http://www.mindswap.org/blog/2007/01/05/ontology-for-the-intelligence-community-part-i/ Weblog for the Mindswap research group at University of Maryland Tue, 2 Dec 2008 21:59:12 +0000 http://wordpress.org/?v=2.0.10 http://www.mindswap.org/blog/2007/01/05/ontology-for-the-intelligence-community-part-i/#comment-17702 Thu, 22 Feb 2007 14:24:19 +0000 http://www.mindswap.org/blog/2007/01/05/ontology-for-the-intelligence-community-part-i/#comment-17702 [...] The problem of ontology alignment presents an interesting diversion from this sort of dialectic. Ontology alignment, as a discipline, focuses on algorithms for alignment. There is in fact a competition, held each year, to find more effective algorithms, in terms of precision and recall. Precision, in information retrieval terms, is the extent of correct matches the algorithm reaches against some pre-established result; recall is the extent of correct matches the algorithm reaches as a percentage of the total matches it finds. Irrespective of how the algorithm operates - and there are a number of techniques it can exploit - it has no knowledge of conceptual schemes. It merely operates on the symbols available to it. The question then becomes - how is the pre-established result in fact established? Who guarantees that the result is itself sensitive to the possibility of incommensurability? There remains the very likely possibility that precision and recall in fact represent varying degrees of false positives - automated matches which correlate to some predefined match, which for some other interpreter, is no match at all. In this sense a matching algorithm can only be measured against a human interpretation; it is therefore a result which can be totally inverted when faced, at the extreme, with a diametrically opposing interpretation. The incommensurabilist will uphold this likelihood; the commensurabilist will see it as a deviate case, to be consciously avoided by diligent professionals. (Something like these attitudes can be found in an interesting description of ontologies in the intelligence community). [...] […] The problem of ontology alignment presents an interesting diversion from this sort of dialectic. Ontology alignment, as a discipline, focuses on algorithms for alignment. There is in fact a competition, held each year, to find more effective algorithms, in terms of precision and recall. Precision, in information retrieval terms, is the extent of correct matches the algorithm reaches against some pre-established result; recall is the extent of correct matches the algorithm reaches as a percentage of the total matches it finds. Irrespective of how the algorithm operates - and there are a number of techniques it can exploit - it has no knowledge of conceptual schemes. It merely operates on the symbols available to it. The question then becomes - how is the pre-established result in fact established? Who guarantees that the result is itself sensitive to the possibility of incommensurability? There remains the very likely possibility that precision and recall in fact represent varying degrees of false positives - automated matches which correlate to some predefined match, which for some other interpreter, is no match at all. In this sense a matching algorithm can only be measured against a human interpretation; it is therefore a result which can be totally inverted when faced, at the extreme, with a diametrically opposing interpretation. The incommensurabilist will uphold this likelihood; the commensurabilist will see it as a deviate case, to be consciously avoided by diligent professionals. (Something like these attitudes can be found in an interesting description of ontologies in the intelligence community). […]

]]> http://www.mindswap.org/blog/2007/01/05/ontology-for-the-intelligence-community-part-i/#comment-10013 Tue, 09 Jan 2007 13:50:00 +0000 http://www.mindswap.org/blog/2007/01/05/ontology-for-the-intelligence-community-part-i/#comment-10013 "Before rushing to download..." I was looking forward to checking out the 103 slides, but that link and most of the others here seem to start with an "a" start tag that has no attributes, so they don't work as links. Is it possible to download those slides? Thanks... “Before rushing to download…” I was looking forward to checking out the 103 slides, but that link and most of the others here seem to start with an “a” start tag that has no attributes, so they don’t work as links. Is it possible to download those slides? Thanks…

]]>