Lecture – VIII Monojit Choudhury RS, CSE, IIT Kharagpur GRAMMAR – II From Syntax to Meaning Communication Empowerment Laboratory, IIT Kharagpur Lecture – VIII Monojit Choudhury RS, CSE, IIT Kharagpur 14th Feb 2005 CEL, IIT Kharagpur
The Big Picture Source Language Analysis Target Language Generation Text Normalization Text Rendering Morphological Analysis Morphological Synthesis POS Tagging Phrase Generation Parsing Role Ordering Semantic Analysis Lexical Choice Discourse Analysis Discourse Planning 14th Feb 2005 CEL, IIT Kharagpur
Recap Rationalism vs. Empiricism Modelling English using CFG Regular Languages are insufficient Even CFG cannot model English purely Advantages Captures most of the things Efficient parsing algorithms Problems Structural Ambiguity Agreement Constraints 14th Feb 2005 CEL, IIT Kharagpur
But what to do with the parse trees??!!! Parsing Surface Form Deep Structure PARSER Word Tag Pairs All possible parse trees But what to do with the parse trees??!!! Grammar Rules 14th Feb 2005 CEL, IIT Kharagpur
Parse Trees Who like fresh fruits? What do children like? Do children like fresh fruits? VP NP NP N V A N Children like fresh fruits 14th Feb 2005 CEL, IIT Kharagpur
Parse Trees Who like fresh fruits? What do children like? Do children like fresh fruits? VP NP PP Aux V Pre A N N Fresh fruits are liked by children 14th Feb 2005 CEL, IIT Kharagpur
Grammatical Functions Apart from the Phrase Structure in PSG, there are also Grammatical Functions Specifies the relation of the NP with the verb. Subject, Object1, Object2, ... Children like fresh fruits. Their mother gave them some apples. 14th Feb 2005 CEL, IIT Kharagpur
Active vs. Passive voice Fresh fruits are liked by children. They were given some apples. Children like fresh fruits. Their mother gave them some apples. 14th Feb 2005 CEL, IIT Kharagpur
Conclusions In fixed word order languages, Grammatical Functions are dependent on the position of the NP relative to the verb. Grammatical function, therefore, changes with the voice of the sentence. Hence, it does not capture the semantics completely. 14th Feb 2005 CEL, IIT Kharagpur
Theta Roles Semantics is “unambiguously” captured using Theta roles Agent, Co-agent/Patient, Instrument ... The children cut the apples with a knife The apples were cut by a knife 14th Feb 2005 CEL, IIT Kharagpur
“The stone rolled down.” Ambiguity Again! What characterizes Agent-ship/patient-ship? “The stone rolled down.” What should be an ideal set of theta roles? Are Theta roles language independent? The debate is primarily philosophical with some Engineering connotations. 14th Feb 2005 CEL, IIT Kharagpur
Lexical Functional Grammar Joan Bresnan (Ed.) The Mental representation of Grammatical Relations, MIT press, 1982 Why lexical we will come to later. Functional because it captures functional relations c- structure and f-structure 14th Feb 2005 CEL, IIT Kharagpur
c-structure S S VP VP NP NP NP PP N V A N Aux V Pre A N N Children like fresh fruits Fresh fruits are liked by children 14th Feb 2005 CEL, IIT Kharagpur
Children like fresh fruits. f-structure - I Children like fresh fruits. subject Predicate ‘like’ ( subject, object) Tense present Object Predicate Child Number Plural Predicate fruit Adjective fresh Number plural 14th Feb 2005 CEL, IIT Kharagpur
Fresh fruits are liked by children. f-structure - II Fresh fruits are liked by children. subject Predicate ‘like’ ( by-object, subject) Tense present by-Object Predicate fruit Adjective fresh Number Plural Predicate Child Number Plural 14th Feb 2005 CEL, IIT Kharagpur
Formal Model R1: S NP VP subj = = R2: VP V {NP} {NP} PP* obj = obj2 = (pcase) = obj R3: PP prep NP obj = 14th Feb 2005 CEL, IIT Kharagpur
Lexicon Entries the, det spec = the children, noun pred = child number = plural 14th Feb 2005 CEL, IIT Kharagpur
f-structure representation Feature Structures General Concept: A declarative way of specifying constraints and contexts Can model agreement and other context dependent features which are not modeled using CFG (Compare with types in C) Can be defined at the lexical, syntactic, semantic levels 14th Feb 2005 CEL, IIT Kharagpur
Feature Structure FEATURE1 VALUE1 FEATURE2 VALUE2 FEATURE3 VALUE3 . AVM (Attribute Value Matrix) Representation 14th Feb 2005 CEL, IIT Kharagpur
Atomic vs. Nested Feature Structure Category verb Root run TAM Agreement Category verb Root run Tense Past Aspect Cont Modality Default Number Φ Person Φ Tense Past Aspect Cont Modality Default Number Φ Person Φ “was running” 14th Feb 2005 CEL, IIT Kharagpur
Some Concepts Under-specification & Subsumption Lattice DAG representation Feature Path Reentrants Typed Feature Structures and Type Hierarchy Unification 14th Feb 2005 CEL, IIT Kharagpur