Unification-based formalisms, which represent the state of art in natural language parsing, show a real and novel convergence on the way of structuring lexical information, of using this information in the rules, and on the data structures to resort to during the parsing process. This convergence makes it possible to see more clearly which basic software components should be available to build natural language processors. OLMES, a general written in CLOS, has unification-based parser, been designed to be versatile and easily extensible, in order to realize various language processing tools. The implementation of OLMES is being used for teaching natural language processing and parser design. The underlying classes have been specialized so as to provide linguists with improved pattern- matching for compound terms retrieval, or testbed for spelling checkers... Last, the inheritance graph and the combination of methods, an original feature of CLOS, were taken advantage of to mirror the complex interaction in idioms between general syntactic rules and idiosyncrasies.