The everyday puzzle

How can writing feel fluent when the mind has not finished planning what comes next?

  • Writers often start before the full sentence, paragraph, or argument is ready.
  • The fingers keep moving while ideas, words, and spelling are still being prepared.
  • The seminar asks what this tells us about language production.

The central claim is simple: fluent writing depends on parallel planning, and difficulty reveals where that parallel system breaks down.

Three questions for the seminar

  1. How do people pre-plan what they want to say or write? (Part 1 & 2)
  2. What happens when spelling becomes difficult? (Part 3)
  3. How can real-time signals help us support different writers? (Part 4)

Part 1: Planning Ahead

Example 1: planning scope

Papers: Roeser et al. (2019); Roeser, Torrance, et al. (2024); Roeser, Torrance, et al. (2025)

How do people prepare what they are going to say or write before they begin?

  • People do not usually prepare an entire sentence before speaking or typing.
  • But they may prepare more than the first word.
  • The test case here is whether syntactic relations between nouns change how far planning reaches.

The key question is not whether planning happens, but why do people sometimes plan more than the first word before output starts.

“The boy and the dog moved above the kite”

Why syntax matters here (Roeser et al., 2019)

Conjoined start:

The boy and the dog moved above the kite.

Simple start:

The boy moved above the dog and the kite.

  • Longer starts for conjoined phrases show planning beyond the first noun.
  • Eye movements showed that attention moved to “dog” before sentence start.
  • Similar results for both speech and writing.

Writing is not just speech slowed down, but both systems plan ahead when the upcoming dependency makes that useful.

Part 2: Flow

Writing usually flows

When things are going well, writing can be remarkably fluent.

  • The writer knows what they want to say.
  • The words and spellings are available.
  • Typing or handwriting can keep moving.

The puzzle starts when this flow is disrupted: why do writers pause?

Why do we pause?

Keystrokes are a behavioural trace of how planning, retrieval, spelling, and motor execution are coordinated in real time.

  • Serial models: upcoming words are planned during pauses.
  • But if writers always had to stop to plan the next word, writing would not flow.
  • Upcoming words might be planned at the same time as we write.

The modelling question is not just “where are the pauses?”, but “what process generated this timing pattern?”

Serial model of planning

In serial models stages have to wait for one another.

Parallel model of planning

The next idea can start while current idea is being turned into keystrokes.

Parallel consistently beats serial planning

Papers: Roeser, De Maeyer, et al. (2024); Roeser, Conijn, et al. (2025); Roeser & Torrance (2026)

250 copy-task recordings, 1k writers producing 5m interkey intervals.

  • We fit serial and parallel planning models to six natural writing corpora.
  • The serial model treated delays as systematic effects of text location, word/sentence boundaries, and bigram frequency.
  • The parallel model allowed upcoming words and sentences to be planned while typing continued.
  • Word and sentence boundaries often had no pause, including in 126 sixth-graders where serial planning is often expected.

Transitions without pauses show writers were already planning in tandem.

Part 3: Friction

Where the system gets stuck

If writing is parallel, then a problem at one level should affect other levels.

  • Spelling is a useful test case.
  • It is low-level enough to manipulate experimentally.
  • But it can still disrupt higher-level sentence planning.

Spelling difficulty is not just a spelling problem; it can become a planning problem.

Spelling as friction

Difficult spelling delays planning of “and thimble”.

Spelling as friction

Talks: Roeser, Aros Muñoz, et al. (2024); Roeser et al. (2026)

Easy/short

The fly and thimble are above the peg.

Difficult/long

The caterpillar and thimble are above the peg.

Dropping “the” after “and” increases friction: there is less time to plan the next noun in parallel.

Word omission exposes downstream disruption

Part 4: Signals

Applying process research to writing support

  • So far we looked at how cognitive processes are coordinated when planning language.
  • The applied step is to use process data to detect when writing support is needed.
  • Keystrokes, eye movements, and semantic traces can show problems before the final text does.

Support should respond to how writers are composing, not only to what they hand in.

Feedback during writing

ProWrite: Dux Speltz et al. (2022)

Process traces can be turned into feedback while the writer is still composing.

Eye tracking: reading as process data

Eye movements make source consultation visible as behavioural data.

Reading-history model

Part 5: The bigger picture

A single account

Flow

Parallel planning keeps output moving.

Friction

Local difficulty blocks the cascade and slows the text.

Writing is a mixture process grounded in parallel, cascaded computation.

Inclusive academia in action

The applied aim is to make writing support more precise, timely, and fair.

  • Children, adults, multilingual writers, and writers with literacy difficulties may differ in process as well as product.
  • Real-time evidence can help distinguish ability, access, strategy, and momentary difficulty.
  • Cognitive research can therefore inform teaching, assessment, and intervention.

Closing claim

Fluent writing is not the absence of planning. It is planning that has learned to run ahead.

When that running-ahead fails, we see friction. Understanding that friction is how we build better theory and better support.

Thank you!

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

US National Science Foundation (PI: Chukharev)
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”

References

Dux Speltz, E., Roeser, J., & Chukharev-Hudilainen, E. (2022). Automating individualized, process-focused writing instruction: A design-based research study. Frontiers in Communication, 7, 933878. https://doi.org/10.3389/fcomm.2022.933878

Roeser, J., Aros Muñoz, P., & Torrance, M. (2024). Spelling matters: The timecourse of sublexical retrieval in written word production. Experimental Psychology Society Meeting. https://eps.ac.uk/wp-content/uploads/2024/03/EPS-NTU-Programme-v3.pdf

Roeser, J., Aros Muñoz, P., & Torrance, M. (2026). Write here, write now: Spelling difficulty disrupts parallel sentence planning. 39th Annual Conference on Human Sentence Processing. https://hsp2026.org/

Roeser, J., Conijn, R., Chukharev, E., Ofstad, G. H., & Torrance, M. (2025). Typing in tandem: Language planning in multi-sentence text production is fundamentally parallel. Journal of Experimental Psychology: General, 154(7), 1824–1854. https://doi.org/10.1037/xge0001759

Roeser, J., De Maeyer, S., Leijten, M., & Van Waes, L. (2024). Modelling typing disfluencies as finite mixture process. Reading and Writing, 37(2), 359–384. https://doi.org/10.1007/s11145-021-10203-z

Roeser, J., & Torrance, M. (2026). Flow in mind, flow in fingers: Parallelism in written language production. The Cognitive Psychology Bulletin, 11, 42–48. https://doi.org/10.53841/bpscog.2026.1.11.42

Roeser, J., Torrance, M., Andrews, M., & Baguley, T. (2024). No default syntactic scope for advance planning in sentence production: Evidence from finite mixture models. OSF. https://doi.org/10.31219/osf.io/u7v36

Roeser, J., Torrance, M., & Baguley, T. (2019). Advance planning in written and spoken sentence production. Journal of Experimental Psychology: Learning, Memory, and Cognition, 45(11), 1983–2009. https://doi.org/10.1037/xlm0000685

Roeser, J., Torrance, M., & Baguley, T. (2025). Semantic contrast ahead: Contrast guides pre-planning in complex noun-phrase production. Language and Cognition, 17, e64. https://doi.org/10.1017/langcog.2025.10022