A method and apparatus for performing a search for information containing natural language is disclosed which uses a natural language query. The query is input in the form of units of the natural language and this is matched with units in the natural language of the data. Where there are unmatched units in the query and/or the data, context data in the form of one or more unmatched units of the query and/or the data is generated. Each unmatched unit as a predefined linguistic relationship to one of the or each matched unit. Output data is formed as matched units with any respective context data.