In this paper, the Grammatical inference (GI) is considered as an instance of the Constraint Satisfaction Problem (CSP), and a Constraint Logic Program (CLP) is proposed which instantiates automata in a lattice. In this task, the inference engine takes as input a set of individual positive and negative examples of documents and outputs a set of rules that recognises similar documents. In the theoretical outline, the Pattern Recognition and the GI problems are considered in an algebraic framework in which a set of constraints will define the process of generalisation. The CLP implementation discussed here is used in a document handling project in which (paper) documents are typographically tagged and then recognised. The main aim is to extract the physical and the logical structures of a given set of (paper) documents in order to produce a machine readable form like XML, HTML or TeX format.