File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/intro/90/h90-1051_intro.xml
Size: 2,044 bytes
Last Modified: 2025-10-06 14:04:55
<?xml version="1.0" standalone="yes"?> <Paper uid="H90-1051"> <Title>Using Explanation-Based Learning to Increase Performance in a Large-Scale NL Query System</Title> <Section position="3" start_page="254" end_page="254" type="intro"> <SectionTitle> 5. Conclusions and further </SectionTitle> <Paragraph position="0"> directions On the basis of the experiments reported here, we think there are good reasons to take EBL seriously as a practical and generally applicable way of optimizing NL query systems; the speed-ups achieved were very considerable at a low overhead. Even more importantly, it was possible to apply the EBL method despite the target's having several characteristics undesirable from this point of view; our a priori guess at the beginning of the project was that, if it worked here, it would work on most systems. We plan soon to begin implementation of a similar module for a large unification grammar for Swedish, where it should be easy to cover both syntactic and semantic processing.</Paragraph> <Paragraph position="1"> One thing that ought to be studied more is the dependence of access time on the number of learned rules when this number becomes large (over 500, say). It certainly seems reasonable to hope that the pattern-matching algorithm presented here will give approximately logarithmic behaviour, but this is really an empirical question, since it depends on the distribution of the common query-types in terms of their lexical categories. Another important question is the extent to which it is possible to compress the generated rules. Since we are essentially trading space for time, this is likely to define the limits of the method, since we will eventually simply run out of space to store more learned rules, even if we can index them efficiently.</Paragraph> <Paragraph position="2"> In conclusion, it seems to us that application of the EBL method to Natural Language offers a fruitful field for continued investigation of both a practical and theoretical nature.</Paragraph> </Section> class="xml-element"></Paper>