Automatically Determining the Semantic Gradation of German Adjectives

The semantics of gradable adjectives, in particular their order with respect to one another on a given scale, are not sufficiently represented in lexical resources such as wordnets. Thus, it is not clear whether superb expresses the quality of “goodness” equally, more, or less strongly than excellent or great. Sheinman and Tokunaga (2009) determine the relative gradation of English adjectives by applying lexical-semantic patterns that hold between members of pairs of similar descriptive adjectives to corpus searches. The patterns identify one member of such pairs as the one that expresses a stronger, or more intense, degree of the property denoted by the scale than the other member. By iteratively applying these patterns to a set of adjectives, Sheinman and Tokunaga (2009) arrive at a uniform score for each adjective that allows them to place it at an appropriate point on a scale. We extend the AdjScales method (Sheinman and Tokunaga 2009) to some frequent and salient German adjectives to test its crosslingual robustness. Our work has consequences for automatic text understanding and generation, lexicography and language pedagogy.