We can draw on both episodic memories and semantic knowledge to guide behavior in novel situations. Despite growing interest, key questions remain on the interplay between these systems. How do semantic representations guide the retrieval of episodic memories? How are low-level semantic relationships abstracted from everyday observations (episodes)? How are these low-level relationships transformed to higher-order knowledge structures? By bridging insights from memory-neuroscience with findings in memory-based decision-making, we propose that the sequential sampling of related episodic memories offers valuable insights. We first provide a mechanism through which semantic representations can guide the search path of such cascading recall – favoring the subsequent recall of memories with semantically similar features. We additional propose that the cascading recall process itself recall forms an internal source of information to facilitate the Hebbian learning of abstract representations and allows for the bootstrapped learning of higher-order semantic knowledge through the co-activation of lower-level latent representations. We validate these predictions by conducting two online experiments in a paradigm previously designed to elicit the sequential recall of indirectly related experiences. These experiments demonstrate that memories with semantically (even when not visually) similar features are more likely to be sequentially recalled (experiment 1), the sequential recall of memories increases the semantic similarity of their constituent features, and the sequential activation of two semantic groups increases their higher-order similarity (experiment 2). These findings support the role of cascading recall in the interplay between semantic knowledge and episodic memories, with implications for cognitive science, including sleep-related learning and AI research.