Several studies have demonstrated an influence of semantic knowledge on verbal working memory (WM) performance, such as shown by the observation of semantic relatedness (related vs. unrelated words) and word imageability (high vs. low imageability words) effects in working memory. The present study extends these observations by examining in four experiments the extent to which semantic knowledge can protect WM representations against interference. We assessed immediate serial recall performance for semantically related vs. unrelated word lists and for high vs. low imageability word lists, with memory lists being followed by an interfering task after encoding or not. Results show that semantic relatedness leads to a stronger protective effect against interference than word imageability at the item level. Furthermore, the semantic relatedness had a stronger impact on WM performance than word imageability; this was further supported by a meta-analysis of all relevant studies in the field. These results suggest that inter-item associative semantic knowledge can protect WM content against interference, but less so item-level semantic knowledge. This protective effect may result from between-item recurrent reactivation or from reduced cognitive load via the compression of memoranda into conceptual units, as further supported by a series of computational simulations.

多项研究已经证明了语义知识对言语工作记忆（WM）性能的影响，例如通过观察语义相关性（相关与不相关的单词）和单词可成像性（高与低可成像性单词）对工作记忆的影响。本研究通过在四个实验中检查语义知识可以保护WM表示不受干扰的程度，扩展了这些观察结果。我们评估了语义相关与不相关的单词列表以及高与低可成像性单词列表的即时串行召回性能，无论是否进行编码后，内存列表之后都会出现干扰任务。结果表明，在项目级别，语义相关性比单词可成像性导致更强的抗干扰保护作用。此外，语义相关性对WM性能的影响比单词可成像性更强。对本领域所有相关研究的荟萃分析进一步支持了这一点。这些结果表明，项目间关联语义知识可以保护WM内容不受干扰，但项目级语义知识则较少。这种保护作用可能是由于项目间的反复激活或通过将备忘录压缩为概念单元而减少的认知负荷所致，进一步受到一系列计算模拟的支持。