Constructing a data model for a problem area requires identifying and formulating some symbolic representation of the concepts involved, their characteristics, and the relationships between them. Taking a socio-cultural perspective on learning, analysis of classroom dialog is used to identify cognitive challenges met by novice students of data modelling. This paper shows how Vygotskyan theory of concept building sheds light on some psycholinguistic aspects of data modelling. The high-school students in the study displayed a lack of what will be called "metalinguistic consciousness". Many of their problems were related to the conceptualisation process of forming entities and assigning appropriate labels to them. In teaching data modelling, there seems to be a need to focus more explicitly on the four-way relationship between (1) concrete or abstract objects of the world, (2) the terms denoting these objects, (3) the related subjective meaning and (4) the symbolic representation in a data model. (Contains 4 figures.)