Skip to main content
Apply Give
Menu

Comp Cog Sci Group - Zhongyu Hu

A young man in a t-shirt, standing in front of a screen and a chalkboard, delivering a presentation.

Comp Cog Sci Group - Zhongyu Hu

Linguistics Thursday, October 22, 2026 1:00 pm - 2:00 pm Biology-Psychology Building, 1140A

Thursday October 22, Zhongyu Hu leads the Computational Cognitive Science meeting with a presentation of his MA work: “Less-is-More in LSTM Language Models: A Task-Conditional Capacity-by-Input Interaction in Subject–Verb Agreement.” 

In this work Zhongyu tests Elissa Newport's Less-is-More hypothesis by limiting the memory of LSTM language models with a forget-gate penalty, crossing that with simple-to-complex vs. mixed input ordering, and testing the models on subject–verb agreement. So far, the results show that less memory alone does not help learning. However, when it is paired with ordered input, the cost of reduced memory on hard agreement items goes away.

Add to Calendar 10/22/26 13:00:00 10/22/26 14:00:00 America/New_York Comp Cog Sci Group - Zhongyu Hu

Thursday October 22, Zhongyu Hu leads the Computational Cognitive Science meeting with a presentation of his MA work: “Less-is-More in LSTM Language Models: A Task-Conditional Capacity-by-Input Interaction in Subject–Verb Agreement.” 

In this work Zhongyu tests Elissa Newport's Less-is-More hypothesis by limiting the memory of LSTM language models with a forget-gate penalty, crossing that with simple-to-complex vs. mixed input ordering, and testing the models on subject–verb agreement. So far, the results show that less memory alone does not help learning. However, when it is paired with ordered input, the cost of reduced memory on hard agreement items goes away.

 Biology-Psychology Building false