OverviewTerminologyTerminologyJabberwockyCohesionReferential CohesionRhetorical Structure TheoryRhetorical Structure TheoryText SpansRST RelationsThe Concession RelationRST TreesRST TreesTrees are not enough...Why is RST Popular?Issues: Levels of AnalysisIssues: ProliferationIssues: Discourse/Semantics?Uses of RSTExample of GenerationSummaryExampleExampleProperties of Scientific PapersRigidity of global structureFiner StructureDifferent ViewsAnalysing Scientific LiteratureAspects of Discourse MarkupAspects of Discourse MarkupAspects of Discourse MarkupArgumentative ZoningSemantics of LabelsArgumentative ZonesReliability of AZ SchemeAutomatic ClassificationAutomatic ClassificationUses: Summary ListsUses: Citation MapsUses: Citation MapsUses: Citation MapsUses: Citation MapsUses: Citation MapsInformation RetrievalExamplegTheories of Discoursefor modelling Conjunctive CohesionAdvaith SiddharthanResearch AssociateComputer LaboratoryAdvaith Siddharthan May 17, 2006 – p.1/48OverviewgTerminologyCoherence vs CohesionConjunctive vs AnaphoricRhetorical Structure TheoryTheoryExamples and IssuesApplicationsScientific WritingArgumentative ZoningRhetorical Citation AnalysisAdvaith Siddharthan May 17, 2006 – p.2/48TerminologygHalliday and Hasan (1976)...when the interpretation of some element in the discourse isdependent on that of another...Coherent:Reader can make the interpretationCohesive:Dependency is signalled linguisticallyMany other definitions exist...Advaith Siddharthan May 17, 2006 – p.3/48TerminologygCoherenceIs text interpretable by reader?Mental phenomenon (reader-centric)CohesionTextual phenomenonUse of linguistic devices to link discourse elementsContributes to coherenceNecessary?· Definitely makes life easier for readerSufficient?· Requires skill to write cohesive text that is notcoherent for intended readershipAdvaith Siddharthan May 17, 2006 – p.4/48Jabberwockyg’Twas brillig, and the slithy tovesDid gyre and gimble in the wabe;All mimsy were the borogoves,And the mome raths outgrabe.–Lewis CarrollAdvaith Siddharthan May 17, 2006 – p.5/48CohesiongConjunctive (Intentional) ApproachesModel: Rhetorical relations between text spansLinguistic Devices: coordination/subordination/cuephrasesDominant Theory: RST (Mann and Thompson, 1988)Anaphoric (Attentional) ApproachesModel: discourse focus and changes in focusLinguistic Devices: Pronouns, referring expressions,voice, etc.Dominant Theory: Centering (Grosz et al., 1995)Advaith Siddharthan May 17, 2006 – p.6/48Referential CohesiongMain Ideas (Centering Theory)In each utterance, exactly one entity is the Focus/CenterThere is a preference to keep the same entity as Center insubsequent sentencesThe Center is most likely to be pronominalisedUses for Generation by ComputersChoice of Anaphora/Referring ExpressionsSentence Ordering (smooth shifts in focus)Uses for Analysis by ComputersAnaphora ResolutionAssessment of Text CohesionAdvaith Siddharthan May 17, 2006 – p.7/48Rhetorical Structure Theoryg(Examples that follow are adapted from those on the RST website:http://www.sfu.ca/rst/)Mann and Thompson (1988)...for every part of a coherent text, there is some function,some plausible reason for its presence, evident to readers...RST models intentional structureEvery span of text is linked to the rest of the discourse byan RST relationOriginally 24 relationsmotivation, antithesis, background, evidence,concession, condition, elaboration, circumstance,restatement, sequence, contrast...RST does not consider referential relationsAdvaith Siddharthan May 17, 2006 – p.8/48Rhetorical Structure TheorygMain ideas: text organisationText spans are linked by RST relationsText spans are of two types:Nucleus (more essential to author’s purpose)Satellite (often incomprehensible without Nucleus)A Text is a hierarchical structure of text spans connectedby relationsRCONCESSIONTempting asit may be,we shouldn’t embraceevery popular theorythat comes along↓ Arrow points to NucleusAdvaith Siddharthan May 17, 2006 – p.9/48Text SpansgUsual Rule:Independent clause, along with all of its dependenciesOther GranularitiesSubsectionsParagraphsSentencesSub-clausal unitsMy flicking the switch caused the light to turn on.Downward granularity important in GenerationUpward granularity often easier for AnalysisAdvaith Siddharthan May 17, 2006 – p.10/48RST RelationsgTwo Types of Relations:Nucleus–SatelliteRelation Nucleus SatelliteAntithesis ideas favoured by the au-thorideas disfavoured by theauthorBackground text whose understand-ing is being facilitatedtext for facilitating under-standingMulti-NuclearRelation Nucleus SatelliteContrast one alternate the other alternateSequence an item a next itemAdvaith Siddharthan May 17, 2006 – p.11/48The Concession RelationgNucleus Satellitesituation affirmed byauthorsituation which is apparently inconsis-tent but also affirmed by authorConstraint on nucleus: The writer has positive regard for thenucleus.Constraint on satellite: The writer is not claiming that thesatellite does not hold.Constraint on both: The writer acknowledges a potential orapparent incompatibility between the nucleus and satellite.Recognising the compatibility between the two increases thereader’s positive regard for the nucleus.Writer’s intentions: To increase the reader’s positive regardfor the nucleus.Advaith Siddharthan May 17, 2006 – p.12/48RST TreesgRBACKGROUNDELABORATIONLactose is milksugar;the enzyme lactasebreaks it downCONTRASTFor want of lactasemost adults cannotdigest milk.In populations thatdrink milk, adultshave more lactase.Lactose is milk sugar; the enzyme lactase breaks it down.For want of lactase most adults cannot digest milk.In populations that drink milk the adults have more lactase.Advaith Siddharthan May 17, 2006 – p.13/48RST TreesgWhy Trees?ConvenientEasy to represent/understandUseful for GenerationDefinition of Coherence:Incoherent texts are characterised by disjoint treesRST Trees cannot be constructed for many textsParallelism (simultaneous compatible analyses)Crossed dependenciesNodes with more than one parentOther representations: Graphs?Advaith Siddharthan May 17, 2006 – p.14/48Trees are not enough...gExample from Wolf and Gibson (2004)1. Schools tried to teach students history of science.2. At the same time they tried to teach them how to thinklogically and inductively.3. Some success has been reached in the first of theseaims.4. However, none at all has been reached in the second.Advaith