Soft Unification: Towards Robust Parsing of Spontanous Speech

This article describes a modified unification mechanism and its integration into a LR(1)-parser to provide robust spontaneous speech parsing of German language. The model is robust to speech recognition errors and wrongly pronounced morphologically similar words. The modified unification framework presented here is a symmetric, order independent version of unification on feature structures. The difference is that unification of two features structures in this framework will never fail, even if both feature structures have inconstitent values for the same attribute.