From Google Translate to specialist tools, there is now a range of technological options for those seeking to translate text.
However, what value can these bring to a project? Can they ever take the place of a human translator? When is it appropriate to use them?
We look here at the ways machines are involved in carrying out translation tasks and weigh up some of the respective benefits and problems.
Machine translation applies software to translate text from one language to another and programs are gradually becoming more adept at understanding the intricacies of languages, phraseology and idiomatic differences, rather than simply substituting one word for another.
As software advances, it is gradually providing more opportunities to carry out swift and sometimes free translations as and when they are needed, particularly for cases in which concerns such as speed outweigh the need for absolute accuracy.
One of the most well-known machine translation tools is arguably Google Translate, which provides instant translations for any combination of 80 languages.
It uses algorithms to translate words, sentences and even web pages, looking for patterns that will provide the best translation. The detection software makes intelligent guesses and employs information gathered from patterns in documents on which human translators have worked.
Learning from patterns collected from millions of sources to translate text is called statistical machine translation, with systems incorporating feedback from users. Although it uses intelligent guesswork, not all translations will be completely correct and the level of accuracy varies between languages.
Another form is known as rules-based machine translation and involves the use of grammar and language rules in combination with dictionaries. It can be very useful for formal and specialised documents in which professional or specialist terminology is used, but can be quite limited in its scope.
Combining human and machine translation elements is known as computer-aided or computer-assisted translation (CAT) and it can help you to manage high volumes of text by automating part of the process.
Packages may be customised and updated by translators to ensure that they are responsive to the material being fed in, enhancing productivity and carrying out tasks stored in translation memory.
Where machine translation is completely automated and involves no human input, computer-aided translation involves one or more people in the process.
CAT can be particularly useful for technical, financial and legal translation for which consistency of terminology and standardisation is important. Work should be carefully reviewed and checked, but translators can find that computer-aided translation significantly reduces the number of repetitive tasks they have to carry out, thereby improving work rate.
At their most basic setting, CAT tools break down documents by sentences. Once a sentence is fed into the translation machine and processed, the CAT tool will search the document for similar or identical sentences and insert or recommend its translation based on percentage of similarity. This can range from 55% to a 100% perfect match.
Depending on the CAT software, the translation returned may be colour-coordinated to grab the attention of the human translator, who may edit it before sending that entry to the translation machine, for it to ‘remember’ for next time.
By incorporating human corrections into the machine translation process through an interactive framework, it is possible to deliver accurate and high-quality translations within a shorter period of time.
When is machine translation appropriate?
Dating back to the 1950s, machine translation can be very effective when used in situations where text is formalised and tends to adhere to specific patterns that can be identified and fed into algorithms. For example, legal and government documents can be translated using customisable software to provide a relatively accurate version in another language.
If comprehension – rather than 100% accuracy – is the end goal of an exercise, machine translation can provide an effective way of getting rapid results. However, translation of sensitive documents – for which both the meaning and accuracy are essential, and an error could lead to high financial or other costs – is still likely to require significant human intervention.
Furthermore, in cases where grammar needs to be completely accurate or where there is a less formal, more conversational style being translated, then machine translation can fall short.
Specific problems with machine translation include disambiguation, where words have one or more meanings, named entities, and the translation of non-standard language, such as slang or colloquialisms – particularly where there may not be a direct translation of terms.
Crowdsourcing and machines
The internet is also providing a means to crowdsource translation, with some sites using algorithms to determine which translations are the most accurate.
Crowdsourcing is the process by which the capabilities of a large pool of people are leveraged to get work done or secure funding, and usually takes place online.
It is now being put to use in the field of translation to produce fast, cheap translations, as it means that dozens of people are able to work on a piece of text at once. In order to deliver quality, users of a site can rate each other’s work, ensuring that the quality of a translation improves.
However, crowdsourcing is of no use when the material that needs to be translated is confidential or sensitive, or in cases where the accuracy and attention to detail that a professional or specialist translator can provide is needed.
Additionally, there are concerns that novice translators might confirm rather than correct one another’s mistakes and required levels of experience for translating text are not always there when tasks are completed in this way.
In conclusion, machine translation is here and its reach is expanding, with a growing number of situations in which it provides affordable, instantaneous translation. However, there will continue to be cases where it is unsuitable and only a human translator will suffice.