Computational Grammars Azadeh Maghsoodi
History Before First 20s 20s World War II Last 1950s Nowadays
Before 1800 Traditional Grammar Correct Speech of a specific language Not scientific Rejected Useful issues: POS
Indian-European languages Language vs. Other languages Language vs. its history
Early 20s Enough Philology! Language in a specific time
20s America & Western Europe Intellectual Pattern Understanding Processes in human being
World War II Math. Logic as a study tool Computer invention caused new App Abstract Mind model ends Behaviorism
Late 1950 Chomsky is coming! Formal Language Theory “Syntactic Structures” Language Categories – Type 0: Natural (Irregular) – Type 1: Context sensitive – Type 2: Context free – Type 3: Regular
Late 1950 (continue) Chomsky followers professes: – Generative grammar: Accurate and definite enough for testing Generative Grammars – Goal: Unaware knowledge of users – Biologic and inborn basis for linguistic abilities Universal Grammar Shared structures
Nowadays Motives – Discover human mind structure – Language process technology Applications – Word processors – MT – Word predictors – Text predictors – UFIs / DB Queries – Information retrieval
Syntactic Model Grammars Parse Algorithms
Computational Grammars Generative Grammars – Caused by Natural Language Theory – Introduced by Chomsky – Accurate and definite structures – Transformational grammar (TG) – Constraint-Based Lexicalist grammar (CBLG)
TG Less computational efficiency Theoretical basis Complex rules Simple lexicons
TG (continue) Chomsky hierarchy & First TG Standard Theory (1965) Extended Standard Theory Government & Binding Theory ( )
Standard Theory Sentence – Deep structure – Surface structure Generative TG – Basic part Produce deep structure CFG – Transformational part Transformational Rules
Convert deep structure to surface structure Transformational Rule ~ Transformation Example: (same deep structures) – (i) The boys place the book on the table. – (ii) The boy has placed the book on the table. – (iii) Did the boy place the book on the table?
Transformational Rules (example) A deep structure: S NP N the D boy VP Aux will V place NP The book
Transformational Rules (example) To produce yes/no question: – Using a Move Transformation – S[NP VP [AUX V NP]] S[AUX NP VP[V NP]] S NPVP AuxVNP S AuxNPVP VNP
Government and Binding Theory (GB) Universal grammar theory Learning a language = confirming a small set of parameters + learning lexicons Move α: deep structure to surface structure ‘Move α’ moves anything to anywhere Some constraints correct ‘Move α’
GB (continue) Lexicons Deep Struct Surface Struct Logical FormPhonological Form Move-α LF Move-α Stylistic & Phonological Rules
GB (continue) Minimalist Program (MP) – Choose the best candidate instead of direct production – Under study
CBLG Based on TGs Increase computational efficiency of grammars Simple rules Complex lexicons Psychological Computational
CBLG (continue) Constraint-Based architecture – Constraint satisfaction more important than transformational derivation Strict lexicalism – Lexicons: syntactic atoms of a language – Independent Internal structure from syntactic constraints
CBLG (continue) Surface structures are produced directly Most computational grammars are CBLG
Computational Grammars Unification grammar (UG) Categorical grammar (CG) Dependency grammar (DG) Link grammar Lexical/Functional grammar (LFG) Tree Adjoining grammar (TAG) Generalized Phrase Structure grammar (GPSG) Head Driven Phrase Structure grammar (HPSG)
Unification Grammar (UG) Lots of CBLs are UG Augmented CFG – CFG can’t recognize long distance dependencies – A generalized form of CFG + A set of features – Augmented Transition Network (ATN) – Definite Clause Grammar (DCG) Unification Grammars
UG (continue) Unification Grammars – Feature structures are extended – No need to CFGs – Grammar ~ A set of constraints between feature structures – Key concept: Subsumption relation
UG (continue) CAT verb ROOT cry CAT verb ROOT cry CAT verb VFORM present VFORM present (Unificator)
UG (example) S NP VP Unification grammar: X0 X1 X2 CAT 0 = 5 CAT 1 = NP CAT 2 VP AGR 0 = AGR 1 = AGR 2 VFORM 0 = VORM 2
UG (continue) More grammar information are stored in lexicons Less grammar rules Using DAGs
ATN Grammar Transitive network ~ Expanded Finite-State machine ATN Grammar ~ A set of transitive networks Features Constraints
Categorical Grammar (CG) Lots of bases are omitted No difference between lexicons and none- lexicons Part Of Speech is replaced by some complex category NP/S : NP is on the right NP\S : NP is on the left
CG (example) Peter : NP Likes : (NP\S)/NP Peanuts : NP Passionately : (NP\S)\(NP\S) Peter likes peanuts passionately.
CG (example) S NP Peter NP\S (NP\S)/NP Likes NP peanuts (NP\S)\(NP\S) passionately
Dependency Grammar (DG) American linguists Based on TGs Dependencies between words Dependency tree V N boys playAdv well
Link Grammar Planarity phenomenon Legal sequence of words: – Satisfy local necessities (satisfaction) – No crossed conjunctions (planarity) – One connected graph (connectivity) CFG Lexical grammars – Grammar is distributed between words Probability models Voice recognition Hand-written recognition
Link Grammar (example) linking requirements:
Link Grammar (example) linking requirements are satisfied
Link grammar (example) Not part of a language
Lexical-Functional Grammar (LFG) Unification grammar Not TG ATN research and its deficiencies introduced LFG Group structures 4 structures
Tree Adjoining Grammar (TAG) Between CFG and CSG Grammar rules are a set of initial trees Initial trees are anchored trees Two main operations: – Substitution – Adjoin High accuracy
TAG (example) S VP S NP VP + VP ADV NP VP V NP VP ADV V NP
TAG (continue) High accuracy Apps in NLP – MT – Information retrieval – …
Generalized Phrase Structure grammar (GPSG) Only CFLs CFG Rules – Immediate Dominance (ID) – Linear Precedence (LP)
Head Driven Phrase Structure grammar (HPSG) Lexical grammar Based on unification Increase computational potency of GPSG Simple CFG Complex lexicons
Applications
Parse Algorithms Top-Down parsing Bottom-Up parsing (*)
Parse Algorithms Top-Down parsing Chart parser – Dynamic Programming Recursive Transition Network (RTN) – ATN grammar LR parser – Shift-Reduce algorithms Cocke-Younger-Kasami parser (CYK) – Dynamic Programming – CNF grammar
Efficient Algorithms Chart parser CYK parser
Questions???