This paper discusses work on data collection for Swiss German dialects taking into account the continuous nature of the dialect landscape, and proposes to integrate these data into natural language processing models. We present knowledge-based models for machine translation into any Swiss German dialect, for dialect identification, and for multi-dialectal parsing. In a dialect continuum, rules cannot be applied uniformly, but have restricted validity in well-defined geographic areas. Therefore, the rules are parametrized with probability maps extracted from dialectological atlases.