Sections from Castles, Rastle & Nation (2018): Ending the Reading Wars: Reading Acquisition from Novice to Expert,

Chapter 6: Studying Eye Movements (pp. 168–170)

Why study eye movements?

  • Eye movements provide online, moment-by-moment evidence of language processing.

  • Unlike reaction times, they show where and when processing difficulty occurs during reading.

  • Based on the assumption that eye position closely reflects attention.

Key components of eye movements in reading

  • Fixations

    • Brief pauses (≈200–250 ms) where visual information is extracted.

    • Most linguistic processing occurs during fixations.

  • Saccades

    • Rapid eye movements between fixations.

    • Vision is largely suppressed during saccades.

  • Regressions

    • Backward saccades to earlier text.

    • Often indicate processing difficulty, ambiguity, or integration problems.

Measures used in eye-tracking studies

  • First-fixation duration

    • Sensitive to early lexical processing (e.g., word frequency).

  • Gaze duration

    • Sum of all fixations on a word before moving on.

    • Reflects lexical access + early integration.

  • Regression path duration

    • Time from first fixating a word until the eye moves past it to the right.

    • Sensitive to syntactic/semantic difficulty.

  • Total reading time

    • Includes re-reading; reflects later, integrative processes.

What eye movements reveal

  • Readers do not process all words equally:

    • Short, frequent, and predictable words are often skipped.

  • Processing difficulty shows up immediately, not only after sentence completion.

  • Strong evidence against strictly serial, post-lexical models of reading.

Chapter 6: Models of Visual Word Recognition (pp. 192–198)

Core problem

How does the reader:

  1. Identify a visual pattern

  2. Activate the correct word representation

  3. Do so rapidly and accurately, despite noise and ambiguity?

1. The Logogen Model (Morton)

  • Each word has a logogen (a word detector).

  • Logogens accumulate evidence from:

    • Visual input

    • Context

  • When activation exceeds a threshold, the word is recognised.

Key strengths

  • Explains:

    • Word frequency effects (lower thresholds for frequent words)

    • Context effects

Limitations

  • Vague about:

    • Letter-level processing

    • Competition between words

  • Largely descriptive, not computational.

2. Interactive Activation Model (IAM) (McClelland & Rumelhart)

Architecture

  • Three levels:

    1. Feature level

    2. Letter level

    3. Word level

  • Excitatory connections between compatible units.

  • Inhibitory connections between competitors at the same level.

Key principles

  • Processing is parallel.

  • Information flows bottom-up and top-down.

  • Word recognition emerges from competition.

Explains

  • Word superiority effect (letters recognised better in words than isolation).

  • Frequency effects.

  • Contextual facilitation.

Importance

  • One of the most influential models in psycholinguistics.

  • Basis for later models of spoken word recognition (e.g., TRACE).

3. Dual-Route Cascaded (DRC) Model

  • Two routes for reading aloud:

    • Lexical route: whole-word recognition

    • Non-lexical route: grapheme-to-phoneme conversion

Explains

  • Reading of:

    • Irregular words (e.g., yacht)

    • Non-words (e.g., blint)

Criticism

  • Less successful at explaining:

    • Semantic effects

    • Graded frequency effects

  • More modular than interactive models.

Chapter 6: Effects of Meaning Frequency & Prior Context (pp. 202–205)

⚠ This section contains more detail than required — key take-home points are highlighted.

Word frequency effects

  • High-frequency words:

    • Shorter fixation durations

    • More likely to be skipped

  • Frequency effects occur:

    • Early (first fixation)

    • Persist even with strong contextual support

Meaning (sense) frequency

  • Words with multiple meanings:

    • Dominant meaning accessed faster than subordinate meanings.

  • Subordinate meanings cause:

    • Longer fixations

    • Increased regressions

Prior context effects

  • Context facilitates word recognition but does not eliminate frequency effects.

  • Strong context:

    • Speeds recognition

    • Reduces competition

  • Evidence supports interactive models, not purely bottom-up ones.

Theoretical implications

  • Lexical access is:

    • Probabilistic

    • Influenced by prior knowledge

    • Sensitive to distributional properties of language

Appendix: Connectionism (pp. 481–483)

What is connectionism?

  • A class of models where cognition emerges from:

    • Many simple processing units

    • Operating in parallel

    • Connected by weighted links

Also known as:

  • Parallel Distributed Processing (PDP)

  • Neural network models

Core principles

  1. Simple units

    • Units pass activation, nothing more.

  2. Weighted connections

    • Positive weights excite

    • Negative weights inhibit

  3. Emergent behaviour

    • Complex cognition arises from simple interactions.

Interactive Activation & Competition (IAC) models

  • No learning (weights are hand-coded).

  • Used extensively in word recognition.

  • Include:

    • Excitation between compatible units

    • Inhibition between competitors

Processing

  • Activation spreads through cycles.

  • Network settles into a stable pattern.

  • Best-matching word remains active.

Learning vs non-learning models

  • IAC models: explain how processing works

  • Learning models (e.g., back-propagation): explain how knowledge is acquired

  • Back-propagation details are beyond the required scope

Why connectionism matters

  • Explains:

    • Frequency effects

    • Generalisation to novel items

    • Graded, probabilistic behaviour

  • Avoids explicit symbolic rules.

  • Strongly influenced modern psycholinguistic theory.

I