The American Association for Applied Linguistics recently hosted a webinar on AI in applied linguistics. The topic greatly captivated me; hence, I decided to watch the webinar and share some of the key takeaways I learnt. During the talk, the two guest speakers were: Dr. Vajjala from the National Research Council of Canada and Dr. Burstein from Duolingo.
To start, Dr. Vajjala described Generative AI as a type of AI “that can process and generate content for a range of input-output forms”. She also introduced the concepts of deep learning as well as neural networks. I understood some of the ways that Generative AI learns, such as by predicting what comes next based on a previous string of words and trying to align its responses with human preferences. Linking to language learning, Generative AI can even support learners with writing and speaking support tools and aid with test generation and scoring. It can also help generate more personalised and explicit feedback for learners. The key to working with generative AI for applied linguistics though, is learning how to “prompt” it to do things relevant to the field. This opens the avenue possibilities for research. For example, generative AI can act as a coding assistant, and programming knowledge can be useful for applied linguists and their research. Lastly, Dr. Vajjala made some suggestions on how we can continue to improve generative AI. She proposed developing language tests to understand its current limitations, exploring it for applied linguistics beyond English and developing more guidelines for the appropriate use of generative AI.
Dr. Vajjala’s last suggestion went hand in hand with Dr. Burstein’s talk on the Duolingo English Test’s Responsible AI standards. She first defined Responsible AI as the practice of creating and using AI that’s useful and safe, grounded in ethical principles. She then raised the example of facial recognition, which I had personally never considered in depth. Although the technology was designed with good intentions and is very convenient, the underrepresentation in the training data led to poorer accuracy for faces with darker skin. I researched this at the end of the webinar, and the statistics were shocking. While some facial recognition technology has accuracy rates of over 99% when recognising white male faces, the error rate for faces of colour (particularly black women) can be as high as 35%. Next, Dr. Burstein brought up the case study of the Duolingo English Test (DET). While I have never taken the test myself, as a regular user of the Duolingo app, I was curious to see if I could draw connections between some of the elements she brought up and my experience with the platform. The test is fully AI-powered, with automated content, scoring and plagiarism detection. A fun fact that I learnt is that in addition to being more affordable and convenient than many other tests, the DET is actually accepted by over 5000 programs in more than 100 countries! Furthermore, the DET prioritises using AI responsibly. Hence, Duolingo has published multiple responsible AI standards to the public that were developed through collaboration with experts across all related fields (e.g. machine learning, psychometrics and language assessment). In addition, Duolingo employs a fairness and bias review process as well as a plagiarism detection process, which all combine AI with human involvement to make appropriate final decisions and reviews.
Overall, this webinar not only gave me a better understanding of Generative AI in a broad sense, but more specifically its connections to applied linguistics and the importance of using/developing AI responsibly. The latter is such a significant concept that the government of Canada even published their own “guiding principles for the use of AI in government”! Ultimately, if the ideas I raised in this blog post interest you, I strongly recommend that you watch the talk yourself to understand even more.
Access the webinar here: https://www.aaal.org/webinar-recordings.
Credit: IBM
Works Cited
“Guiding Principles for the Use of AI in Government.” Government of Canada, 30 May 2024, www.canada.ca/en/government/system/digital-government/digital-government-innovations/responsible-use-ai/principles.html. Accessed 31 May 2024.
Hassanin, Nada. “Law Professor Explores Racial Bias Implications in Facial Recognition Technology.” University of Calgary News, 21 Aug. 2023, ucalgary.ca/news/law-professor-explores-racial-bias-implications-facial-recognition-technology. Accessed 31 May 2024.
“Webinar Recordings.” American Association for Applied Linguistics, www.aaal.org/webinar-recordings. Accessed 31 May 2024.