Machine Translation: Does Not Compute?

Running the numbers

Many businesses today looking to save money on translation services will consider using machine translation tools.  At first glance, it seems ideal: it’s quick, inexpensive, and machines are more exact than humans, right?  Wrong, not all that glitters is gold!  Today we will examine the pros and cons of machine translation.  We will look at how it functions, who uses it, and why.    

It is our goal to present a balanced perspective.  This isn’t about *not* using machine translation but using it when appropriate and with proper safety measures like post-editing by a professional translator.  If the results are of good quality and suitable to the application and the client’s needs, then machine translation can be used to great effect.

What is machine translation?

Machine translation involves running language data through a computer program, whether specialized or basic, to produce translated text.  Machine translation is often used with post editing, in which a professional translator edits the data, correcting for linguistic problems, nuances, grammar, and localization.  Now Artificial Intelligence, AI, is propelling the world of machine translation to a new place, Neural Machine Translation, or NMT.  An NMT engine enhances the process of machine translation using Artificial Intelligence Networks to predict the likelihood of a sequence of words in the translated text from the analyzed source text.

It is hard to overstate how much machine translation technology has advanced in recent years.  Not long ago, it was generally unhelpful and of poor quality.  Now, advanced programs have emerged for large institutions and universities with capabilities for specialized terminology and regular updates.  For the end consumer, there are apps based on both an existing body of data and real-time crowdsourcing.  Where do these bodies of data come from?

Building blocks of machine translation

A machine translation program draws upon a corpus, which is a large body of texts and audio.  These can be monolingual, bilingual, even multilingual.  Two major corpora are the British National Corpus (BNC) and the European Corpus Initiative (ECI).  There are open corpora that are still being built, and closed corpora, which are complete.  These bodies of language form the basis of several types of machine translation: 

Statistical machine translation uses statistical science to produce output based on a corpus. 

Rule-based machine translation draws from dictionaries and lexicons.

Example-based machine translation compares two languages in a bilingual corpus.

Hybrid machine translation combines the use of corpora with checking the data against the rules and grammar.

Post-editing the output consistently not only improves the translated copy but trains the machine translation engine to improve its capabilities. 

What are the pros of using machine translation?

Machine translation is often the most convenient option short of putting the task to a bilingual employee, which is not a great idea if you are unsure of your employee’s capabilities. 

Cost is a factor.  Sadly, most of the time you get what you pay for.  Seeking savings on projects can quickly become even more expensive and time-consuming if the output is not revised. 

A computer program is often faster than a human translator so higher volumes can be addressed.

AI can be taught to reuse collected data and update itself.  Options to improve quality like post–editing and training capabilities for engines are available as well.

What are the cons of using machine translation?

Machine translation is inexact and can be dangerous if used in some industries, like medicine. 

Grammar is a major concern.  There are languages with or without gendered words, and plenty of different sentence structures.  Homophones, words that are spelled the same but have different meanings, present a challenge when using machine translation.

Some language pairings, such as Spanish to English, produce better results than language pairs that are very distant.  Compare this to translating English to Malay, or Arabic, both of which are challenging. 

Nuance is often lost with machine translation.  Cultural differences and lack of localization can ruin a machine-translated project. 

The output of a machine translation needs to be evaluated by human professional translators to validate that the quality is good enough for post-editing or even to be used as is.

How is machine translation used? 

Machine translation is useful for the discovery phase in the legal process.  Attorneys search through many documents for resources, choose what they need, and run it through the translation program.  It can then be post-edited by a human translator skilled in legal translation. 

Businesses responding to requests for quotes or proposals from foreign clients use machine translation.  They require speed as there are other bidders but quality because the translation must be accurate.  Post-editing done by a professional human translator shows its merit here.

Large ecommerce sites benefit from machine translation.  Quality programs learn from recycling similar words and phrases and are routinely updated.

Governments large and small publish high volumes of content online and use machine translation and post-editing to make it accessible to a multilingual constituency. 

Universities use machine translation and post-editing for volumes of written and audio content, from textbooks to transcripts.  Universities are putting a lot of brainpower and time into research to make strides in the world of machine translation.

The world of travel and tourism can get accurate, attractive content from a quality machine translation program for use on websites, brochures, and travel guides.  Machine translation also comes in handy for consumer-generated data like reviews.

What to do with the data?

Machine translation is gaining in accuracy and ease of use through the development of specialized computer programs or engines and strong corpora.  However, the risks are real and cannot be taken lightly.  Machine translation is of greatest benefit when used judiciously evaluated, improved with post-editing and localizing done by a professional translator.  For serious projects with unacceptable risk levels, use human translators start to finish.

This is the only way to ensure the success of a translation project so that you can be confident in the product.  A professional translation agency can guarantee accuracy and localization to improve your clients’ experience and your brand’s image.