Abstract. SAPIRisapeer-to-peermultimediainformationretrieval

systemthatcanindexstructuredandunstructuredtext,stillandmoving

images,speech,andmusic.Thesystem’sfeatureextraction component,

whichanalyzesdocumentstopreparethemforindexing,isimplemented

usingUIMA.Ithandlescompounddocumentsusinganarchitectureof

(potentiallynested)splittersandmergerswithinaUIMAaggregate.For

example,themovingimagefromavideoissplitintoanumberofrepre-

sentativevideoframes,eachofwhichisprocessedbythesameanalysis

engineusedforstillimages,andthenmergingtheresultsto formaunifi ed

representationofthevideo.Theoutputofthefeatureextractionmod-

uleisadocumentdescr iptioninarepresentationbasedontheMPEG-7

standard.