The sequencing problem

How do people decide what comes next when they speak or write?

  • Language is built word by word, not all at once.
  • Each choice reflects both past experience and current production pressures.

In four connected topics, we address the role of prediction, chunking, interruptions and monitoring in real-time language processing.

Topic 1: Experience shapes what comes next

Contributors: Francesco Cabiddu, Robyn Griffiths, Sofia Tsitsopoulou

  • People are sensitive to patterns in the language they encounter.
  • These regularities create expectations about upcoming information.

Time-course learning figure

Sequence learning is a candidate for efficient chunking.

Topic 2: Familiarity buys processing capacity

Contributors: Harriet Smith, Paula Stacey

Words that co-occur become organised into larger units: chunks.

Efficiency gained from a familiar sequence also benefits the words around it.

Sequence Example sequence
High frequency F1 · party leader · F2
Zero frequency F1 · surface leader · F2

Topic 3: Difficulty ripples through sentence

Contributors: Pablo Aros Muñoz

Spelling difficult can interrupt the flow of production, not just locally!

Topic 3: Difficulty ripples through sentence

Topic 4: Real-time automatic writing support

Contributors: Evgeny Chukharev, Abram Anders, Rianne Conijn, Emily Dux Speltz, Wren Bouwman

  • Students usually receive feedback on the finished product, not on the moment-by-moment process that produced it.
  • Intelligent tutoring system that supports source-based writing without waiting for a completed draft.

Eye movements can be used to model which parts of a source remain active as a writer reads, rereads, and begins composing.

Topic 4: Reading-history model

One picture of language sequencing

  1. People learn statistical patterns from experience.
  2. Repeated patterns can become efficient chunks.
  3. Current production pressures can disrupt this efficient system.
  4. Behavioural signals let us observe and respond to those disruptions.

Language sequencing continuously balances learned expectations with the constraints of the present moment.

Thank you!

The Leverhulme Trust
Grant RPG-2024-197 · “Comprehensively understanding the time-course of statistical learning”

Economic and Social Research Council – New Investigator
Grant ES/W011832/1 · “Can you use it in a sentence?: Establishing how word-production difficulties shape text formation”

US National Science Foundation
Grant 2302644 · “SourceWrite: Real-time, biometric, intention-informed scaffolding of source-based writing processes”
Grant 2016868 · “ProWrite: Biometric feedback for improving college students’ writing processes”