Summary by Dr. Linh Phung
I recently attended a webinar in early December of 2024 with Carol Chapelle around the book that she and her colleagues published as an open resource: Exploring AI in Applied Linguistics. As an FYI, Carol Chapelle is an applied linguist and professor at Iowa State University, past president of the American Association for Applied Linguistics, former co-editor of Language Testing, and former editor of TESOL Quarterly. The webinar was organized by the CALICO's AI SIG and attracted 150 colleagues from all over.
I was in a noisy room in an open office, sitting next to someone who talked very loudly on the phone (I thought some people thought the world liked to hear what they said), so I didn't follow all of the ideas shared, but I did introduce myself and Pangea Chat and asked about its potential for language learning. She commented on how AI would just rewrite your language for you but if the tool was built to assist them in the language production process, then it sounded very helpful. I said that is exactly what we are building with Pangea Chat: allowing students to say what they want to say while drawing their attention to language form (a focus-on-form approach). Pangea Chat now supports 49 target languages and over 100 base languages (L1).
Colleagues also shared a lot of resources in the chat, and I ended up reading Generative Artificial Intelligence and Applied Linguistics by Sowmya Vajjala from National Research Council, Canada. It gives a broad overview of generative AI and is quite interesting indeed. She summarizes recent research on generative AI in areas related to applied linguistics (such as Natural Language Processing) and identifies the three categories of language learning technologies:
Content and test generation
Assessment
Assistive tool development (My side note: Pangea Chat and Eduling Speak would fall into this category)
The challenges identified include:
Brittleness of the prompt-based querying process (small change in the prompt leading to changes in the output, lowering predictability)
Hallucinations
Distinguishing between machine and human-generated output (although there's some development in watermarking the output)
Compromise of the integrity of computer-based testing scenarios (This is interesting as I now think at-home testing like in the Duolingo English Test now faces this challenge).
Want to explore more?
Link to the article: https://jalt-publications.org/sites/default/files/pdf-article/jj46.1-art3.pdf?
Link to Chapelle et al.'s book: https://www.iastatedigitalpress.com/plugins/books/154/
Comments