Imminent Report: Collective Intelligence Is Found in Cultural and Linguistic Diversity

Translated Imminent Report

Translated, the company that brought ModernMT to the language industry market, is known as a language technology innovator. But as AI dominates the conversations in the language industry in 2023, the company is also keen on demonstrating that the technological questions are not solvable without a deep understanding of the power of human intelligence.

More precisely, Translated wants to explore how people engage in collective intelligence, the driving concept behind Translated’s Imminent Research Center’s May 2023 report. The report, launched during a live event on May 31, 2023, is aptly subtitled “Word Wide Wisdom.” It is indeed a compendium of collective wisdom captured in thought-provoking articles from authors whose backgrounds are as diverse as astrophysics, localization, hospitality, neuroscience, and language technology.

The Imminent Report begins with an introduction by Luca De Biase, Director of the Imminent Research Center, in which he asks a few interesting questions: whether it is possible to improve understanding between people who speak different languages and improve their ability to collaborate in a smarter way, and whether a multilingual team can surpass the performance of a monolingual team and just do better things.

De Biase delves into these concepts and possible answers in an article that expands upon the roles of humans and machines in multilingual and multicultural contexts. He also ponders the possibility of what he calls “multilingual normality” and “multilingual collective intelligence,” arguing that perhaps the reason humanity is not there yet is rooted in a history of national identities.

De Biase hypothesizes that “forms of collective intelligence can emerge in multilingual human groups if they are designed as human-machine systems with balanced components.” To explore this hypothesis, forms of collective intelligence are explored as well, and applied to “the design of human-machine systems for multilingual groups.”

The author also clarifies that “we cannot understand collective intelligence by studying the individual intelligence of the people involved in the groups under consideration. Likewise, we cannot consider artificial intelligence a separate entity from the human process involved.”

“We cannot understand collective intelligence by studying the individual intelligence of the people involved in the groups under consideration. Likewise, we cannot consider artificial intelligence a separate entity from the human process involved.” — Luca De Biase, Director, Imminent Research Center

Multilingual intelligence is another dimension of collective intelligence. De Biase points out that some form of collaborative design is needed to create a multi-stakeholder, multilingual, multidisciplinary, multicultural group that works. He adds that in any group that is intelligent, regardless of language and other factors, members respect each other, are empathetic, actively listen to each other, are more often women than men, and are relatively diverse.

Linguistic and Cultural Diversity, The Brain, MT, and AI

Salvatore Giammarresi, head of localization at Airbnb, Daniel Pimienta, manager of the Observatory of languages and cultures on the Internet, Silvia Currò, a professional English/German to Italian translator, and Helga Nowotny, professor emerita of science and technology studies at ETH Zurich explore the pervasiveness of English as a lingua franca in an increasingly multilingual world. 

A lingua franca, understood as a common language used by diverse groups of people who have different native tongues, can be useful in helping people communicate, but several issues arise when it is implemented. First, in a group, there are bound to exist different proficiency levels. Second, individuals who are native speakers of the lingua franca will have an unfair advantage and might dominate.  

On the other hand, adopting a lingua franca with a coherent plan in which groups achieve a uniform level of proficiency can position a global business, for example, to reach further to international customers with no intermediary. According to Giammarresi, for example, having a lingua franca “sensitizes the company’s leaders and employees to the linguistic and cultural commonalities and differences.”

Patrizia Boglione, Brand and Creative VP at Translated, Astrophysicist Ersilia Vaudo, and Paolo Venturi, director of AICCON and The FundRaising School, as well as professor at the University of Bologna, delve from their unique perspectives into the subjects of inclusiveness and diversity, both concepts understood as the acceptance and active participation of individuals in collective endeavors where their uniqueness is valued.

Boglione captures the essence of this premise as she points out how the youngest generations are more and more diverse and brands that do not embrace inclusiveness will become irrelevant. She says that “Inclusiveness … will mean enhancing the value of the uniqueness of disparate communities and intimately connecting with them.” Language is but one dimension of diversity, and all dimensions matter for true inclusiveness.

Kirti Vashee, Language Technology Evangelist at Translated, begins his first of two articles in the Imminent report with an account of how machine translation has failed to match the promises made by developers over the years to reach human parity. 

According to Vashee, there was also a lack of evidence (not enough data and/or no validation) to substantiate some of the claims made in the past about the progress made in machine translation. That is, until Translated was able to conduct a long-term study and gather translation production data (presented at the Association for Machine Translation in the Americas conference) that provides a massive evidentiary sample, validated by a large group of professional translators in multiple languages.

Vashee argues that progress with MT “can also mean that we are that much closer to reaching AGI” (Artificial general intelligence, i.e., a machine can understand or learn any intellectual task that a human being can). Measured in Time To Edit (TTE), in Translated’s study actual progress made in MT points to vast improvements. This evaluation was done by professional translators (over 130,000), not by automated MT scoring systems.

In his second article in the report, Vashee goes deeper into the subject of MT, specifically, the role of this technology in collective intelligence. Vashee says that “over the last decade, the dominance of English continues to decline as more and more new content is introduced in other languages.”

“Over the last decade, the dominance of English continues to decline as more and more new content is introduced in other languages.” — Kirti Vashee, Language Technology Evangelist, Translated

But English has in fact dominated, and it will take some time for data in other languages to catch up. It is also the case when it comes to AI technology. Per Vashee, “English has had a head start and has had more investment over decades, especially in science, technology, and general knowledge, building a large foundational core that is not easily matched by any other language.”

Guido Vetere, an AI professor at University of Guglielmo Marconi in Italy and head of Isagog Srl, Jacob Browning, from NYU’s Computer Science Department, and Yann LeCun, NYU professor and chief AI scientist at Meta, discuss what AI can and cannot do. AI structures, like ChatGPT, “convey a vision of words as pure combinatorial elements,” statistical in nature, argues Vetere.

These authors coincide in that machines can learn, but what they learn is how words combine and how to predict what comes next. Machines are also better able to translate if they learn from vast amounts of multilingual text. But large language models (LLMs) are not able to reach a human level of understanding, and limitations exist also in the nature of language itself. 

AI systems may be able to chat about anything, but that is a knowledge representation, a mimicry of human language and a show of effective predictability. This is very different from the deep understanding humans acquire from experiencing the world and “interacting with culture and other people.”

Marco Trombetti, founder and CEO at Translated, remarked on how humans are still the ones to tell whether machine translation (MT) is improving, and how humans can get great results when working with tools.

At the same time that MT does indeed improve, new opportunities emerge as a result of several factors. The MT improvement helps translators do their work, and editing takes less and less time. The quality gap diminishes as volumes increase and machine learning efficiency is much higher.

Using a quality assurance parameter called “Errors per Thousand (EPT) words,” Translated has measured “about 50 linguistic errors per 1000 translated words” in MT. EPT is expected to be reduced over time as MT continues to improve and budgets remain in check.

“It will be possible to translate more, faster, and much better,” states Trombetti, adding that demand will grow as a result of growth in multiple sectors, including trade, industry, migration, culture, and entertainment. 

“An industry that manages to improve productivity and enjoys growing demand is a healthy and prosperous industry because it is always innovating and developing new markets.” — Marco Trombetti, Founder and CEO, Translated

Humans make an essential contribution to the translation industry, including those who improve themselves and the way they work “to capitalize on the opportunities offered by machine translation,” Trombetti reckons, mentioning additional value-added opportunities such as copyright services, cross-cultural research, and stylistic updates to translations.

The Translated CEO concluded that, “an industry that manages to improve productivity and enjoys growing demand is a healthy and prosperous industry because it is always innovating and developing new markets.”

Neuroscientists Carlotta Barelli, Valentina Rava, Nereo Kalebic, Elena Taverna, and Martina Ardizzi explore the connections in the human “language-ready brain” and how it has evolved. The authors also touch on culture, an element the authors cite as the ability of humans to imitate, and how language is critical both for that imitation and the transfer of knowledge. 

Some conclusions drawn from these analyses are that modern neuroscience helps study the mechanisms that shape the language-ready brain, and that the more people’s brains synchronize, the better a team works, contributing to collective intelligence. 

The Imminent Report also includes a section in which seven experts in diverse fields answer the question of how collective intelligence can be improved in a multilingual context. Collective intelligence can be used to bring together diverse knowledge, cultures, experiences, data, expertise, and languages to foster innovation and address societal challenges. This hinges on effective communication across languages, which in turn means making the best use of technology for an equitable contribution. 

The report closes with a summary of annual Imminent Research Grants, valued at EUR 100,000, and three summaries on interesting countries to keep an eye on: Estonia, Indonesia, and Kenya, written by Imminent’s staff with the help of Marion Jõepera, Rachmadhina Insan Widyapianissa, and Winnie Ngimor.