The improvement of AI-based translation software poses a number of risks regarding understanding and exchanges between different cultures
At the beginning of April, a short video circulated widely on social media in which the mayor of Venice Luigi Brugnaro informs tourists about the paid entry system into the city foreseen on some days of the year and in force for a few months. In the film, admittedly created by reworking another video with artificial intelligence software, Brugnaro speaks in fluent English, in his voice, and in an apparently realistic way.
For some time, various technology companies have offered integrated translation, dubbing and video editing tools which, starting from a film in which a person speaks, allow you to create an equivalent one in which the same person expresses the same concepts speaking another language. They are practical and relatively cheap tools for the end user, but their increasing efficiency and popularity have stimulated a parallel debate about the risks they imply. We discuss not only those relating to possible abuse and unauthorized use, but also the risk of a progressive disappearance of the need to learn new languages except for specific study reasons.
The doubt shared in the debate is that the spread of artificial intelligence software applied to voice translation could reduce the practical incentives for language learning. Samsung, for example, has been promoting for months an instant phone call translation function available on one of its famous smartphone models. In September, Spotify started with OpenAI, the company that produces the ChatGPT software, an experimental project for the vocal translation of podcasts into other languages while maintaining the original voices of the hosts.
The latest automatic speech translation systems have been made possible by rapid improvements in the field of artificial intelligence and in particular in speech processing. machine learning, that is, the activities of computers learning from data. They are consequently systems that work very well with so-called languages with high availability of resources, such as English but also Italian, for which large quantities of data exist.
However, they work less well with African languages such as Swahili, or Indo-Aryan languages such as Urdu, the official language in Pakistan and parts of India. They are languages spoken by hundreds of millions of people around the world, but little spread in digital format. The lack of data has a negative impact on the availability and quality of translations from these languages obtained through artificial intelligence programs, which are also used in very risky situations: the translation of testimonies from asylum seekers or people in war contexts, for example.
– Read also: Is there too much English in scientific research?
Without directly associating it with the spread of translation and dubbing systems via artificial intelligence software, theAtlantic reported a recent decline in the number of people studying foreign languages in several Western countries. In the United States, they decreased by 29.3 percent from 2009 to 2021. In Australia, the amount of high school students learning a foreign language in 2021 was the lowest ever (8.6 percent). And in South Korea and New Zealand, universities are closing French, German and Italian departments. English proficiency has also declined among young people, according to a report by EF Education First, an international company that organizes English language courses and cultural exchanges around the world.
Regardless of various factors that might explain the phenomenon, including school disruptions during the pandemic and reduced funding for humanities programs, many people are not actually learning new languages at a time characterized by the widespread availability of tools with where they can talk to them without knowing them.
One of the main risks in the perspective of a world in which foreign languages are used without studying them is considering them all equivalent: which is a very reductive way of understanding them. Over a century ago, helping to define the modern concept of linguistic relativity, linguists such as Wilhelm von Humboldt first and Edward Sapir and Benjamin Lee Whorf then theorized that language is not a means of transmitting thought, but a way of interpreting reality itself . Learning a new language is equivalent, in many respects, to learning a new way of seeing the world and thinking.
Listen too: Does language shape the way we think?
“As technology normalizes, we may find that we have allowed deep human connections to be replaced by technically competent, but ultimately empty communication,” wrote theAtlantic.
Recently the use of artificial intelligence software has affected not only the translation and dubbing of voices, but also language learning itself. On social networks like Instagram and Facebook you may come across advertisements for services that offer language teaching based on an interaction with a bilingual person who does not exist in reality. In some cases the lack of human evaluation of progress and possible errors, therefore of judgment, is explicitly described as a value.
However, the human part of relationships is the most important factor not only in the study of languages, but in any exchange between different cultures. Gabriel Nicholas, researcher at the Center for Democracy and Technology, a US NGO that promotes democratic values and individual rights in the development of technologies, toldAtlantic that part of the problem with artificial intelligence programs applied to languages is that they lead users to think that translation is something “neutral”. And they do not develop the fundamental ability to “move from one language to another”, taking on a different perspective from time to time depending on the circumstances and intentions.
In other words, although there is not just one right way to translate a sentence from one language to another, many software works as if there was actually only one right way: even more so in the case of languages with low resource availability. In language teaching courses with native speakers, however, it happens that, when students ask how to say a certain sentence in another language, most of the time the teacher replies to change perspective and turn the sentence differently, depending on the case. .
According to several linguistics scholars, given the growing diffusion of artificial intelligence software, greater investments are needed in a different study of languages. It should be more based on the development of cultural skills and an in-depth understanding of the beliefs and practices widespread in the populations of other countries, which also differ depending on the social background, the age of the speakers and the communication context.
At the same time, it would be necessary to invest in the study and collection of information relating to languages with low availability of resources, to try to counterbalance the preponderance of online texts in English, the main source of data for software training. Almost 90 percent of the sites are written in 10 languages: apart from English, which accounts for more than 50 percent, they are Russian, Spanish, German, French, Japanese, Turkish, Portuguese, Italian and Persian. Of the more than 7,000 languages spoken in the world, Google Translate supports 133, and the most popular chatbots are even more limited, because most advanced language models serve between eight and ten languages.
Explaining toAtlantic the importance of the intercultural components of language learning, Paula Krebs, executive director of the Modern Language Association, an association that brings together American scholars and experts in modern languages and literatures, cited an episode of the science fiction series from the 1990s Star Trek: The Next Generationtitled Darmok.
In the episode, the crew aboard the Enterprise is unable to clearly decipher a communication from a spaceship orbiting a planet called El-Adrel IV, inhabited by the Tamarians. It seems like an invitation, but the unknown structure of the language does not allow us to understand it. Through a kind of universal translator the crew manages to understand the syntax and basic semantics of the Tamarians’ language, but not the deeper meaning of their expressions. As the episode continues it becomes clear that the language of the Tamarians revolves around allegories rooted in their history and their unique practices, translatable but incomprehensible without a sharing of those non-formal aspects of the language.
«More than 30 years later, something similar to a universal translator is developing on Earth. But like that it doesn’t have the power to bridge cultural divides in the way humans can,” concluded theAtlantic.
– Read also: Is music really universal?
Continue on the Post
Tags: prospect world foreign languages longer studied Post
