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.
• What this tells us about language production?

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 / write?

• 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 boy and the dog moved above 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.

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.

A pause is not just empty time!

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

• Mainstream view: 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.

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.
• 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.

If writing is parallel, then a problem at one level should not only delay the difficult word but disrupt planning downstream.

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

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

• We looked at how cognitive processes are coordinated when planning language.
• Keystrokes, eye movements, and text content can show problems before the final text does.
• We use process data to detect when writing support is needed and to distinguish different writing problems.

The goal is not to replace human feedback. It is to make tailored support available at the moment when it may be most useful.

ProWrite: Dux Speltz et al. (2022)

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.
• That matters if we want writing support that is fair rather than one-size-fits-all.
• Cognitive research can therefore inform teaching, assessment, and intervention.

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

@sentwrite.bsky.social