Genuine Enthusiasm About Machine Translation in the ATA Chronicle

It is never appropriate to submit unreviewed, unrevised machine translation output to a client as a finished translation.

(This is an actual highlighted quote from an article called “Questions and Answers about Machine Translation” published in the November/December 2011 issue of the ATA Chronicle).

There was an interesting article in today’s Washington Post by Paul Farhi about the fact that despite a law against it, “stealth commercials frequently masquerade as TV news”. The newspaper used the example of Alison Rhodes who “is one of a small army of hosts and reviewers” of products offering to Fox viewers advice on what to buy. According to the article, Rhodes sees no issue in accepting payment for her recommendation because “her enthusiasm is genuine”.

Presumably, the more they pay her, the more genuine the enthusiasm. There are laws precisely against this kind of peddling and huckstering of products masquerading as news, but these laws are not enforced. In many respects we are already living in the post-legal reality of the corporate America at the beginning of the 21st century. Laws are enforced or ignored depending on who is breaking them.

I don’t watch Fox News, but I read, or try to read,  the ATA (American Translators Association) Chronicle magazine. One of the three featured articles in the November/December issue of the ATA Chronicle is an article written by Laurie Gerber titled “Common Questions about Machine Translation”.

Although Laurie Gerber states at the end of the lengthy article “Please know I have no ties (other than professional friendship with former colleagues) to any MT developer/seller listed below”, as far as I can tell, instead of providing a useful analysis of the strengths and weaknesses of machine translation (MT), the article was written to promote further acceptance of MT by translators and “LSPs” (language service providers, which is the newspeak for translation agencies), see my post here, because all resistance to MT on the part of translators would be as futile as resistance to Star Wars-like or Terminator-like machines that were sent by a highly sophisticated and extremely malevolent culture of space aliens or intelligent machines to colonize the Earth.

As far as I can tell, the article is a long advertisement for a product, namely MT. It has perhaps two thousand words, with a handy list of vendors of MT products at the end of the article.

For example, this is how Laurie Gerber deals with one of the objections that shortsighted translators such as myself might have against incorporating MT in human translations:

Q: “Isn’t it easier to do a good translation from scratch rather than edit a bad translation?”

A: “This, plus MT’s inability to handle language nuances and culture, are condemnations of MT that most translators find quite satisfying. I offer three ideas for your consideration below.

1. Post-editing MT can be unpleasant, which makes it seem like it takes much longer.”

[Mad Patent Translator’s Note: These dumb translators must be thinking that it takes longer because they are mentally unstable. They are not even smart enough to realize that they have to start using MT to make more money].

“At the Translation Automation Users Society (TUAS) conference in October 2010, Mirko Plitt, of Autodesk, described Autodesk’s experience. He explained that while translators often feel that post-editing MT takes longer than translating from scratch, timing their work showed that post-editing MT actually yielded a 30% – 40% productivity increase.”

Etc., and so on, and so forth …. If you don’t use MT, you are a Luddite and a loser. But, the author cautions ernestly, “It is never appropriate to submit unreviewed, unrevised machine translation output to a client as a finished translation.”

How very true. It is also never appropriate to kill your husband and children in their sleep, although that may sound like an easy solution too.

Anybody can design an MT test that will prove anything about machine translation, or the exact opposite, depending on how you design the test. I use MT quite frequently because it saves me time as I don’t have to use dictionaries or look for words online. But I happen to know that it is absolutely not just my imagination that it is much faster to translate from scratch than to “post-edit MT”. I would be post-editing MT myself if it did work the way Laurie Gerber and Autodesk say it works. Machine translations of patents are now available for free to anybody on a number of websites (the website of Japan Patent Office, European Patent Office, World Intellectual Property Office, and other websites).

I will now try to demonstrate what I am saying above on my own test of MT of a Japanese patent. True, some short and simple paragraphs translated by MT almost make perfect sense, for example this one: “If you use an alkali metal fluoride as a catalyst, and the manufacturing method, etc. Specific examples of alkali metal fluoride is supported as described in the first embodiment.”

You could edit something like that without wasting much time. It might even not take much longer than translating the whole thing from scratch.

But most longer paragraphs are completely butchered beyond recognition, see for example how specialized MT designed for patents handled “language nuances and culture” in this paragraph:

“Before the mind complex acidification matter wa, 2 or more の metal elements wo constituent element と su ru composite acidification things で thou ri, the metal element と し て は, patients え ば, カ Hikaru シ ウ ム, マ Jewellery Neko シ ウ ム, ケ イ factors, ア Hikaru ミ ni ウ ム, チ Tatari ン, ji ru コ ni ウ ム, ni オ ブ, Getting ン Tatari ン, grades posted リ ウ ム, ni ッ ケ Hikaru , マ ン ga ン, 亜 lead, su ズ, lead, and other ga Ju げ ン リ ら れ ru. Acidification of the former compound material wa mind, one kind with い て の も み wo shi い Certified, and use of more than two wo shi て も Certified い.”

What the hell happened there ↑? I have no idea. And in all probability, neither do the MT programmers who created this freak of nature, I mean of silicon.

Under most circumstance, with most texts for which anybody would pay a translator any money to translate, post-editing of MT takes longer than translating from scratch, often much longer.

But according to what Laurie Gerber says in her article in the ATA Chronicle, translators simply need to get used to editing somewhat imperfect translations like these because  “MT post-editing is a cultivated skill. Embarking on any new activity where we lack “fitness” is frustrating, embarrassing and feels very uncomfortable.”

I think that machine translation could be in fact a very useful tool not only to translators like me who don’t have to open up dictionaries as much anymore when they translate if they are armed with a free MT product, since MT already is or soon will be also very useful to so called LSPs (translation agencies).

I can think of at least two reasons why some agencies, especially the larger ones who believe in the sanctity of the corporate model, could have warm and fuzzy feelings about MT.

1. MT will put the Fear of God into the hearts of translators. It is easier to negotiate rates down for human translation if you carry the big stick of MT. All you have to do next time if a translator dares to ask for a higher rate is to remind him of the big stick hanging over his head. Given that he is competing with MT, he should be glad he’s paid anything at all.

2. It should be possible to design a model in which instead of paying translators a rate based on the word count, you pay them a lower hourly rate because they are no longer translators. They are mere MT post-editors, and most people who have some familiarity with a foreign language should be able to edit MT product. If you can reclassify translators as MT post-editors, you should be able to pay them for instance 30 dollars an hour and then sell the post-edited MT product for more than twice as much.

MT is in fact a very useful tool in the hands of translators who can use it as an additional resource for human translators. But if the same translators are not smart enough to realize that the same tool is not really very useful to people who can’t translate, it is also a tool that can be used by smart business people to keep dumb translators in their place.