Semantics of 'deep' grammars - its applicability and evaluation

Presenting advances in semantic specification using Minimal Recursion Semantics embedded in an HPSG grammar, the paper addresses the question how the semantic performance of such grammars can be evaluated. We suggest that if a grammar can contribute to correct steering of an artificial agent in a text-based instruction application, then this is as close as one can get to a correctness criterion. However, this is a type of situation we can still only imagine in this paper, we show examples of semantic representations which stand a chance of providing the explicitness required in a steering scenario, and comment on the prospect of more systematically, and with more coverage, producing such representations using the type of grammar in question.