There are a number of potential pitfalls involved in working with Translation Memory (TM).
The first, most general, doubt people have about TM is that it seems to go against conventional wisdom about translation; surely, people say, translators should translate the sense or core meaning of a text, rather than individual sentences? This is undoubtedly applicable in the case of, say, a novel, or even marketing collateral. But with more technical documents, consistency is of the essence, and there is much less room for personal interpretation. Although it is true that there is rarely only one correct translation of a document, or even a sentence, in the case that the translation of a technical term or concept has been established as a result of legislation or industry-wide consensus, it is vitally important that this standard is adhered to.
A more serious potential issue is that a translator would start to think in terms of sentences, or parts of sentences, rather than the text as whole. Naturally, it is essential for the translator to have in mind the unity of the document he or she is working on, and they must take care to avoid missing the wood for the trees, as the adage has it. This is a potential problem for all translators, however, not just those who use TM, and part of the art of translation is to think simultaneously at both a micro- and a macro- level; that is to say, about the precise, grammatical or terminological details as well as the broader thrust of a document. In fact, TM makes it easier to do both.
On a related note, some translators dislike CAT software because it can change the nature of the translation process. The reason they give is that, working from scratch, the translator recasts the document in its entirety. With a TM, they say, it is less about establishing an authentic syntax in the target, and more about adapting the TM’s suggestions until they are close enough to authentic to be passable. As with the other possible pitfalls mentioned above, there is undoubtedly a risk of this happening. Given a match of 70% similarity, even a particularly conscientious translator is unlikely to go back to the drawing board and rebuild the sentence from the ground up, not least because they will find it hard to get the fuzzy match the software has already suggested out of their head.
This challenge, like the others, can be overcome with experience and training. If someone has been working with CAT for a significant portion of their professional life they will have developed strategies to deal with it.
Another potentially prohibitive factor to be considered is the cost of a TM, both in terms of the initial outlay on the software, which tends to be expensive, and the training it necessitates. Although open-source translation software is available, it has so far failed to gain a foothold in the market and most Language Service Providers (LSP) pay for a fully-licensed package. And, although TM use has become a component in most translation courses, money and time still have to be invested in training new translators to work with it. The differences between the various TMs on the market also mean that, even if you have worked with a TM in the past, you may need to ‘re-learn’ a new system.
There is no getting round these initial investments. However, we believe that the huge advantages to both translator and customer that TMs offer make the initial outlay worthwhile.