Speech production-perception relationships in children with speech delay

Kyoko Nagao, Mark Paullin, Vilena Livinsky, James B. Polikoff, Linda D. Vallino, Thierry G. Morlet, N. Carolyn Schanen, H. Timothy BunnellInterspeech 2012

TL;DR

This study investigates the intricate link between how children with speech delay (SD) produce sounds and how they perceive them. By subgrouping children based on their specific articulation errors, the researchers found that some children with SD exhibit significant perceptual difficulties, even for sounds they don't misproduce, suggesting a more complex underlying issue than previously understood.

What Problem Does This Paper Address?

Speech delay (SD) is a common developmental disorder affecting a significant percentage of school-aged children, often leading to broader language and learning challenges. While it's understood that SD involves difficulties in producing speech sounds, the relationship between a child's speech production abilities and their auditory perception of those same sounds is not fully clear. Previous research has yielded mixed results, with some studies showing a direct link between production and perception errors, while others find little correlation. This ambiguity makes it challenging to accurately diagnose and effectively treat children with SD. This paper aims to clarify this relationship by examining speech perception in children with SD, specifically looking at how different types of articulation errors might influence their ability to distinguish between similar speech sounds.

Methods

The researchers employed a categorical perception experiment to assess how children distinguish between speech sounds. They synthesized continua of sounds, meaning they created a series of auditory stimuli that gradually changed from one phoneme to another (e.g., from /s/ to /S/). Children were presented with these continua and asked to identify which sound they heard by selecting corresponding pictures.

The key innovation here is the subgrouping of children with speech delay based on their specific articulation errors. Instead of treating all children with SD as a single group, they were categorized based on whether they misarticulated the /s/ sound, the /S/ sound, both, or neither. This allowed for a more nuanced analysis of how different production patterns relate to perceptual abilities.

Standard techniques like LPC (Linear Predictive Coding) analysis were used to synthesize the stimuli, ensuring a controlled and consistent experimental setup. Statistical analysis involved 2-way repeated measures ANOVAs to compare the identification functions of different subgroups against a control group.

Datasets

The study utilized data from 63 children (ages 5-10 years) who participated in the Nemours Genetics of Speech Delay (SpGen) Project. These children were part of families where at least one child had been diagnosed with delayed speech development. A control group of typically developing children was also included.

The Goldman-Fristoe Test of Articulation 2 (GFTA-2) was used to assess the children's articulation errors. The perception experiment itself generated the "dataset" of identification responses for each child across the synthesized sound continua. No new benchmark datasets were introduced; rather, existing methods were applied to a specific clinical population.

Key Insights and Ideas

  • Heterogeneity of Speech Delay: The study strongly reinforces the idea that speech delay is not a monolithic disorder. Children with SD exhibit diverse patterns of both production and perception difficulties.
  • Production Errors Don't Always Equal Perception Deficits: While some children with SD showed clear perceptual issues related to their production errors, others did not. This highlights that a child might struggle to produce a sound without necessarily having trouble distinguishing it from another sound.
  • The "Both-Error" Group is Particularly Vulnerable: Children who misarticulated both /s/ and /S/ sounds showed significantly impaired perception of these sounds, exhibiting a near-linear identification function. This suggests a fundamental difficulty in processing these phonemes, rather than just a production issue.
  • Unexpected Perceptual Deficits in /k/-/g/ for /S/-Error Group: Surprisingly, children who only had articulation errors on the /S/ sound (but not /s/) showed perceptual difficulties with the /k/-/g/ contrast. This suggests that difficulties with one set of sounds might be linked to broader perceptual processing issues that can affect other sound contrasts.
  • Naturalistic Stimuli Matter: The authors suggest that the use of more naturalistic synthesized stimuli in their study, compared to older research, might explain some discrepancies in findings, particularly regarding boundary shifts.

Key Results

  • /s/-/S/ Contrast:
    • Children with articulation errors on both /s/ and /S/ ("both-error" group) showed significantly poorer performance than controls across most steps of the /s/-/S/ continuum, indicating a non-categorical perception.
  • Children with only /s/ errors ("/s/-error" group) had identification functions very similar to the control group.
  • Children with only /S/ errors ("/S/-error" group) showed a less steep slope and more variability than controls, with a significant difference found at the end-point stimulus.
  • /k/-/g/ Contrast:
    • The "both-error" group again showed significantly poorer performance than controls at several steps, indicating perceptual difficulties even though they did not misarticulate these sounds.
  • The "/S/-error" group also showed significantly different identification scores from controls at specific steps, suggesting perceptual issues with /k/-/g/ sounds.
  • The "/s/-error" group did not show significant differences in /k/-/g/ perception compared to controls.

Limitations and Future Work

  • Stimulus Quality: While the study used more naturalistic stimuli, the authors acknowledge that differences in stimulus quality across studies might explain variations in findings, particularly regarding boundary shifts.
  • Subgroup Size: Some of the subgroups, especially the "both-error" group, were relatively small, which can limit the statistical power of the findings.
  • Scope of Articulation Errors: The study focused on specific phonemes (/s/, /S/, /k/, /g/, /r/, /w/, /b/). Future work could explore a wider range of articulation errors and their perceptual correlates.
  • Acoustic Modeling: The authors suggest that developing acoustic models that parallel perceptual judgments on misarticulated sounds is crucial for better classification of children with speech delay.
  • Further Analysis: The paper mentions ongoing analysis of misarticulated /s/ sounds to understand if errors are due to distortion rather than substitution, which could explain why the /s/-error group didn't show perceptual deficits. This line of research is promising.
  • Temporal vs. Spectral Domains: The study hints at potential deficits in both spectral and temporal domains for some subgroups, suggesting future research could disentangle these specific perceptual processing challenges.

Summary updated

BibTeX
@inproceedings{nagao2012speech,
  author = {Kyoko Nagao and Mark Paullin and Vilena Livinsky and James B. Polikoff and Linda D. Vallino and Thierry G. Morlet and N. Carolyn Schanen and H. Timothy Bunnell},
  title = {{Speech production-perception relationships in children with speech delay}},
  booktitle = {Interspeech},
  year = {2012},
}
Figure