August 9, 2021 By Olena Polissky Comments are Off

“Language is the source of misunderstandings.”
Antoine de St. Exupery, The Little Prince

In our daily work, we often see our clients confusing CAT tools with machine translation, so in this article, we would like to explain what exactly CAT tools are and how they are different from machine translation.

Let’s first look at machine translation.

What is Machine Translation

Machine translations (MT) are translations that are carried out entirely by automated computer programs. While these have the advantage of being quick and easy to produce, they are often dreadfully inaccurate, as computers, unlike humans, are unable to create meaning from combinations of words that they have never seen before.

Let us translate the following two sentences to explore this weakness:

“What happens when a sentence is translated multiple times by a machine? Let’s take a simple example and find out.”

These should translate seamlessly into French, a language similar in structure to English. However, the second of these sentences is conversational in tone and contains the prepositional phrase “find out.” When the text is translated into French, and the French translation is translated back into English, we get:

“What happens when a sentence is translated several times by a machine? A simple example to know.”

The distinction between “multiple” and “several” collapses in French, resulting in its absence in the reverse translation. Even worse, the meaning of the prepositional phrase has been lost in translation, resulting in a sentence fragment. After a second iteration of translating into French and back into English, we get:

“What happens when a sentence is translated several times by a machine? A simple example of knowledge.”

While the first part of the text remains unchanged, the meaning of the second section has not only been distorted beyond recognition but has taken on a new dimension. The machine translation has more or less captured the syntactic meaning of the inputted words, but it did not capture their collective sense.

Considering how terribly awry an elementary machine translation may turn out, it is clear that using machine translations for longer texts can lead to disaster.

When to Consider Machine Translation

Even though machine translation has improved dramatically lately, the results it produces depend on the language combination, the type of content, and its level of technicality, as well as numerous other factors.

For example, machine translation doesn’t work well with marketing or legal texts, i.e., anything where there are layers of meaning; also, texts with lots of references to UI elements in bilingual form, e.g., “Press the Save button”->”Appuyez sur le bouton Save (Enregistrer)“.

Machine translation works very badly with the latter since they are non-deterministic (the engine will translate X->Y one segment, and then X->Z the next).

Also, machine translation won’t work well on documents for which there isn’t a good corpus of high-quality translations.

On the other hand, straight, dry technical documents are the best candidate for machine translation. Especially documents that are periodically revised (e.g., car owner’s manual, or the product documentation for instance), so the machine translation is already well-trained.

It is important to approach machine translation with caution. A neural engine is only as good as its training corpus, and the best results are produced with a custom MT engine trained on a good (i.e., high-quality translations) corpus, specific to the client. And even then, it still has its deficiencies and cannot be used on its own without thorough Machine Translation Post Editing conducted by a qualified and experienced human translator.

What the Translators Say About Machine Translation

We have collected some feedback from the translators who have experience with Machine Translation Post Editing (MTPE). They share their thoughts about the quality of machine translation.

“There are many misconceptions about MTPE. People think it is very time-saving, that it is an efficient way to cut expenses, and think we can work on MTPE projects at a proofreading rate. That is not the case at all. All MTPE material requires attentive revision, and often times complete restructuration, which is time-consuming. Not to mention that the research part still needs to be done, which is also very time-consuming.”
Ninon, English to French translator

“I’ve worked on projects where the end client switched to MTPE to save money, so basically skimped on the training corpus, the custom engine, etc. as well. It was faster to translate from scratch than to try to post-edit this hot mess!”
Martin, English to French translator

“MT is most valuable for any kind of technical vocabulary, usually quite good, but often not so good on usage – placement of verbs, agreement of verbs, etc. So, I would say sentence structure and style are what require the most human intervention.”
MacLeod, Spanish to English translator

“The quality of translation from NMT is extremely poor. Every sentence has to be retranslated.”
Hélène, English to French translator

According to the Canadian Translators and Terminologists and Interpreters Councils, as recognized by machine translation software designers,

“… even today’s most sophisticated software, however, doesn’t … possess the skill of a professional translator. Automatic translation is very difficult, as the meaning of words depends on the context in which they’re used. … it may be some time before anyone can offer human quality translations”

What Are CAT Tools and What They Do

Computer-assisted translations are not nearly as similar to machine translations as they may sound. A computer-assisted translation is a human translation that is produced with the assistance of Computer-Assisted Translation (CAT) tools, the systems which help the translator to manage their translations, edit and store them for future use, thus greatly improving the quality of translation and streamlining the efficiency of human translation.

CAT tools serve multiple functions in the translation process. Firstly, they treat sections of texts as segments, saving both the source text and the translated text as a translation unit in a translation memory. Every time a translation unit reappears in a text, the same segment of translated text pops up, saving the translator time.

The second major function of a CAT tool is, with its terminology management software, to enable translators to search for specific terms from a word bank in order to ensure consistency of terminology.

Another important benefit of CAT tools is their preserving formatting of the source document.

Let’s have a closer look at these benefits.

The Benefits of Computer-Assisted Translation (CAT) Tools for Human Translation

1. Translation Memory for Increased Translation Efficiency

One of the most important features of a CAT tool is Translation Memory (TM). Translation Memory is a database of the segments in the source language and their translations in the target language.

While working on the project, the translator has an option to reuse previous translations which significantly improves productivity and quality of translation as well as ensures consistency of terminology with previously translated material.

The important thing to understand is that it is a human translator who decides whether to use a previous translation, modify it, or replace it with a different translation.

The screenshot below is what a translator typically sees when working on the project using Translation Memory.

2. Termbase for Quality Terminology Management

According to statistics, terminology discrepancies are the most common cause for translations needing to be redone, causing the release to be delayed.

“Managing terminology improves consistency, and that means the quality of your product and the quality of your customers’ experience. It can have an immediate impact on the success of your product or services in the marketplace.”
Uwe Muegge, Head of Terminology, Global Marketing at Facebook

Termbase is a terminology management tool used to store terminology specific to each organization. It is not a dictionary but rather a reference system indicating what is the preferred translation of such and such term, phrase, trademark, etc. is.

It may also indicate which terms or brand names should never be translated. The termbase reminds the translator of specific ways of handling certain terms and whether they should or should not be translated. As a consequence, the termbase helps to produce a more accurate translation.

The screenshot below is what a translator typically sees when searching a termbase.

You can find more information on terminology management in this article.

3. Handling Multiple File Formats in the CAT Tools

Most CAT tools today allow the translator to process the files of different formats without having the software associated with those formats. This presents a huge benefit of using CAT tools for translation as it significantly saves time and effort for post-translation Desktop Publishing (DTP).

In the screenshot below you can see some file types supported by SDL Trados Studio, one of the most commonly used CAT tools.

What the CAT Tools Are Not

It is important to note that CAT tools do not replace human input in computer-assisted translations. It is the human translator who chooses whether or not to use the segment of text that the translation memory proposes, and it is a human who creates the original translation units and uses the terminology management software. The computer itself serves only to memorize translation units and enable the human translator to quickly search for specific terms in the word bank.

The table below sums up what CAT tools do and what they are not.

As you can see, the difference between machine translations and computer-assisted translations is simple: the former provide low-quality, inaccurate translations that fail to capture tone, idiomatic language, and dialect, while the latter are human translations made more efficient through the use of modern tools, consideration of geographic variations in language use, and creative selection between multiple possible variations of a given section of text.

At Art One Translations we understand the significant gains to be made through the use of CAT tools, which contribute to our quality translations without removing the human element from them.

We use a number of the most sophisticated translation tools to assist our expert human translators to achieve the ultimate result – an accurate translation that sounds like an original.

Comments are closed.