Reference Resolution inDiscourseJamie Samdal12/4/2007Chapter 18, Section 1Introduction Overview of Reference Resolution Lappin and Leass’s Algorithm for PronounResolution ExampleReference ResolutionJenny bought a new CD for Joseph. He already hadone. He tried to exchange it for a movie. The storewould not take it without a receipt. How can a computer know what each pronoun refersto in the above example?– Number agreement– Gender agreement– ProximityLappin and Leass’s Algorithm System for Pronoun Interpretation– Simple weighting scheme– Recency and syntactically based preferences– No semantic preference except for agreement– Non-reflexive, third-person pronouns Basic Idea:– Discourse Model Update– Pronoun ResolutionDiscourse Model Update Noun phrase evokes new entity– Representation of entity added to discoursemodel– Degree of Salience (Salience Value) computedfor the new entity Sum of weights of salience factors Many noun phrases may refer to samereferent– Associate equivalence class– Weight for each factor highest of sentence usage– Also include prior sentence usageSalience Factors Sentence Recency 100 Subject Emphasis 80 Existential Emphasis 70 Direct Object Emphasis 50 Indirect Object 40 Non-Adverbial Emphasis 50 Head Noun Emphasis 80 Role Parallelism 35 Cataphora -175Pronoun Resolution1. Collect potential referents– Up to 4 sentences back2. Remove potential referents that do not agree innumber or gender with pronoun3. Remove referents that do not pass basic syntacticcoreference constraints– Ex. Reflexives4. Compute total salience value of the referent5. Select referent with highest salience value– If there is a tie, select the closest referentExampleJenny bought a new CD for Joseph. He already hadone. He tried to exchange it for a movie. The storewould not take it without a receipt.2308050100Joseph280805050100CD310805080100JennyTotalHeadNounNon-AdvInd-ObjObjEx.SubjRecExampleJenny bought a new CD for Joseph. He already hadone. He tried to exchange it for a movie. The storewould not take it without a receipt.230/2 = 115{Joseph}Joseph280/2 = 140{a new CD}CD310/2 = 155{Jenny}JennyScore (for next round)PhrasesReferentExampleJenny bought a new CD for Joseph. He already hadone. He tried to exchange it for a movie. The storewould not take it without a receipt.280805050100one310805080100HeTotalHeadNounNon-AdvInd-ObjObjEx.SubjRecExampleJenny bought a new CD for Joseph. He already hadone. He tried to exchange it for a movie. The storewould not take it without a receipt.(115+310)/2 = 212.5{Joseph, he1}Joseph(140+280)/2 = 210{a new CD, one}CD(155/2) = 77.5{Jenny}JennyScore (for next round)PhrasesReferentExampleJenny bought a new CD for Joseph. He already hadone. He tried to exchange it for a movie. The storewould not take it without a receipt.2308050100movie280805050100it310805080100HeTotalHeadNounNon-AdvInd-ObjObjEx.SubjRecExampleJenny bought a new CD for Joseph. He already hadone. He tried to exchange it for a movie. The storewould not take it without a receipt.(230)/2 = 115{a movie}Movie(212.5+310)/2 = 261.25{Joseph, he1,he2}Joseph(210+280)/2 = 245{a new CD, one, it1}CD(77.5)/2 = 38.75{Jenny}JennyScore (for next round)PhrasesReferentExampleJenny bought a new CD for Joseph. He already hadone. He tried to exchange it for a movie. The storewould not take it without a receipt.2308050100receipt280805050100it310805080100StoreTotalHeadNounNon-AdvInd-ObjObjEx.SubjRecExampleJenny bought a new CD for Joseph. He already hadone. He tried to exchange it for a movie. The storewould not take it without a receipt.310{the store}Store230{a receipt}Receipt115 + 280{a movie}Movie245 + 280 + 35{a new CD, one, it1}CDScore – This round!PhrasesReferentExampleJenny bought a new CD for Joseph. He already hadone. He tried to exchange it for a movie. The storewould not take it without a receipt.310{the store}Store230{a receipt}Receipt115{a movie}Movie560 – wins!{a new CD, one, it1 ,it2}CDScore – This
View Full Document
Unlocking...