Presentation Karaoke Discourse Coherence Theories and Modelling Department of Computational Linguistics, Saarland University April 29th, 2013 (CoLi Saarland) Discourse Coherence April 29th, 2013 1 / 10
(1) Coherence • coherent text → common topic • individual units of information → meaningful relationships • coreference between referring expressions • semantic & pragmatic relations (e.g. Cause) between adjacent units • understanding → uncovering those relationships Example (1) The motherboard has one PCI slot, (2) so it provides only little space for extensions. → (1) is the Cause for (2) (CoLi Saarland) Discourse Coherence April 29th, 2013 2 / 10
(2) Content Zone • continuous portion of a text that fulfills a functional role for the text as a whole • contributes to the overall message / purpose • characteristic for the genre of the text Example Teufel’s Argumentative Zones in research papers Film reviews → descriptive / comment zones (CoLi Saarland) Discourse Coherence April 29th, 2013 3 / 10
(3) Topic-induced text structure • = a sequence of non-overlapping text segments that completely covers the text (partitioning). • each unit = one or more sentences that address a common topic • topic segmentation → largely matter of intuition, hard to create gold standard Example Johann Sebastian Bach was a German composer, organist, harpsichordist, violist, and violinist of the Baroque period. Bach was born in Eisenach, Saxe-Eisenach, into a very musical family; his father, Johann Ambrosius Bach was the director of the town musicians, and all of his uncles were professional musicians. (CoLi Saarland) Discourse Coherence April 29th, 2013 4 / 10
(4) Cohesion • sentences and clauses are connected by a variety of linguistic means • can be identified at the text surface • noting cohesion does not require deep understanding of the text Example • connectives (‘and then’) • word repetition • related words • pronouns (CoLi Saarland) Discourse Coherence April 29th, 2013 5 / 10
(5) Anaphora and Coreference Resolution Anaphora Resolution • task: find an antecedent for each anaphor in a text. • anaphor → its discourse referent can only be identified when its antecedent is interpreted. • anaphora = irreflexive, non-symmetrical relation. Example The monkey took the banana and ate it. Coreference Resolution • partitioning of mentions of discourse referents in a text into ‘classes’ (chains) corresponding to those referents. • coreference = equivalence relation Example The monkey sat under the palm tree . It took a banana and ate it . Finally, it climbed on it . (CoLi Saarland) Discourse Coherence April 29th, 2013 6 / 10
(6) Elementary Discourse Unit (EDU) • span of text • clause • (nominalization) NP • ... • sentence • denotes a single event or type of event • complete, distinct unit of information • the subsequent discourse may connect to an EDU • may be structurally embedded in another EDU Example (1) The motherboard has one PCI slot, (2) so it provides only little space for extensions. (CoLi Saarland) Discourse Coherence April 29th, 2013 7 / 10
(7) Penn Discourse TreeBank • large-scale corpus on top of Penn TreeBank • information related to discourse structure and discourse semantics • annotation of Explicit discourse relations : lexically grounded (connectives) Example Michelle lives in a hotel room, and although [ ARG 1 she drives a canary-colored Porsche ] , [ ARG 2 she hasn’t time to clean or repair it ] . • annotation of Implicit discourse relations : (insert connective) Example [ ARG 1 The third principal in the S. Gardens adventure did have garden experience. ] Implicit = because for example [ ARG 2 The firm of Bruce Kelly/David Varnell Landscape Architects had created Central Park’s Strawberry Fields and Shakespeare Garden. ] (CoLi Saarland) Discourse Coherence April 29th, 2013 8 / 10
(8) Coherence Relation • = a specific relationship on the semantic/pragmatic level between adjacent units of text • definitions can be given: • in semantic terms • in terms of speaker intentions (e.g., Rhetorical Structure Theory) • granularity of relation set can differ Example causality, similarity/contrast, and contiguity (temporal or other) Rhetorical Structure Theory (RST) Discourse relations hold between: Example: E LABORATION relation • nucleus (more important discourse unit) Arts-and-Craft jewels tend to be elaborate. • satellite (less important discourse unit) They are often mass-produced. (CoLi Saarland) Discourse Coherence April 29th, 2013 9 / 10
Thanks! (CoLi Saarland) Discourse Coherence April 29th, 2013 10 / 10
Recommend
More recommend
