There Is Really Nothing Fuzzy About the Logic Or the Concept of Fuzzy Matches

Fuzzy logic is a term that has been used for decades now to describe a branch of logic designed to allow degrees of imprecision in reasoning and knowledge, typified by terms such as `very’, `quite possibly’, and `unlikely’, to be represented in such a way that the information can be processed by computer.

It is a very useful method because it allows computers to make a decision when the information presented to them is not quite clear (i.e. not a simple choice between two alternatives). Most humans have the ability to make such decision quite easily, but since machines can only respond to a clear command, they have to be instructed what to do in case of uncertainty.

Fuzzy logic thus makes it possible to prevent or reduce occurrences of the status known as “freezing”, which is what happens when a computer does not know what to do.

The term “fuzzy matches” is much younger. I came across this term for the first time when a translation agency sent me a “Confidentiality Agreement” that specified the pitiful, fractional payments for “full matches” and “fuzzy matches” about five years ago. Because I had no idea what “fuzzy” and “full matches” meant, I called the company. “Oh, it’s for people who use Trados or other CATs”, was the response. Since I did not then and do not  now use any computer memory tools, I was told not worry about it.

We are always told not to worry about the fuzziness of the one-way communication between the very important people who run everything and the rest of us who have to do the real work, such as the fuzziness of methods used to calculate statistical data.

For instance, unemployment statistics do not count as unemployed people who work part-time because there is no full-time work for them, or who no longer bother to contact the unemployment office because they have not been able to find employment for a long time, and inflation statistics do not include prices of food, medicine, and fuel (because they are “too volatile”!), etc. Actually, they are not volatile at all. Volatile means that they may frequently go up or down, but these price go over time basically only in one direction. Or have you noticed that the price of any of items such as milk, bread, meat, rice or potatoes at your store, or the price of gas at your gas pump, has gone down over the years? I have noticed the opposite.

The real reason why these prices are excluded from statistics is to prevent people from knowing what the real inflation numbers are.

But let’s get back to the interesting subject of fuzzy matches.

I received an e-mail this morning which said:

Dear Steve:

This is XYZ, a project manager at ABC Company, a US-based LSP.  I am currently putting a team together for our client. The project consists of translating a manual for [insert the name of the gizmo]. We have some Trados translation memory matching – there are 5,868 new words, 8,232 100% matches, 573 repetitions, and 2534 fuzzy matches.

I’m afraid I was quite rude in my response to this project manager, although not nearly as rude as the project manager was to me, without even realizing it, by assuming that my price per word is only a springboard to fabulous discounts that I would be glad to give to this agency (excuse me, I meant LSP) for the privilege of finally having some work.

Based on what I discovered mostly by reading comments on my blog and blogs of other translators, I am hardly the only translator who does not use Trados or any other computer memory tool. At least 40% of translators and probably more do not use these tools, because although they may be very suitable for example for translating repetitive updates of software or computer manuals, they would be mostly or completely useless in other translation fields, namely fields that require creativity, experience and good writing skills, while other translators will probably not acknowledge that they do use them even if they do … because they don’t want to be forced to give discounts.

**************

 It so happens that a few days ago I translated two very similar long patents, each about 13 thousand words. One described a device or an apparatus to manufacture a product, while the other one described the method used with this apparatus to manufacture said product.

I don’t use computer memory tools because I don’t need them. As long as I remember what it is that I am translating, I can always find repetitive passages in my translation of the first patent quite easily, cut and paste them into my second translation, and then carefully proofread everything to make sure that I did not miss anything.

And I did give the client, not a translation agency, but a scientist working for the manufacturer, my standard discount of 10% because the cutting and pasting technique does save time. I have been giving discounts like this to my clients for years if not decades before the first computer memory tools appeared on the market.

But in this case it is up to me, not up to a translation agency, to decide whether I will in fact provide a discount, and how much of a discount it will be. Had I decided not to offer a discount, it would still be OK because the agreement between myself and my client is based on the actual number of words in my translation, not on a fuzzy concept of “full, partial, and fuzzy matches” that some translation agencies are trying to wield like a mighty sword beating down any hopes that translators might be able to make a decent living if they accept the Shylockian logic of this concept.

According to this logic, if you just spent half an hour researching on the Internet a very complicated term that may be crucial for your translation, you get paid whatever few cents you were supposed to get paid for that one word once you know beyond any doubt the author of the text did not make a mistake and that you indeed have the right term.

But you would get those few cents only for the first occurrence of this word.

After that, “Trados translation memory matching” would reduce your remuneration based on “5,868 new words, 8,232 100% matches, 573 repetitions, and 2534 fuzzy matches” to …. a fraction of what it would be based on normal human logic.

As I said in the introduction, fuzzy matches have nothing to do with fuzzy logic, these are two completely different things.

Instead, the logic of “fuzzy matches” is really the logic of getting something for nothing, a very old concept also known as stealing.

UPDATE

Note how protective of “fuzzy matches” are most of the translators in the Proz discussion group linked in the comments. They no longer seem to understand the concept of a world in which a translator is a well paid professional who is translating complicated texts for direct clients, and who would simply never agree to be controlled by CATs counting “fuzzy matches” or “full matches” to generate automatic discounts, even though these complicated text may have many repetitions in them.

To me, this is very similar to what is known as Stockholm syndrome.

Hostages under the influence of Stockholm syndrome also no longer understand the concept of not cooperating with their captors.