Under-resourced or overloaded? Rethinking working memory deficits in developmental language disorder.

Area of Science:

• Neuroscience
• Computational Linguistics
• Developmental Psychology

Background:

• Dominant theories link developmental language disorder (DLD) to working memory capacity deficits.
• An alternative perspective suggests working memory in DLD is overloaded by poorly discriminable speech representations.

Purpose of the Study:

• To propose and test a computational model where DLD arises from overloaded working memory due to degraded speech representations.
• To investigate the role of low-level auditory processing deficits in shaping internal speech representations and their impact on DLD characteristics.

Main Methods:

• Deep convolutional neural networks trained on degraded and typical spoken word spectrograms.
• Analysis of spoken word recognition accuracy, predictive probability, and entropy.
• Mean-field-theory based manifold analysis to assess representation dimensionality and classification capacity (attentional control proxy).

Main Results:

• Degraded auditory input led to high-dimensional internal speech representations.
• Low representation separability exhausted classification capacity, mimicking attentional control deficits.
• Simulated deficits replicated DLD characteristics like reduced accuracy and increased uncertainty (high entropy, low max probability).

Conclusions:

• Working memory overload, not capacity limitation, is a viable explanation for DLD speech processing deficits.
• Auditory processing deficits can lead to representational and control issues underlying DLD symptoms.
• Computational modeling provides a framework for understanding DLD beyond traditional working memory capacity theories.