File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/intro/98/m98-1008_intro.xml
Size: 3,038 bytes
Last Modified: 2025-10-06 14:06:28
<?xml version="1.0" standalone="yes"?> <Paper uid="M98-1008"> <Title>Description of the American University in Cairo's System Used for MUC-7</Title> <Section position="1" start_page="0" end_page="0" type="intro"> <SectionTitle> INTRODUCTION AND BACKGROUND </SectionTitle> <Paragraph position="0"> Portions of the American University in Cairo's MUC-7 system, MUC7-Plink, have participated in every Message Understanding Competition since MUC-4. The Plink parser was developed at the University of Michigan where it formed the core of the systems entered in MUC-4 #5B2#5D and MUC-5 #5B1#5D. Recently, the Plink parser was added to GATE #5B6#5D to facilitate interaction between language processing modules. Most of the modules used in MUC7-Plink were already in GATE having been imported from the LaSIE system used in MUC-6 #5B8#5D.</Paragraph> <Paragraph position="1"> GATE provides an environment that greatly simpli#0Ces the reuse of existing natural language models. When the call for participation in MUC-7 was made, I was a faculty member at the American University in Cairo, and had several students who were considering participating along with me in MUC-7. I could have easily divided the tasks and had, for instance, one student work on the Gazetteer, one work on coreference and perhaps a small group work on discourse interpretation. Along with the existing Plink parser this would have comprised a largely new system. Unfortunately, I left Cairo, and had only a very small amount of time to develop the system. Furthermore, I had to develop the system at home on my PC. Fortunately, GATE already had all of the modules that I needed, and ran #28albeit slowly#29 on my PC. I did have to modify some things, but with a very small amount of e#0Bort, I developed a working MUC-7 system.</Paragraph> <Paragraph position="2"> Sadly, due to the lack of resources, the results of the system were poor, and by no means re#0Dect the ceiling of the technology. They do however showhow easy it is to perform relatively well with virtually no development time.</Paragraph> <Paragraph position="3"> The MUC7-Plink system is largely that of the She#0Eeld system. It di#0Bers largely by the use of an on-line lexicon, and the use of a di#0Berent parser. This parser was used for the University of Michigan's MUC-5 entry, but the grammar and parsing heuristics have been rewritten to take advantage of the on-line lexicon, the Gazetteer, and the automated part of speech tagger. The parser also produces signi#0Ccantly di#0Berent output for the XI discourse interpreter #5B17#5D. In the rest of this document I will #0Crst describe the system; this will include a module by module description of the components, and a brief description of GATE. I will then describe the performance of the system; this will include a summary of MUC7-Plink's scores on the TE, TR and ST tasks, a brief summary of how development time was spent, and a walk through of the sample article. I will conclude with a few observations.</Paragraph> </Section> class="xml-element"></Paper>