We consider the problem of characterisation of sequences of heterogeneous symbolic data that arise from a common underlying temporal pattern. The data, which are subject to imprecision and uncertainty, are heterogeneous with respect to classification schemes, where the class values differ between sequences. However, because the sequences relate to the same underlying concept, the mappings between values, which are not known ab initio, may be learned. Such mappings relate local ontologies, in the form of classification schemes, to a global ontology (the underlying pattern). On the basis of these mappings we use maximum likelihood techniques to handle uncertainty in the data and learn local probabilistic concepts represented by individual temporal instances of the sequences. These local concepts are then combined, thus enabling us to learn the overall temporal probabilistic concept that describes the underlying pattern. Such an approach provides an intuitive way of describing the temporal pattern while allowing us to take account of inherent uncertainty using probabilistic semantics.