<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-3205">
  <Title>Exploring variant definitions of pointer length in MDL</Title>
  <Section position="5" start_page="38" end_page="38" type="concl">
    <SectionTitle>
4 Conclusion
</SectionTitle>
    <Paragraph position="0"> The overall purpose of this paper has been to illustrate what was for us an unexpected aspect of using Minimum Description Length theory: not only does MDL not specify the form of a grammar (or morphology), but it does not even specify the precise form in which the description of the abstract linkages between concepts (such as stems and signatures) should be encoded and quantitatively evaluated. We have seen that in a range of cases, using binary strings instead of the more traditional frequency-based pointers leads to a smaller overall grammar length, and there is no guarantee that we will not find an even shorter way to accomplish the  morphologies using pointers versus binary strings (French corpus) same thing tomorrow11. Simply put, MDL is emphatically an evaluation procedure, and not a discovery procedure.</Paragraph>
    <Paragraph position="1"> We hope to have shown, as well, that a systematic exploration of the nature of the difference between standard frequency-based pointer lengths and binary string based representations is possible, and we can develop reasonably accurate predictions or expectations as to which type of description will be less costly in any given case.</Paragraph>
  </Section>
class="xml-element"></Paper>