Patients with schizophrenia commonly exhibit a symptom of disorganized thought, including delusions and incoherent speech, which are characterized by impaired connections between concepts. Interestingly, these patients show seemingly contradicting behaviors depending on the level of abstraction, such as concreteness and over-inclusion in their verbal reports. However, the fundamental pathology underling the difficulties that patients face in selecting an appropriate level of abstraction remains poorly understood. In this study, we applied an encoding modeling method to functional magnetic resonance imaging data to obtain cortical representation of various word concepts. We then utilized persistent homology analysis to examine the topological structure of the word relationships in schizophrenia patients, healthy controls, and random data, across different levels of abstraction by varying similarity scales in the representation space. We observed abnormalities in the semantic network of the patients’ brains, where word concepts are homogeneously connected at every hierarchical level, and the topological structures shift toward randomness. The observed topological characteristic of word relationships, spanning from semantically close to distant concepts in schizophrenia, likely underlies the contradictory behaviors observed in verbal reports: the tendency toward concreteness can be attributed to a reduction in contextual modulations due to the homogenization of word relationships at the level of semantically close concepts. The absence of clear boundaries between groups of related words, resulting from the homogenization of the word relationships at the level of semantically distant concepts, leads to over-inclusion.