Last weekend was Multilingualism Day at the European Institutions. This is a great initiative to showcase why multilingualism matters and what institutions do to make it possible “that all EU citizens can follow the work of directly-elected representatives in any of the 24 official EU languages.” While this effort is commendable, a simple reality check of the world around us tells us that this proposition applies only to the EU. For the wider world, when people interact in a multilingual setting, they are restricted in their communication abilities to the use of a few languages. This is the case in public offices, in international conferences, and the list goes on. To be realistic, this is not even the case for the European institutions at large—not even for the Day of Multilingualism. Their program is the best example to showcase this reality. While the EU is the prime example of multilingualism at work, their conference to celebrate multilingualism was available only in Dutch, English, French, German, and Italian.
Obviously, what the organizers did is multilingualism by definition. However, it is a form of asymmetric multilingualism, i.e., the ability to support multilingualism only for a subset of people who have the advantage and privilege to use their own native language to communicate. In practice, I argue, it is also a form of discriminatory multilingualism, i.e., a situation where those not belonging to a specific linguistic group are repeatedly discriminated against from using their native language. Discriminatory multilingualism happens constantly because the languages offered in a multilingual context are always the “major” ones, depending on the context and geographic areas, and never the “smaller” ones. Slovenian and Czech people, to name just a few from the same official European languages, do not have the privilege to use their mother tongue in almost any international contexts. Exceptions apply. It is quite common to attend international conferences where the guise of multilingualism and accessibility is limited to, let’s say, English and French.
Now, there are several reasons for this. Asymmetric or discriminatory multilingualism happens because the effort to offer accessibility to speakers of different linguistic backgrounds is dictated by choices related to habits, logistics, budget, availability of professionals, and number of speakers. In other words, it is an intrinsic feature of multilingualism at large: It can not be other than this. When multilingualism is catered for solely by humans—be they translators, interpreters, or multilingual staff—it is simply impossible to offer true, i.e., non-discriminatory, access to and for all languages. Again, there are exceptions to this. Think of an event where only two language groups are in contact. Okay, if you want to foster multilingualism you can have an interpreter. Think of the plenary assembly of the European Parliament. Okay, you put up the biggest (and most expensive) language service in the world, and try to do your best to cater for true multilingualism. But the vast—and I really mean the vast—majority of multilingual events and encounters are by definition discriminatory. The United Nations have a discriminatory language regime (6 languages: either take it or leave it). International conferences adopting only English, French, and the local language are discriminatory. Discrimination is born from the factual limitations of human-crafted work. And you cannot do anything against it with traditional means.
A possible solution—the only one, to be more precise—to achieve non-discriminatory multilingualism is offered by automating the translation process by means of AI tools for live speech translation. Only by scaling these tools, making them available everywhere, anytime, and in the largest number of languages possible, will we alleviate these limitations. While it is not foreseeable that in the immediate future the quality of machine interpreting systems will match the flexibility and contextualization abilities that humans can achieve, the progress is very fast, and the quality has improved dramatically over the past few years. There is no doubt that improvements will continue over the next few years too, even without any major paradigm change in ML and AI. Human interpreters will continue to be relevant—as I have argued several times, this will be the case even on the day AI reaches some sort of parity with humans—but it is only through AI interpreters that we will be able to offer access to conferences, events, etc., in any language of choice for every language group. This is what I call, for lack of a better term, non-discriminatory multilingualism.
Are we there already? Certainly not. There are still many hurdles. The technology is only emerging now, and it will take time to spread, get used, understand where it should be used and where not, etc. Quality needs to improve, obviously, especially in its ability to be flexible in different situations (it works already at a very high level in certain, more formal, domains). While language coverage is already remarkable, covering dozens of larger and smaller language groups, many languages are still not covered. And many things need to be debated. How much error tolerance will be okay. How to deal with differences in performance among languages. The list can go on. Complete non-discriminatory multilingualism might note be achievable. But we are moving towards this goal. And the journey is already a worthwhile goal. This is what keeps people like me awake at night.