This paper presents a novel fuzzy clustering algorithm that can cluster sentence-level text into overlapping clusters. Unlike hard clustering methods, fuzzy clustering allows sentences to belong to multiple clusters to different degrees. As sentences can relate to more than one topic, this fuzzy approach is appropriate for sentence clustering. The algorithm uses a graph representation and operates in an expectation-maximization framework to identify semantically related sentences based on their pairwise similarities. It has potential applications in various text mining tasks. Results show it can effectively cluster sentences into overlapping topics.