Jump to content

Recommended Posts

My experience over a good few years has been this; my boss rushes up in a state of barely concealed excitement with a box that has cost anywhere from 50,000 yen to 250,000 yen. He says he has heard good things about this translation software, and clearly has a fantasy of getting rid of the expensive foreign translator. Will I try out the software and write a report about how best to introduce it?

 

In this case, I could actually get out and submit the first report that I wrote about this in 1998, but being a diligent and naturally curious person, I check out the software anyway -- then edit the names and dates in the 1998 report and submit it. The report says that we must introduce a rigid system in which our clients give us texts that have been written to be 'machine translatable', avoiding all idiomatic phrases, sophisticated grammatical constructions, and any of several kinds of ambiguity whatsoever. The software must then be trained by someone who has read the manual and understood it. The output will be reviewed and usually rewritten by a native speaker. Even with this system in place, only perhaps 5% of any texts that the company may handle can be treated in this way. Even manuals and patents that naturally follow a certain pattern require extensive human intervention before they're usable.

 

The box then gets put on the shelf, and the boss occasionally casts a wistful glance at it, then glares balefully at the translators.

 

What is available online is probably worse than what is packaged as a 'professional solution', and it cannot be trained.

 

I do however use translation memory for everything I do. This makes a database of the Japanese and English translated pairs (human translated), and whenever you translate a similar sentence, it presents the previously translated sentence and shows you which parts of the sentence are different. The vocab is also stored in a database and can be plopped into place with one click. This results in huge savings of time and effort. I only wish that the effort spent on making crappy machine translation software was spent on making translation memory software more intelligent, a perfectly realizable goal.

Link to post
Share on other sites

Long termer Japanese-English translators have lived in fear of machine translation being "five years away" for at least fifteen years now.

 

I have used machine translation software, but it was incapable of producing anything like usable text. You could use the thing to translate all of the vocab, but in my experience that was no big deal. I found just having to move the cursor between the usuable bits (nouns mainly) and the unusable bits (the rest) to be extremely troublesome.

 

I don't use translation memory. In the field in which I work, few people use it or at least admit to using it. I don't take on a wide variety of work, so I spend very little time looking things up. What sometimes takes time is working out what the original wants to say as opposed to what is written on the page. In such cases, a machine is not going to be of any use. The productivity tools I use are text expansion, which can really speed up your typing (pro typists use it too), and a separate search tool, which can scan all of my files for character strings in seconds if need be. The text expansion I use is Word's autocorrect function. There are better dedicated solutions out there, but Word does the job. The search tool I use is called KWIC finder. It's much more effective than searching in Win2k.

 

In certain fields like software localization, the use of translation memory is becoming a prerequisite for getting work. There is a free program you can dabble with called Wordfast, though it may crock your preferences in Word. Read up before using it.

 

IBM supposedly have an advanced English to Japanese machine translation system that they use for their own work with favourable results, but I have heard nothing about it being commercialized. Ultimately more information flows into Japan than out, so I would expect to see more progress being made in the English to Japanese direction than in Japanese to English.

Link to post
Share on other sites

The first such software I reviewed was IBM's Translation Manager and Nova's Transer which both cost some obscene amount when my company bought it. IBM no longer sells its TM, but I'm not surprised if they do use it in-house. IBM is the sort of company that would have the knowhow and motivation to develop workflows where TM and even MT would deliver results.

 

The BBC article is a very shallow look at the issue unfortunately. They just couldn't resist squeezing it for that zany, low-level humour that results from these kind of initiatives.

 

One of the problems over and above the issue of translation is how poorly many people express themselves in the first place. I get texts in Japanese and I think to myself "I know what you're trying to say, even though you're not actually saying it. And when I've translated it, you still won't be saying what you're trying to say. But you're hiring me to translate, not to edit. So you will not ultimately be understood." In many cases, editing at the sentence level would not make things clearer, even if you had a mandate for that. Of course machine translation software will never feel the pathos of being in this position, and will churn out its garbage unfeelingly until the end of time.

Link to post
Share on other sites

After just spending a week of my time re-writing a fellow researchers' paper to a level acceptable for a top international science journal, I am now after an engrish (Japlish) to english translation software package. Now how much would that one cost I wonder? ;\)

Link to post
Share on other sites
×
×
  • Create New...