The author starts with a review of the use of computers in the analysis of natural languages and then describes and evaluates developments from formal models through early NLP to the current situations. The author points out the difficulties which the processing of 'open text' poses for explaining the way that words make meaning, at the same time indicating a way forward through the construction of what he calls 'local grammars' based on observed patterns in textual data. Illustrating the local grammar approach in relation to the word 'trouble' based on the evidence of the 500-million-word bank of English corpus, he demonstrates that neither NLP techniques nor conventional grammatical and semantic constructs are necessary to the analysis.