Perceptual Processes in Reading
- Template Matching
- A Computational Analysis of Reading
- PANDEMONIUM
- The Interactive Activation Model
- Acoustic Recoding
- Reading Units
- Reading and the Brain
Template Matching
- What is a template?
- Problems with Templates
- Pre-Processing
What is a template?
- A template is a stored pattern.
- According to template matching theories, letters are recognized by finding a template
that matches them.
Problems with Templates
- Orientation
- Size
- Novelty
- "O" versus "Q"
Preprocessing
- A preprocessor corrects the size and orientation of the image before it is compared to
the stored templates.
- Circularity Problem
- The preprocessor has to know what the pattern is to know its correct size and
orientation.
A Computational Analysis of Reading
- What information is available?
- What is the goal of the computation?
- What strategy is used to achieve the goal with the available information?
PANDEMONIUM (Selfridge, 1959)
- The Architecture
- Representational Assumptions
- Processing Assumptions
- Processing Characteristics
- Empirical Tests
THE ARCHITECTURE
Representational Assumptions
- Features
- Demons
- Demons represent hypotheses.
- Demons are connected to other demons (input, output).
- Levels of Representation
- Decision Demon
- Cognitive (Letter) Demons
- Feature Extraction Demons
- Image Demons
Processing Assumptions
- Each demon has an activation level.
- The activation level begins at baseline (0).
- When a demon receives input, its activation level rises.
- As the activation rises, confidence in the hypothesis rises.
- This is a very straightforward spreading activation process.
Processing Characteristics
- Bottom-up, top-down, or interactive?
- Parallel or sequential within levels?
- Parallel or sequential between levels?
- Automatic or controlled?
- Symbolic or sub-symbolic?
- Modular?
Empirical Tests (Neisser, 1967)
- Visual Search
- Search for 2 targets as quickly as one?
- Similar background harder than dissimilar?
The Interactive Activation Model (McClelland & Rumelhart, 1981)
- The Architecture
- Comparisons to PANDEMONIUM
- An Empirical Test
- A Problem with the IAM
The Architecture
Comparisons to PANDEMONIUM
- Levels of Representation
- Letter/Feature Position
- Activation Levels
- Frequency determines baseline level of activation.
- Recognition occurs when threshold is reached.
- Excitation versus Inhibition
- Horizontal versus Vertical Connections
- Competition Among Competing Alternatives
- Advantages of Simulation
An Empirical Test
- A Demonstration
- The Word Superiority Effect
- Pronounceability versus Gang Effects
The Word Superiority Effect
- It is easier to identify a letter when it occurs in the context of a word.
- Some non-words (e.g., "SLET" but not "XLQJ") also facilitate letter
recognition.
- Why?
Pronounceability versus Gang Effects
- Pronounceability is usually used to explain this pseudo-word superiority effect.
- The IAM suggests that this effect is due instead to the fact that only some non-words
share letter combinations with real words.
- These shared letter combinations create a gang effect.
- Rumelhart & McClelland (1982)
- Pronounceable with Shared Letters ("SLET")
- Unpronounceable with Shared Letters ("SLNT")
- Unpronounceable without Shared Letters ("XLQJ")
A Problem with the IAM
- The entire network must be crafted by hand and that’s not how people learn to read.
- A self-organizing artificial neural network solves the problem.
- This gives rise to distributed representations and sub-symbolic processing (McClalland
& Rumelhart, 1986).
Phonological Recoding
- When people read silently, they automatically activate the sounds of words.
- The Tongue Twister Effect
- Functional Imaging
- Single Route versus Dual Route Theories
- How would you test this?
Reading Units
- Approaches to Reading Instruction
- Frequency and Automaticity
- Healy (1976)
- How could we modify the IAM to accommodate these results?
Reading and the Brain
- Single Cell Recordings and Features
- Neural Style Computation
- The 100 Step Limit
- Functional Imaging and Acoustic Recoding
- Dyslexia
- Peripheral Dyslexias involve visual analysis.
- Central Dyslexias occur at levels beyond visual analysis.
- Can the IAM account for these dyslexias?
Peripheral Dyslexias
- Attentional Dyslexia
- Patients can identify letters/words in isolation but not in context.
- Letter-By-Letter Reading
- Patients read words one letter at a time.
- Fixation time increases with word length.
- No "Pop Out"
- Neglect Dyslexia
- Patients can’t read one side of words (usually the left).
Central Dyslexias
- Surface Dyslexia
- Patients are unable to read irregular words (e.g., "pint" or
"broad").
- Has the direct route been lost?
- Phonological Dyslexia
- Patients can’t read pronounceable non-words.
- Has the indirect route been lost?
- Deep Dyslexia
- Patients make meaning-based word substitutions (e.g. "duel" -->
"sword").
The End!
|