The Need for Speed – Olympics Edition
The Olympics were great, weren’t they? For 3 weeks we watched athletes from all over the world push the limits of what is humanly possible. For many Olympians, such as sprinters, speed is the ultimate goal and they train for thousands of hours and invest countless resources in the hopes of shaving mere milliseconds off their times. In fact, between 1896 and 2012 the world record for the 100 meter dash improved by only 2.37 seconds. That’s right; in 116 years and despite incredible innovation in nutrition, technology, and sports medicine, sprinters only managed to improve their times by an average of two one-hundredths of a second per year.
Don’t get me wrong. The point here is not to malign the best athletes on the planet, but rather to demonstrate that when we’re talking about human endeavor, there are very real physical limitations that constrain what we can do.
Now, since you find yourself reading this blog post, the topic of speed may have you thinking about your organization’s translation needs. And, hopefully, the topic of human endeavor has you thinking about the translators behind those translations. If you didn’t make that second connection, perhaps it’s not surprising. Because most buyers of these services purchase them from language service providers (“LSPs”) rather than individual translators, it’s easy to forget that there are people behind your translations, and like all humans, there are limits to how many words an individual linguist can translate per day.
The Laws of Physics Apply to Both Sprinters and Translators
When considering how fast linguists can translate, it’s helpful to conceptually divide their work into concrete and abstract components. The physical, concrete aspects of a translator’s job are, in basic terms, reading the source text they need to translate and then writing the translation down. Everything else (i.e., the abstract components) is a function of how much thought they put into translating each word, sentence, or paragraph.
Let’s look at the physical limitations of these components. According to Wikipedia, the world record for typing speed was set in 1946 when Stella Pajunas typed 216 words in one minute on an electric typewriter. As of today, the fastest typists in the world can sustain average rates of about 170 words per minute (peaking at about 212 wpm). The average typist, on the other hand, clocks in at only about 41 words per minute and makes 8 mistakes for every 100 words typed.
With respect to reading, Wikipedia says the average reading speed in 17 different languages is about 184 words per minute. Reading more than 900 wpm is physically impossible due to the anatomy of the eyeball, and even though speed readers can achieve speeds of around 700 wpm using “skimming” techniques, there is a lot of debate as to how much comprehension is affected. In any case, it is probably safe to assume that most buyers of translation services don’t want to risk that critical information was lost in translation because their translator only “skimmed” the source text.
In theory, then, the average person needs at least 5 minutes to read and comprehend 1,000 words and then another 25 minutes to type the equivalent… or, put another way, a minimum of 30 minutes to “translate” 1,000 words, assuming they put zero thought into what they’re doing. The point here is that there exists a minimum baseline for how fast a translator can physically work, albeit totally impractical.
More importantly, this theoretical minimum for translation time creates a good conceptual contrast for the real work that translators do. Hopefully it doesn’t come as a surprise that translators are doing far more than simply reading and typing, because the truth is that their value is not in word processing. Instead, what translators really do is constantly deliberate on how to best convert your source text into another language. Making these decisions requires a substantial amount of thought and is what takes up the biggest amount of time a translator spends. It is also the single biggest factor in translation quality.
Think about your own process for writing an email, letter, or report. If you are like most people, you probably create a draft and then go back and polish it. And, if it’s for your boss, you probably polish it again, and then polish it some more. The same principals apply to translators. The very best translators polish their work too…and the more they polish, the better it will be.
PEDs – Performance Enhancing Devices
Since life in today’s world means we all need information as quickly as possible, how can we speed up the translation process if there are finite limits to how fast a linguist can translate?
To start, let’s quickly nix the idea that you can simply tell the translators to “work faster.” Sure you can tell them that, but what are you really asking them to do? If we assume that translators are already reading and typing as fast as they naturally can, then the only real variable is how much thought they put into their work and how much time they spend polishing their work. Or, as a cynic might say, asking a translator to translate faster is sort of like asking them to pay less attention to what they’re doing.
Obviously, if quality and accuracy matter at all, this is not an option in most situations. Fortunately, there are many steps an LSP can take to turn your translations around faster while still providing the quality you need.
CAT Tools and Translation Memories
So far we’ve discussed this topic in simplified terms, and as such we’ve assumed that a translator would have to start from scratch with each and every translation job. You may have already remarked to yourself “…but we re-use a lot of content. And sometimes we’re just updating files that have already been translated in full!” The intuition that there is a more efficient way to handle repeated text is what spawned the idea of Translation Memories (TM) and the CAT tools that enable them. A TM is a database that stores already translated content side by side with its translation. Typically, any new content you send for translation will be analyzed against this database so that translations for sentences that have already been processed before can be re-used with minimal processing time.
These tools have become more and more powerful over time, and today they are indispensable to just about any translation workflow. That said, the extent to which a TM saves time (and cost) is dependent on whether: a) the new content has been translated before, and b) the previous translations are good enough to reuse. It doesn’t save any time if translators have to carefully read and “fix” the matched content. And of course, if none of the new content has been processed before, there’s nothing in the TM that will give a translator a head start.
You know what can read, write, and think faster than any human on the planet? A computer. If you truly have no time to devote to translation, your only option may be instantaneous machine translation (MT). Now, before you ask why this article is necessary if MT is a viable option, let’s turn back to the topic of quality. There are very, very few scenarios in which raw MT output (meaning no human has ever looked at it or modified it) is good enough to be used out of the box. Don’t believe me? Here’s an example of some out of the box MT output taken from a recent article on BBC.com:
'Imagine a far flung land where you can catch a ride from the Jackie Chan bus stop to a restaurant called Translate Server Error, and enjoy a hearty feast of children sandwiches and wife cake all washed down with some evil water.'
If you’re ok with risking that your translated content reads this way, then MT is for you. Otherwise, for virtually any MT driven translation program, there will also be a human “post-editing” step needed to make it even marginally acceptable. What is post-editing (PE)? Well, to re-use some of my own words from above, you could describe it as “deliberating about how to deal with the information in the source text and best render it in another language.” So again, while the computer may successfully eliminate the time needed to physically type the translation, your MT + PE time lines are still going to be dictated by how much thought a translator puts into the translations. Often times it can take just as much time to translate from scratch as it can to fix MT translations like the example above.
In 2012, the Jamaican Men’s track team set a world record for the 4x100 relay with a time of 36.84 seconds. By contrast, the world record for the individual 400M (for men) is 43.18 seconds. The point here is that things go faster when you break them down into manageable pieces and divide the work among capable people. Translation is just the same, even if it’s not really analogous to a relay. Since translators can work concurrently, 2 translators could potentially tackle the same 10,000 words in half the time it would take a single translator.
This is, by far, the safest way to speed up translation work, but it is not without its risks and limitations. First of all, you have to have enough content to make dividing it up worthwhile. For some projects, it could take more time to break apart, organize, assign to multiple translators, and re-combine some text than it could to just give it to one translator to begin with. Second, there are limits to how many translators can be reasonably assigned to a project. For example, if you have a 100 page document that needs to be translated overnight, your LSP could theoretically give each page to 100 different translators...but you probably wouldn’t like the result.
Fortunately, LSPs like LanguageLine are experts at building custom solutions to turn around translation projects within the time frame needed at the quality levels required. Glossaries and CAT-Tool enabled Term Lists can ensure that translation teams use the same words for important concepts. They also employ copyeditors and proofreaders to edit the translation so that it reflects a common style and voice when multiple translators are used.
The above notwithstanding, buyers of translation service should understand that in most cases, the very highest quality will be achieved with one translator, one editor, and one proofreader. Translation is a complex task fraught with subtleties, and how an individual translator deals with these subtleties can manifest in real stylistic differences compared with other translators who may also be working on the same content. Thus, even though dividing work among multiple translators is the safest way to speed up translation without sacrificing quality, when the need for consistency is paramount, you will want to speak with your LSP about what solution is best for you.
The Closing Ceremony
Ask anyone in the industry what “standard” throughput rates are for translation and you’ll likely get different answers with endless caveats. Most experts agree, however, that under the right conditions you can expect to achieve between 2,000 and 2,500 words per day of new translation. It’s not worth wasting time listing the caveats that accompany a figure like this because the exact words per day rate you can expect for translation is not the point of this article. The point is that there are very real limits to how fast a linguist can translate, so unless you want to completely sacrifice quality, it is imperative to work with an LSP that can build custom solutions to meet your quality and timeline goals.
(By the way, using the metrics we discuss above I estimate it took me about 35-40 minutes to physically type this ~2,300 word blog post, yet it took me at least 8 hours to write it. How long do you think it should take to translate it?)