The 4 real differences between AI translation and human translation
Author: Gabriela Vornicu
Romanian translator & researcher
Updated 6 January 2024
In this article, you'll find 4 differences between an AI translation and a professional human translation, namely in the following areas:
Note: As I am a medical translator, you'll find examples related to healthcare, so inhale, exhale and let's get started.
Most translation problems are caused by ambiguity in the source text, which can only be corrected by human judgement.
And when a text has 2 interpretations, readers must decipher the intended meaning, which slows down their reading speed and discourages them from finishing the lecture.
Ambiguity caused by bad syntax
Ambiguity caused by unclear terminology
AI vs. human approach to issues of ambiguity
AI will either leave the ambiguity in the translated text OR randomly choose one interpretation, so the probability of AI deciphering the ambiguity correctly is ~ 33%.
Professional translators, on the other hand, will look for additional information or request more data from the author to arrange the words so that one - and only one - interpretation is possible.
The way AI translates terms is based on frequency and probability.
However, I have often found texts that contain specific or newly created terms that are difficult to find online. As you can imagine, mistranslating such terms can only lead to failure.
AI vs. human approach to terminology issues
In these cases, the AI will either translate the terms literally or leave them in English, both of which are 100% failures.
Professional translators, on the other hand, will research the terms or contact healthcare professionals to find the best medical and linguistic equivalent.
In translation, localisation refers to the adaptation of certain aspects (such as dates, institutions, product names) to the target culture.
As you might expect, AI has a limited ability to recognise that these elements are different in other languages, and an even more limited ability to find the right translation.
AI vs. human approach to terminological issues
AI rarely knows how to adapt these things to the target culture, so what it does is translate them word for word.
Localisation is more than just finding the equivalent of a substance. Done correctly, it involves creating a translation strategy that covers the translation of proper names, adaptation of treatments and diagnoses based on national protocols, etc.
As a marketing translator, I often improve the translation using the principles of psycholinguistics (also known as the psychology of language) to keep the reader's attention.
In short, if a translation is easy to understand → the information will be perceived as logical → the user will like it and perceive the text as truthful.
This is how I make a translation easier to process:
I vary the length of sentences to keep the reader's attention, as better illustrated in this post by Gary Provost:
I simplify the translation by:
Removing set-up verbs
Adding conjunctions (but, therefore, so, as a result, that is why, consequently) to create logic and make the text easier to understand
Replacing passive language with active language to make the text easier to process
Replacing abstract advantages with concrete ones
AI vs. human approach to psycholinguistic problems
AI doesn't understand the meaning of sentences deeply enough to improve them. It can simplify the text, but only in a limited way.
Also, human psychology is constantly changing and only a human mind can keep up with the trends.
As long as language is changing (and it is), AI translation will never be better than human translation because it simply cannot deal with issues such as ambiguity, new terminology and localisation.
And if we consider marketing translations, where human psychology has to be taken into account, it becomes even more difficult to achieve the quality of a human translation.
1. King, D., & Auschaitrakul, S. (2020). Symbolic sequence effects on consumers’ judgments of truth for brand claims. Journal of Consumer Psychology, 30(2), 304-313 2. Ehrlich, K., & Johnson-Laird, P. N. (1982). Spatial descriptions and referential continuity. Journal of verbal learning and verbal behavior, 21(3), 296-306. 3. Kamalski, J. (2007). Coherence Marking, Comprehension and Persuasion on the processing and representations of discourse (Doctoral dissertation, Netherlands Graduate School of Linguistics). 4. Langer, E. J., Blank, A., & Chanowitz, B. (1978). The mindlessness of ostensibly thoughtful action: The role of "placebic" information in interpersonal interaction. Journal of personality and social psychology, 36(6), 635.
About the author
Romanian translator & researcher
I'm an English to Romanian medical translator and an independent researcher into the psychology of language. I help medical companies with highly specialised medical translation and research services, as well as healthcare marketing agencies with highly creative adaptations and SEO services.