The “exact” match, aka duplicate, repetition, internal match. When a document is presented for translation, the first and best type of match hoped for is this one. This type of match means that an exact match of the entire source segment has been located in the memory, and a corresponding translation is present. The more of these types of matches, the better the savings! But wait…
What if the translation memory is useless / harmful as pointed out above? What if a post-translation edit was introduced to those segments but not retro-fit into the memory? Translation agencies have learned to point this out to the client and admit that it is an industry-wide flaw. Clients have learned to accept that flaw. Both sides usually compensate with some kind of “review” charge. Basically, an “exact” match usually never provides 100% ROI to the client in a segment-based search engine.
Ironically, in a term-based search the presence of a flawed translation is glaringly evident, as is the correct translation. For example, this illustration shows how a search on “purchase order” results in dozens of matches (only 7 shown). A quick glance at the list will reveal any inconsistency with that term in most, if not all, contexts.
Potentially inconsistent or questionable terms can be brought to a team leader’s attention where decisive and thorough action can be taken. Verbingo, for example, allows terms to be managed via a dashboard accessible not only by the translator but also by the client, client reviewer, project manager, etc.
When the translation memory is accessible in a term-based manner the ownership of it can be transferred to the client, where it belongs. Once an agency delivers the translation of a project, the client can then spot-check or review the translation as desired, and a certain level of maintenance can be agreed-upon (think “review”), but the ultimate ownership can now lie directly in the client’s hands. Terms can be changed by reviewers or end-users as needed without the approval, or even involvement, of the translation agency. A simple search/replace can assure all high-level, important terms are kept consistent and accurate.
By taking control of the translation memory, the client can now expect three things: 1) 100% ROI on exact matches, meaning 2) no more automatic “review” charges resulting not only in a cost savings, but also 3) a potentially drastic time savings. And if this is the case, we translation agencies MUST accept that there should be NO charge whatsoever to a client for exact matches.
So where does that leave the “fuzzy” match? I’ve always viewed fuzzy matches as big boulders teetering on pointed mountains. One tiny, little push can result in a big effect, either positive OR negative. After all, the entire concept of fuzzy matching is based on the edit of an existing translation, and as pointed out above, most translation memories are poorly maintained at best. Clients and agencies have been accepting the flawed concept of “exact” matches for years; what hope can there be for accepting the editing of flawed matches? Clients and agencies need to stop wasting their time and money on these processes. As with exact matches, the value in fuzzy matches is never 100% of what it appears to be in segment-based engines.
And for the same reasons mentioned above, if translation professionals and clients can accept that even flawed translation memories have some value in term-based searches, then perhaps we can begin to apply less value to the “fuzzy” match and its inherent flaws and instead apply more value to the overall ability of the translation tool to enable the best possible reference for every term in a segment. In other words, it’s easier and cheaper to maintain a translation memory on a term-by-term basis, than it is on a segment-by-segment basis.
Consider the example below. If a translator were to rely on a fuzzy match to assist him with reference for improved accuracy and/or speed, he would be greatly disappointed as the translation for (and valuable reference stored in) the segment “Your database has now been configured and is ready for use.” will not be found by any current segment-based search. Only a term-based system will, and should, allow “is ready for use” to find a valuable match in the memory.
Up Next: The Next Phase
