Original title: Typological Feature Prediction with Large Language Models: An In-Context Learning Approach
Qianwen Wang, York Hay Ng, Aditya Khan, En-Shiun Annie Lee · September 2026
Languages have specific traits, like how they build words or arrange sentences. Multilingual computer systems use these traits, but many languages are missing this information. Older methods try to guess missing features, but they do not explain their answers. Researchers tested large language models to solve this. Instead of retraining the models, they gave them examples directly in the prompt using data from URIEL+ and Glottolog. Simply asking the models with no clues failed. But when the prompts included facts about related languages and nearby geographical neighbours, the models beat every baseline method. This approach worked well even for low-resource languages. Most importantly, the models wrote explanations that matched the provided evidence, making their predictions understandable.
Imagine trying to guess if a new shop will be open on Sundays. If you have no information, you're guessing blindly. But if you learn that it's in a shopping district where most shops are open Sundays and owned by a chain that always operates then, your guess becomes informed. In this paper, the system predicts language features by using evidence from geographically close or historically related languages, much like using neighbor patterns to predict shop hours.
This helps build better language technologies for underserved languages. Having explainable predictions makes the results more trustworthy and usable in real-world applications.
The abstract doesn't say explicitly, but the work is limited to predicting typological features using evidence from URIEL+ and Glottolog, and it doesn't explore other types of evidence or all feature categories.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Typological Feature Prediction with Large Language Models: An In-Context Learning Approach — Qianwen Wang, York Hay Ng, Aditya Khan et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.