Presented at the 39\(^\text{th}\) HSP Annual Conference on Human Sentence Processing at MIT, Cambridge, MA, Thursday 26\(^\text{th}\) March 2026, 4 – 6.05 pm.

Funded by UKRI ESRC (ES/W011832/1) – Project title: “Can you use it in a sentence?”: Establishing how word-production difficulties shape text formation

Poster: rpubs.com/jensroes/hsp-2026 (or use QR code)

Introduction

• Expressing ideas in language involves a cascade of processes starting with an intended message, moving on to retrieving words, their spelling or phonology, and finally outputting them in speech, onto paper or screen.
• Our understanding of the timecourse of linguistic planning and spelling retrieval in writing is limited.
• Upcoming linguistic information may be planned serially at linguistic edges (Flower and Hayes, 1980) or in parallel to production (Roeser et al., 2025).
• We tested to what extent difficulty with orthography delays planning of upcoming information.
• Importantly, only the parallel view predicts that disruptions delay planning of upcoming information.

Conclusion

• Spelling difficulty disrupts parallel planning! When writing the N1 and N2 … N2 is retrieved while spelling N1.
• Hesitations related to N1 spelling affect N2 retrieval when there was not enough time for parallel planning.
• Written sentence planning unfolds in parallel to production: difficulty with spelling creates a planning bottleneck with knock-on effects on higher level planning.

Methods

Design

Task: Describe the image arrangement from left to right.

Figure 1: Example item (see e.g. Roeser et al., 2019).

Design: N1 type – see Fig. 1 – where long is \(\ge\) 2 syllables and difficult is \(H_\text{spell}\) > 0.5. DV: inter-key intervals.

Prediction: Retrieval of N2 (clown) is delayed if N1 doesn’t allow for parallel planning when N1 is difficult to spell (broccoli) or too short (wok) .

Materials, Procedure & Participants

Thirty-six items with auditory prime (simple, conjoined subject NP) controlling for N1 retrieval; see Fig. 2.

Figure 2: Example window sequence.

• 24 fillers with structures different form items.
• Sentence-picture match: YES for items; NO for fillers.

Ten ppts per counterbalancing list. Exp. 1: 56 ppts (6 lists), Exp. 2: 80 ppts (9 lists), Exp. 3: 96 ppts (12 lists).

Writing timecourse analysis

Hesitation probability \(\theta\) was inferred from inter-key intervals (iki) using Bayesian mixed-effects mixture models (Roeser et al., 2025, 2024);

\[\begin{equation*} \scriptsize \begin{aligned} \left(\log\left(\text{iki}_\text{ij}\right) - \text{shift}_\text{keyloc[i]}\right) \sim \ & \theta_\text{keyloc,n1type} \times \mathcal{N}\left(\beta_\text{keyloc} + \delta_\text{keyloc} + u_i + w_j, \sigma_{e'}^2\right)\ + \\ & \left(1 - \theta_\text{keyloc,n1type}\right) \times \mathcal{N}\left(\beta_\text{keyloc} + u_i + w_j, \sigma_{e}^2\right)\\ \end{aligned} \end{equation*}\]

also illustrated in Fig. 3.

Figure 3: Hypothetical mixture of fluent (faded) and hesitant (solid) inter-key intervals.

Tutorial: rpubs.com/jensroes/mixture-models-tutorial

Experiment 1

Finding: Spelling difficulty increases hesitations before, during, and immediately after typing N1.

Figure 4: Posterior hesitation probability with 95% probability intervals (PIs).

Experiment 2

Prime without N2 determiner: The N1 and N2 are above the N3 (vs The N1 and the N2 \(\dots\))

Finding: When N1 does not allow parallel planning, “the”-omission increases post-N1 hesitations.

Figure 5: Posterior hesitation probability with 95% probability intervals (PIs).

Experiment 3

(Auditory) prime with N2 preview: The N1 and clown are above the pipe (vs The N1 and bow)

Finding: Post-N1 hesitations are associated with N2 retrival.

Figure 6: Posterior hesitation probability with 95% probability intervals (PIs).

References