Can artificial intelligence and translation go hand in hand?

In November 2022, OpenAI launched ChatGPT, an advanced conversational AI based on the GPT-3 model of language comprehension and generation. The launch of this deep learning tool caused a major stir in social media, given that it is capable of doing large number of activities, and refining them through learning: whether its responding to questions of all kinds, generating texts… or even translating.

These types of technological advances lead us to once again think about the role of artificial intelligence in our lives. Keep in mind that artificial intelligence is “the ability of a machine to display human-like capabilities such as reasoning, learning, planning and creativity”.[1] Does this mean that artificial intelligence is capable of replacing people in certain tasks?

[1]European Parliament. “What is artificial intelligence and how is it used?” (2021). <> [Access date: 03/04/2023]

Artificial intelligence and translation

First of all, we should point out that AI has formed a part of the translating world since 2017. In our post “The pros and cons of machine translation” we already mentioned the 3 main types of machine translators (click here to read the post:

A few examples of MT tools based on artificial intelligence include DeepL, Amazon Translate, Google Translate, Microsoft Translator and Meta’s AI model, NLLB-200.

These systems are notable for their machine learning ability and their nearly instantaneous speed when translating, plus they can handle a large number of languages simultaneously.

They are highly useful tools, especially for texts that have low-medium complexity or that are repetitive and have a low level of specialisation.

The most common errors of machine translation

Despite AI’s great algorithm-based learning ability, there are still some aspects that are beyond the grasp of AI. At Siens, we have done an analysis to summarise the most notable aspects, based on our experience:

  • AI does not detect errors in the source text.
  • There is a lack of knowledge and training on the specialised terminology of sectors or industries.
  • AI doesn’t know your policies or style guides.
  • Consistency errors (AI sometimes uses both the formal and informal forms of “you” in Spanish texts, and it sometimes uses synonyms as the translation for the same term).
  • AI isn’t able to recognise the intention behind the source text, so it can’t always keep the same tone or style.
  • AI cannot localise text, meaning it doesn’t know the cultural conventions of the speakers of a target language and doesn’t recognise what could be culturally inappropriate.
  • Sometimes transcreation has to be used in the translation process, which requires a high level of creativity and of cultural and social knowledge.
  • AI does not have information about the context of a text.
  • AI is unaware of neologisms.
  • Finding natural equivalents in a target language is difficult when where there is a high level of emotion, such as with sarcasm or puns.

Shall we look at a few examples?

Error: use of the wrong noun due to a misunderstanding of context.

The source sentence refers to a “bank” but AI uses “bench” as it is a polysemic word and it doesn’t understand the context.

Error: unnatural translation.

Our proposal would be: “So don’t hesitate to contact us by calling”.

Error: literal translation.

In this case, “Pisa fuerte” should be translated as “Be strong” or “Be confident”, but it has been literally translated.

Please note!

Machine translations are also better suited to certain language combinations than others. For example, they are not highly advisable for translations of Nordic languages or so-called exotic languages (Arabic, Japanese and Chinese, among others).

The relationship between professional translators and machine translations

Considering all the aforementioned, we can draw the following conclusions: machine translation based on artificial intelligence is a very useful tool that can be used as support for saving time and that can be combined with other technologies (translation memories, databases, etc.). Professional translators now regularly work with these tools, but always conducting a meticulous post-editing process that ensures the inclusion of any aspects that AI misses or can’t cover