Did you write your essay?
Nope. I have to be in the right mood for that.
Whats the right mood?
Last-minute panic.
In one of the Calvin & Hobbes strips by Sam Watterson
This is not a guide that could help you choose a good CAT program. You will not find many technical details likely to help anyone make a decision to buy a program. In fact, you may not even have to buy any program of that kind, as many translation companies will be all too happy to sub-license their software to you. Consider the comments below as the voice from down at the bottom end of the food chain, the first impressions of the end user who gets the software at the same time as the translation assignment. The length of my experience with these programs varies from four weeks to three years, but basically I have always had to start NOW on a job using a program I had never seen before.
Evaluating Star Transit would be a difficult job for me, because the occasional job I did with it required installing the program twice or three times and working with it was very frustrating, but other colleagues who regularly work with Transit sound quite happy with it. Maybe Im denser than average, or maybe I really lacked time to explore the program and get familiar with it. So if there is a way to translate repeats without having to get back to their previous occurrence and copy/paste them, I didnt find it. The same applies to a feature that would have enabled me to use one translation to complete other files. This was obviously the clients job.
During this short and frustrating time, I did spot a few positive points: the Termstar dictionary is easy to search, edit and update. And the program provides a count of the number of words translated and left to translate, a feature that is not present, or not as visible, in other programs, and is much easier than a percentage to convert into the usual word-count metrics used to assess ones workload.
IBM TM/2 has been much less trouble to learn. Controls are easy, although you may regularly forget to activate the segment you want to change. With the usual mode of extra-fast hands-on learning of CAT programs, I didnt find a way to do automatic search+replace and spellchecking outside the active segment, but some colleagues say you can do that. Controls are very easy, and the few keystrokes required are quickly mastered. The matching function is quite efficient, with a good hit rate for fuzzy matches. I havent tested the new version yet, but they say it allows you to compare various versions of one sentence if it has been translated in different ways, and provides better edition of translation memory files. This is a program I would also recommend to someone interested in buying one, provided the filtering functions with word processing software are easy to handle.
Joust, a.k.a. TSS (Translation Support System) is Alpnet Internationals proprietary translation program. This tool is pretty similar to TM/2, and just as convenient. A version for Windows has finally been developed, after years of suffering in OS/2 without even being able to use the mouse.
This program is basically easy to use, once youve learned to use the F5 (Save), F8 (translate next) and F10 (paste source text into target line) function keys. It is hardly a joke, because a lot of work can be done using about 12 key combinations. The dictionary is also easy to update, and an unlimited number of entries can be added and, with a little juggling in the File Manager, used for later work.
The matching function is quite good, fuzzy matches being within a word. The insertion of matched sentences is automatic, with possible fuzzy matches highlighted (though the fuzzy may result only from a lower/upper case difference). However, this automatic feature may be slightly annoying when figures from a table are included in the text. If, for example, one line goes like this :
1 to 2 inches
the translation (say, into French) will read :
1 à 2 pouces
But if the next line is :
2 to 5 inches
the matching function will automatically insert the
1 à 2 pouces
line again and make a fuss because the match is fuzzy. This is obviously less of a problem with TM/2, as the program demurely waits for the user to insert the matched sentence, instead of overzealously inserting fuzzy matches (which, in the case of a 4-word/figure phrase, may be 50% fuzzy). But this is a minor disadvantage. On the other hand, if you find out that a sentence comes up every two or three lines and youre tired of hitting F8 to insert it, you have the possibility of backing up the project without the synchronization file coordinating the target and repetition files, then reload and resegment it, and then all full matches will be automatically completed during resegmenting, and you may gain a few percent points on the completed segments counter.
The repetition file is also a precious help in case of power and/or program failure : the target file containing the translation seems to be more sensitive, so that it is easily corrupted or even (in the late OS/2 version) destroyed. Since the source and repetition files usually survive any incident, it is easy to resegment the project without losing any work.
The major problem, both with TM/2 and TSS, is caused by improper segmenting of text, which may be done without taking the order of words in the target language into account. Staying with the English-French language pair, the program may filter in a table with columns reading :
Corrosion | Wear |
resistance | resistance |
If segmented in an intelligent way, the phrases should read corrosion resistance and wear resistance, which would be no problem, but if they are filtered as they are in the table they will be chopped up to:
Corrosion
resistance
Wear
Resistance
The French will invert the words in each term, so that the first will become, e.g.:
Résistance
à la corrosion
Alas, the resistance = à la corrosion match is not correct, and translating a table without a hard copy of the text (and a lot of patience to correct mismatches) can be catastrophically misleading.
Finally, we have had a mind-curdling experience with Trados Workbench: not the program itself, but what the clients do with it. In fact, some of the latest comments about Workbench among members of the FLEFO forum have been quite enthusiastic, so it seems that those who do have the complete software are quite happy with it. But some of our European clients, both translation companies and industrial clients, use Workbench to provide pre-translated text that can be delivered as RTF files and worked with any word processing software. The text then comes as in the following example:
{0>Die Abmessungen des Portalausschnitts im Wagenkasten sind in der Zeichnung "Meßblatt Seitenwand, Endwagen", Nr. {0>2 16106.0.00.040.101<0} dargestellt.<}0{>
Die Abmessungen des Portalausschnitts im Wagenkasten sind in der Zeichnung "Meßblatt Seitenwand, Endwagen", Nr.{0>2 16106.0.00.040.101<0} dargestellt.<0}
{0>Folgende Randbedingungen müssen gegeben sein: <}0{>
Folgende Randbedingungen müssen gegeben sein:<0}
The text in blue is the source segment, that in black the target. None of the small <}0{> or other codes should be deleted, so that even though the source is hidden text, hiding it to work with a clear view of the target text is out of the question, and so is automatic search+replace of repetitive items because every second occurrence must not be translated. The text to deliver must look like this:
{0>Die Abmessungen des Portalausschnitts im Wagenkasten sind in der Zeichnung "Meßblatt Seitenwand, Endwagen", Nr. {0>2 16106.0.00.040.101<0} dargestellt.<}0{>
Les dimensions de louverture de la porte dans la caisse de la voiture sont présentées sur la planche « Cotes de la paroi latérale, voiture pilote », numéro {0>2 16106.0.00.040.101.<0} <0}
{0>Folgende Randbedingungen müssen gegeben sein: <}0{>
Les conditions secondaires suivantes doivent être respectées:<0}
The client will then filter out the information, prepare more pre-translated text, and the translator will soon be ready for the loony bin. Id rather wish nobody knew Im lunatic already, so my partners and I have decided to refuse to work in this way, which we consider tedious, irritating, time-consuming and absolutely counterproductive. Workbench may therefore be a good piece of software to work with, but no translator should ever accept a pre-translated file if he/she doesnt have the program.
A couple of general remarks to finish: memory space used to be a concern, but with the newer computers it is no longer so. I run TM/2 and WinJoust on a 486 PC with a 16 MB RAM, that also has Word, Excel and PowerPoint for Windows and an assortment of Internet and e-mail programs and plenty of CD-ROM stuff as well. As far as I remember, this machine also worked fine before it was upgraded from 8 MB. The projects are compound files made up of a number of unit files (source, target, repetition/memory, etc.), and may take up 200 KB to 1 or 2 MB. This is normally not a problem if your processor is fast enough (even 100 MHz is fine).
None of these programs is, in the strictest sense, a computer translation program. They are computer-aided translation programs designed to make repetitive text less boring (or, to put it more professionally, to allow efficient translation of repetitive text), and therefore work best with such texts, e.g. user manuals, instructions of all kinds, catalogues, etc. They do not perform better than the flesh-and-blood translator for general texts with few repeats.
One huge advantage of this approach to translation is that, for once, the client is technically forced to provide some terminology. The glossaries of the projects may not be that big, but they still help, as well as the text possibly contained in translation memories or repetition files. One pitfall about these pre-translated items is that the programs often skip them to go to the next empty segment, so that the translator may not always (remember to) look at potentially useful reference text. In this case, reading the whole file after translation is completed, and not only the freshly translated parts, is an essential step to ensure coherence.
One problem with these programs is that they usually come at the same time as the first job, and the deadline for delivery obviously doesnt take training into account. The translator rarely has the time to read the users manual (if this is provided at all), and sometimes cant even install the program properly on the first (and even second) attempt (although the client did give instructions). This is one point to take into account before agreeing to work with a program that the client is offering to provide.
Another problem may be eye strain. Because the text is segmented, hard copies are not much help, as segments do not always end with full stops. Therefore the text on paper can only be used for revision purposes, not during the initial translation work, so that one has to look at the screen all the time. My screen may need upgrading and my eyes may be sensitive, but working two or three days with these programs is enough to make me spend two weeks on eye drops. You may not want to wear sunglasses whenever you work with CAT programs, but a really good monitor and a UV filter are essential prerequisites.
See also the review of Atrils Déjà Vu in this issue of the Translation Journal.
© Copyright 1997 Translation Journal and the Author Send your comments to the Webmaster URL: http://accurapid.com/journal/03TM2.htm Updated 12/24/97
|