Book Review: Is That a Fish in Your Ear?

Is That a Fish in Your Ear is a book about translation, so it should be no surprise that I picked it up. But it’s not so much a book about the mental process of translation as it is a book about the business of translation and its role in society at large. So it certainly has some tidbits that are of interest to translators, but not a lot of insight. It isn’t the endless buffet of food for thought that Le Ton Beau de Marot by Douglas Hofstadter is.

Early in the book, the author, David Bellos, discusses how many different words the Japanese language has for “translation,” wandering perilously close to snowclone territory—though he later tackles the idea of snowclones head-on.

He does talk about the size of the international translation market, which is interesting, but his source is UNESCO’s Index Translationum, which only covers books. This is only a part of the translation industry, and as far as I can tell, only a small part (also, digging through UNESCO’s statistics reveals some very messy data). There are some interesting facts about the flow of translation—English is overwhelmingly dominant as a source language, exceeding the second-most common source language, French, almost sixfold. Japanese, the language I translate from, comes in at 8th, with only about 1/8th as many works translated as French, and only about 1/47th as many as English.

English is much less popular as a target of translation, coming in fourth to German, French, and Spanish (with only about half as many works as German). Japanese comes in fifth, with only about 15% fewer works translated into Japanese than into English. This would suggest that English speakers are not that interested in hearing what the rest of the world has to say.

But I am sure these figures are not representative of commercial translation. In patents, for example, I have read that English is the dominant target language, and the most frequently translated source languages for English are Japanese, German and French. Not sure where those statistics come from, but they don’t surprise me. It should be relatively easy to get statistics on patents, because they wind up being collected by a few government agencies. Much less so all the newspaper clippings, press releases, user manuals, depositions, specifications, clinical drug trials, and so on that make up the bulk of commercial translation, some of which are never intended for public consumption (and some of which is probably hardly read at all, in any language).

There are also different market forces at work here: a lot of patent translation is “push” translation (by the authors), whereas book translation is “pull” translation (by literary agents in the target language that are proxies for the eventual audience).

As to some of the standard problems of translation—He does talk about translating poetry (with reference to Hofstadter’s book), and to a lesser extent, humor, but these were fairly superficial treatments. He talks about the problem of translating culture-specific features to some extent and some of the solutions. He points out that it is not hard to translate a high social register satisfactorily, but translating uneducated-sounding regional dialects is much more problematic, and in the last pages of the book, he actually gets around to an interesting observation that might explain why: he makes the argument that language in its original form did not emerge to communicate ideas, but to reinforce social bonds, like monkeys grooming each other. And the aspects of speech that are peculiar to one community—which reinforce those social bonds—are exactly what can’t be translated to another, and seem glaringly misplaced when “translated” to the speech habits of a different community.

Is the book worth reading? It’s light. It gives a mile-high view of translation. It doesn’t grapple much with translation problems, much less offer solutions to them. It feels like it’s more intended for people who are interested in knowing about translation than in actually translating (which is fair, since there are probably more of them). It won’t have a lot of keen insights and revelations for translators, but it is interesting.


Brent Simmons writes about gamification, saying

you could look at this trend and say, “As software gets simpler, it gets dumbed-down — even toddlers can use iPads. Users are now on the mental level of children, and we should design accordingly. What do children like? Games.”

I’ve been thinking about gamification a little for a while now, and I think it’s actually more sinister than that. Look at a website like Stack Overflow. They’ve got it set up with this treadmill of meaningless rewards to keep you engaged in the site, asking and answering question. In addition to increased ad impressions (which is cynical enough, the sole point of a game like Farmville, which has no rewards that I recognize as such), your labor makes the site more valuable: a good “answer site” like Ask Metafilter (which is a cool community, not an exploitative business play) gets very high Google rankings—Stack Exchange clearly want to cash in on that action getting strong Google rankings for their own site, leading to more pageviews, and the circle of life continues. For your efforts you get a gold star. A virtual gold star. But they’ve figured out that points and achievements activate some hindbrain reward center that they cynically play off of.

In my own vocation of translation, there’s been an increasing trend toward uncompensated crowdsourcing (another hot-button word) as an alternative to professional work, and I fully expect to see gamification tactics applied to that as well before long.

The most unusual dictionary I own

Cover of National Security Agency Japanese-English Technical Terms Dictionary, Volume I

Back before every translator worth his salt had an always-on, high-speed Internet connection, we had to be more self-reliant in the way of reference sources. So we bought dictionaries. Lots of dictionaries. When I lived in Japan, I’d stop in Jimbocho at least once a month. There was a used bookstore that specialized in technical and scientific books, and I’d buy dictionaries there on the off chance that someday I would translate something relevant.

Finding translations

Wil Shipley wrote about how coders can organize their software projects for localization. But what happens on the other end—once those files of strings to be localized get handed off to those “errrr… whatchacallums. ‘Polyglots!'”

I’m a translator. I’ve done a small amount of coding, so I can appreciate the issues involved on that side, but it’s not my strength. I’ve done a little software localizing, and a lot of translations of software manuals.

Suppose you’ve got an application that you want to make available in other languages in addition to your native language. You’ve got the technical aspects of the problem solved. How do you get it translated? I will assume that you want every localized version to produce the same delight and satisfaction in speakers of other languages that you are trying to create with your own-language version, and that you are willing to go to some trouble to do so.

Your first decision, from which many other tradeoffs will follow, is whether to give the project to a translation agency or to act as your own translation coordinator. There are pros and cons to each approach.

When you hand a job off to an agency, they will act as a buffer between you and the translators. In fact, they will insist on it. If the translators have any questions in the course of the work, they’ll footnote it and the agency will try to resolve the question either on their own or by contacting you.

Most large agencies will claim to handle any language combination; for any relatively common language combination, they probably have access to dozens or hundreds of translators. The coordinators at these agencies often have no ability to evaluate the quality of translation work themselves. Some of these large agencies farm work out on a first-come, first-serve basis, some farm them out to whichever translator is in their database as handling that subject matter at the lowest rate. Large jobs on tight deadlines will likely be split up among multiple translators, perhaps with some effort at harmonization after the fact. Editing work is also usually farmed out using the same methods.

One point that Shipley makes is not to use pictures of words, since they are so much more work to localize. While this is unquestionably good advice, bigger agencies may have tools for stripping text out of Photoshop/Illustrator files to be translated, and substituting the translated text back in (if it’s styled text, this could obviously get sticky).

There are smaller agencies that specialize in only one language, or only a few. These typically have bilingual people on staff, do the editing in-house, have some kind of personal relationship with the freelance translators they work with, and have a pretty clear understanding of each translator’s strengths and weaknesses. These agencies will also act as buffers between you and the translator.

Finally, there is dealing with individual translators. This will allow (or require) you to work more closely with the translator, which should in theory permit them to do higher quality work. Dealing directly with the translator will expose you to a wider range of rates: on the one hand, you’re not paying the agency’s vig (which can be anything from 15% to 200% in my experience). On the other, you may be dealing with translators who have priced themselves out of agency work. Dealing with individuals does place you at risk if one of them flakes out (rare but possible), and will take some work just to find suitable candidates. You should also arrange for someone to edit the work. You should be able to ask the translator to do this, although some translators might be reluctant to let their colleagues see their work. Any freelancer worth his salt who is working for a direct client will do a meticulous editing job, but the more eyes that look at it the better. Mistakes can always slip by.

Let’s assume that, one way or the other, your localizable strings file has reached the translator. I’ve seen agencies send these as Excel files, with each text chunk in its own row, the source in one column and the target to be translated in another. I hate doing extended typing in Excel, but this at least has the virtue of clarity. I usually wind up moving the table into Word, which is a friendlier place to type.

The problem with these chunks of text is that they are disembodied. Out of context. And context matters. Knowing that a chunk of text is going to appear as a menu item, as opposed to a dialog box, tooltip, or whatever, will influence how the translator expresses that idea. For that matter, knowing that several menu items are all clustered in the same menu, between separator bars, would also influence how each of them was translated. A keen-eyed translator will often be able to figure out some context based on hints the text provides, but there are limits. The result of this lack of context is that the localized version will be prone to seem a little disjointed to native speakers. Screenshots of the most common interface elements would be one way to ameliorate this. Actually giving the translator the software to play around with (in addition to the strings file) would be even better, but this restricts you to working with translators who use the same platform as you (which wouldn’t be a bad idea, since they are more likely to be conversant with platform-specific lingo), and probably rules out working through an agency.

Let’s jump ahead and assume that you’ve received the translated strings file. Now what?

Well, you could blindly assume that the work has been done to your satisfaction, and you might get lucky. Or not. Regardless of whether you are working with a big agency, small agency, or individual translators, you should line up native speakers of the target language to critique the translation and make sure that if you incorporate it into your product, you’ll be happy with the way it appears in that language.

It would be prudent, regardless of who you’re working with, to request an early delivery of a fraction of the work, and have that go past your native informants to make sure the work is up to snuff. Sometimes, prospective translation clients will ask an agency to have a sample translation done on spec. In cases like this, some sleazier agencies have had a known-good translator do the sample (or many translators do a sample so they can pick and choose), and then pass the job on to the cheapest translator. Many reputable translators resent being asked to do spec work at all, so the use of spec work as a selection method is clearly fraught with problems. Better to suck it up and gamble on 10% of the project.

You should also provide as much information up front about the job as possible. Make a list of special terminology that needs to be handled consistently, and give a glossary if possible. Explain your expectations regarding style and voice.

Update: See also the response by Ryan Ginstrom, a real-live programmer and translator.

The machinery of intermediaries

My fellow translator Ryan Ginstrom yesterday wrote about one of those problems that can arise in the translation business when one is dealing with both the ultimate consumer of the translation and an intermediary.

As luck would have it, I was having a conundrum at the same time that was so perfectly complementary I can’t help but write about it.

I had previously corresponded with a potential client about some work. I didn’t get the job at hand, but he said he’d like to work with me in the future. Shortly after, he approached me about some new work. I dealt with him and a new co-worker of his. They sent me an NDA, which I signed and returned, provided them with an estimate, and they gave me the green light. This happened over the course of one day.

A few hours after I started working on this job, I got a phone call from someone who I had never heard of, at a company I had never heard of, who explained to me that all the independent contractors who work for this end client have to go through him. On the one hand, it was obvious that he must have gotten my name and number, and some facts about the job from the end client, so presumably there was something to what he said. And as soon as he realized that I had not been brought up to speed on this situation, he backed off and agreed that I should clear it up with my contacts at the end client. On the other hand, having a complete stranger insist that he was going to interpose himself in the deal after the fact was unsettling and seemed a little dodgy.

I did indeed talk with the end client, and he did indeed assure me that everything was copacetic—that his company pays off agencies to shield themselves from the unspeakable horror of being seen as dealing directly with freelancers—even though that is exactly what they are doing, and the agency serves as a flimsy fig-leaf against some kind of legal exposure.


Translating a press release about a new server with a new Intel chip, which describes all the buzzwords the chip is compliant with. Because the release is in Japanese, I’m not sure how these buzzwords are supposed to be rendered in English. Google to the rescue. They are:

  • Turbo Boost
  • QuickPath
  • Hyper-Threading

Got that? One is written as two words, one is InterCapped, and one is hyphenated. Get it together, guys.

Finally, something I can point to

One of my disappointments as a commercial translator is that I seldom see my own work in its final form. I’ve been translating for over 20 years, and I’ve seen my own work in print only two times that I can think of. The overwhelming majority of my work has been consumed internally within one company, or distributed to a very limited audience.

These days, some amount of stuff that I’ve translated is also made available online, and I’ve had occasion to see some of it in its finished form a few times, which brings me to another disappointment (which may be particular to Japanese-English translation): seeing how my work gets butchered. I’ve done hundreds of press releases for a large Japanese electronics company. I do them for an intermediary company, which then passes them back to the end-client. I try to do a good job on these, and my direct client is happy with my work, but somebody at the end-client is not: they routinely rewrite my translations to be literal, awkward renderings of the Japanese.

So it is a rare pleasure when I get to see something that I’ve done and also can take pride in pointing to it and saying “I did that.” In this case, it’s one of the meatiest and most gratifying jobs I’ve done in a long time, a guide for developing Firefox addons. (Here’s the Japanese original. The English version has been updated for Firefox 3 since I translated it.) Working on the job was a pleasure: it was well-written source material on a subject that I understood in depth. I felt like I had the latitude to do the best job possible, as opposed to the best job the client would let me get away with.

My gripes about translation memory

I recently tweeted that I was experimenting with OmegaT, a translation-memory tool. When asked by one of its proponents how I liked it, I responded

@brandelune do not like omegaT. really only works with plain text. ugly. burdened w/ typical java on mac shortcomings. not customizable.

That barely begins to cover what I don’t like about OmegaT. I’ve been thinking about what I would like in a translation tool for a while now. My desires break down into two categories: the translation-memory engine, and the environment presented to the translator.

Unreasonable expectations

Yesterday morning I was contacted by a large, well-known translation agency about a very large job. About 250,000 words spread out over about 60 documents of various types, due this afternoon—about 30 hours after the initial inquiry was sent.

Obviously they didn’t expect one translator to do all of this. In fact, very large jobs on very short deadlines are the MO for this agency—let’s just say their name rhymes with “trance perfect”—but this is an extreme case, made more extreme in that the job is due Christmas Eve.

I’m not an especially fast translator. I’m faster than some, slower than others. I can do 2,500 words in a day without breaking a sweat under reasonably good conditions, and I can manage 4,500 if I’m highly motivated and working with good source material. Let’s say I’m around the middle of the bell curve.

I didn’t accept this job and didn’t see the source material (not because of the time of year, mostly because I knew the job would be a rathole), but I can make a few guesses about it. It’s discovery material for litigation. It’s internal company information containing a lot of insider lingo that is not explained anywhere and will often pick up where another document left off. The job is being parceled out willy-nilly to a lot of different translators, by a coordinator who quite possibly does not read Japanese, so there may be no attempt to give each translator a cohesive package (assuming that would be possible at all).

In short, not great source material.

As to motivation, this agency pays OK, but not magnanimously. They do nothing to cultivate personal relations with translators. So there’s no special motivation for a translator to go the extra mile in terms of quality or quantity. If anything, the reverse. Although I don’t know much about the agency’s inner workings, I get the impression they do not have much of a translation-checking process. And for a job this big, on such a ridiculous deadline, it would be impossible to enforce consistency across all the translators, so even a translator inclined to do a top-notch job would know that the work would wind up looking like a hodge-podge anyhow.

So all around, I’m guessing they’d need 100 translators who are available on Christmas Eve. If they’ve got a checking process in place, 20 or more checkers. It takes a hell of a lot of chutzpah to get a call from a client who asks for 250,000 words in a couple of days and say “OK.”

I wonder if they’re trying to compete against machine translation. That’s a chump’s game. I wonder what it would take for them to say “no” to an unreasonable request from a client.

Finally, I wonder why the client was making this unreasonable request in the first place. It is possible that there really is such a short window between when the client received the documents and when they need them translated, but I’m doubtful. It’s possible that the client has been trained to have unreasonable expectations through previous contact with this agency. It’s also possible the client said “as soon as possible” just out of habit, and the agency has responded by treating it as an urgent request.

Characters per word

In Japanese-English translation, it’s useful to be able to estimate how long the English translation will be based on the number of Japanese characters in the source document. If you bill by the output word, you want to be able to give the client a cost estimate upfront, and if you estimate your productivity by the output word, you want some idea of how long a job is going to take.

Early in my career, I arrived at 2.2 Japanese characters per English word as a rule of thumb; I get the impression that a lot of JA-EN translators use a number around this. If a client asks me for an estimate, I’ve used 2 characters per word just to cover my ass.

Back when I started out, most jobs arrived by fax, so it was a lot of work to get an accurate character count on the source text. These days, most of my work arrives in some sort of live text, so it’s a lot easier. For the past year, I’ve been tracking the length of both the source and target documents when possible (my desire to track data like this is part of the reason I’m dissatisfied with every job-tracking app out there), and have come up with some average values for characters per word that surprise me.

  • Client A (newspaper articles, corporate communications): 2.66 characters per word
  • Client B (press releases): 2.92 characters per word.
  • Client C (discursive programming documentation): 3.10 characters per word
  • Client D (verbose corporate procedures): 3.16 characters per word
  • Client E (variety): 2.28 characters per word

There are a few jobs in my archive with more than 4 characters per word. I’m not sure if I’ve become a more economical translator with age—I’d like to think so—but perhaps I need to rethink my rule of thumb. I’d like to know what numbers other translators come up with.