Human Translation vs Machine Translation: The Numbers


Photo by William Brawley on Flickr

Faced with translating Atlas Shrugged (565,000 words) or Casanova’s memoirs (1.2 million words), or letting a machine do it with human post-editing, which would some folks choose? Depends on if you’re a reader or the company paying the human translator(s). Human translation vs machine translation: The debate rages.

But is it really either/or? And does it really justify any rage? Machine translation, and particularly Enhanced Machine Translation, will change how human translation works, but is probably one of the few career choices that won’t be threatened by those dreaded machines. For extremely critical text like legal documents or medical instructions, or for the subtle emotional and meaning-laden content in a novel or an autobiography, we may never be able to rely on MT in any form. Human language changes and evolves so quickly that what something means today might not tomorrow. Until the 1970s, ‘gay’ meant happy and merry.

Cultural differences, values and histories also make a difference. A cultural joke or comment that relies on the receivers’ knowledge and understanding of Buddhist thought (for example, in Thailand or Japan) falls flat with cultures unfamiliar with it. The translation challenges faced by popular media projects demonstrate how it’s more than just knowing the right word. Sometimes there isn’t a ‘right word.’

 

(Hu)man vs the Machine

Man with hands over his face in frustration

You want me to translate the entire Harry Potter series? Into Tagalog? BEFORE CHRISTMAS???
Photo by Allan Rotgers on Flickr

The challenge organizations think they face when needing to translate large bodies of content is human translation vs machine translation. In fact, it’s both: They need to consider their budget and adjust for humans accordingly because even if they pick any form of MT, including Google’s GNMT (Google Neural Machine Translation, touted to be more accurate – eventually – than other forms of MT) it will still need to be post-edited by humans.

Human translation may make more sense and be cost-effective for companies with a small, finite amount of content, but for much larger amounts – like 200,000 words in eCommerce product descriptions that require five language translations (one million words total) – the time and cost will be too prohibitive. A good second-level-or-higher machine translation job combined with human post-editing will be far more cost-effective, without sacrificing quality.

 

Related: Post-edited machine translation among the fastest-growing segments of the language industry

 

The four levels of translation

  • Machine Translation. The basic short-text capabilities of ad-hoc online translators and widgets.
  • Enhanced Machine Translation. Real-time translation run through multiple translation engines and corrected for spelling, accents and abbreviations, and contextual and lexicon rules applied.
  • Machine Translation and Post-Editing. EMT-translated text required light post-editing, MT-translated will require full post-editing.
  • Human Translation. For projects in which meaning, nuance, and above all accuracy is of supreme consideration.

Because MT is still evolving, it has never met science’s breathless promises of being indistinguishable from human translation. As such, how much to charge for post-editing vs. regular translation?  It’s still a matter of debate for translators, but part of the math is easy enough to do: You can read through more pages of text in an hour than you can translate. This is assuming that you started with an MT system more competent than your typical on-line free translator; otherwise your page count dwindles even further. That type of free translation is okay for shorter, more straightforward text but even then it has famously messed up the end product. The language pair is important too; Italian to French will be easier for the machine to translate compared to a pairing like English to Korean.

The demand for translation services, especially at the enterprise level, far outpaces the supply of human translators and it’s not expected there will be enough of them in the future to fill it. Machine translation combined with human post-editing, therefore, is an area of high growth opportunity for Language Service Providers (LSPs). (Common Sense Advisory, 2016)

Human translation vs machine translation: Do the math

Let’s see how much it costs to translate a million words for an eCommerce site with an experienced translator, who  can handle 5,000 of an estimated 3-5,000 words a day.

Number of words:                                           1,000,000 x

Average translation cost per word:                  .06

Total human translation cost:                    $60,000.00

How long will it take?

1,000,000 words / 5,000 =             200 days of work

200 days x 8 hours =                     1,600 hours of work (6.5 months)

Have you got the budget? Can you wait that long?

Related: Language services and technology market to surpass US $40 billion in 2016

 

With machine translation and post-editing

There are generally two types of machine translation post-editing:

Light post-editing – Fixing the most obvious errors and making the text easier to read, if not perfect. LPE can cover up to 20,000 words per day.

Full post-editing – For producing human-quality output. If you don’t want your website to look as sloppy as your competitors’, you’ll choose this option.

Since our theoretical experienced translator can already translate 5,000 words a day on her own, we’ll figure on 8,000 post-edited words a day for her.

1,000,000 words / 8,000 =             125 days of work (a little over four months)

125 days x 8 hours =                      1,000 hours of work

1,000 hours x $45/hr =                  $45,000

Forty-five dollars an hour is a suggested full post-editing base standard rate.  Our intrepid post-editor can do the job for $45,000 in three-and-a-half months. That’s a 25% decrease in cost.

That’s a very speculative number based on an ideal speedy, knowledgeable translator. Just to name a few actual unknown factors for a given translator include individual competency and experience; electronic vs hard copy source text; actual hourly rates based on whether the translator works for an agency or freelances; and the fact that post-editing for machine translation is still in its infancy.

So when it comes to human translation vs machine translation, when you have a large body of work that doesn’t require the extreme attention to detail, meaning and nuance that fiction and autobiographies do, MT + post-editing may well be more cost-effective and accurate.

 

 

Yappn Corp is an enhanced machine translation company offering translations in 67 different languages who utilizes several translation engines rather than just one, and expects never to send translators to the unemployment line. For desperate cries for help please contact sales@yappn.com or call us at +1.905.763.3510 x246.

 

Download our new white paper!   WATCH YOUR LANGUAGE! The benefits of machine translation over single-translation engines

 

Never miss out on another Yappn blog post! We love language, translation, and the cool machine translation technology driving it. Not to mention eCommerce, Asian growth, and how we’re all learning to communicate better with each other. Sign up for our new post email notifications today!

 

 

 

Written by Nicole Chardenet, Sales Development Rep at Yappn.

Share This Post