3/30/2011 CSE842, Spring 2009, MSU 1CSE 842Natural Language ProcessingLecture 20: Computational Discourse3/30/2011 CSE842, Spring 2009, MSU 2Pragmatics and Discourse• Pragmatics: context dependent meaning• Discourse: anything longer than a single utterance or sentence, a group of sentences– Monologue– Dialogue: • May be multi-party• May be human-machine• Topics– Discourse segmentation– Text coherence– Reference resolution– Coreference resolution3/30/2011 CSE842, Spring 2009, MSU 3Discourse• Discourse: anything longer than a single utterance or sentence, a group of sentences• Discourse coherence– Cohesive device: linguistic devices such as pronouns, etc. – Coherence: • John hid Bill’s car keys. He was drunk• John hid Bill’s car keys. He likes spinach3/30/2011 CSE842, Spring 2009, MSU 4Discourse Segmentation• Separate a document into a linear sequence of subtopics• Unsupervised approaches:– Cohesion-based approach– Cohesion: the use of certain linguistic devices to link or tie together textual units (e.g., repetition of words, hyponyms, synonyms, etc.)– Approaches: TextTiling, clustering. • Supervised approaches: – Train classifiers based on annotated data3/30/2011 CSE842, Spring 2009, MSU 5TextTiling Algorithm1. Create pseudo sentences3/30/2011 CSE842, Spring 2009, MSU 6TextTiling Algorithm2. Calculate lexical cohesion score between sentences based on similarity of text before and after the gap3/30/2011 CSE842, Spring 2009, MSU 7TextTiling Algorithm3. Calculate depth score for the valley 3/30/2011 CSE842, Spring 2009, MSU 8DS Evaluation: WindowDiff)0),(),((1),(1≠−−=∑−=++kNikiikiihyphypbrefrefbkNhyprefWindowDiffWhere b(i,j) is the number of boundaries between positions i and j in a text. WindowDiff returns a value between 0 and 1.3/30/2011 CSE842, Spring 2009, MSU 9Text Coherence(a) John hid Bill’s car keys. He was drunk.(b) *John hid Bill’s car keys. He likes spinach. Coherence: the meaning relations between two textual units.Example of relations (Hobbs’79)Result: The Tin Woodman was caught in the rain. His joints rusted. Explanation: John hid Bill’s car keys. He was drunk.Parallel: The Scarecrow wanted some brains. The Tin woodman wanted a heart. Elaboration: Dorothy was from Kansas. She lived in the midst of the great Kansas prairies. Occasion: Dorothy picked up the oil-can. She oiled the Tin Woodman’s joints3/30/2011 CSE842, Spring 2009, MSU 10Coherence Relations (Hobbs)• Result: Infer that the state or event asserted by S0 causes or could cause the state or event asserted by S1. – John bought an Acura. His father went ballistic• Explanation: Infer that the state or event asserted by S1 causes or could cause the state of event asserted by S0– John hid Bill’s car keys. He was drunk.3/30/2011 CSE842, Spring 2009, MSU 11Coherence Relations (Hobbs)• Parallel: Infer p(a1,a2,…) from the assertion of S0 and p(b1, b2,…) from the assertion of S1, where ai and bi are similar, for all i– John bought an Acura. Bill leased a BMW• Elaboration: Infer the same proposition P from the assertions of S0 and S1. – John bought an Acura this weekend. He purchased a beautiful new Integra for 20 thousand dollars at Bill’s dealership on Saturday afternoon.3/30/2011 CSE842, Spring 2009, MSU 12Coherence Relations (Hobbs)• Occasion: a change of state can be inferred from the assertion of S0, whose final state can be inferred from S1, or a change of state can be inferred from the assertion of S1, whose initial state can be inferred from S0. – John bought an Acura. He drove to the ballgame.3/30/2011 CSE842, Spring 2009, MSU 13S1: John went to the bank to deposit his paycheckS2: He then took a train to Bill’s car dealership.S3: He needed to buy a carS4: The company he works for now isn’t near any public transportationS5: He also wanted to talk to Bill about their softball league. Explanation Coherence Relation3/30/2011 CSE842, Spring 2009, MSU 14S1: John went to the bank to deposit his paycheckS2: He then took a train to Bill’s car dealership.S3: He needed to buy a carS4: The company he works for now isn’t near any public transportationS5: He also wanted to talk to Bill about their softball league. ParallelExplanationCoherence Relation3/30/2011 CSE842, Spring 2009, MSU 15S1: John went to the bank to deposit his paycheckS2: He then took a train to Bill’s car dealership.S3: He needed to buy a carS4: The company he works for now isn’t near any public transportationS5: He also wanted to talk to Bill about their softball league. ExplanationExplanationParallelCoherence Relation3/30/2011 CSE842, Spring 2009, MSU 16S1: John went to the bank to deposit his paycheckS2: He then took a train to Bill’s car dealership.S3: He needed to buy a carS4: The company he works for now isn’t near any public transportationS5: He also wanted to talk to Bill about their softball league. OccasionExplanationParallelExplanationCoherence Relation3/30/2011 CSE842, Spring 2009, MSU 17Text CoherenceS1: John went to the bank to deposit his paycheckS2: he then took a train to Bill’s car dealershipS3: He needed to buy a car. S4: The company he works for now isn’t near any public transportation.S5: He also wanted to talk to Bill about their softball league. 3/30/2011 CSE842, Spring 2009, MSU 18Other Discourse Theories• Rhetorical Structure Theory (Mann and Thompson, 1988)• Grosz and Sidner’s attention, intention and structure of discourse (1986)3/30/2011 CSE842, Spring 2009, MSU 19Rhetorial Structure Theory (RST)• RST is a descriptive theory of text organization based on the relationships that hold between parts of text (Mann and Thompson, 1987)– I love to collect classic automobiles. My favorite car is my 1899 Duryea.– I love to collect classic automobiles. My favorite car is my 1999 Toyota. 3/30/2011 CSE842, Spring 2009, MSU 20Rhetorial Structure Theory (RST)• Nucleus: central segment of text• Satellite: more peripheral segment. • Example relations:– Elaboration: the satellite has additional details concerning the content of the nucleus.– Contrast: multi-nuclear– Condition: the satellite presents sth that must occur before the situation presented in the nucleus can occur– Purpose…– Sequence– Result3/30/2011 CSE842, Spring 2009, MSU 21Rhetorical Structure Theory (RST)• 23 rhetorical relations that can hold between spans of text (e.g., evidence, elaboration, contrast, etc.)– Nucleus: