The cognitive mechanisms underlying the processing of non-adjacent syntactic dependencies are critical for the understanding of human language processing. For instance, a verb needs to be syntactically and semantically integrated with its subject, or a reflexive like himself needs to be syntactically bound by its antecedent before it can be assigned any meaning. Thus, when processing the second part of a syntactic dependency, the parser needs to retrieve the corresponding first part of this dependency. The mechanisms underlying these syntactically triggered retrieval processes have drawn considerable attention in psycholinguistic research. Lewis and Vasishth (2005) (LV05) developed a model of sentence processing which is based on the general cognitive architecture ACT-R (Anderson et al., 2004). This model assumes a content-addressable memory in which cue-based retrieval processes are subject to similarity-based interference from (partially) cue-matching distractors. The LV05 model has widely been used to explain interference effects observed in the processing of syntactic dependencies such as reflexive-antecedent or subject-verb dependencies. Although the model is able to capture some of the empirically observed effects, there is a range of data the model is unable to explain. We propose to extend the LV05 model by two independently motivated assumptions, namely dynamic cue confusion and activation-sensitive interference. We demonstrate that this extended model explains a wide range of empirically observed effects the original LV05 model does not account for.