210 likes | 310 Views
WG2 PWI24617-5 SemAF - Discourse Structure. 20101014, Berlin HASIDA Koiti hasida.k@aist.go.jp AIST, Japan. Introduction (Concerns). annotation, production, translation, etc. of documents discourse structures not only in linguistic content but also in (possibly silent) video, etc.
E N D
WG2 PWI24617-5SemAF - Discourse Structure 20101014, Berlin HASIDA Koiti hasida.k@aist.go.jp AIST, Japan
Introduction (Concerns) annotation, production, translation, etc. of documents discourse structures not only in linguistic content but also in (possibly silent) video, etc. documents without predefined total temporal ordering of presentation, such as hypertexts and games organization of discourse structures consisting of eventualities (or what represent them, such as sentences, clauses, phrases, video scenes, and so on) and discourse relations among them to minimize the set of discourse relations by attributing presentational information to other parts of discourse structures If the discourse structures of speech and other linguistic data contained in motion pictures were fitted to this scheme, then multilingual subtitles to these pictures could be composed for a reduced cost by means of some standardized tool for multilingual translation.
Scope typically sentences factual and inferential To describe how discourse constituents (eventualities) are combined through (possibly implicit) discourse connectives (discourse relations) to constitute a discourse (its semantic content). Elaboration, etc. Criteria for including certain relations
Policies • Minimize the set of discourse relations while addressing semantic differences. • Concentrate on semantic content representation, minimizing presentational aspects (such as importance: nucleus/satellite distinction) and maximizing the versatility of the representation. • Maximally accommodate polymorphism. • Use discourse trees to encode presentational aspects.
Terms Event???: event (possibly dialogue act) or state or process or their abstraction (type) discourse relation: semantic relation among eventualities discourse graph: graph representing discourse semantics in which nodes represent eventualities and links among them represent discourse relations discourse tree (discourse annotation?): annotated tree structure of linearly-ordered discourse representing presentational structures of the discourse besides its semantic content
Discourse Relation • relations among eventualities and/or their types • [I worked hard {to pass the exam}]. • factual and/or inferential. • [[Tom came] {because [Mary came]}]. • = [[I guess Tom came] {because [Mary came]}]. event ($1) event type ($2) purpose conclusion -inference reason result -causes cause
Discourse Graph realize ubiquitous information service purpose purpose huge amount of content is necessary retrieval must be quick and easy inference authoring of content must be easy inference inference semantic annotation is necessary
Wrapped Arguments/Metonymy • In previous annotation practices, discourse relation may concern not the whole apparent argument but its core wrapped in an attitude report, a modal operator, etc. Remember all those vegetables you slipped under the table ? you slipped under the table causes Maybe that’s why Sparky lived so long. Sparky lived so long
Discourse Graph is Explicit remember? content you slipped those vegetable under the table maybe causes object Sparky lived so long
Discourse Tree A minus sign is the inverse operator. A pair of curly brackets is a discourse constituent headed by a discourse connective. discourse connective discourse relation The arg. of a discourse connective is the 2nd arg. of the discourse relation. A discourse connective depends on the 1st arg. of the discourse relation. [ [Semantic annotation is necessary {-inference because {conjunction [2 retrieval must be quick and easy] and [3 authoring of content must be easy]}}]. [2 Retrieval must be quick and easy {purpose in order to [0 realize ubiquitous information access]}]. [3 Authoring of content must be easy {-inference because [1 huge amount of content is necessary {purposein order to [0 realize ubiquitous information access]}]}]. ]
Discourse Tree (cont.) • Encodes presentational aspects including importance (nucleus/satellite distinction) possibly partially. • The current syntax is not a serious proposal. • It should be easy to come up with a LAF-based representation of DTs. • Do we have to standardize it? • Harmonization requirements? • with SynAF and other annotation practices
Importance • Abstract importance (nucleus/satellite distinction) away from discourse relation, as it’s a matter of presentation. • [1 {Although its rooms are small}, the hotel is large]. [{So1} Tom will stay there]. • [2 {Although the hotel is large}, its rooms are small]. [{So2} Mary won’t stay there]. Tom will stay there the hotel is large inference symmetric conflict its rooms are small Mary won’t stay there inference
Importance (cont.) • Unification between inverse relations: • means vs. purpose • cause vs. result • reason vs. conclusion • attribution vs. content • general vs. specific • whole vs. part • Any criterion under which to choose names and directions of these relations?
Object/Eventuality similar dissimilar general-specific set-member whole-part example restatement comparison attribution-content means-purpose comment-topic Temporal Projection circumstance before-after until simultaneous Instance/Type purpose conditional unconditional Semantics/Pragmatics enablement Polymorphism, Metonymy, and Projection domain=range
Object/Eventuality Some relations concern not only eventualities but also objects. • comparison • [Tom is taller {than Mary is tall}]. • attribution-content • [I believe {that he’s right}]. • [the belief {that he’s right}] • means-purpose • [cut it {with this sword}] • [cut it {by using this sword}]
Instance/Type Some relations may concern both instances and types of eventualities. • purpose • [I used this sword {to cut it}]. • conditional • [{If you’re going to school}, it’s eight o’clock].
Semantics/Pragmatics • enablement • [1Here’s coffee.] [{So1} drink it]. The fact that here’s coffee enables the precondition for the imperative. • [{Since here’s coffee}, it’s possible that you drink it]. dialog act here’s coffee enables Drink it. sem. content here’s coffee enables you drink it
Temporal Projection • `time’ and `circumstance’ may be unified. • [Tom came {at 8 o’clock}]. • [Tom came {when Mary came}]. time (semantic role) equality or Projection? equality or projection? circumstance (discourse rel.)
Taxonomy • Ted Sanders’ 3 (out of 4) dimensions • additive vs. causal • positive vs. negative • factual vs. inferential Cf. the other dimension concerns linear order • basic vs. non-basic
additive • positive • Elaboration: specific, part, step, object, member, example, extraction, minimum, detail, restatement, definition • Attribution: content • Background: background, circumstance • Comparison: similarity, proportion • Complement: supplement • Additive: coordination, addition • Manner: manner • negative • Contrast: contrast, dissimilarity, disjunction, substitution • Complement: constraint • Comparison: comparison, preference • causal • positive • Causality: causes, motivates, triggers • Enablement: purpose, enables • Inference: inference, explanation • Evaluation: evaluation, interpretation, comment • Condition: conditional • negative • Concession: conflict • Condition: otherwise, unconditional, compromise