Louise, when you started with post-editing (PE), were there any hurdles for you? Tell us about your initial fears and issues – if there were any?
I had only seen Google Translate and had a natural fear of being made redundant – that a machine would be doing my job, and a poor one at that. I was afraid that quality would no longer be important. As we could see the engines learn and adapt and follow the evolution of machine translation (MT), this was replaced by a genuine feeling of "yay – more help for me".
To me, machine translation is just another tool in my toolbox. Most translators already use CAT tools with translation memories and term bases and pre-translate the files to leverage previous translations – this has been taken one step further by having a machine take care of the more simple and repetitive translation work and using algorithms to create a new sort of fuzzy matches. I save so many key strokes by having a lot of words already typed for me. I sometimes compare it to cooking: most of us wouldn't do without our food processors or hand mixers to do the tough work – we can then focus on the more delicate work.
What engine are you using when doing post editing and are some engines better than others?
I normally work with client or area specific engines, meaning that they know the preferred terminology and phrasing from previous translations or common terminology within a field. There are generic engines available that will help you with generic translation; these require a little more work as they do not know what terms or phrasings to prefer over others, but they will still offer basic translations.
What was the most complicated part for you?
Learning when to leave well enough alone. As with all editing, we all have our own preferred phrases and style, and sometimes you need to put your preferential changes in a locked drawer. You need to acknowledge that this is an editing job, not a translation from scratch, and if you keep changing every little thing, you will lose the advantage of having machine assistance.
Was there anything that went more smoothly than you first thought it would?
I hadn't expected the term consistency that MT offers (when using specific engines). And sometimes the volumes and deadlines mean splitting work between several translators, and it can be very good to give them a common baseline to work from – the editor's job of streamlining work from three different translators with different preferences and styles will be a little easier when the framework is already outlined.
Are there different types of PEMT?
There are two flavours of post-editing of machine translation (PEMT): light and full. Light means editing the output from the engine to make sure that the translation is complete (nothing is added and nothing is missing). This is perhaps the most challenging form of PEMT for a translator; here you need to just do a basic edit – spellcheck (although machine translation is usually free of spelling errors, the translator may introduce typos during edit), quality check (numbers, tags) and grammar check (grammar and punctuation does not have to be 100% correct, just to a degree that the meaning is not distorted). For the diligent linguist, it can be difficult to accept that it doesn't have to be perfect. Anything that is not for print or end customer facing could benefit from light post-edit, if deadlines are tight or someone just needs a basic understanding of the content.
I mostly work with full PEMT, which means that one linguist amends the output from the machine and transforms it into a translation of the same quality as a "normal" translation, and another linguist reviews this translation. The result should be as good as anything else you deliver; the difference is that you type less and don't look-up words in the dictionary quite as often. This frees up time to focus on the finer details or the difficult parts.
Would you say it is better for some texts than others?
Whenever you can use segment-based CAT tools for translation, you can benefit from machine translation to varying degrees. Basic manuals for printers, mobile phones, etc. will often be an absolute delight; the outcome really depends on the source text, which also goes for normal translation – if the text has been written in or translated into English by a non-native English speaker before translation into your language, you may be in for a quite challenging ride. If the text is written in a standardised way, using simple, straightforward phrasing, the results of machine translation can be impressive.
Texts with a lot of user interface references (names of buttons, menus, dialogue windows, etc.) that need to be kept bilingual can also be a challenge (although sometimes the results are quite good). Complex text with many sub-clauses can be challenging as well, as they are in normal translation.
As a rule of thumb, you can say that if the translation requires a specialised translator, PEMT should be done with a specialised engine.
Many translators are reluctant to work with post-editing due to the quality of the different MT engines. What would you like to say to them?
I would say that I wish they could see the output from machine translation in CAT tools today, and hope that they will try it out – and try again if the first few jobs didn't meet their expectations. As with everything, there is a learning curve, and this goes for suppliers, translators and engines. We all get better with practice.
How do you see yourself working in 10 years' time?
There is a Danish saying: "Predicting is difficult, especially the future", but I think that in 10 years, all non-literary translation will be machine assisted, and the role of the translator will have morphed into a linguistic advisor or localiser focusing on tweaking or fine-tuning translations and ensuring correctness in context.