week 2

PSYC0004: Language and Cognition – Week 2

The Science of Reading (Prof. Jenni Rodd)

1. Overview and Aims of the Lecture

This lecture introduces reading as a learned cognitive skill and uses visual word recognition as a central case study to explore broader questions about:

  • How cognitive processes are organised

  • How learning shapes the language system

  • How written words are recognised and mapped onto meaning

The lecture integrates evidence from behavioural experiments, eye‑tracking, and computational modelling, with implications for real‑world debates about how reading should be taught (the Reading Wars).

2. Reading as a Learned Skill

2.1 Why Reading Is Special

  • Reading is not biologically innate (unlike spoken language)

  • It is a recent cultural invention (only a few thousand years old)

  • It is not universal:

    • ~20% of 15‑year‑olds in developed countries have low literacy skills

  • Illiteracy has major consequences:

    • Costs the global economy >$1 trillion per year (World Literacy Foundation, 2015)

    • Impacts health, hygiene, and access to information

2.2 Why the Science of Reading Matters

  • Most children do not learn to read naturally

  • Explicit instruction is typically required

  • Reading underpins all later educational achievement

  • There is major disagreement about the best way to teach reading

  • Scientific evidence is essential to inform educational policy

3. What Is Visual Word Recognition?

3.1 Definition

Visual word recognition refers to the processes that allow a reader to:

  1. Identify letters accurately (e.g. p vs b)

  2. Encode letter order (e.g. form vs from)

  3. Match the visual input to a specific word in the mental lexicon

Key point: recognising a word means uniquely identifying which word is present, not just perceiving letters.

3.2 Why It Is Remarkable

  • Usually easier than spoken word recognition

  • Highly accurate despite noisy or degraded input

  • Extremely fast (see below)

4. The Time Course of Visual Word Recognition

4.1 Speed of Recognition

Within 500–600 ms, readers can:

  • Read a word aloud (naming task)

  • Decide if a letter string is a real word (lexical decision)

  • Make semantic judgments (e.g. living vs non‑living)

Other key facts:

  • Words can be recognised with presentations <100 ms

  • During normal reading, fixations last ~250 ms

Almost all experimental evidence comes from printed text, not handwriting.

5. Eye Movements in Reading

5.1 Basic Properties

  • Reading involves rapid eye movements called saccades

  • Between saccades, the eyes pause in fixations

  • Only the fovea (central retina) provides high‑resolution input

5.2 Visual Constraints

  • Only 4–5 letters fall within the fovea

  • Useful preview extends:

    • ~15 characters to the right

    • ~3–4 characters to the left

  • Direction reverses in right‑to‑left languages (e.g. Hebrew)

5.3 Typical Eye‑Movement Statistics

  • Fixations: 200–250 ms

  • Saccades: ~8 letter spaces

  • Short words (2–3 letters): fixated ~25% of the time

  • Long words (8+ letters): almost always fixated

  • 10–15% of eye movements are regressions (backwards movements)

5.4 Importance for Skilled Reading

  • Eye movements are not random

  • Skilled readers optimise eye movement patterns

  • Readers slow down for difficult or unfamiliar words

  • Eye‑movement control is a core component of fluent reading

6. Factors That Influence Word Recognition

Large‑scale evidence (e.g. the Megastudy approach) shows that many variables affect reading speed and accuracy:

  • Word frequency (house > shark)

  • Word length (rat < hippopotamus)

  • Age of acquisition (giant < data)

  • Semantic ambiguity (bark)

  • Orthographic neighbourhood density (sink vs yacht)

  • Concreteness (book > hope)

These effects strongly constrain theories of word recognition.

7. Models of Visual Word Recognition

7.1 Core Theoretical Question

How does the visual input activate the correct word?

Two broad possibilities:

Serial Processing

  • Words checked one at a time

  • Like dictionary lookup

  • Example: Forster’s Serial Search Model (1976)

Parallel Processing

  • Many words activated simultaneously

  • Competition between candidates

  • Examples:

    • Morton’s Logogen Model (1969)

    • Connectionist models

8. Connectionist (PDP) Models

8.1 What Are Connectionist Models?

  • Computational simulations of cognitive processes

  • Also called:

    • Artificial Neural Networks (ANNs)

    • Parallel Distributed Processing (PDP) models

  • Inspired by neural organisation in the brain

  • Massively parallel processing

  • Behaviour emerges from simple units + learning

Developed primarily in the 1980s (McClelland & Rumelhart)

9. Interactive Activation and Competition (IAC) Model

9.1 Architecture

Three levels:

  1. Visual features

  2. Letters

  3. Words

Connections:

  • Excitatory between levels

  • Inhibitory within levels (competition)

  • Activation spreads bidirectionally (bottom‑up and top‑down)

9.2 What the Model Explains Well

Parallel Processing

  • Demonstrates feasibility of parallel word recognition

Word Superiority Effect

  • Letters are recognised more easily within words than non‑words

  • Example: identifying R in CARD vs CQRD

  • Explained by top‑down feedback from word units

This interactivity remains theoretically controversial.

10. Limitations of the IAC Model

10.1 Letter Position Coding

  • Uses slot‑based coding (position‑specific letter units)

  • Example: C1 ≠ C2 ≠ C3

Consequences:

  • CLAM and CALM share only 50% overlap

  • No tolerance for letter transpositions

  • Predicts identical activation for SEVRICE and SEDLICE

10.2 Empirical Evidence Against This

Readers are tolerant of letter order errors:

  • Transposed‑letter priming effects

  • SERVICE recognised more easily after SEVRICE than SEDLICE

Thus, slot‑based coding is unrealistic.

11. The Role of Sound in Reading

11.1 Why Phonology Matters

For alphabetic languages:

  • Letters map systematically onto sounds

  • Sound‑to‑meaning mappings already exist before reading

  • Print‑to‑sound mapping is easier than print‑to‑meaning

Phonics instruction exploits this structure.

11.2 Two Routes to Reading

Indirect (Phonological) Route

  • Essential for learning

  • Allows reading of novel words (e.g. wug)

Direct Route

  • Important for skilled reading

  • Supports irregular words (e.g. yacht)

Both routes remain active in adulthood.

12. The Reading Wars

Debate over phonics instruction:

  • Phonics is necessary but not sufficient

  • Concerns about reducing enjoyment of reading

  • Evidence supports phonics as foundational, not exclusive

13. Word Meaning Access

13.1 From Word Recognition to Comprehension

  • Recognising a word is only the first step

  • Many words are ambiguous

  • Correct meaning selection is critical for comprehension

Example: multiple meanings of RIGHT (noun, verb, adjective, adverb, politics, direction)

14. Cognitive Mechanisms of Meaning Selection

14.1 Core Findings

  • Multiple meanings activated automatically

  • Rapid selection of context‑appropriate meaning

  • Occasional reinterpretation when context arrives late

14.2 Reordered Access Model (Duffy et al.)

  • Meaning access depends on:

    • Context

    • Meaning dominance (frequency)

Evidence primarily from eye‑tracking studies.

15. Cues That Guide Meaning Selection

Cue 1: Sentence Context

  • Immediate disambiguation reduces processing cost

Cue 2: Long‑Term Experience (Dominance)

  • High‑frequency meanings accessed faster

  • Individual differences exist

Example: rowers develop new dominant meanings for words like catch

Cue 3: Recent Experience

  • Word‑meaning priming effects

  • Recent encounters bias interpretation

  • Effects decay quickly unless reinforced

16. Learning and Flexibility in the Language System

Key conclusions:

  • Meaning preferences are highly flexible, even in adults

  • Short‑term experience produces transient effects

  • Repeated exposure produces long‑lasting changes

  • Learning optimises comprehension by biasing likely meanings

This flexibility likely characterises the entire language system.

17. Big Picture Takeaways

  • Reading is a learned, cognitively demanding skill

  • Visual word recognition is fast but complex

  • Eye movements are central to skilled reading

  • Computational models help test theoretical assumptions

  • Phonology plays a crucial role, especially in learning

  • Word meaning access depends heavily on learning and experience

End of Week 2 Notes