Claire Tian

LALC officer Raeanne Li (10) listens to Dr. Allen Romano’s talk about language models. Raeanne organized the speaker event and moderated the audience questions and discussion.

Natural language processing (NLP) expert Allen Romano explained the uses of large language models (LLMs) when analyzing and understanding classical texts in his talk at the Language and Linguistics Club (LALC) via Zoom on Thursday.

Dr. Romano currently works as an engineer for Khanmigo, the educational LLM chatbot by Khan Academy. Before he entered the tech industry, Dr. Romano graduated from Stanford with a Ph.D. in classics and was a professor at Florida State University, where he conducted LLM research. 

“The challenge of LLMs is: can we, entirely through language, replicate some form of rationality or thinking?” Romano said. “Although these language models are essentially next word completion, they can come up with something that sounds sensible.”

Dr. Romano explained the applications of LLMs in linguistics research and their advantages over traditional NLP models. Although he could access his institution’s resources, Dr. Romano noted that the versatility of LLMs benefits researchers outside of a formal academic setting.

“For a long time, you used to have to build custom tools,” Dr. Romano said. “With the scale of a lot of LLMs and the new techniques developed, you now have a more generalizable tool across certain tasks. It’s a Swiss Army Knife that, even in the short term, is changing the nature of language study and research.”

LALC officer Raeanne Li (10) organized the event. After hearing his Junior Classical League seminar, she invited him to speak at LALC for his experience in the rapidly expanding field of LLMs.

“Right now, everywhere you go there are large language models,” Raeanne said. “Everyone’s using Chat GPT, and a lot of companies are trying to incorporate chatbots into their interface. It was an interesting topic to let him speak about to the club.”

Part of Dr. Romano’s research was training language models on ancient languages. He discussed how language models could detect subtle insights in ancient syntax and semantics that are difficult for non-native speakers to parse.

“I was looking at sociolinguistic markers of ancient speech,” Dr. Romano said. “Imagine you have an ancient drama. Based on the characters’ speech, how do you tell the difference between men and women? There are interesting findings; for example, female goddesses in ancient Greek speak like they are men rather than women in sociolinguistic terms.”

After hearing the talk, LALC club member Timothy Deng (11) grew interested in NLP’s potential to preserve languages that are in danger of extinction.

“I think NLP is a useful tool, especially given the increased globalization and anglicization of the world,” Timothy. “It’s important to know how to be able to converse with people speaking other languages. It’s also important to be able to preserve and safeguard our cultural heritage as people change languages and people drift away from the old vernaculars.”