German 1

Human translation in the era of machines and AI

Since the first experiments with machine translation in the 1950s, attempts to create a “fully automatic high-quality machine translation system” have been met with hopes as well as fears. Debates around the possibility of replacing human translators with computers go to the very heart of what language is – and what it is to be human. While for decades there’s been an undercurrent of anxiety in the translation industry that human translators might one day become obsolete, this hasn’t happened yet…Why is this, and why do professional human translators continue to outperform computers at almost every level? In this article, we’ll take a look at what machine translation is and how it works, the benefits and drawbacks of machine translation, and how to weigh up whether you need professional human translation, machine translation or a combination of the two for your localisation strategy.


What is machine translation and how does it work?


As the name suggests, machine translation is translation of a text from one language to another, performed automatically (and practically instantaneously) by a computer. There are several types of machine translation, but the main types used today rely on a large number of pre-existing translations arranged as parallel texts, known as a “corpus”. The computer analyses this corpus to find the probability that a certain string of words in the source language (called an “n-gram”) will be translated with a given string of words in the target language, choosing the most likely combination as its final translation. More recently, neural machine translation tools have been developed that have the ability to “learn” and improve their performance as time goes by. Note that “computer-assisted translation” isn’t the same as machine translation: this refers to a range of software tools used by human translators to ensure consistency and efficiency.


What are the benefits and drawbacks of machine vs human translation?


Computers can translate large volumes of text within seconds, so when speed is your main priority, machine translation is hard to beat. The pace at which a human translator is able to translate varies depending on factors such as the complexity of the text, but on average a professional will translate around 2000–2500 words a day. This also means that machine translation generally costs less, because even when you factor in a “machine translation post-editing” stage (known as MTPE, a check carried out by a human translator), the overall time taken to reach the final translation is considerably reduced.


On the other hand, because machine translation relies on a corpus, the quality of the translations it produces depend on the quality of the corpus. This also means that it will replicate any biases in the corpus, potentially leading to issues such as sexism in the translation output. For example, if you take a gender-non-specific term such as “the doctor” in English and use a popular machine translation engine to translate it into Italian, you get “il dottore”, the male doctor. Do the same thing with “the nurse” and you get “l’infermiera”, the female nurse. Problematic, right?


Another problem with the corpus-based approach of machine translation is the fact that it relies on frequency of terms and phrases, which means that the computer will struggle with anything novel or niche. Creative language, neologisms, or unusual technical terminology are all significant stumbling blocks for machine translation. Once again, we can use a well-known free machine translation tool to illustrate this: take the catchphrase from the TV show RuPaul’s Drag Race, “condragulations” – a play on the words “congratulations” and “drag”. Google translates this into Russian as “pozdravleniya”, which is simply the usual word for “congratulations” and has none of the humour, creativity or wordplay of the source text. The human translators who create the Russian subtitles for the show, in contrast, have come up with the perfect solution: “pozdragliayu”. In more technical fields, a human translator specialising in the relevant field will have the background knowledge to know exactly which word to use to translate a niche term, unlike a machine translation engine trained on general texts, which is likely to use entirely the wrong term or not be able to provide a solution at all.


Machine translation also lacks what we call “common sense”, which means it sometimes comes up with some nonsensical – and often amusing – translations. To give a real-life example, we recently edited a machine translation where the German word “Kater” had been translated as both “male cat” and “hangover” in the same text. Both valid translations in some situations, but not in content related to veterinary medicine! This wouldn’t cause any issues for a human translator, who would instantly know which word to use based on the context, but the statistics-based machine translation tool is prone to mixing terms up given that it has no real-world knowledge or understanding of the bigger picture.


What does the future hold?


Despite the issues we’ve described, machine translation is here to stay – and is likely to become a key part of the toolkit of many international companies in future, thanks to its speed and low cost. Used wisely, it can be a brilliant way to lift barriers to international communication at scale. While human involvement is still needed in most cases, the level of intervention depends on a number of factors, including the characteristics of the source text, the quality of the machine translation tool and the purpose of the final translation.


In a similar way, generative AI is already widely used outside the realm of translation and localisation, for example for content writing, website and app design and support chatbots. As with machine translation, human input is still needed in the majority of cases, particularly where anything non-routine, critical or complex is involved. As a result, new services are emerging such as post-editing of AI-generated copywriting.


Only time can tell how the interaction between machines and humans will develop, but the possibilities of this emerging field are certainly exciting. At ecls translations, we have years of experience in leveraging the potential of machine translation while deploying the strengths of a team of specialist human translators, so we’re perfectly placed to advise you on the best approach for your unique business needs in a rapidly changing environment.


Human translation or machine translation?


There are certainly times when machine translation is a useful and appropriate solution: when you need to get the gist of a text in a foreign language, for example, or when you need a fast translation of a straightforward, non-specialist text. While we would always recommend getting a professional translator to check machine translations for accuracy, they can be a great, fast and affordable option in the right circumstances, offering a way to translate large volumes of text at speed. If this sounds like something you might be interested in, we offer a machine translation post-editing service that is a quick and cost-effective solution while still offering the peace of mind that a professional human translator brings in terms of accuracy.


However, if you need to translate a technically complex, creative or humorous text, we would usually recommend working with a human translator. This is also the case for documents with legal validity or anything intended for high-visibility publication. Our team of professional human translators are selected for their specialist experience and knowledge and can be trusted to create high-quality, accurate and fluent translations that will resonate with your target audience.


Whether you’ve decided which service you need and want a quote, or you’re still not sure whether you need a human translator, we’d be delighted to talk you through the options: just get in touch with the ecls team at











Tags: No tags

Comments are closed.