A translator’s personal experience [of CAT and MT]
My name is Henrietta. I am a translator specialising in legal and technical texts, working from French into English. My natural inclination is to work simply with word documents but translation job offers increasingly require the translator to use CAT tools, and invariably, in my case, SDL Trados Studio. CAT tools are an expensive outlay and are all the more off-putting given that their use means the translator receives less pay per project than they would receive without using a translation memory. Having said that, I invested in Trados Studio 2017 when an agency offered to contribute 50% of the cost and bulk bought packages for their translators at a discounted price. If you haven’t already bought the software, it may be worth asking a favourite agency whether they would consider this option.
What do you think are the advantages and disadvantages about CAT tools?
Now that I have experience of using a CAT tool, I definitely find it useful when translating long documents with repetitions that produce segments matches, which are frequent in legal and technical documents. I always find that where a segment starts and ends is logical, so the translation units are coherent.
When I began using Studio, I thought that the matches would only occur at segment level. A text may contain a recurring term or string of words, but a translation would not be generated for these, unless they constituted a whole segment, and did not merely form part of a segment. I then discovered my favourite features of Studio 2017 – the ability to run a termbase alongside your translation and use the Term Recognition function, which recognises terms even when part of a longer string of words.
• Using a termbase enables you to add and edit terms, synonyms or word strings as you go along. These functions are great for avoiding repetition, where elegant variation is desirable, or, on the contrary, ensuring repetition, where consistency or emphasis are key. The termbase can be based on an existing glossary that you convert to a termbase, or you can start from scratch. You can also indicate the field to which a term belongs, so that if a term has a variety of meanings, you can select the right translation for the particular context.
• The only limitation is that once the termbase has been set up, you can only add terms to it as you translate. It would be tremendously useful to be able to add terms independently of the translation and build up a large termbase that can be used again and again for different projects. I like to be able to use one termbase and keep adding to that as I translate.
The value of CAT is also apparent in the review stage: As it requires you to confirm each segment, it ensures that you miss nothing out, and encourages you to re-read each segment as you go along. It also helps avoid punctuation and spelling errors as these are immediately redlined. Then, because the target segment has been checked for accuracy, when a source segment is repeated, the translation generated will be consistent and accurate. You can also preview the file in the format in which it was originally produced and review it in the same way you would review a word document. Reviewing the target file in its original format is essential. I have often made amendments at this stage even when I was happy with the target text segments in the bilingual file. The end-user will read the text in its original format and the translator needs to view it that way too, otherwise there is a risk that tone and flow will be compromised.
As I said, I prefer working with word documents. There’s no need to set anything up, the formatting and layout are evidently in place, you can overwrite text and retain what you want to keep, such as names, addresses, figures and logos. However if you work from a pdf document, there is definite scope for error in transcribing such information and here Studio can be very helpful. Pdf files can be converted and translated in Studio, whether they are editable or scannable.
Machine Translation or MT?
Another tool in the translator’s kit box is machine translation, freely available to all. While useful to translators, MT is designed, ultimately, to remove the translator completely, as it does not inherently involve the translator’s input. In 2016, Google moved on from MT and devised “Google Neural Machine Translation”, which it uses in its Google Translate service. While MT translates fragments of text and, using programmed codes, rebuilds phrases, the results are often flawed. Google Neural, however, develops neural pathways, much like a human brain, to “learn” translations, from enormous quantities of data, and provide its own solutions without these having been programmed. It can translate whole sentences at a time and the results are more natural. It is even capable of translating between language pairs that it have never seen in that combination: based on its “learning” from translations from Language X → English and English → Language Y, it can produce a translation from X → Y. At the moment it is available for a range of languages to and from English, is improving all the time and is constantly learning context, collocation, syntax and expressions, and can select the appropriate translation of a polysemous term: Google “bâtiments tertiaires“ and you get “commercial buildings”. The word order is correct, and the selection of “commercial” from a range of possible meanings of “tertiaire” is appropriate in the context of buildings.
Do you think that MT is a reliable tool?
However, while MT is helpful, it is currently unreliable. Common expressions are sometimes translated well: “coup de foudre” is rendered as “love at first sight”, while “coup de coeur” results in “heart stroke”, an expression which, if retained as a translation in advertising material, could seriously hinder your marketing efforts! The internet is awash with examples of funny MT mistranslations. MT cannot produce high-quality translations without humans to correct and improve them. I use MT for single terms or short word strings, but often go on to check the result in context. For texts of greater length, I don’t even attempt to use it, although friends who are not translators appreciate it for the gist translations it can provide. The use of MT has led to an increase in demand for Post Editing Machine Translation jobs, but I steer clear of these if possible. The less said about the pay the better as you have to roughly translate the text anyway and the quality of the translation sometimes involves having to translate from scratch where the translation is unintelligible. I am happiest when in the act of translating rather than editing a piece of work produced by someone/something else. However, sometimes needs must and it is always good to add a string to your bow, particularly if the translation industry is moving towards the increased use of MT and PEMT. When I have taken on PEMT work, it has been from a trusted client, and I have always asked to first see the target text to check its quality as well as the source text to make sure that I could translate it without needing to spend much time, if any, on research.
CAT tools, unlike MT, are designed to assist translators, not supplant them, and have a wonderful role to play in an industry that can grow through the use of such technology and therefore engage more translators, not fewer. I hope that this is not wishful thinking!