We propose a model for a statistical representation of the conceptual structure of a restricted subset of spoken natural language. The model is used for segmenting a sentence into phrases and labeling them with concept relations (or cases). The model is trained using a corpus of annotated transcribed sentences. An understanding system is being built around this model, allowing for unconstrained spoken input in a database retrieval task. The results on a test set of 148 sentences show that almost 97% of cases were correctly assigned.