Rational inattention: A new theory of neurodivergent information seeking.

This paper presents rational inattention as a new, transdiagnostic theory of information seeking in neurodevelopmental conditions that have uneven cognitive and socio-emotional profiles, including developmental language disorder (DLD), dyslexia, dyscalculia and autism. Rational inattention holds that the optimal solution to minimizing epistemic uncertainty is to avoid imprecise information sources. The key theoretical contribution of this report is to endogenize imprecision, making it a function of the primary neurocognitive difficulties that have been invoked to explain neurodivergent phenotypes, including deficits in auditory perception, working memory, procedural learning and the social brain network. We argue that disengagement with information sources with low endogenous precision (e.g. speech in DLD, orthography-phonology mappings in dyslexia, numeric stimuli in dyscalculia and social signals in autism) constitutes resource-rational behaviour. We demonstrate the strength of this account in a series of computational simulations. In experiment 1, we simulate information seeking in artificial agents mimicking an array of neurodivergent phenotypes, which optimally explore a complex learning environment containing speech, text, numeric stimuli and social cues. In experiment 2, we simulate optimal information seeking in a cross-modal dual-task paradigm and qualitatively replicate empirical data from children with and without DLD. Across experiments, simulated agents' only aim was to maximally reduce epistemic uncertainty, with no difference in reward across information sources. We show that rational inattention emerges naturally in specific neurodivergent phenotypes as a function of low endogenous precision. For instance, an agent mimicking the DLD phenotype disengages with speech (and preferentially engages with alternative precise information sources) because endogenous imprecision renders speech not conducive to information gain. Because engagement is necessary for learning, simulation demonstrates how optimal information seeking may paradoxically contribute negatively to an already delayed learning trajectory in neurodivergent children. RESEARCH HIGHLIGHTS: We present the first comprehensive theory of information seeking in neurodivergent children to date, centred on the notion of rational inattention. We demonstrate the strength of this account in a series of computational simulations involving artificial agents mimicking specific neurodivergent phenotypes that optimally explore a complex learning environment containing speech, text, numeric stimuli, and social cues. We show how optimal information seeking may, paradoxically, contribute negatively to an already delayed learning trajectory in neurodivergent children. This report advances our understanding of the factors shaping short-term decision making and long-term learning in neurodivergent children.

Do children really acquire dense neighbourhoods?

Children learn high phonological neighbourhood density words more easily than low phonological neighbourhood density words (Storkel, 2004). However, the strength of this effect relative to alternative predictors of word acquisition is unclear. We addressed this issue using communicative inventory data from 300 British English-speaking children aged 12 to 25 months. Using Bayesian regression, we modelled word understanding and production as a function of: (i) phonological neighbourhood density, (ii) frequency, (iii) length, (iv) babiness, (v) concreteness, (vi) valence, (vii) arousal, and (viii) dominance. Phonological neighbourhood density predicted word production but not word comprehension, and this effect was stronger in younger children.

A maturational frequency discrimination deficit may explain developmental language disorder.

Auditory perceptual deficits are widely observed among children with developmental language disorder (DLD). Yet, the nature of these deficits and the extent to which they explain speech and language problems remain controversial. In this study, we hypothesize that disruption to the maturation of the basilar membrane may impede the optimization of the auditory pathway from brainstem to cortex, curtailing high-resolution frequency sensitivity and the efficient spectral decomposition and encoding of natural speech. A series of computational simulations involving deep convolutional neural networks that were trained to encode, recognize, and retrieve naturalistic speech are presented to demonstrate the strength of this account. These neural networks were built on top of biologically truthful inner ear models developed to model human cochlea function, which-in the key innovation of the present study-were scheduled to mature at different rates over time. Delaying cochlea maturation qualitatively replicated the linguistic behavior and neurophysiology of individuals with language learning difficulties in a number of ways, resulting in (a) delayed language acquisition profiles, (b) lower spoken word recognition accuracy, (c) word finding and retrieval difficulties, (d) "fuzzy" and intersecting speech encodings and signatures of immature neural optimization, and (e) emergent working memory and attentional deficits. These simulations illustrate many negative cascading effects that a primary maturational frequency discrimination deficit may have on early language development and generate precise and testable hypotheses for future research into the nature and cost of auditory processing deficits in children with language learning difficulties. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

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

Dominant theoretical accounts of developmental language disorder (DLD) commonly invoke working memory capacity limitations. In the current report, we present an alternative view: That working memory in DLD is not under-resourced but overloaded due to operating on speech representations with low discriminability. This account is developed through computational simulations involving deep convolutional neural networks trained on spoken word spectrograms in which information is either retained to mimic typical development or degraded to mimic the auditory processing deficits identified among some children with DLD. We assess not only spoken word recognition accuracy and predictive probability and entropy (i.e., predictive distribution spread), but also use mean-field-theory based manifold analysis to assess; (a) internal speech representation dimensionality and (b) classification capacity, a measure of the networks' ability to isolate any given internal speech representation that is used as a proxy for attentional control. We show that instantiating a low-level auditory processing deficit results in the formation of internal speech representations with atypically high dimensionality, and that classification capacity is exhausted due to low representation separability. These representation and control deficits underpin not only lower performance accuracy but also greater uncertainty even when making accurate predictions in a simulated spoken word recognition task (i.e., predictive distributions with low maximum probability and high entropy), which replicates the response delays and word finding difficulties often seen in DLD. Overall, these simulations demonstrate a theoretical account of speech representation and processing deficits in DLD in which working memory capacity limitations play no causal role. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

Prediction Cannot Be Directly Trained: An Extension to Jones and Westermann (2021).

In January 2021, we published an article titled "Predictive Processing and Developmental Language Disorder" in the Journal of Speech, Language, and Hearing Research. The current commentary provides an important extension to this work. Specifically, we aim to head off the suggestion that a child's "predictive capacity" may be trained independently of improving the quality of their long-term speech representations.

Predictive Processing and Developmental Language Disorder.

Purpose Research in the cognitive and neural sciences has situated predictive processing-the anticipation of upcoming percepts-as a dominant function of the brain. The purpose of this article is to argue that prediction should feature more prominently in explanatory accounts of sentence processing and comprehension deficits in developmental language disorder (DLD). Method We evaluate behavioral and neurophysiological data relevant to the theme of prediction in early typical and atypical language acquisition and processing. Results Poor syntactic awareness-attributable, in part, to an underlying statistical learning deficit-is likely to impede syntax-based predictive processing in children with DLD, conferring deficits in spoken sentence comprehension. Furthermore, there may be a feedback cycle in which poor syntactic awareness impedes children's ability to anticipate upcoming percepts, and this, in turn, makes children unable to improve their syntactic awareness on the basis of prediction error signals. Conclusion This article offers a refocusing of theory on sentence processing and comprehension deficits in DLD, from a difficulty in processing and integrating perceived syntactic features to a difficulty in anticipating what is coming next.

Density and Distinctiveness in Early Word Learning: Evidence From Neural Network Simulations.

High phonological neighborhood density has been associated with both advantages and disadvantages in early word learning. High density may support the formation and fine-tuning of new word sound memories-a process termed lexical configuration (e.g., Storkel, 2004). However, new high-density words are also more likely to be misunderstood as instances of known words, and may therefore fail to trigger the learning process (e.g., Swingley & Aslin, 2007). To examine these apparently contradictory effects, we trained an autoencoder neural network on 587,954 word tokens (5,497 types, including mono- and multisyllabic words of all grammatical classes) spoken by 279 caregivers to English-speaking children aged 18-24 months. We then simulated a communicative development inventory administration and compared network performance to that of 2,292 children aged 18-24 months. We argue that autoencoder performance illustrates concurrent density advantages and disadvantages, in contrast to prior behavioral and computational literature treating such effects independently. Low network error rates signal a configuration advantage for high-density words, while high network error rates signal a triggering advantage for low-density words. This interpretation is consistent with the application of autoencoders in academic research and industry, for simultaneous feature extraction (i.e., configuration) and anomaly detection (i.e., triggering). Autoencoder simulation therefore illustrates how apparently contradictory density and distinctiveness effects can emerge from a common learning mechanism. OPEN RESEARCH BADGES: This article has earned an Open Data badge for making publicly available the digitally-shareable data necessary to reproduce the reported results. The data is available at https://osf.io/2qk5j/. Learn more about the Open Practices badges from the Center for Open Science: https://osf.io/tvyxz/wiki.

Neighborhood Density and Word Production in Delayed and Advanced Learners.

Purpose This study reexamines the claim that difficulty forming memories of words comprising uncommon sound sequences (i.e., low phonological neighborhood density words) is a determinant of delayed expressive vocabulary development (e.g., Stokes, 2014). Method We modeled communicative development inventory data from (N = 442) 18-month-old children, with expressive lexicon sizes between 0 and 517 words (Mdn = 84). We fitted a Bayesian regression model in which the production of each communicative inventory word (N = 680) by each child was predicted by interactions between that child's expressive lexicon size and the word's (a) phonological neighborhood density, (b) frequency in child-directed speech, (c) length, (d) babiness, and (e) concreteness. Results Children with larger expressive lexicons were more likely to produce words comprising uncommon sound sequences than age-matched children with smaller lexicons. However, the magnitude of the interaction between expressive lexicon size and phonological neighborhood density was modest relative to interactions between expressive lexicon size and word frequency, length, babiness, and concreteness. Conclusion Emphasis on a difficulty with the memorization of low-neighborhood density words as a determinant of slow vocabulary growth may be unwarranted, and the current evidence base in this direction is not robust enough to strongly support the development of possible interventions for late talkers.

Auditory Lexical Decisions in Developmental Language Disorder: A Meta-Analysis of Behavioral Studies.

Purpose: Despite the apparent primacy of syntactic deficits, children with developmental language disorder (DLD) often also evidence lexical impairments. In particular, it has been argued that this population have difficulty forming lexical representations that are detailed enough to support effective spoken word processing. In order to better understand this deficit, a meta-analysis of studies testing children with DLD in the auditory lexical decision task was conducted. The objective was to provide summary effect size estimates for accuracy and response time measures for comparisons to age- and language-matched control groups. Method: Two thousand three hundred seventy-two records were initially identified through electronic searches and expert consultation, with this cohort reduced to 9 through duplicate removal and the application of eligibility and quality criteria. The final study cohort included 499 children aged 3;8-11;4 (years;months). Results: Multivariate analysis suggests that children with DLD were significantly less accurate in the auditory lexical decision task than age-matched controls. For the response time estimate, however, confidence intervals for the same group comparison crossed 0, suggesting no reliable difference between groups. Confidence intervals also crossed 0 for language-matched control estimates for both accuracy and response time, suggesting no reliable difference between groups on either measure. Conclusion: Results broadly support the hypothesis that children with DLD have difficulty in forming detailed lexical representations relative to age- though not language-matched peers. However, further work is required to determine the performance profiles of potential subgroups and the impact of manipulating different lexical characteristics, such as the position and degree of nonword error, phonotactic probability, and semantic network size.