In a Linear Associative Net (LAN), all input settles to a singlepattern, therefore Anderson, Silverstein, Ritz, and Jones (1977)introduced saturation to force the system to reach othersteady-states in the Brain-State-in-a-Box (BSB). Unfortunately,the BSB is limited in its ability to generalize because itsresponses are restricted to previously stored patterns. We presentsimulations showing how a Dynamic-Eigen-Net (DEN), a LANwith Short-Term Plasticity (STP), overcomes thesingle-response limitation. Critically, a DEN also accommodatesnovel patterns by aligning them with encoded structure. We traina two-slot DEN on a text corpus, and provide an account oflexical decision and judgement-of-grammaticality (JOG) tasksshowing how grammatical bi-grams yield stronger responsesrelative to ungrammatical bi-grams. Finally, we present asimulation showing how a DEN is sensitive to syntacticviolations introduced in novel bi-grams. We propose DENs asassociative nets with greater promise for generalization than theclassic alternatives.