Google launched its popular translation tool in 2006. Since then it has become the reference translator for hundreds of millions of people who use it, occasionally or regularly, and Google has extended its capabilities and functionality to other applications with which it has integrated, such as Google Lens. During 17 years has maintained an undisputed reign, despite notable competitors such as deepL, but that have not obtained the same implantation. Among other reasons, because the almighty Google is not behind it.
That situation has started to change this year with the success of generative artificial intelligence chatbots. Although it is not its main purpose, both ChatGPT and the most recent Bard, to focus on the best known, have revealed themselves as notable translators. Better than Google? Depending on the point of view from which you look at it, but if we look at their ability to understand the context and better refine the translations and their nuances, without a doubt.
Bard is just out, ChatGPT is not yet a year old, and Google has been working on its translator since 2006, but a quick comparison yields much more promising results for the former, especially bard.
Google Translate is no stranger to artificial intelligence. In 2016, I replaced statistical machine translation with neural machine translation using deep learning and neural networks. However, these technologies predate architecture. transformer introduced in 2017 by Google and from which language models such as ChatGPT (GPT 3.5) or Bard (LaMDA) have been developed.
What is Google Translate better at?
As a tool, Google Translate is much more polished than its current competitors. Supports 133 languages and has another 100 in development while Bard arrives at 100 and ChatGPT stays at about fifty. The function of play the words of a translation by voice it is available both on the web and in the Google Translate apps, while in ChatGPT it is only available in the app, already available for iOS and from this week for Android, but not on the web. Bard, which for now only has a web version, does have it.
Google Translate has dedicated buttons to easily translate, in addition to text entered by the user, images, documents and web pages and can even be used, through the app, to communicate with a correspondent in another language and in real time, which is not allowed by ChatGPT or Bard. They also function more rudimentary in which you have to indicate to the chatbot the type of task that is required and the target language.
How ChatGPT and Bard are better translators
But it’s not all good news for Google Translate. This does not allow you to enter more than 5,000 characters when making a query, just like in Bard, while with ChatGPT you can work with texts of about 8,000 words. But beyond the fact that Google Translate, which is used by more than 500 million people every day, is a more finished and complete tool than its competitors, its value lies in its quality as a translator. And here the surprises begin.
Both Bard and ChatGPT offer a better understanding of the context and a deeper knowledge of the specific forms of each language. Note, not all languages are at the same level in any of the applications we are talking about, but if we focus on Spanish and its translation into English, the results are clearly better.
To verify this, we have focused on a handful of examples of colloquial phrases in Spanish whose literal translation into English ensures that our interlocutor will be left wondering what we wanted to say.
Translating colloquial expressions from Spanish to English
For example, the expression “be even in the soup” he Google translator it solves it with a literal “being even in the soup” without adding any explanation or context. ChatGPT It offers two options that make sense to an English speaker, such as “to be everywhere” and “to be all over the place”, in addition to offering a brief explanation of the meaning of the expression. Bard It delves deeper and in addition to offering five translation options for as many colloquial expressions in English in the same sense and explaining in which cases it is better to use one or the other.
If we use “sing 40“, he Google translator interprets it as “sing the 40”. ChatGPT He gives a more forceful answer but misses the shot equally when he understands the expression as if it were “cantar la traviata” and translates it as “to spill the beans” or “to tell all”. Bard It falls into the same error as ChatGPT and, although it proposes more alternatives, it is seen that “singing 40” is an expression that artificial intelligence still cannot understand.
Introducing “he has two newscasts left“, he Google translator Once again, he bets on the most absolute literalness and turns it into “he has two newscasts left”. ChatGPT He understands it perfectly and, in addition to translating it correctly (“he/she doesn’t have much time left”/ “he/she doesn’t have long to live”), he explains the meaning of the expression. The only handicap is that she gives the entire explanation in English even though the question has been asked in Spanish. Bardwhich does not fall into this error and returns all the questions posed in Spanish, hits again with the proposed English idioms (“he’s on his last legs” and “”he’s not long for this world”) and the explanation.
Based on these and other tests that it would be repetitive to introduce, chatbots have a better understanding of the nuances of languagesat least when we talk about Spanish and English, and return a higher quality translation than the classic Google Translate. Not necessarily perfect, but with enough information added so that the user knows if the system has succeeded or not. Considering that Bard is from Google, hopefully it won’t take too long for the company to introduce its Translator capabilities to equip chatbots.