Document scaling techniques have been widely used in political science to infer partisanship measures and to rank documents on a scale of ideal points, based on bag-of-word approaches. These approaches typically underestimate the semantic and syntactic patterns contained in the corpus. Recent advances in natural language processing, particularly semantic search models, offer an improved topic coherence due to a semantic space of embedded words and documents, whose structure is able to identify topics without setting their number as a hyperparameter. We propose a scaling technique, namely TopicShoal, that extracts meaningful topic vectors using a semantic search technique (Top2Vec) and scales partisanship among speakers or parties using a Bayesian factor analysis on the document-topic distances, thereby enabling a semantic explanation of the ideal points’ variations. This novelty, suited for both monolingual and multilingual corpora, addresses the bag-of-word constraint by capturing the narrative signals in the corpus and exploiting a coherent and independent topic vector structure. Applied to a corpus of German party manifestos and Deutsche Bundesbank executive board members’ speeches, TopicShoal successfully identifies discourse-level differences among parties and speakers via topic intensities, whose projection on the ideal points’ scale reveals common debated themes and other sideline interests that differentiate parties and speakers.