There are a lot of differences between expressions used in written language and spoken language. It is one of the reasons why speech synthesis applications are prone to produce unnatural speech. This paper represents a method of paraphrasing unsuitable exp
InordertoparaphraseUESintoSES,thispaperpro-posesamethodoflearningparaphrasepairsintheform
of‘UES
SES’.ThekeynotionofthemethodistodistinguishUESandSESbasedontheoccurrenceprob-abilityinwrittenandspokenlanguagecorporawhichareautomaticallycollectedfromtheWeb.Theprocedureofthemethodisasfollows:1
(step1)Paraphrasepairsofpredicates2arelearnedfrom
adictionaryusingamethodproposedby(Kajietal.,2002).(step2)Writtenandspokenlanguagecorporaareauto-maticallycollectedfromtheWeb.(step3)Fromtheparaphrasepairslearnedinstep1,
thoseintheformof‘UESSES’areselectedusingthecorpora.
Thispaperdealswithonlyparaphrasepairsofpredicates,althoughUESincludesnotonlypredicatesbutalsoothercategoriessuchasnouns.
Thispaperisorganizedasfollows.InSection2relatedworksareillustrated.Section3summarizesthemethodofKajietal.InSection4,wedescribethemethodofcollectingcorporaformtheWebandreporttheexperi-mentalresult.InSection5,wedescribethemethodofselectingsuitableparaphrasespairsandtheexperimentalresult.OurfutureworkisdescribedinSection6,andweconcludeinSection7.
2RelatedWork
Paraphrasesaredifferentexpressionswhichconveythesameoralmostthesamemeaning.However,therearefewparaphrasesthathaveexactlythesamemeaning,andalmostallhavesubtledifferencessuchasstyleorformal-ityetc.Suchadifferenceiscalledaconnotationaldif-ference.Thispaperaddressesoneoftheconnotationaldifferences,thatis,thedifferenceofwhetheranexpres-sionissuitableorunsuitableforspokenlanguage.
Althoughalargenumberofstudieshavebeenmadeonlearningparaphrases,forexample(BarzilayandLee,2003),thereareonlyafewstudieswhichaddressthecon-notationaldifferenceofparaphrases.OneofthestudiesisaseriesofworksbyEdmondsetal.andInkpenetal(EdmondsandHirst,2002;InkpenandHirst,2001).Edmondsetal.proposedacomputationalmodelwhichrepresentstheconnotationaldifference,andInkpenetal.showedthattheparametersofthemodelcanbelearnedfromasynonymdictionary.However,itisdoubt-fulwhethertheconnotationaldifferencebetweenpara-phrasesissuf cientlydescribedinsuchalexicalre-source.Ontheotherhand,Inuietal.discussedread-1NotethatthispaperdealswithJapanese.2
Apredicateisaverboranadjective.
ability,whichisoneoftheconnotationaldifferences,andproposedamethodoflearningreadabilityrankingmodelofparaphrasesfromataggedcorpus(InuiandYa-mamoto,2001).Thetaggedcorpuswasbuiltasfollows:alargeamountofparaphrasepairswerepreparedandan-notatorstaggedthemaccordingtotheirreadability.How-ever,theyfocusedonlyonsyntacticparaphrases.Thispaperdealswithlexicalparaphrases.
Thereareseveralworksthattrytolearnparaphrasepairsfromparallelorcomparablecorpora(BarzilayandMcKeown,2001;Shinyamaetal.,2002;BarzilayandLee,2003;Pangetal.,2003).Inourwork,paraphrasepairsarenotlearnedfromcorporabutlearnedfromadic-tionary.Ourcorporaareneitherparallelnorcomparable,andareusedtodistinguishUESandSES.
Thereareseveralstudiesthatcomparetwocorporawhichhavedifferentstyles,forexample,writtenandspo-kencorporaorBritishandAmericanEnglishcorpora,andtryto ndexpressionsuniquetoeitherofthestyles(Kilgarriff,2001).However,thosestudiesdidnotdealwithparaphrases.
Bulykoetal.alsocollectedspokenlanguagecorporafromtheWeb(Bulykoetal.,2003).ThemethodofBu-lykoetal.usedN-gramsinatrainingcorpusandisdif-ferentfromours(thedetailofourmethodisdescribedinSection4).
Inrespectofautomaticallycollectingcorporawhichhaveadesiredstyle,Tambouratzisetal.proposedamethodofdividingModernGreekcorpusintoDemokitiandKatharevoua,whicharevariationsofModernGreek(Tambouratzisetal.,2000).
3LearningPredicateParaphrasePairs
Kajietal.proposedamethodofparaphrasingpredi-catesusingadictionary(Kajietal.,2002).Forexample,whenade nitionsentenceof‘chiratsuku(toshimmer)’is‘yowakuhikaru(toshinefaintly)’,hismethodpara-phrases(1a)into(1b).(1)
a.ranpu-gachiratsukualamptoshimmerb.
ranpu-gayowakuhikarualamp
faintly
toshine
AsKajietal.discussed,thisdictionary-basedparaphras-inginvolvesthreedif culties:wordsenseambiguity,ex-tractionoftheappropriateparaphrasefromade nitionsentence,transformationofpostposition3.Inordertosolvethosedif culties,heproposedamethodbasedoncaseframealignment.
Ifparaphrasescanbeextractedfromthede nitionsen-tencesappropriately,paraphrasepairscanbelearned.Weextractedparaphrasesfromde nitionsentencesusingthe
3
Japanesenounisattachedwithapostposition.
Paraphrasing Predicates ... 8页 免费 A Spoken Language Transl... 5页 ...written language, including ungrammatical (but acceptable for human) expressions...
Paraphrasing Predicates ... 8页 免费 Corpus of Spoken Slovak ... 暂无... Spoken language identification using the SpeechDat corpus Current language ...
Paraphrasing Predicates ... 8页 免费 Tagging Romanian Texts a... 暂无... Tagging children’s texts using written language statistics Sylvana Sofkova ...
Paraphrasing Predicates ... 8页 免费 Insights into spoken lan... 4页 ... REMOVING HUM FROM SPOKEN LANGUAGE RESOURCES Hartmut R. Pf itzinger ...
Abstract Paraphrasing and Aggregating Argumentative ... which produces natural language proofs from proofs... as well as to logical functions or predicates....
But these pure categories are predicates of things...Paraphrasing: using synonyms Translating: equivalents... written, spoken Referent: Things: objects, ...
Abstract Paraphrasing and Aggregating Argumentative ... which produces natural language proofs from proofs... as well as to logical functions or predicates....
热门文档