Published on August 20, 2007
Discourse: Discourse Read J andamp; M Chapter 18.1 More than One Sentence at a Time: More than One Sentence at a Time The alphas have a long-standing hatred of the betas. Their leaders have decided that the time has come to launch an attack. They are considering the Thanksgiving weekend. Suppose we are trying to do text extraction and we want to fill in records such as: attack (status andlt;planned, imminent, active, completedandgt;) (attacker ) (attackee ) (location ) (mode-of-attack ) (time-of-attack ) (credit-claimer ) Or Dialogue Management: Or Dialogue Management Reference resolution: andgt; Who planned the October attack? The UMPA guerrillas. andgt; Who is their leader? Coherence: http://www.alicebot.org/ Or Machine Translation: Or Machine Translation English: John saw a new bicycle at the store. He wants it. French: John a vu une nouvelle bicyclette au magasin. Il le/la veut. from Babel Fish: John a vu une nouvelle bicyclette au magasin. Il le veut. Reference Resolution: Reference Resolution Indefinite NPs: A boy and a girl came in. They were happy. The girl laughed. Definite NPs: The city council will vote tomorrow. John went to the store/hospital. (brit) John went to hospital. Pronouns: Pronouns John went wrote a new book. It is a bestseller. John went to the store with Bill. He bought a new TV. John was the first person to get there. Bill arrived a bit later. He brought the cookies and tea. John parked his car in his garage. It is incredibly messy, with old bike and car parts lying around everywhere. John parked his car in Beverly Hills. It is incredibly messy, with old bike and car parts lying around everywhere. Nancy ate her oatmeal this morning because she heard that it lowers cholesterol. Nancy hates raccoons because they ate her corn last year. If a farmer owns a donkey, he beats it. If a farmer doesn’t own a donkey, he beats it. Why don’t they ever fix the roads? One Anaphora: One Anaphora I just saw the new Hondas. I want one. Reserve a seat for me on the flight. Reserve one for Jack too. John has a red jumper and a blue one. I saw two bears in the woods. Bill saw some in the parking lot too. Is there a flight to Melbourne before 7am? No but there is one at 9. A Discourse Model: A Discourse Model A realistic model has three layers: Linguistic Discourse (pegs) Knowledge base Examples: There are a lot of books in the room. Pick one. Look at the title page. It should contain both the author’s name and the title. John wants Bill to know about the movie. He thinks he would like it. A Blackboard Model: A Blackboard Model Constraint Sources: Constraint Sources Each constraint source has four components: Modeller Constraint poster Proposer Evaluator The Process: The Process Each CS updates its model if necessary. Each CS posts constraints if necessary. Example: He saw him. Each CS has the opportunity to propose candidates. Evaluation: Kinds of Constraint Sources: Kinds of Constraint Sources Finite set generators (e.g., reflexive pronouns) Fading infinite set generators (e.g., recency) Filters (number agreement) Preferences (semantic coherence) Coherence: Coherence Mary gave Sue her favorite book. It was hard to part with it, but she knew it would be appreciated. Mary gave Sue her favorite book. She’ll always be grateful for that. Mary gave Sue her favorite book. She’s very tall. The problem with coherence: It appears to be an AI complete problem.