Most of the research aimed at improving the times of the retrieval component of parsing uses statistical methods that require training. During grammar development, the time spent for the entire edit-test-debug cycle is important. Therefore, a method that requires considerable time for gathering statistical data by parsing a training corpus could burden the development process. Indexing methods that are time efficient for the entire grammar development cycle are a necessary alternative.
Current techniques (such as the ``quick-check'', [Malouf et al.2000]) reduce the parsing times by filtering out unnecessary unifications based on statistically derived filters. Widely used in databases [Elmasri and Navathe2000] and automated reasoning [Ramakrishnan et al.2001], indexing presents the advantage of a more principled, yet flexible, approach. Compared to simple filtering, an indexing structure allows for searches based on multiple criteria. It also provides support for efficiently processing complex queries that are not limited to membership checking [Manolopoulos et al.1999], such as range queries (only when the index is organized as a tree, not as a hash). The flexibility is reflected in the ease of adapting an existing parser to different grammars: an indexed parser needs only changes in the search criteria (a process that can be entirely automated), while a statistically optimized parser needs re-training.