Language Learning and Theories of Acquisition

Language

• Language is a system to transfer complex information between minds using marks on paper, squeezing air through your throat and mouth, or moving your hands.

Language Learning

• Language learning requires phonetics, phonology, morphology, syntax, semantics, and pragmatics
• Learning language from sound is possible without explicit teaching of linguistic concepts
• Language learning is a complex process

Language Acquisition: Nativism and Empiricism

• Nativism suggests innate, specialized cognitive modules exist for language processing
• Empiricism suggests learning is based on experience and general cognitive abilities

Theories of Language Acquisition

• Universal Grammar (UG) theory proposes an innate language-specific module for grammar processing
• The Anti-UG view suggests language learning is entirely based on general cognitive abilities

Arguments for Universal Grammar

• Poverty of the stimulus argument: children learn complex grammar despite limited exposure to complex sentence structures.
• Linguistic Universals argument: commonalities across languages suggest innate constraints on possible languages.
• Emergence of structured languages in environments without structured language suggest innate grammar modules.

Arguments Against Universal Grammar

• Occam's Razor suggests children might learn complex grammar through simplified explanations rather than innate modules.
• Universal language features might have counter examples in some languages contradicting the assumption of innate constraints.
• Modern language models demonstrate learning complex structures without innate constraints, but models are trained on significantly more data than humans.

Why Don't Animals Use Language?

• Animals can be taught to use "words", but lack complex sentence construction and displacement (referencing things not present)

Theories of Animal Language

• Chomsky: Universal Grammar mutation is unique to humans
• Terrace: humans learn words through shared attention

Up Next

• First paper due tomorrow at 11:59pm
• Grading rubric is available on Courseworks
• Example papers emailed on October 1st
• Thursday class: Introduction to the brain

Mario Example

• Mario's position is compared to the barrels' positions
• If Mario has a hammer and his position matches a barrel, it signifies a specific condition

Hardware

• Uses IDT6116SA/LA 2Kb C-MOS RAM for storing positional information

Personal Space

• Interpersonal distance violation negatively impacts the evaluation of one's partner
• Expected distance is 3 ft with adjustments based on familiarity
• Degree of response is determined by the difference between expected and actual distance
• Striatum tracks expected and actual distances and computes a prediction error

Ultimatum Game

• Player 1 proposes a division of 100 units.
• Player 2 can accept the division or reject it, resulting in both players receiving 0 units.
• The game illustrates the dynamics of fairness and cooperation in decision-making.

Computational Level Theory

• Focuses on explaining players' behavior in the Ultimatum Game
• Attempts to describe the strategies and rational choices involved
• Analyzes the probabilities of rejection based on the proposed division.

Algorithmic/Representation Level

• Aims to represent and compute the maximum share for Player 1
• One representation uses binary numbers for each share option
• Another approach involves continuous-valued intensities that compete against each other over time.

Implementation Level

• Focuses on physically implementing the representation and algorithm
• Explores how these concepts can be realized in the real world
• Relates to the physical mechanisms that support the decision-making process.

Marr's Three Levels

• A framework for understanding computational systems
• Used to analyze the "how" and "why" of mental processes.
• Levels: Computational, Algorithmic, and Implementational
• Example: cash register.

Neurons and Computation

• Neurons can be analyzed at the computational level
• Hubel and Wiesel's work on visual processing demonstrates how neurons "compute" specific information about stimuli.
• The response of a neuron can be described in terms of its preferred stimuli and computational function.

High-Speed Microscopy

• Advances in microscopy allow detailed observation of brain activity
• Captures fleeting signals related to brain function, such as the firing patterns of individual neurons.

Donkey Kong Simulation

• Computer simulations can be used to model cognitive processes
• The brightness of pixels in a Donkey Kong game can be used to represent different information

Abstraction Levels

• Different questions require different levels of abstraction
• Marr's Three Levels provide a framework for understanding the nature of explanations at different levels.

Levels of Explanation

• To understand how the mind works, we need to examine it at different levels of abstraction.
• One way to think about this is by asking different kinds of questions about a computational system:
  • What are the goals of the system?
  • What are its inputs and outputs?
  • What representation does the system use?
  • How are these representations transformed?
  • How are these operations physically implemented?

Marr’s Three Levels

• Marr's Three Levels offer a framework for understanding a computational system:
  • Computational (goals/properties): This level focuses on the overall goals and properties of the system.
  • Algorithmic (representations/processes): This level describes the specific algorithms and representations used to achieve the system's goals.
  • Implementational (physical hardware): This level explains how the algorithms and representations are physically implemented in the real world.

Examples:

• Cash Register: The first example is the cash register.

  • Computational: The goal of a cash register is to calculate the total price of items.
  • Algorithmic: The cash register represents prices in base-10 and uses addition to calculate totals.
  • Implementational: Digits in the cash register may represent quantities to be added, and their positions may correspond to different powers of 10 (e.g., ones, tens, hundreds).

• Donkey Kong: The second example is playing Donkey Kong on a computer.

  • Computational: When Mario hits a barrel with a hammer, the barrel disappears.
  • Algorithmic: The game checks if Mario's position matches the barrel's and if Mario has the hammer.
  • Implementational: The game stores position information in a specific type of RAM chip (IDT6116SA/LA 2Kb C-MOS RAM).

• Personal Space: This example describes how we perceive and respond to violations of personal space.

  • Computational: Violations of personal space lead to negative evaluations of the partner.
  • Algorithmic: We maintain an expected distance based on familiarity and liking. The degree of response depends on the difference between expected and actual distance.
  • Implementational: The striatum (part of the brain) tracks expected and actual distances, computing a 'prediction error'.

• Ultimatum Game: The final example showcases the Ultimatum Game, a game where players negotiate over a sum of money.

  • Computational: Players' decisions can be explained by maximizing their own share.
  • Algorithmic: We can represent value choices using binary numbers or continuous-valued intensities.
  • Implementational: The brain can implement these algorithms by representing values as neural activity and using neural networks to process them.

This quiz explores the intricate processes of language learning and acquisition theories, including nativism and empiricism. Delve into the concepts of Universal Grammar and the arguments supporting it. Test your understanding of how we acquire and process language.