Information Load Predicts Stuttering Events

Area of Science:

• Speech and Language Pathology
• Computational Linguistics

Background:

• Stuttering variability is a complex phenomenon.
• Information load, measured by word predictability, is a potential factor influencing stuttering.
• Word surprisal quantifies information load based on linguistic context.

Purpose of the Study:

• To investigate if words with higher information load (word surprisal) are more prone to stuttering.
• To control for linguistic variables affecting stuttering.
• To examine if stuttering severity and impact moderate the relationship between word surprisal and stuttering.

Main Methods:

• Analysis of spontaneous speech from 35 adults who stutter (FluencyBank corpus).
• Computation of word surprisal using Generative Pretrained Transformer 2 (GPT-2).
• Statistical control for six linguistic predictors of stuttering and moderation analysis with severity/impact measures.

Main Results:

• Increased word surprisal significantly predicted a higher likelihood of stuttering.
• Linguistic covariates were controlled for in the analysis.
• No significant moderation effect of stuttering severity or impact was found.

Conclusions:

• Higher information load in words contributes to stuttering events during spontaneous speech.
• This may be due to increased cognitive or communicative demands of high-surprisal words.
• Individual differences in managing communicative pressure, not solely clinical severity, likely shape stuttering patterns.