J Speech Lang Hear Res. 2014 Apr 1;57(2):467-83. doi: 10.1044/2013_JSLHR-L-12-0254.
PURPOSE In this study, the authors used neural network modeling to investigate the possible mechanistic basis of developmental language delay and to test the viability of the hypothesis that persisting delay and resolving delay lie on a mechanistic continuum with normal development. METHOD The authors used a population modeling approach to study individual rates of development in 1,000 simulated individuals acquiring a notional language domain (in this study, represented by English past tense). Variation was caused by differences in internal neurocomputational learning parameters as well as the richness of the language environment. An early language delay group was diagnosed, and individual trajectories were then traced. RESULTS Quantitative variations in learning mechanisms were sufficient to produce persisting delay and resolving delay subgroups in similar proportions to empirical observations. In the model, persisting language delay was caused by limitations in processing capacity, whereas resolving delay was caused by low plasticity. Richness of the language environment did not predict the emergence of persisting delay but did predict the final ability levels of individuals with resolving delay. CONCLUSION Mechanistically, it is viable that persisting delay and resolving delay are only quantitatively different. There may be an interaction between environmental factors and outcome groups, with individuals who have resolving delay being influenced more by the richness of the language environment.
目的 在这项研究中，作者使用神经网络建模来探究发展性语言延迟的可能机制基础，并检验持续延迟和解决延迟与正常发育存在机制连续性的假设的可行性。
方法 作者使用群体建模方法研究了 1000 名模拟个体在习得一个概念语言领域（在本研究中，用英语过去时表示）中的个体发展速度。变化是由内部神经计算学习参数的差异以及语言环境的丰富度引起的。诊断出一个早期语言延迟组，然后追踪个体轨迹。
结果 学习机制的定量变化足以产生与实证观察相似比例的持续延迟和解决延迟亚组。在模型中，持续的语言延迟是由处理能力的限制引起的，而解决延迟是由低可塑性引起的。语言环境的丰富度并不能预测持续延迟的出现，但确实可以预测解决延迟个体的最终能力水平。
结论 从机制上讲，持续延迟和解决延迟只是数量上的不同是可行的。环境因素和结果群体之间可能存在相互作用，解决延迟的个体受到语言环境丰富度的影响更大。
