Automatic Recognition of Theme, Focus and Contrastive Stress

Theme, focus and contrastive stress are categories which are necessary for the solution of several problems in NLP, including word order determination, scope recognition and anaphora resolution. However, they largely depend on the context, and analysing the discourse structure of a text is di cult and time-consuming. As context has an impact on the order of constituents in context-embedded sentences, it is possible in many cases to identify theme, focus and stress at a sentential level without referring to the context. This paper describes the complex German word order mechanism and presents an algorithm to identify the three categories automatically. With certain language-speci c changes, this algorithm can be applied to other languages than German.