The ability to learn about regularities in the environment and to make predictions about future events is fundamental for adaptive behaviour. We have previously shown that people can implicitly encode statistical regularities and detect violations therein, as reflected in neuronal responses to unpredictable events that carry a unique prediction error signature. In the real world, however, learning about regularities will often occur in the context of competing cognitive demands. Here we asked whether learning of statistical regularities is modulated by concurrent cognitive load. We compared electroencephalographic metrics associated with responses to pure-tone sounds with frequencies sampled from narrow or wide Gaussian distributions. We showed that outliers evoked a larger response than those in the centre of the stimulus distribution (i.e., an effect of surprise) and that this difference was greater for physically identical outliers in the narrow than in the broad distribution. These results demonstrate an early neurophysiological marker of the brain’s ability to implicitly encode complex statistical structure in the environment. Moreover, we manipulated concurrent cognitive load by having participants perform a visual working memory task while listening to these streams of sounds. We again observed greater prediction error responses in the narrower distribution under both low and high cognitive load. Furthermore, there was no reliable reduction in prediction error magnitude under high-relative to low-cognitive load. Our findings suggest that statistical learning is not a capacity limited process, and that it proceeds automatically even when cognitive resources are taxed by concurrent demands.

了解环境规律性并预测未来事件的能力是适应性行为的基础。先前我们已经证明，人们可以隐式编码统计规律并检测其中的违例，这反映在对带有独特预测错误签名的不可预测事件的神经元反应中。然而，在现实世界中，关于规律性的学习通常会在竞争性认知需求的背景下发生。在这里，我们问统计规律的学习是否受到并发认知负荷的调节。我们比较了与对窄音或宽高斯分布采样频率的纯音声音响应相关的脑电图指标。我们发现，异常值所引起的反应要比刺激分布中心处的反应大（例如，的结果），并且在物理上相同的异常值在窄范围内比在宽分布范围内更大。这些结果证明了大脑在环境中隐式编码复杂统计结构的能力的早期神经生理学标记。此外，我们通过让参与者在听这些声音流的同时执行视觉工作记忆任务来操纵并发的认知负荷。我们再次在较高和较低的认知负荷下，在较窄的分布中观察到较大的预测误差响应。此外，在相对于低认知负荷的情况下，预测误差幅度没有可靠的降低。我们的发现表明，统计学习不是能力受限的过程，