certain language pairs, it’s useful for content that is syntactically “light” or predictable, and where terminology and vocabulary are controlled. Polysemous words and complex syntax are still roadblocks.
I also think MT becomes beneficial in situations where humans aren’t available to take on the translation work that’s needed. This is often the case for languages of limited diffusion or in crisis scenarios where time is of the absolute essence and lives are at stake.
4. What should students know about machine translation as it relates to project workflow?
The place of machine translation in project workflows is not one-size-fits-all and the level of its application will vary from one project to the next. Literary translation is a domain where MT might not make much sense, although we are starting to see interesting research to suggest otherwise.
Some projects might call for raw, unedited MT. This is where clients might benefit from the gist of the content in a more granular sense than what some of the popular free, web-based applications can provide. When post-editing, translators might be working with already-established strings of text or entire texts in the target language as a point of departure. In other situations, they might be making use of adaptive MT proposals when generating translations using translation memories.
As previously mentioned, translation and localization considerations might be taken up during an earlier phase of project planning, where the anticipated use of MT might have an impact on the authoring of content and the establishment of work breakdown structures.
5. How does Kent State approach machine translation in its courses for students?
We weave aspects of machine translation into all of the courses in our curriculum.
For example, students learn how to use it in our Terminology and Computer Applications course and often apply it in our specialized translation practice courses. In our Project Management course, students learn about its various positions in project workflows. In our Editing for Professional Translation course, students post-edit MT output. In our Theory of Translation course, students approach MT as it applies to various approaches to translation as well as from an ethical standpoint.
By the time they graduate, students know what MT can and cannot do, when it helps and when it hinders.
6. Are there any other aspects of machine translation you’d like to touch on?
Machine translation is basically here to stay, and it’s rapidly improving. Here at Kent State, we train our students to successfully use it as future language industry professionals, and to be aware of contexts where it works and where it doesn’t. It’s certainly changing the language industry, and we feel, generally, for the better.
Contact Kent State
If you’re interested in translating text through machine translation and want to earn your M.A. in Translation to advance your career, reach out to our team today!