inform dextract system urdu resourc dpoor languag discov voter prefer blog mixtur topic model opinionmin machin learn system web opinion mine extract name entiti recognit improv retriev translat chines document trec univers buffalo legal blog track gener hypothes web automat score short handwritten essai read comprehens test comprehens test infoxtract customiz intermedi level inform extract engin inform extract system assist analyst assimil inform electron document paper focus task design support inform discoveri applic inform discoveri impli examin larg volum document drawn sourc situat anticip priori requir system breadth depth impli domain independ system easili custom specif domain end user tool custom system impli defin intermedi level task richer subject verb object svo tripl produc shallow system complex domain specif scenario defin messag understand confer muc paper describ robust scalabl engin design purpos describ task entiti profil concept base gener event repres realist goal term accomplish term provid action inform task facilit correl output engin exist structur data benchmark result core engin applic util engin present improv knowledg discoveri document collect combin text retriev link analysi techniqu text mine model hypothesi gener automat score handwritten essai autom task score handwritten student essai challeng problem ai goal assign score compar human scorer human machin handwrit recognit achiev perfect transcript research attempt base coupl ai technolog optic handwrit recognit autom essai score test bed essai written children statewid read comprehens test school process involv imag level oper remov pre print matter segment handwritten text line extract word constraint provid read passag question answer rubric recogn handwritten word method essai score base vector space model machin learn approach learn score paramet set human score sampl system perform compar score human rater test test bed handwritten answer system perform compar automat score base perfect manual transcript rank cross document evid trail hypothesi gener mine concept associ knowledg discoveri concept chain queri automat extract nomin mention event bootstrap probabilist classifi approach event extract focu mention anchor verb mention event surfac noun phrase detect increas recal event extract provid foundat detect relat event paper describ weaklysupervis method detect nomin event mention combin techniqu word sens disambigu wsd lexic acquisit creat classifi label noun phrase denot event event classifi bootstrap probabilist gener model context event event context lexic anchor semant depend relat np method dramat improv bootstrap comfort outperform lexic lookup method base larger handcraft resourc autom score handwritten essai base latent semant analysi abstract handwritten essai wide educ assess classroom instruct paper concern design autom system perform task take input scan imag handwritten student essai read comprehens test produc output score answer analog provid human scorer system base integr technolog optic handwrit recognit ohr autom essai score ae ohr system perform pre process step form remov rule line remov segment text line word final recognit step tune task read comprehens evalu primari educ set perform lexicon deriv passag read ae system base approach latent semant analysi set human score answer determin score system paramet machin learn approach system perform compar score human rater test small set handwritten answer system perform compar automat score base manual transcript knowledg discoveri document concept chain queri unappar inform revel concept chain graph approach minim document set retriev detect inform document stream import compon potenti applic work novelti detect approach base identif sentenc level inform pattern propos inform pattern concept novelti detect present emphasi inform pattern gener topic queri simpli turn specif question answer specif name entiti ne elabor analysi sentenc level inform pattern data trec novelti track includ sentenc length name entiti sentenc level opinion pattern analysi guidelin appli pattern novelti detect gener topic final unifi pattern base approach present novelti detect gener specif topic method deal gener topic focu experiment result show propos approach significantli improv perform novelti detect gener topic perform topic trec novelti track weakli supervis learn cross ddocument person disambigu support inform extract document represent dclass svm abstract motiv appli text categor sort web search result paper describ extens experiment studi impact bagof word document represent perform major classifi na ve bay svm vote perceptron knn text repres short web page descript dmoz open directori web page ontolog transform input data stem normal logtf idf dimension reduct found statist signific improv degrad effect classif perform measur classic metric accuraci precis recal emphasi studi determin document represent correspond classifi describ effect individu transform classif mutual relationship incorpor prior knowledg weight margin support vector machin ub trec genom track abstract paper describ experi state univers york buffalo trec particip genom track submit offici run adhoc retriev task approach languag model ir system develop hous present unoffici result triag task categor task orthograph case restor supervis learn manual annot featur select text categor imbalanc data number featur select metric explor text categor inform gain ig chi squar chi correl coeffici cc odd ratio consid effect cc side metric ig chi side featur select side metric select featur indic membership featur select side metric implicitli combin featur indic membership posit featur nonmembership neg featur ignor sign featur neg featur valuabl ensur optim combin kind featur imbalanc data work investig us explicit control combin propos featur select framework multinomi na ve bay regular logist regress classifi experi show great potenti actual merit explicitli combin posit neg featur optim fashion imbalanc data featur select text categor imbalanc data number featur select metric explor text categor inform gain ig chi squar chi correl coeffici cc odd ratio consid effect cc side metric ig chi side featur select side metric select featur indic membership featur select side metric implicitli combin featur indic membership posit featur nonmembership neg featur ignor sign featur neg featur valuabl ensur optim combin kind featur imbalanc data work investig us explicit control combin propos featur select framework multinomi na ve bay regular logist regress classifi experi show great potenti actual merit explicitli combin posit neg featur optim fashion imbalanc data bootstrap approach name entiti classif success learner approach name entiti ne classif approach requir common noun pronoun seed correspond concept target ne type man woman person ne entir bootstrap procedur implement train success learner decis list learn pars base high precis ne rule ii hidden markov model train learn string sequenc base ne pattern learner train corpu automat tag learner expert lexicon approach identifi english phrasal verb phrasal verb import featur english languag properli identifi basi english parser decod relat structur phrasal verb challeng natur languag process nlp sit borderlin lexicon syntax tradit nlp framework separ lexicon modul parser make difficult handl problem properli paper present finit state approach integr phrasal verb expert lexicon shallow pars deep pars handl morpho syntact interact precis recal combin perform benchmark consist phrasal verb identif problem basic solv present method exploit syntact structur queri languag model approach ir orthograph case restor supervis learn manual annot bootstrap name entiti tag concept dbase seed case restor approach name entiti tag degrad document paper describ approach name entiti ne tag degrad document ne tag process identifi salient text string unstructur text name peopl place organ time date ne tag typic part larger inform extract process applic improv search inform retriev system post process result ocr system focu degrad document case insensit document lack orthograph inform exampl includ output speech recognit system mail tradit approach involv retrain ne tagger degrad text cumbersom oper paper describ approach text restor implicit case sensit form subsequ process origin ne tagger result show approach lead precis loss ne tag degrad document featur select framework text filter ed dsupport vector machin sequenti minim optim incorpor queri term depend languag model document retriev introduct recent advanc inform retriev base statist languag model slm repres document evalu relev user queri languag model lm explor natur languag task includ machin translat speech recognit lm approach document retriev document view languag model md queri document rank base probabl md languag model gener queri lm approach inform retriev motiv di erent perspect experi smooth unigram languag model assum term independ estim document languag model gram specif bigram languag model captur context provid previou word perform unigram model biterm languag model ignor word order constraint bigram langu ub trec hard genom track knowledg supervis perceptu group base qualit build detect monocular aerial imag locat normal inform extract ambigu high locat name citi name buffalo countri name canada brazil china citi name usa citi main street broadwai ambigu handl refer locat name visual relat extract event paper present hybrid approach locat normal combin lexic grammar driven local context constraint ii graph search maximum span tree iii integr semi automat deriv default sens focu resolv ambigu type locat name island town citi provinc countri biterm languag model document retriev introduct statist languag model lm natur languag process task includ speech recognit machin translat recent languag model explor framework inform retriev basic idea view document languag model model queri gener process document rank base probabl languag model gener queri document fix entiti inform retriev languag model document su er spars data problem smooth unigram model demonstr perform languag model vector space probabilist retriev model document retriev song croft propos gener languag model combin bigram languag model good ture estim corpu base smooth unigram probabl improv perform observ combin bigram ub trec batch adapt filter method measur complex imag databas subspac morph theori appear base object identif keyblock approach content dbase imag retriev question answer system support inform extract question answer system support inform extract hybrid approach name entiti dtype tag paper present hybrid approach name entiti ne tag combin maximum entropi model maxent hidden markov model hmm handcraft grammat rule innat strength weak combin result high precis tagger maxent includ extern gazett system categori gener discuss model multimod inform retriev find inform larg multimod document collect www encount numer fals posit pose challeng multimod inform retriev system mmir gener model multimod inform retriev propos user inform express composit multimod queri weight combin index techniqu determin machin learn approach order satisfi inform focu improv precis recal mmir system optim combin text imag similar experi present demonstr util individu index system improv averag precis keyword multimod inform retriev contentbas retriev imag retriev multimod queri machin learn applic imag understand semant dorient multimedia inform retriev show semi dautom imag annot system intellig index semant retriev multimod document find inform larg multimod document collect www encount numer fals posit pose challeng multimedia inform retriev system mmir research address problem find pictur fact imag isol accompani collater text exploit independ exist techniqu pictur retriev text base ii imag base method limit research present gener model multimod inform retriev address follow ing issu user inform ii express inform composit multimod queri iii determin weight combin index techniqu order satisfi inform machin learn approach propos focu improv precis recal mmir system optim combin text imag similar experi present demonstr util individu index system improv averag precis face detect applic intellig focus imag retriev paper present face detect techniqu applic imag retriev face detect method high fals posit low detect rate compar dedic face detect system literatur imag understand due simpl fast natur shown system appli imag retriev focus applic domain applic exampl combin face detect index collater text imag retriev human be combin face detect convent similar match techniqu imag retriev similar background experiment result show propos approach significantli improv imag retriev precis exist search engin focus applic domain kei word face detect imag retriev collater text similar match focus domain spatial color histogram content dbase imag retriev color histogram import techniqu color imag databas index retriev paper tradit color histogram modifi captur spatial layout inform color type spatial color histogram introduc annular angular hybrid color histogram experi show proper trade granular color dimens spatial dimens perform histogram outperform tradit color histogram exist histogram refin color coher vector keyword color histogram spatial color histogram content base imag retriev annular color histogram angular color histogram hybrid color histogram inform extract support question answer paper discuss inform extract system textract questionansw qa track recent held trec test major object examin ir inform retriev applic qa studi show provid solid support qa ii low level name entiti tag compon handl type question iii robust natur languag shallow parser structur basi handl question iv high level domain independ extract multipl relationship gener event expect bring breakthrough qa introduct natur languag qa question answer ideal test bed demonstr power inform extract vision natur oper ir inform retriev import question type support ir qa trec experi suni buffalo document rank exploit multimod context imag retriev index retriev sigir workshop summari find pictur context multimedia index retriev summari report speech input imag interpret annot retriev larg vocabulari recognit dline handwritten cursiv word critic featur comput system interfac user led develop user interfac technolog mous touchscreen penbas input devic handwrit familiar commun media pen base interfac combin automat handwrit recognit offer easi natur input method pen base interfac essenti mobil comput scalabl recent advanc pen base hardwar wireless commun influenti factor renew interest line recognit system line handwrit recognit fundament pattern classif task object input pattern handwritten signal collect line digit devic classifi pre set word system lexicon exact recognit difficult lexicon constrain recognit output vocab gener edit distanc incorpor domain inform handwritten text recognit case studi paper damerau levenshtein string differ metric gener wai accur compens type error present script recognit domain basic dynam program method comput measur extend merg split letter substitut edit oper refin categori effect visual appear word set recogn independ constraint develop reflect sever inform lost due oper constraint solv assign specif cost oper experiment result corrupt string lexicon word show higher correct rate origin form keyword string distanc string match spell error correct word recognit correct text edit script recognit post process introduct goal text recog automat index content dbase retriev caption photograph multimedia input autom imag annot content dbase retriev research explor interact linguist photograph inform integr text imag databas util linguist descript pictur speech text input coordin point refer pictur extract inform aspect imag interpret imag retriev imag interpret phase object region mention text identifi annot imag store databas futur incorpor techniqu previou research photo understand accompani text system piction identifi human face newspap photograph base caption imag retriev phase imag match natur languag queri present user rank order phase combin output imag interpret annot phase ii statist text retriev method iii imag retriev method colour index system bo combin text imag inform content dbase retriev control structur incorpor pictur dspecif context imag interpret paper describ effici control mechan incorpor pictur specif context task imag interpret knowledg base vision system gener domain context reduc comput burden imag interpret knowledg effort explor pictur specif collater inform assum constraint pictur gener natur languag understand modul process descript text accompani pictur develop unifi framework exploit constraint object locat identif label stage describ techniqu incorpor constrain search context base vision final demonstr effect approach piction system caption label human face newspap photograph introduct solv inher constrain task imag multimedia input autom imag annot content dbase retriev research explor interact linguist photograph inform integr text imag databas util linguist descript pictur speech text input coordin point refer pictur extract inform aspect imag interpret imag retriev imag interpret phase object region mention text identifi annot imag store databas futur incorpor techniqu previou research photo understand accompani text system piction identifi human face newspap photograph base caption imag retriev phase imag match natur languag queri present user rank order phase combin output imag interpret annot phase ii statist text retriev method iii imag retriev method colour index system bo lexic syntact techniqu recogn handwritten text output handwritten word recogn wit noisi due factor order compens behaviour choic wr ini tialli consid case handwritten sentenc phrase recognit linguist constraint appli order improv result wit paper discuss statist method appli linguist constraint output wit input consist sentenc phrase base colloc promot lower rank word choic propos word markov model syntax base syn tactic categori tag word case show improv word recognit rate result appli constraint comput model integr linguist visual inform survei paper survei research develop comput model integr linguist visual inform begin discuss system implement continu computation motiv theori human cognit exist research span disciplin natur languag understand comput vision knowledg represent applic area import contribut paper categor exist research base input object final kei issu relat integr inform divers sourc outlin relat exist research kei issu address correspond problem associ visual event word vice versa introduct necess link languag vision order system exhibit intellig behaviour win wal complet natur languag und caption collater text understand photograph incorpor syntact constraint recogn handwritten sentenc output handwritten word recogn hwr noisi due factor order compens behaviour choic hwr initi consid case handwritten sentenc phrase recognit linguist constraint appli order improv result hwr paper discuss statist method appli syntact constraint output hwr input consist sentenc phrase method base syntact categori tag word pure statist method hybrid method combin higher level syntact inform hypertag statist inform transit hypertag introduct recent great deal interest handwrit recognit system system refer line line system light pen special tablet writer en piction system caption label human face newspap photograph paper describ effici control mechan incorpor pictur specif context task imag interpret knowledg base vision system gener domain context reduc comput burden imag interpret knowledg effort explor pictur specif collater inform assum constraint pictur gener natur languag understand modul process descript text accompani pictur develop unifi framework exploit constraint object locat identif label stage describ techniqu incorpor constrain search context base vision final demonstr effect approach piction system caption label human face newspap photograph introduct solv inher constrain task imag newspap imag understand 