Multilingual A-Maze experimental software creation

Supervisors: Cui Ding and Junlin Li (Hong Kong Polytechnic University)

A Maze task is a tool to collect incremental reading data. Participants need to select between the target words and the distractors for the reading to continue. 
A-Maze is Maze with automatically generated distractors using LMs, originally designed for English. 

It is difficult to use for other languages because it requires finding a language-specific LM that does not use a subword tokenizer, e.g., BPE. 

Our goal: 

•     Develop a new generation of Multilingual A-Maze using AI agents, which is more flexible. 
•     A demo for Chinese is already written. The results are much better and faster in the text we tested than the existing methods. 
•     Validate this method with several languages. 

Your task: 

•     Understand the demo, and run it for other languages (need a bit adaption, maybe) 
•     Replicate two classic studies (controlled experiment and naturalistic reading) to validate the pipeline. 
•     Analyze the data 
•     Writing 

Python, R, Intermediate NLP Knowledge