Path indexing is an extension of the positional indexing presented in Section 6.1. Its functionality is related to quick check: extract a vector of types (of feature values) from a mother (that will become and edge) and from a daughter, and test the unification of the two vectors before attempting to unify the edge and the daughter.
Path indexing differs from quick-check in two major aspects. First, as mentioned in Section 4.5.2, quick check needs statistical training to decide from which nodes to extract the quick check vectors. Path indexing identifies these nodes by a static analysis of grammar rules, performed off-line and with no training required. Second, path indexing is built on top of positional indexing, therefore the vector of nodes used as a pre-test for unification can be different for each pair of mother-daughter that can unify. This does not appear to be the case for quick-check, as it was conceived of in [Malouf et al.2000].
This section will first introduce the theoretical foundations of the static analysis used to select nodes for inclusion in the path indexing. Next, details about path indexing will be given.