Language, Cognition & Computation

by Albert Major

Course Focus & Shift in Methodology

• The course studies language as a communication system: how it’s represented, processed, and learned by humans.
• The second half of the course focuses on computational modeling as a method for studying human language cognition.
• Key question: Can AI models (like ChatGPT) tell us something about how humans learn language?

Human Language vs. AI Language Models

• AI systems (like GPT-4, Siri, Alexa) now produce very natural, human-like language.
• But: Do they learn/process language like humans do?
  • Machines use huge amounts of data.
  • Human children learn language from very limited exposure.
  • The difference suggests machines and humans likely learn differently.

Studying Human Cognition

• Traditional method: controlled lab experiments or natural observation (e.g., watching how kids learn new words).
• Example: Fast mapping experiment:
  • A child is shown a known object (apple) and an unknown object.
  • When asked “Can you give me the dax?”, the child gives the unknown object.
  • Why? They infer the unfamiliar word refers to the unfamiliar object.
  • This shows children use inference and elimination to learn word meanings.

The Brain as a Black Box

• We can’t see inside a working human brain.
• So we treat it as a black box:
  • Provide input → observe output → infer what’s inside based on behavior.

Modeling Human Cognition

There are 3 core components:

1. Environment (stimuli, input)
2. Cognitive Process (the black box)