LFG-DOP is a powerful, hybrid model of language processing where the tree representations of Data-Oriented Parsing (DOP) are augmented with the functional representations of Lexical Functional Grammar (LFG). The result is a robust parsing model which generates linguistically informed output. However, difficulties arise in the accurate implementation of fragmentation and sampling in this model. Due to these unresolved issues, there is currently no satisfactory implementation of the LFG-DOP model. In this thesis, we propose a backing-off to Grammatical Feature-DOP (GF-DOP). The GF-DOP model differs from Tree-DOP and LFG-DOP in that the trees are annotated with selected features extracted from the f-structure, rather than explicitly linked to corresponding f-structure units. In this way, we rnake use of the irlformation available to us in the f-structure, while avoiding the problems inherent in the implementation of LFG-DOP. We aim to improve the quality of the parses generated by modeling additional functional and feature information. Experiments on the HomeCentre corpus have shown this model to be a valuable middleground between the two alternative models. GF-DOP has been shown to outperform the Tree-DOP model, as a result of its ability to identify and make use of grammatical features, while maintaining the integrity of the probability model.