Article in partnership with Day Interpreting.

In a 2023 YouTube video, Wired looked at human and machine interpreting. In the introduction to the video, the voiceover mentions that, in order to get an accurate translation, the interpreter must convey elements such as intonation, speed, spontaneity and emotion. The video aimed at testing whether AI was better than humans at interpreting languages. Personally, I would have set the scene of the video by saying that we would be doing a comparative study to evaluate human and machine language interpretation and then analyse the results. To say that your starting assumption is whether AI is better at interpreting is leading the audience to have a preconceived idea that machine translations have a competitive advantage, for example with regards to the speed of rendering translated speech.

​The Wired Language Interpreting Tests

Wired used three tests to look at the performance of human interpreters and AI using the same speeches.

First Test

The first test looked at emotion, more specifically at how the Spanish King Felipe made an announcement about the Covid crisis in 2020.

A human interpreter highlighted that in this particular speech it was important for the Monarch to connect with his people and show appreciation for the work of healthcare staff as well as the cooperation of the entire country to contain the spread of the disease.

The human interpreter also analysed the response from AI and its choice of words.

Second Test

The second test was spontaneity with a speech from Colombian President Gustavo Petro.

The human interpreter looked at the President’s body language to build a clearer picture of the speech, as well as he studied previous speeches to understand the President’s mannerism and style including speed in his delivery; however, the human interpreter also praised AI’s ability to capture each and every word, which sometimes can be missed by a human interpreter. In fact, the human interpreter called AI “effective” in this instance.

Because the President quoted various figures in a short timeframe, AI could keep up with the speed and could capture all the financial information accurately.

The human interpreter admitted that he had missed a couple of details as there weren’t enough pauses in the speech and he had to keep up with what was being said. On the flipside of that, the pauses that were interspersed in the speech were what made it more powerful and impactful to the audience, as the breaks conveyed the importance of what was being said. AI did not take these pauses into account, rendering a series of uninterrupted sentences and giving them equal importance and gravitas. Natural speech is often interspersed with fillers and pauses; at times the speaker may start a sentence but then change direction and improve on an initial thought, maybe leaving a trail of incomplete sentences.

Third Test

Then, the third test was on speed. In this test the language intepretation comparison looked at a 2021 speech from the President of El Salvador, Nayib Bukele. This case was different from the previous talks because the President not only was speaking from a teleprompter and the text was prepared in advance, but also the speed of delivery was higher than in the previous two examples.

The human interpreter admitted that he found this particular speech challenging due to its speed. He mentioned that a professional interpreter has to extract the essential information and this could be at the expense of the completeness of information. As long as the translation is workable, an interpreter has done his job.

AI’s interpretation covered each and every word without missing anything. In this case, AI provides better completeness but included mistakes particularly grammar errors and expressions that native speakers wouldn’t use.

AI picks words without processing them in the context of a sentence and of the whole talk.

Human professional interpreters concluded that AI is better suited in informal situations but are not to be relied upon particularly in the medical and legal sector and most importantly when crucial decisions that can affect people’s lives are involved.

​Conclusions

The conclusion is that AI is not a replacement for human translators but a tool that can be used in some specific situations that are non-critical.

On a related note, users of AI tools should ideally be sophisticated enough to monitor and rectify errors contained in computer-rendered content.