Why Argos TSP refuses to use machine translation

Everyone has used Google Translate or Reverso at one time or another. These machine translation tools can be very useful as they allow anyone to have a basic conversation with a Latvian or decipher a text in Hungarian or even write a short message in Korean. However, their reliability, although improved in recent years, remains low as they produce sometimes amusing, often ridiculous, and always amateurish errors.

This article is not intended to deny the usefulness of these tools. While their use may be appropriate in many situations, we consider them unsuitable for professional translation in the fields in which we specialise and, more generally, for any translation intended for publication in any form. It follows that these tools should only be used for informal conversations or for skimming a text written in a foreign language.

Before continuing, we need to define two key terms in order to better explain what we are talking about:

So why doesn’t Argos TSP include machine translation in its translation process and provide post-editing services to its clients?

The main pitfalls of machine translation

The quality of machine translation is closely linked to the quality of the translations used to “train” the computer program

A machine translation system must analyse thousands of existing translations in order to generate an output. However, the input fed into these tools varies wildly in quality. As they use translations that are often too literal or even wrong, we consider the translations produced by these systems to be below professional standards.

The computer program misunderstands the source language

Machine translation systems fall into most of the usual translation traps. Complex concepts, atypical expressions, industry-specific terms, puns, jokes, innuendoes and idioms are endless sources of errors that only a good human translator can avoid.

The document to be translated is not always perfect

The author of the original document rarely asks themselves whether their text will easily cross the language barrier. Content producers are mainly concerned with the message they want to convey in their own language, even if it means sometimes speaking in a deliberately vague way so as not to say too much… and translating vague ideas is often an impossible task without the help of an expert. In addition, original documents sometimes contain syntax, grammar or spelling errors that machine translation systems are incapable of detecting and correcting.

The computer program does not necessarily take into account the context

Since many words have multiple meanings, it is common for the same term to be translated in different ways depending on the context. However, the program does not typically capture the context of each sentence.

Machine translation has difficulty with concepts it has never encountered

If you live or work in a multilingual environment, you may have noticed that it is almost impossible to find a satisfactory translation for certain concepts, in certain contexts, even after long searches on the Internet. If you can’t find them, it’s because only experienced translators have the solution which is not available through a cursory search on the Internet. Just like any non-professional translator, machine translation systems have no choice but to generate literal translations that a person who does not speak the source language may not understand.

Machine translation can be extremely repetitive

When writing a text, and this is especially true for translation, it is important to avoid excessive repetition and redundancy which requires being able to deftly vary the vocabulary used. If you examine a computer-generated translation, you’ll find that the computer program has not done what a professional translator would have been sure to do. The computer program translates one sentence after another, taking no or too little account of the context.

The computer program does not ensure the consistency of important terms

The program does not know which terms are important and must remain consistent throughout a document or even across multiple documents. If you have ever had the opportunity to read a computer-generated translation, you will surely have noticed that some key terms are translated in several different ways. These inconsistencies can create problems of clarity and even comprehension. It can be a bit confusing when, in the same document, different terms are used to refer to the same technical concept.

Post-editing is not such a good idea

Now that we have outlined the main reasons why machine translation is generally unsuitable in a professional context, it may be tempting to think that simply having a professional translator proofread a machine translation will quickly result in a good translation. Could the solution be post-editing? In our opinion, this is a bad idea.

Translators are often misled by machine translation

It’s tempting to think that if a machine translation contains errors, a translator will just simply correct them. This is not necessarily true. Each and every error in meaning made by the computer program sets a trap for the translator. It is inevitable that errors will be missed, even after the document has been proofread several times. Some machine translations can even seem natural, creating a false sense of security. As it is a tremendous source of errors, we believe that the post-editing model is not suitable for professional translation.

Correcting a machine translation often means redoing it from scratch

For professional purposes, the quality of computer-generated translations is so low that it is better to start from scratch and redo the entire translation. Correcting each sentence so that it accurately reflects the meaning of the original and sounds just as natural simply takes more time than re-translating it. The time savings that would justify the use of post-editing is therefore only a mirage. Despite those who think that machine translation is the future of the industry, a good translation is only possible if you have a good translator.

Underpaid professionals have little motivation to do a good job

The downward price pressure exerted by traditional translation agencies and the numerous corrections that would be required to obtain a translation of “acceptable” quality make post-editing projects unprofitable for translators. It is therefore understandable that translators concentrate on improving the style of machine translation (if they manage to do so without rewriting everything) and forget to correct the meaning in order to spare themselves the indispensable research that is part of our job. Post-editing gives often poorly paid translators the feeling of being only half responsible for the translation. It is not in their interest to do a meticulous job due to a lack of time and motivation.

It is clear that post-edited translations are not up to professional standards. Why then do traditional agencies continue to provide this service?

Post-editing allows translation agencies to increase their margins by simply having poor quality machine translations proofread by underpaid translators while still charging their clients high rates, often without their knowledge. This poses the same ethical problem as when a restaurant serves supposedly home-made food when it is actually store-bought frozen food.

Sometimes it is the client who, being unfamiliar with the translation market, pushes translation agencies to lower their rates, leading them to resort to machine translation or post-editing. Procurement departments that regard translation as one of the expenses that needs to be reduced are rarely aware of the danger their policies pose to their company’s image. We feel it is necessary to make company executives aware of this. Quality translation has a cost—translation agencies have to be able to pay their human translators and human proofreaders properly and cover their operating costs. If a translation is underpriced, there must be a problem somewhere, namely hidden post-editing, no proofreading, inexperienced translators, sloppy work, etc.

Who are the biggest losers in the post-editing boom?

Are good translators an endangered species? If they continue to be exploited by a machine translation post-editing process that allows large translation agencies to maximize their profits, good, highly qualified translators could well choose different professions that properly reward their work and expertise.

The widespread use of post-editing is certainly problematic for translators, but the biggest losers are the clients who use translation services! You have lost in terms of quality and this trend will continue. In our opinion, your company’s image deserves better than that, and your staff’s time is far too valuable to be wasted on endless proofreading that wouldn’t be necessary if you used a boutique agency like Argos TSP.

Does this mean we should stop using computers to translate?

Certainly not! Computer-assisted translation (CAT) is an extraordinary tool. Argos TSP remains committed to a model based on the expertise of human translators using the very best translation tools.

It will come as no surprise that we do not rely on post-editing at Argos TSP. We do not underpay translators based in countries where labour is cheaper to proofread computer-generated translations. This does not mean that we haven’t kept up with the times. We do indeed use tools that help us to ensure the consistency of our translations. Simply put, we continue to provide our loyal customers and those using our services for the first time with documents that are translated with care and proofread by experienced professionals equipped with the latest technology.

If you too would like to learn to trust a translation company again, contact Argos TSP!