What ChatGPT means for linguistic diversity and language learning
Some have noted that once ChatGPT and other AI-assisted tools are more widely used in education, the gap between the languages where such tools work well and those where it does not will become larger, and in some scenarios create a new global educational and digital divide based on language.
In particular, as AI-assisted writing takes hold in some parts of the world, in other parts of the world – such as some Asian countries with complex writing systems – students could be left behind if AI companies do not swifty develop other language versions of ChatGPT.
Jieun Kiaer, a professor of linguistics at the University of Oxford in the United Kingdom and an expert on Asian languages, is rushing out a book on ChatGPT this month. She warns that tools such as ChatGPT could reinforce the global dominance of English and some other European languages. It could also reduce the richness of language.
A range of language limitations
And there are other limitations.
“Because of social media English is absorbing new words from other languages faster than ever. ChatGPT’s dataset is capped at 2021, so would not know any of the words that have emerged and-or been added to the dictionary in the last year,” Kiaer told University World News.
“This is another of AI’s linguistic issues. AI must be able to keep up with all the world’s languages, and the new words constantly emerging in each language, too.”
“ChatGPT appears to be less proficient in Asian languages,” Kiaer says in her forthcoming e-book Doing Language with ChatGPT: A linguist’s response, which uses material from her students’ use of ChatGPT in digital humanities workshops.
Kiaer says the book was written “in collaboration with ChatGPT” as she quizzed it on its own capabilities and other aspects of language learning.
“ChatGPT is most effective in English and so it is not well suited to becoming a local platform in all countries. Other regions will be developing their own AI technology,” she quotes ChatGPT as responding at one point.
Because ChatGPT relies on huge language databases, it is most proficient in English. “Everyone uses English, it is the world’s lingua franca,” she notes. Training ChatGPT in other languages requires a huge amount of time and money to get good results.
“AI operates in English on a global scale and while it might be simple for it to be translated into other European languages, AI does not tend to function well in non-European languages,” Kiaer says in her forthcoming book, extracts of which she shared with University World News.
“What is more, AI is not yet able to replicate the delicate and complex pragmatic meanings that humans portray through their language use instinctively. As such, AI still has its limitations.”
Pragmatic variety in linguistics refers to the way language is used in context to convey meaning and includes irony, metaphor and sarcasm.
Asian language versions being developed
Kiaer explains that she tested ChatGPT’s capabilities earlier this year with Korean, Chinese, Japanese, Czech, Slovak, Russian and Polish and found it can be trained to work well with some central European languages.
“If you train it, even for half an hour, incrementally, it works much better,” she told University World News. For example, with Czech, ChatGPT “provides amazing output that is consistent” after some training.
She points out that the ability of ChatGPT to learn is a major advance for students and researchers. “You cannot train Google to be clearer,” she told University World News.
However, some Asian languages like Chinese and Japanese have a different writing system that can make them more challenging for AI models to understand, although both China and Taiwan are working on ChatGPT-like tools for Chinese character-based languages.
AI models for a specific language require a larger amount of data and resources, so it may take a while before all the languages are covered. Additionally, many Asian languages have a large number of characters and variations that can make them more difficult for AI to process.
East Asian countries such as Japan, Korea and China in particular have their own search engines such as Naver for Korean and Baidu for Chinese. In addition, with ChatGPT banned in China, no information from China feeds into ChatGPT. By contrast, China’s Baidu search engine is developing a ChatGPT-type bot for use in Chinese that has reportedly trained on both Chinese and English engines.
Kiaer believes that in a globalised world, lack of diversity is a serious limitation.
“ChatGPT cannot service well if it is missing data from whole continents,” she says. “This is a serious challenge that AI creators need to think about – how better datasets can be constructed so that AI bots like ChatGPT can provide a wide range of information from diverse sources that support the diverse world in which we live.”
Need for hybrid models of teaching using AI
Kiaer is against banning tools like ChatGPT in schools and universities but believes AI developers, teachers and policy-makers need to work together to work out how best to use such tools in education and research.
“We need to work out a healthy hybrid model [of teaching] rather than getting rid of it by banning it in education,” she explains.
“If AI develops to the point of us interacting with digital humans, then we may be able to practise speaking and writing with digital humans, which could help to ease foreign language anxiety and improve the psychological well-being of learners. The signs are all pointing to the fact that AI will revolutionise language learning, allowing us to learn more easily and efficiently than ever,” she says.
She points to the huge amount of money spent by parents in East Asia, particularly in her native country Korea, on sending their children to cram schools to learn English.
“If AI can take on the basics of any area of study, then perhaps students need not spend so painfully long memorising, as technology puts all the answers at our fingertips. Instead, they will need to learn skills of critical thinking, analysis and real-world application from an earlier age,” she says.
“Being able to use knowledge stored on online places is the only real ‘school’ that will be needed in the future and curricula need to begin adapting to that reality now,” Kiaer says, adding: “A revamp of the school curriculum has been needed since the invention of Google.”
Exams will need to change
She notes that exams and how students are assessed will also have to change because AI-assisted tools like ChatGPT can do fact-checking, summarising and-or analysing better than students can. “Questions will have to be reformulated,” she notes.
And standardised exams such as the highly competitive college entrance exams prevalent in countries like China, Japan, Korea and Vietnam as well as elsewhere in the world, will have to be rethought. Mega-exams such as the Test of English as a Foreign Language (TOEFL) and the Test of English for International Communication (TOEIC) “will become useless”.
According to Kiaer, eventually AI will be able to develop personalised exams tailormade to individuals and which will more accurately measure their levels of learning, particularly with language learning.
It can also help reduce the financial burden on parents and learners by making language learning more accessible and reducing the need for expensive immersion programmes or year-abroad programmes, she says.
Schools should embrace language learning using AI, she believes. “With English language education in Asia and many parts of the world, AI will feel more like bliss than a disaster.”