Is anyone still looking for proof in 2023 that the future of multilingual communications is at the intersection of human and AI translation? If the rise of voice AI and a multilingual ChatGPT left room for doubt, this week brings compelling new evidence to dispel it, highlighting in especially human terms how technology-driven localization can be a force for good.
Axios reports Climate Cardinals, a worldwide nonprofit giving voice to young climate activists of more than 40 nationalities, has allied with machine learning experts at Google Translation Hub to help to bridge gaps in language access for would-be participants of some of the world’s top environmental forums and summits. Specifically, the tech giant lent its computational brawn to help an international network of volunteer linguists build a multilingual library of documents totaling roughly half of a million words that they sourced from top climate change resources in English. As well as giving non-English speaking advocates a leg up in dozens of languages, the effort saw foundational reports like the UN’s newest Intergovernmental Panel on Climate Change Report translated into upwards of 20 languages. Not least, it engaged Yale Climate Connections to provide a vast Spanish-language translation intended to help US communities that primarily speak Spanish engage with emerging information on climate and environmental issues.
The news follows closely on Meta’s late May announcement that its Massively Multilingual Speech (MMS) model is facilitating language preservation for some of the world’s scarcest and most endangered traditions, marking a shift from the metaverse to the real world for Meta’s multilingual LLMs. Building with the foundations of its 2022 engine for speech-to-text translation that covered the world’s top several hundred languages, the model was able to identify over 4000 of the world’s most seldom spoken languages and deliver translations between text and speech for roughly one quarter of these after training on an average of 32 hours of speech recordings in each. Perhaps more interesting, the model extrapolated from those 1,100 known languages to distinguish the rest with no prior information of what or who it might be listening to, or where the audio came from – most often, spoken orations in local languages of the world’s most widely read text, the Bible. This fact itself underscores the problem Meta may well have put a dent in: that so much of the world is in terribly little contact with the exchange of information emerging at the forefront of a changing global landscape. When ancient scripture is the best resource for documenting its languages, a wealth of information commonly accessed in the world’s largest economies awaits translation for the benefit of its long tail of over 7000 distinct linguacultural traditions.
Expanding Language Access: Where Localization and AI Translation Meet
While disparate in focus, both developments correctly highlight the nature of AI tools in language services against a backdrop of misconceptions. AI translation does not replace what people do, but instead helps with problems of a different scale and nature to enable new undertakings for human linguists. While it is unclear if the machines delivered clarity this time – a question ultimately for professional, in-country linguists – it is a welcome sight when the same tools we fear could replace people are instead helping to keep them involved in a changing global landscape that threatens to leave countless ways of life behind. With the expansion of available resources in new regions, demand for multilingual communications in diverse subject matter areas will only continue to grow and underscore the need for professional localization with best practices for human-AI translation.
Are you interested in keeping up with developments at the forefront of language, technology, and localization? Follow us to learn more at blog.csoftintl.com or visit csoftintl.com to learn more about CSOFT’s technology-driven translation solutions.