next up previous contents
Next: Evaluation using the unconstrained Up: Experimental Evaluation Previous: Prolog Data Structure   Contents


Experiments

In this section the preliminary experimental evaluations of the indexing methods proposed in Chapter 6 are presented. All experiments were carried out using the MERGE grammar described in Section 7.1. The performance was timed on a Sun Workstation with 1024 MB of memory, running an UltraSparc v.9 processor at 440 MHz with 2MB of cache. The parser was implemented in SICStus 3.10.1 for Solaris 8.

The baseline used in these experiments was the EFD parser. As shown in [Penn and Munteanu2003], a parser that reduces the copying of edges is already faster than several other available parsers. Also, the chosen Prolog encoding of TFSs proved to be very efficient when used in the EFD parser. All experiments were carried out using both the unconstrained and constrained versions of the MERGE grammar.

Figure 7.3: Parsing times for EFD, EFD with positional indexing, and EFD with path indexing applied to the unconstrained MERGE grammar. Sentences were numbered according to the ascending order of parsing times for path indexing.
\includegraphics[width=\textwidth]{merge_new_tests_all.eps}



Subsections
next up previous contents
Next: Evaluation using the unconstrained Up: Experimental Evaluation Previous: Prolog Data Structure   Contents