Whose canoe? ‘Moana 2’ debate exposes AI threat to Pacific languages

Fears AI like ChatGPT may misrepresent Pacific knowledge and languages by losing their essence and nuance

Image released by Disney showing the character Moana on her canoe in a scene from "Moana 2."

By Jope Tarai

(BenarNews)--As Moana sailed the Pacific late last year in her traditional canoe on the big screen, a heated debate raged online over the ownership of this potent symbol that unites Pacific islanders. The discussion of her canoe in turn influenced, in a very short time, how artificial intelligence chatbots responded when asked about its origins, highlighting the technology’s concerning implications for Pacific knowledge and languages.

The highly anticipated “Moana 2” movie saw some Papua New Guineans argue the outrigger resembled one of their traditional lagatoi designs. That’s despite the representation being guided, in part, by the milestone involvement in the Hollywood production by Solomon Islands creative artist Millicent Barty. Some keen PNG scholars and bloggers, particularly on the moderated Academia Nomad platform, argued that, while the canoe appeared similar, it was rightfully a Solomon Islands design. 

As more voices piled in on the impassioned online debate, and the longer it continued, the choppy sea of words generated began to influence AI learning.

At the height of the exchanges, AI’s dominant and prominent chatbot at the time, ChatGPT, began providing responses that the “Moana 2” canoes were from both the Solomon Islands and PNG.

This gives rise to a significant potential risk for oral traditions and languages in the Pacific, particularly when the majority of what is sometimes known, and scraped by AI, is written text by those not of and from the culture. By doing so, AI may exacerbate the ongoing colonial or neocolonial legacies that privilege written texts over oral histories in languages. This in turn may escalate these legacies and dominate language learning to eventually re-represent or misrepresent Pacific languages. 

Out of the 6,000 estimated spoken languages globally, the Pacific is home to at least a quarter of them. Of this, PNG has the largest linguistic diversity in the world, with more than 800 Oceanic indigenous languages.

AI’s progress in creating human-like responses to questions, via language processing, is a massive technological turning point in recent years. ChatGPT, or Chat Generative Pre-Trained Transformer, was developed by OpenAI in late 2022. It learns through available algorithms to comprehend and process human language - called large language models (LLMs) - by reviewing and studying available text data to pre-empt the likely next set of words in a sentence, called “reasonable text continuation.”

The more text data AI consumes, the greater its predictive capabilities and outputs, which in turn creates more data. To moderate the variety of contexts in which language is used, AI uses models that mimic the human brain - or neural networking - to recognize patterns and adjust accordingly. This means Pacific languages can become subject to algorithmic imperatives rather than measured and rigorously reviewed content. 

ChatGPT admittedly has limited capability with Pacific languages, although it has demonstrated it can comprehend and generate text in a few, including Tok Pisin, Bislama, Te Reo Maori, Samoan, Tokelauan, Tongan and Fijian. As ChatGPT use continues, it is likely to absorb more text data around these given languages and also expand its interpretative capability. 

DeepSeek, the Chinese counterpart to ChatGPT, launched in late 2023 operates on a similar, if not the same, premise of LLMs as ChatGPT, but is currently touted as more efficient. DeepSeek’s added competition can only heighten the potential for Pacific language access.

The quality and capacity of face-to-face language learning cannot be entirely replaced. The human or conversational format makes self-directed and paced learning a lot easier. AI provides potential for greater access to translations of Pacific languages through chatbots. This creates scope for learning or relearning Pacific languages, especially for diaspora communities that have now settled away from the islands.

Given the oral traditions of the Pacific, which negate primary textual formats, AI may not be able to provide access to all Pacific languages anytime soon. This may be subject to the evolution of AI from textual to audio data processing, which could then see a shift, but this remains limited for now.

AI, like most technological advancements, comes with its benefits and risks for the Pacific and its languages. The uncritical language processing of incorrect or non-factual notions related to Pacific cultures – as illustrated by the “Moana 2” debate – and languages may diminish as AI evolves and discounts the influence of momentary online interpretations or misinterpretations.

The pace of progress with AI is unprecedented but caution must remain on a number of fronts over its application to Pacific languages. Most significantly the loss of human, face-to-face driven learning, may dilute the essence and nuance of language that is unique and special within specific contexts. One can learn to speak a new language, but to embrace the weight of its meaning from its people, within Pacific contexts and mannerism, can never be replaced. 

Jope Tarai is an Indigenous Fijian scholar pursuing a Ph.D. in Digital Politics at the Australian National University's Department of Pacific Affairs. The views expressed here are his, not those of BenarNews/RFA.