Valuable research by non-native English speakers gets missed, suggests Adrian Wallwork, because of the cost of translation services. A more enlightened approach to the use of LLMs could help solve the problem.
Despite becoming the hottest year on record, in early 2023 researchers in emerging economies writing about how drought and flooding were affecting their countries’ staple diets breathed a sigh of relief — finally LLMs could give them a much greater chance of publishing their research. By using chatbots they could correct their English, save time, and have more funds left for equipment, reagents etc. Mid 2023: Backlash. ChatGPT and other bots are increasingly found to generate non-existent sources of academic papers, universities around the world question whether chatbots are ethical, bans are put into place, and AI gets itself a reputation as being untrustworthy with strong racial and gender biases.
A year later and much research is still not finding its way into international journals. Why? Despite massive advances in the reliability of LLMs, many non-native English speaking researchers don’t trust them or cannot access them; The cost of having a paper edited is way above what researchers in the global south can afford.
Result? I have a strong suspicion, derived from 15 years of lecturing agronomy PhD students from around the world, that research which could possibly help combat the effects of climate change, e.g. by increasing the yield of staple crops, is not being published in English and not reaching the audience it needs to.
Reliable editing agencies in the US charge nearly US$550 for a typical 4,000-word research article. Agencies in Europe are smaller and may charge half the US rate, but this is still more than most researchers, even in relatively rich countries such as Italy, can afford to pay.
I contacted some of my past international PhD students to see how they are managing to get their research published and whether, due to continued high editing costs, they are switching to AI.
Hussam, in Iraq, wrote that his fellow researchers now use Indian colleagues to edit their articles. They have also increased the number of contributors in their articles – consequently the cost per author is reduced. But Hussam himself works individually and bears the entire cost of editing – 17% of his monthly salary. He doesn’t trust AI.
Alya Batool, from Pakistan, reported that the budget for education in her country is approximately 1.7% of the total GDP (4.2% UK) and that “this creates initial difficulties even in conducting research work, plus the additional costs of professional English editing, along with publication charges for articles”. So Pakistanis are now opting for local editing services, which cost a sixth of the price of US services. Again, little use of AI.
Researchers in former British colonies benefit from already knowing English, yet still need to have it checked. Frashia Ndungu from Kenya: “We have greater ease of access to tech compared to other African countries. Most of us tend to use free tools such as basic Grammarly and ChatGPT. But our budgets do not extend to buying any premium versions of AI tools.” Being an English speaker, Frashia is also more likely to learn about new chatbots and the advances being made in AI, as such information is always available in English first.
The fact that AI is underused is a shame because chatbots such as Perplexity, Poe, Mistral, Copilot, Llama and ChatGPT are generally excellent at correcting English.
Unfortunately, AI detectors have put a spanner in the works by becoming a barrier to publication. Researchers who use chatbots to correct their English can potentially get caught by an AI detector and be forced to rewrite part of their paper. This is clearly absurd as these researchers are not using AI to write their papers from scratch, but to correct them or generate ideas for what to include. The result is that some are now paying editors, not to correct the English of their manuscripts, but to make them AI free (there are AIs to do this too, but they often produce gibberish) – and even then such text may still get erroneously identified as 100% AI.
One service that has the potential to revolutionise the world of drafting papers in English is Curie, which edits an entire paper in Track Changes. I have verified Curie’s claim that the corrections it makes are 95% correct, but it sometimes introduces mistakes by itself. For instance in the previous paragraph it changed ‘spanner in the works’ to ‘AI detectors have been used as spanners’!
Curie’s marketing department informed me that they “are looking into region-specific pricing, and discounted or free access via various partnerships”. This is a move in the right direction, and if Curie manages to improve its tool, the benefits to non-native researchers will be enormous.
Solutions? First, researchers whose first or second language is a major language could write directly in that language, modify the structure of sentences so that they reflect English syntax, use DeepL or Google Translate, and finally use a bot to correct the result.
Researchers who use chatbots to correct their English can potentially get caught by an AI detector and be forced to rewrite part of their paper. This is clearly absurd as these researchers are not using AI to write their papers from scratch, but to correct them or generate ideas for what to include
Second, journals need to limit the use of AI detectors (checking for plagiarism seems fine, but checking for AI is just penalising the authors); and I recommend that researchers keep copies of all the drafts of their manuscripts to prove that AI was only used to correct the English, not to write it.
Third, reviewers should stop obsessing about grammar and typos, and focus on readability and whether key findings are highlighted sufficiently.
Fourth, students and researchers need courses on how to prompt and how to critically judge the bots’ output.
Finally, scientific publishers, who have long profited from not paying the scientists who produce articles for their journals and instead making them pay for the honour, should consider using some of their vast profits to offer discounted editing services for those who would not otherwise be able to afford them, and donate funds so that every university can have access to AI tools and human editors.
With these measures in place, researchers such as my ex-student Joaquim Mbasa Molo, who left his home in the DR Congo to study AI in Italy, would thus have access to accredited, trustworthy editing agencies in his country (currently there are none), would work for universities in the DRC that have the funds to subscribe to online AI services, would not be penalised from being born in a Frenchspeaking country, and would not have to find extra funds to counteract the whims of currency restrictions and exchange rates.
The AI tools to help non-native English speakers already exist. Let’s just make them affordable for everyone.
Adrian Wallwork is author and editor of Springer Science’s English for Academic Research series and lectures PhD students in scientific communication. His latest book is AIAssisted Writing and Presenting in English
