These forest rescoring algorithms have potential applications to other computationally intensive tasks involving combinations of different models, for example, head-lexicalized parsing (Collins, 1997); joint parsing and semantic role labeling (Sutton and McCallum, 2005); or tagging and parsing with nonlocal features. A common improvement on this architecture is to pass k-best lists between processing stages, for example (Sutton and McCallum, 2005; Wellner et al. , 2004). So we might want to postpone some disambiguation by propagating k-best lists to subsequent phases, as in joint parsing and semantic role labeling (Gildea and Jurafsky, 2002; Sutton and McCallum, 2005), information extraction and coreference resolution (Wellner et al. , 2004), and formal semantics of TAG (Joshi and Vijay-Shanker, 1999). Recent negative results on the integration of syntactic parsing with SRL (Sutton and McCallum, 2005) provide additional evidence for the difficulty of this general approach. (2005) and Sutton and McCallum (2005) performed a different approach by learning a re-ranking function as a global model on top of the base SRL models.