A method and apparatus for identifying the focus of a document, in a natural language processing application, the natural language processing application comprising a hierarchical concept tree having a plurality of nodes, each node being associated with a term, the method comprising the steps of: mapping an input document to nodes in a concept tree to determine a number of occurrences of a term in the input document which also occur at a node in the concept tree; weighting each node in the concept tree, depending on the determined number of occurrences of the term in the input document and a determined value assigned to each node in the concept tree; traversing the concept tree to identify a heaviest weighted path, in dependence on the weighting of each node in the concept tree; and determining the focus of the input document by identifying a node having the heaviest weight along the most heavily-weighted path.