AI won’t replace teachers anytime soon, but it can be used for language learning and teaching, explains Rob Reynolds.
We are in the midst of the AI revolution. At least, that’s what some are calling it. Artificial intelligence is changing the way we write, program, search the web, and more—but how will it change how we teach? Assistant Professor Rob Reynolds (Linguistics, Digital Humanities), in a lecture on January 31, argued that AI tools can be useful for students and teachers looking to streamline and facilitate the language-learning process. His lecture was titled “Using Artificial Intelligence to Help Teach, Learn, and Assess Russian.”
Artificial “Intelligence”
For anyone not on the cutting edge of computer science, generative AI might seem a bit like magic or real human intelligence: how can computer software read and write just like (or even better than) a real human being? For example, when prompted to write a limerick about Brigham Young University, the program ChatGPT returns the following in a matter of seconds:
Not bad for a computer!
Reynolds, a researcher in natural language processing (NLP), explained that ChatGPT and other generative AI programs are simply the product of advanced predictive computation. “These models are really good at giving you the most probable words, but they have no idea what they are doing,” he said. Essentially, they don’t understand the meanings of the words they generate; their writing is based on patterns they identify in the language they were trained on.
AI Language Learning: Probabilistic NLP versus Rule-Based NLP
For its apparent strengths, probabilistic AI (such as ChatGPT) has some significant shortcomings. “Probabilistic NLP is not good at being explainable,” Reynolds said. “If it makes an error, you have no idea why it made that error. It just happened, and the only way to fix that error is to train on more data and hope that error goes away.” This unpredictability means that ChatGPT isn’t ideal for helping language learners pick up the rules of a language.
Because of this drawback, Reynolds is more interested in building tools in a different area of AI: rule-based NLP, which relies on hard-coded rules to process text. The tools that Reynolds works with can’t generate a Shakespearean sonnet, but they have the benefit of being explainable and predictable while harnessing the benefits of computing power. These tools, specially designed for students and teachers of Russian, offer the benefits of AI for language learning.
RuMOR
The most recent of Reynold’s projects is RuMOR (Russian Mentor for Orthography), which takes Russian writing as an input and generates information on the errors in the writing. It is able to identify word and grammar errors, suggest corrections, and provide explanations.
To make RuMOR as accurate as possible, Reynolds incorporated two important computational features: a finite-state transducer and a constraint grammar. In a nutshell, Reynolds explained, “The finite-state transducer will generate every possible reading and the constraint grammar will remove any that it can based off of context.” This way, the AI language-learning tool can reliably identify what feature the writer was aiming for and provide correct suggestions, which can prove invaluable for language learners getting familiar with Russian spelling and grammar.
ICALL
Another tool that Reynolds has worked on is ICAll, which automatically generates grammar exercises based on web pages in Russian. Though the program was originally built by other NLP researchers for other languages, Reynolds built on its Russian capabilities.
Unfortunately, the program is currently down, Reynolds said. Such is the impermanent nature of digital tools like these. But when he gets it back up and running, it will provide language learners with a powerful tool to turn any web page into grammatical exercises—from multiple choice questions to fill in the blanks to identifying parts of speech.
FLAIR
The final tool that Reynolds discussed promises to ease a common issue in language learning and teaching: finding texts at appropriate reading levels. It’s called FLAIR (Form-focused Language-Aware Information Retrieval), and it functions essentially as a grammatically aware search engine. It allows users to perform a search on the web and sorts the results based on customizable parameters such as text difficulty, text length, and presence of linguistic features (such as tense, sentence types, and different parts of speech).
This tool provides perhaps the most significant benefit for language teachers. It aims to reduce the intensive process of finding good texts for students by letting a computer perform the processing.
Final Thoughts
The point of these tools, Reynolds explained, is to “automate the boring stuff.” AI language-learning tools are not meant to replace teachers or make language learning irrelevant; they are meant to supplement the processes of teaching and learning. “There are parts of language learning, teaching, and assessment that can be tedious, monotonous, or just very difficult that computers are actually quite good at,” he said. “By building tools that can do those things for us, we can focus on the learning.”
Learn more about how AI language-learning tools and other digital humanities projects are changing the way we teach and learn at the BYU Office of Digital Humanities website.