The formation of linguistic dependencies is subject to memory interference. In this study, we ask whether memory representations are fixed, or whether they can be distorted and updated after their initial encoding. Models of Cue-Based Retrieval assume that memory representations are fixed. However, representational interference and rational inference models assume that memory contents can be edited. To examine this, we test how reflexive attraction is affected by preceding verbal agreement in Hebrew. Cue-Based Retrieval suggests that agreement on the verb can exaggerate the reflexive’s sensitivity to the distractor. In contrast, we propose that if memory representations can be edited, verbal agreement can alter the representation of the subject. This process would reduce vulnerability to distortions originating from the distractor. In two self-paced reading experiments and one forced-choice completion we find (i) decreased reflexive attraction when (grammatical or ungrammatical) agreement cues were available on the preceding verb; and (ii) a preference for reflexive forms matching the verb over forms matching the subject when the sentence included ungrammatical verbal agreement. These results suggest that comprehenders use featural information from the verb to recover properties of the subject. The findings are therefore consistent with a memory model where representations can be distorted and updated, as well as with rational inference about memory disruption.

语言依赖性的形成会受到记忆的干扰。在这项研究中，我们询问记忆表征是否是固定的，或者它们在初始编码后是否可以被扭曲和更新。基于提示的检索模型假设内存表示是固定的。然而，表征干扰和理性推理模型假设内存内容可以编辑。为了检验这一点，我们测试了希伯来语中先前的口头协议如何影响反射性吸引力。基于提示的检索表明，对动词的一致可能会夸大反射对干扰因素的敏感性。相比之下，我们建议，如果记忆表征可以被编辑，那么口头协议就可以改变主体的表征。这个过程将减少干扰因素带来的扭曲的脆弱性。在两个自定进度的阅读实验和一个强制选择完成实验中，我们发现（i）当前面的动词有（语法或非语法）一致提示时，反射性吸引力降低；(ii) 当句子包含不合语法的口头协议时，优先选择与动词匹配的反身形式而不是与主语匹配的形式。这些结果表明理解者使用动词的特征信息来恢复主语的属性。因此，这些发现与表征可以被扭曲和更新的记忆模型一致，也与关于记忆破坏的理性推理一致。