Working Memory Training Transfer-CaiLab-202309

Whether and how the effect of working memory (WM) training can transfer to untrained tasks has been under intensive debate. Few studies have clarified training transfer boundaries or distinguished sources of training improvements. In our study, we trained participants with a delayed estimation task for locations and systemically tested its transfer boundaries with three types of untrained WM tasks: tasks with the same paradigm and changed stimuli (i.e., delayed estimation tasks for colors and letters), tasks with changed paradigms with the same stimuli (i.e., complex span and n-back tasks for locations), and tasks with combined changes in paradigms and stimuli (e.g., a complex span task for colors). We found decreased recall errors in both the trained task and the delayed estimation task for colors. In particular, among the participants with larger training improvements, we observed transfers in both the color delayed estimation task and the location complex span task but not in the color complex span task or others. These results indicated that training could transfer to tasks with either stimuli or paradigm changes, but there was an extra transfer boundary for tasks with combined changes. Furthermore, we adapted model fittings to estimate WM quantity and quality separately in delayed estimation tasks for locations (trained) and colors (untrained). Our results revealed that both increased WM quantity and quality contributed to the recall improvement in the trained task, and individuals with lower baseline gained larger training benefits. For the untrained task, we found that the recall improvement was underlain by increased quality and an optimized quality-quantity trade-off strategy. Additionally, participants’ baseline performance positively predicted the transfer benefit. Our findings suggested that training improved the stimulus-specific WM resource and optimized the paradigm-specific WM strategy. These results extended our understanding of WM training transfers and shed light on future studies.

工作记忆（working memory, WM）训练的效应能否以及如何向未训练任务发生迁移，一直是学界广泛争论的议题。鲜有研究明确界定训练迁移的边界，或是区分训练提升的来源。在本研究中，我们采用针对位置的延迟估计任务（delayed estimation task）对被试进行训练，并系统检验了其迁移边界，共设置三类未训练的WM任务：范式相同但刺激材料改变的任务（即针对颜色与字母的延迟估计任务）、范式改变但刺激材料相同的任务（即针对位置的复杂广度任务（complex span task）与n-back任务（n-back task）），以及范式与刺激材料均发生改变的任务（例如针对颜色的复杂广度任务）。我们发现，在训练任务以及针对颜色的延迟估计任务中，被试的回忆错误均有所降低。具体而言，在训练提升幅度更大的被试群体中，我们观察到了颜色延迟估计任务与位置复杂广度任务上的迁移效应，但未在颜色复杂广度任务或其他任务中发现迁移。上述结果表明，训练可向仅改变刺激材料或仅改变范式的任务发生迁移，但对于范式与刺激材料均发生改变的任务，迁移存在额外边界限制。此外，我们通过模型拟合分别估算了训练用位置延迟估计任务与未训练用颜色延迟估计任务中的WM容量与质量。结果显示，WM容量与质量的共同提升促成了训练任务中的回忆表现改善，且基线水平更低的个体能获得更大的训练收益。对于未训练任务，我们发现其回忆表现的提升源于质量提升以及优化后的质量-容量权衡策略。此外，被试的基线表现可正向预测其迁移收益。本研究结果表明，训练可提升针对特定刺激的WM资源，并优化针对特定范式的WM策略。上述发现深化了我们对WM训练迁移机制的理解，可为后续相关研究提供启示。