Given the present state of work in natural language processing, this address argues first, that advance in both science and applications requires a revival of concern about what language is about, broadly speaking the world; and second, that an attack on the summarising task, which is made ever more important by the growth of electronic text resources and requires an understanding of the role of large-scale discourse structure in marking important text content, is a good way forward.