There is almost no area in life that is not being affected by the rise of AI. Language interpretation is no exception. On the contrary, due to the nature of AI being built on large-language-models, interpretation is particularly prone to be affected.
Does this mean that AI will replace all human interpretation in the near future? We don’t think so. At LiveVoice, we are offering both, human interpretation and AI Voice Translation. Both have their unique strengths and limitations, which we want to outline in this article.
1. Quality and context
While some people overestimate the power of artificial intelligence in translation and interpretation believing that AI has already surpassed the skills of human interpreters, others consider AI merely as good as Google translate in its early years. The truth is somewhere in the middle.
Currently, AI live translation is already at a really good level, but it cannot measure up to human interpreters. This is mostly due to the fact that for highest quality translation, context and cultural nuances are key.
Here is an example: The German homonym “Opfer” has two meanings in English: sacrifice and victim. While human interpreters base their choice of words on cultural background knowledge, artificial intelligence will choose the word that is used more commonly in the datasets it relies on, which in the end may lead to misunderstandings.
There are ways to teach the AI specific terms, hence at LiveVoice we are creating glossaries that help AI to work with the desired range of terms. But currently human cultural sensitivity and judgment are still superior to any computer trained solution.
Having said this, also human interpreters are not fault free. And their performance depends on their level of preparation, which is why interpreters need to make sure they are provided with material beforehand.
2. Vocabulary & understanding
One of the strengths of AI is complex and specific vocabulary. While human interpreters have limited capacities in terms of the quantity of vocabulary, computers can be trained enabling them to have the right word by hand no matter how complicated or unusual.
Something that AI is not as good at as human interpreters, is when speakers have poor pronunciation. For example, if two words sound similar, AI might detect the wrong word (like hut and gut), which will lead inevitably to a wrong translation. Human interpreters can make up for that or at least correct themselves if they have misunderstood a word.
The question we often get is: What’s the accuracy rate of the LiveVoice AI Translation? The answer: It depends. Generally, we can suggest an accuracy of 85-95% but it depends on the circumstances like input audio quality, speaker quality, dialect, background noise along with many other factors.
3. Speed
Another difference is speed. Human interpreters usually have some “décalage” as it is called in the professional world, meaning there is some delay between the speaker’s output and the interpretation. At the same time interpreters often make predictions based on the first part of a sentence or the general topic of a talk and can therefore keep the décalage quite little.
In contrast, the AI needs to first “hear” the whole sentence before it can start translating. If the AI processed only half of the sentence, the quality of the translation would be poor, because it doesn’t know the underlying purpose of a speech.
Now, if you have speakers that like to use long, complicated sentences, the machine needs to wait quite some time until it can actually start translating. This inevitably leads to a longer delay than you can expect from human interpreters.
Another factor is how fast a speaker speaks and what the language combination is. There are languages that generally use more words per sentence than others. All of this will be compensated by a human interpreter – part of why their job is really hard! For LiveVoice AI Translation, we have created some algorithms that will also take this into account. But of course you won’t get the same result as with human interpreters.
In our experience, AI Voice Translation works great for use cases like keynotes, where you have one speaker at a time. For panel discussions, where people need to understand each other quickly in order to give answers, and where speakers switch a lot, human interpreters will be the better choice.
4. Voice and emotion
Artificial voices used to be terrible. This is not the case anymore. They sound more and more natural and meanwhile it is absolutely okay to listen to AI voices even for a longer period of time. But of course, human interpreters can still adjust way more to the emotions of a speaker than the computer-based voice.
On the other hand, AI voices have the advantage that you can choose from many different versions. LiveVoice AI Translation currently gives you the option of male and female, but there will be more options in the future. At some point you might even be able to copy the voice of the original speaker…
5. Translation vs. interpretation
There is a reason it is called “simultaneous interpretation” and not “translation” when human beings are doing the job. Because their job is not just translating word by word, but they interpret what is said. They try to convey meaning rather than individual words. Hence, commonly, they use other words, circumscribe or paraphrase what has been said. They might even need to shorten chunks of sentences to get the idea across still in time.
In contrast, automatic translation mostly translates word by word. This is also why we called it “AI Voice Translation” and not “AI Voice Interpretation”. In the future AI will certainly also move towards interpretation, but for now we believe it’s important to stress this difference and we want to be transparent about it.
6. Flexibility & Availability
Providing interpretation at events has become way more flexible in recent years. It is common practice nowadays that interpreters work from anywhere (RSI, remote interpretation). Hence, offering translation at events even short term has become a lot easier than in the past.
However, AI brings flexibility on a new level, as its availability is unmatched. AI is available 24/7. And since machines do not need mic checks there is no need for coordination of rehearsals or any other preparation appointments. Adding more languages does not add more complexity. And there is no need for making reservations for interpreters long term. You can add and book AI basically any time – it’s always ready for you.
Another great advantage: In contrast to human interpreters, artificial intelligence does not need to “see” anything while it is translating. You only need to send audio, not video. This enables you to add live translation even to guided tours, making them multilingual with a few clicks.
7. Costs
Well, now this is of course where AI has its biggest advantage. Using AI Voice Translation is by far cheaper than hiring professional interpreters. Human interpreters usually work with a partner taking turns, since language interpretation is a high concentration task. As a consequence, you have to pay two interpreters per language.
As mentioned above, professional interpreters commonly invest a lot of time in preparation, which is the reason why hiring them is not cheap. Usually, it is only possible to book half days or full days, as the bulk of the work is done beforehand no matter if an event takes an hour or seven.
In contrast, LiveVoice AI Translation is billed by the minute. With only 30 € per hour per language, making events multilingual is possible basically for any budget.
What is the right choice for me?
So, what is the right choice? Your decision will depend on many factors, but here is a list that can help you choose the right option.
Human interpreters:
- You want highest possible quality
- Important decisions depend on the accuracy of the translation
- Cultural sensitivity is crucial
- Emotions play a big role
- Speed is key, e.g. at panel discussions
AI Voice Translation:
- You require high flexibility
- Short term notice events
- Large numbers of languages are needed
- More formal occasions like keynote speeches
- You have a rather limited budget
Conclusion
Both, AI Voice Translation and human interpretation have their strengths and limitations. Depending on your situation, your choice might look different every time.
If your situation involves crucial decision-making and sensitive political or cultural topics, we recommend human interpreters. If your budget is tight and stakes are not so high, AI is a good alternative. For example, when you have an event that needs translation into many different languages, but each language group is very small, you might not have the budget to provide human interpretation to all of them. In this case you could offer human interpretation to the main language group, and add AI Voice Translation for several other small ones.
Depending on your needs, budget and situation, you should be able to decide for yourself which option makes most sense. With LiveVoice you can have both – or even combine them!
