Despite the hopes and dreams of a number of dedicated programmers, computer translation–also known as machine translation–still falls far short of the quality most clients expect in their translated documents. However, one area in which computers have made significant progress is in the area of translation memory, which relies heavily on pattern-matching new source text against previously encountered source and translated target texts (the latter translated by a human translator), which are stored in a database.
One of the strengths of such systems is that patterns need not match 100% for the software to work. The sentence
My name is Mary is a 75% match for the sentence
My name is John and any translation memory system worth its storage space will recognize that. So not only will translation memory software recognize (and reuse) chunks of text that are identical, but it will also reuse high-percentage matches, presenting them to the translator for editing into final form.
Increasingly, however, agencies are using translation memories to “pre-process” documents. This results in the insertion, into draft translations, of (more or less) matching segments taken from aggregated corporate translation memory files. The reason for doing this has to do with the rather strange relationship between editing and translation in the translation profession: Editing, you see, is paid at a lower rate than is translation, typically between 25% and 33% of the rate paid for translation.
This is curious, if you consider that when a translator translates a document, the work flow cycle can be approximately described as follows:
On the other hand, the work cycle undertaken when a translator edits a document can be approximately described as follows:
I don't know about you, but it seems pretty clear to me that a translation editor's work is significantly more complex than that of the translator of the text being edited. So then, you may ask, why is editing paid at a lesser rate?
I think it is because the editing rate has traditionally been predicated on the translation being of high quality to begin with.
If a translator is really, really good and produces high-quality work with no usage, grammar, spelling, style, or punctuation problems, or awkward sentences, or inconsistent terminology, the editor will generally have a pretty easy time of it when it comes to editing. However, if a translator produces translations that are riddled with such issues, editing such text begins to resemble a retranslation instead of an edit. In fact, when I started in this industry, agencies did their best to refrain from assigning work to translators who produced poor translations.
What's a high-quality translation? Well, back when I worked in-house for one of my clients, we took the criterion used by the American Translators Association for certification of entry level proficiency and adapted it to our company's ISO 9000 quality program. For example, if an editor found, on the average, less than one major error per 500 words of text, said translation was considered acceptable (from the perspective of what an editor had to do to edit it). Finding one major error per 250 words of text was considered cause for concern; finding two such errors per 250 words of text was evidence of incompetent work.
Do the segments in “pre-processed” documents represent high quality? Hardly.
Typically, pre-processed documents will contain segments that represent an “80% match” or better. This means that unless an offered segment is a “100% match,” it contains–at minimum–at least one major translation error as compared against the source text.
As an aside, it should be noted that even a candidate segment offered as a “100% match” doesn't actually guarantee that the segment is usable. One all-too-common problem is that the “100% match” classification is not merited in the first place. (A document I translated/edited some time ago offered “acquisition parameters” as a 100% match for the Russian “технические условия,” whereas the fairly standard rendering of the Russian source term is actually “specification.”) And problems may even be encountered if a “100% match” is perfectly acceptable text, as in the case where the terminology used in a proposed segment is not consistent with that used in the rest of the document.
In the final analysis, if we assume that segments to be “edited” contain exactly one major error per segment, then if we charitably assume an average of 25 words per sentence and do the math, we find that one page of “pre-processed” text presents the editor with ten major errors to correct per page. Some would maintain that such work squarely falls into the retranslation category. (I do, at any rate, but I just work here.)
In the end, having to
edit retranslate gobs of what are, in effect, error-riddled translations at what amounts to a steeply cut rate is bad for one's morale and risks awakening that little voice in the lazy, shirking, goldbricking part of the mind that all of us seem to have, urging us to exert a minimum level of effort when making editing corrections, because even if the result is not perfect, it is head-and-shoulders better than what we started with.
Listening to that voice is a sure-fire route to what motivational speaker Zig Ziglar called “stinkin' thinkin',” and that kind of stuff is fatal if you want to succeed as a freelancer.