We argue that machine translation provides a useful testbed for developing computational theories of humour. Thus
focusing on the notion of ‘translation humour’, where the humourous note arises from two sentences in different languages,
we suggest a system architecture that extends the standard transfer-based machine translation paradigm with
a humour processing component, which analyzes the full syntactic chart of the source sentence. An implementation
within the existing MT system ‘KIT-FAST’ is under way. We illustrate our approach with an example.
Towards a humour switch for machine translation