Compact Representation of Ambiguities. Generating Sentences by Grammar Specialization. Compact Representation of Ambiguities

Similar documents
Review of CFGs and Parsing I Context-free Languages and Grammars. Winter 2014 Costas Busch - RPI 1

Section IV.6: The Master Method and Applications

FOURIER SERIES PART I: DEFINITIONS AND EXAMPLES. To a 2π-periodic function f(x) we will associate a trigonometric series. a n cos(nx) + b n sin(nx),

We will begin by supplying the proof to (a).

Review of the Riemann Integral

,... are the terms of the sequence. If the domain consists of the first n positive integers only, the sequence is a finite sequence.

Error-free compression

: : 8.2. Test About a Population Mean. STT 351 Hypotheses Testing Case I: A Normal Population with Known. - null hypothesis states 0

INFINITE SERIES. ,... having infinite number of terms is called infinite sequence and its indicated sum, i.e., a 1

SOME IDENTITIES BETWEEN BASIC HYPERGEOMETRIC SERIES DERIVING FROM A NEW BAILEY-TYPE TRANSFORMATION

Context-free grammars and. Basics of string generation methods

Riemann Integration. Chapter 1

Merge Sort. Outline and Reading. Divide-and-Conquer. Divide-and-conquer paradigm ( 4.1.1) Merge-sort ( 4.1.1)

Content: Essential Calculus, Early Transcendentals, James Stewart, 2007 Chapter 1: Functions and Limits., in a set B.

Advanced Calculus Test File Spring Test 1

A GENERAL METHOD FOR SOLVING ORDINARY DIFFERENTIAL EQUATIONS: THE FROBENIUS (OR SERIES) METHOD

is infinite. The converse is proved similarly, and the last statement of the theorem is clear too.

Students must always use correct mathematical notation, not calculator notation. the set of positive integers and zero, {0,1, 2, 3,...

Fig. 1. I a. V ag I c. I n. V cg. Z n Z Y. I b. V bg

Chapter 2 Infinite Series Page 1 of 9

Some Properties of Brzozowski Derivatives of Regular Expressions

Containment of Nested XML Queries

( ) k ( ) 1 T n 1 x = xk. Geometric series obtained directly from the definition. = 1 1 x. See also Scalars 9.1 ADV-1: lim n.

Formal Languages Simplifications of CFGs

Uncertainty Analysis for Uncorrelated Input Quantities and a Generalization

PRACTICE FINAL SOLUTIONS

4.1 Bounded integral.

Crushed Notes on MATH132: Calculus

Lesson-2 PROGRESSIONS AND SERIES

King Fahd University of Petroleum & Minerals

Chapter Real Numbers

Discrete Mathematics I Tutorial 12

MTH 146 Class 16 Notes

Applications of Regular Closure

POWER SERIES R. E. SHOWALTER

A GENERALIZATION OF GAUSS THEOREM ON QUADRATIC FORMS

Unit 1. Extending the Number System. 2 Jordan School District

=> PARALLEL INTERCONNECTION. Basic Properties LTI Systems. The Commutative Property. Convolution. The Commutative Property. The Distributive Property

A general theory of minimal increments for Hirsch-type indices and applications to the mathematical characterization of Kosmulski-indices

Cyclic Inputs. Surface Generation (Part II) Ambiguity-preserving Generation. Ambiguity-preserving Generation. Ambiguity-preserving Generation

Linear Programming. Preliminaries

Chapter 5. The Riemann Integral. 5.1 The Riemann integral Partitions and lower and upper integrals. Note: 1.5 lectures

Infinite Series Sequences: terms nth term Listing Terms of a Sequence 2 n recursively defined n+1 Pattern Recognition for Sequences Ex:

Lecture 4 Recursive Algorithm Analysis. Merge Sort Solving Recurrences The Master Theorem

Certain sufficient conditions on N, p n, q n k summability of orthogonal series

Qn Suggested Solution Marking Scheme 1 y. G1 Shape with at least 2 [2]

Chapter 7 Infinite Series

Mathematical Notations and Symbols xi. Contents: 1. Symbols. 2. Functions. 3. Set Notations. 4. Vectors and Matrices. 5. Constants and Numbers

g as the function in which for every element x be the set of polynomials of a degree less than or equal to n with , for each i from 0 to n )

SUTCLIFFE S NOTES: CALCULUS 2 SWOKOWSKI S CHAPTER 11

The Definite Integral

THE GAMMA FUNCTION. z w dz.

EXERCISE a a a 5. + a 15 NEETIIT.COM

Introduction to mathematical Statistics

Similar idea to multiplication in N, C. Divide and conquer approach provides unexpected improvements. Naïve matrix multiplication

It is always the case that unions, intersections, complements, and set differences are preserved by the inverse image of a function.

Solution of the exam in TMA4212 Monday 23rd May 2013 Time: 9:00 13:00

Computability and computational complexity

Remarks: (a) The Dirac delta is the function zero on the domain R {0}.

The Average Complexity of Moore s State Minimization Algorithm is O(n log log n)

DEPARTMENT OF ELECTRICAL &ELECTRONICS ENGINEERING SIGNALS AND SYSTEMS. Assoc. Prof. Dr. Burak Kelleci. Spring 2018

Content. Languages, Alphabets and Strings. Operations on Strings. a ab abba baba. aaabbbaaba b 5. Languages. A language is a set of strings

The Basic Properties of the Integral

CHAPTER 3 NETWORK ADMITTANCE AND IMPEDANCE MATRICES

KNOWLEDGE-BASED AGENTS INFERENCE

Sets and Probabilistic Models

Vectors. Vectors in Plane ( 2

CSC 473 Automata, Grammars & Languages 11/9/10

Relation of BSTs to Quicksort, Analysis of Random BST. Lecture 9

Geometric Sequences. Geometric Sequence. Geometric sequences have a common ratio.

Abel Resummation, Regularization, Renormalization & Infinite Series

Lecture 16: Monotone Formula Lower Bounds via Graph Entropy. 2 Monotone Formula Lower Bounds via Graph Entropy

Applied Databases. Sebastian Maneth. Lecture 16 Suffix Array, Burrows-Wheeler Transform. University of Edinburgh - March 10th, 2016

Formal Languages The Pumping Lemma for CFLs

The Weierstrass Approximation Theorem

arxiv: v1 [cs.cc] 23 Oct 2008

MATH 104 FINAL SOLUTIONS. 1. (2 points each) Mark each of the following as True or False. No justification is required. y n = x 1 + x x n n

Inference on One Population Mean Hypothesis Testing

Section 6.3: Geometric Sequences

Module 9: String Matching

Introduction to Computational Molecular Biology. Suffix Trees

STURM-LIOUVILLE BOUNDARY VALUE PROBLEMS

Notes on Dirichlet L-functions

Sets. Sets. Operations on Sets Laws of Algebra of Sets Cardinal Number of a Finite and Infinite Set. Representation of Sets Power Set Venn Diagram

Lecture 2 Clustering Part II

MAHESH TUTORIALS SUBJECT : Maths(012) First Preliminary Exam Model Answer Paper

lecture 16: Introduction to Least Squares Approximation

1.4 Nonregular Languages

BRAIN TEASURES INDEFINITE INTEGRATION+DEFINITE INTEGRATION EXERCISE I

Digital Signal Processing, Fall 2006

Lesson 4 Linear Algebra

ANALYTIC WAVELETS AND MULTIRESOLUTION ANALYSIS A NOTE ON CERTAIN ORTHOGONALITY CONDITIONS

Probability and Stochastic Processes: A Friendly Introduction for Electrical and Computer Engineers Roy D. Yates and David J.

SOME SHARP OSTROWSKI-GRÜSS TYPE INEQUALITIES

10.5 Power Series. In this section, we are going to start talking about power series. A power series is a series of the form

9.5. Alternating series. Absolute convergence and conditional convergence

{ } { S n } is monotonically decreasing if Sn

The Elementary Arithmetic Operators of Continued Fraction

The Reimann Integral is a formal limit definition of a definite integral

Linear Algebra. Lecture 1 September 19, 2011

Transcription:

Compct Represettio of Ambiguities Geertig Seteces by Grmmr Speciliztio Jürge Wedekid Prsig Chrt: represets ltertive sytctic lyses pcked f-structures (cotexted costrit stisfctio, Mxwell & Kpl 91) Geertio Chrt-bsed pproches (Ky 96, Neum 98; bsed o Shieber 88) Chrcteriztio: 2 phses 1. phse: costructio of compct represettio of ltertive costituet structures/f-structures/strigs 2. phse: eumertio of prticulr lyses 1 2 Compct Represettio of Ambiguities Compct Represettio of Ambiguities Lg s forml chrcteriztio of the 2 phses (for cotext-free chrt prsig) 1. phse: grmmr speciliztio Give: CFG G & iput strig s Costruct: G s = chrt if s L(G) the L(G s ) = if s L(G) the L(G s ) = {s} d G s ssigs s the sme prse trees s G 2. phse: productio of prses by stdrd geertio lgorithm pplied to G s Our pproch to geertio (joit work with Ro Kpl) 1. phse: grmmr speciliztio Give: LFG G & cyclic f-structure f (iput) Costruct: G f (cotext-free) G f is speciliztio of the cotext-free bckboe of G G f derives ll strigs tht G reltes to f G f simultes ll derivtios of f i G 2. phse: productios of strigs by stdrd geertio lgorithm pplied to G f 3 4 LFG Represettios LFG Grmmr Setece, costituet structure d fuctiol structure Rules ( SUBJ)= S NP DET N studet c-structure VP V fell SUBJ studet NUM SG CASE NOM PAST fll<subj> f-structure NP DET N Cotext-free rules defie vlid c-structures ( )= studet Aottios re isttited t tree odes to give f(uctiol) descriptios = costrits o correspodig f-structure F-structure results from the miiml model of f-descriptio 6 1

NP S ( SUBJ)= ( )= studet NP S ( SUBJ)= ( )= studet DET N 6 studet 7 V DET N 6 studet 7 V 1. Well-formedess of costituet structure The whole structure is licesed by the rules 2. Isttitio of the metvribles 7 8 NP S ( SUBJ)= = ( )= studet NP S ( SUBJ)= = ( )= studet DET N 6 studet 7 V DET N 6 studet 7 V 2. Isttitio of metvribles by the licesed dughter 2. Isttitio of metvribles by the licesed dughter 9 10 NP S ( SUBJ)= = = = ( )= studet NP S ( SUBJ)= = = = = ( )= studet DET N 6 studet 7 V DET N 6 studet 7 V 2. Isttitio of metvribles by the licesed dughter 2. Isttitio of metvribles by the licesed dughter 11 12 2

NP S ( SUBJ)= = = = ( SPEC)=INDEF ( NUM)=SG = ( )= studet ( NUM)=SG ( SUBJ)= = = = ( SPEC)=INDEF ( NUM)=SG = ( )= studet ( NUM)=SG DET N 6 studet 7 V ( )= fll<subj> ( )=PAST V ( )= fll<subj> ( )=PAST 2. Isttitio of metvribles by the licesed dughter 3. Costructio of the f-descriptio = uio of the isttited descriptios of the licesig rules 13 14 V {( SUBJ)=, = } { =, = } { = } ( SPEC)=INDEF, ( )= studet, { ( NUM)=SG { ( NUM)=SG ( )= fll<subj>, { ( } )=PAST } } S NP DET 6 N studet 7 V ( SUBJ)=, =, =, =, {( SUBJ)=, = } =, = ( 3 SPEC)=INDEF, ( NUM)=SG, ( )= studet, ( NUM)=SG, { =, = } { = } ( )= fll<subj>, ( )=PAST ( SPEC)=INDEF, ( )= studet, { ( NUM)=SG } { ( NUM)=SG } ( )= fll<subj>, { ( } )=PAST 3. Costructio of the f-descriptio = uio of the isttited descriptios of the licesig rules 3. Costructio of the f-descriptio = uio of the isttited descriptios of the licesig rules 1 16 ( SUBJ)=, =, =, =, ( SUBJ)=, = } =, S = ( DET N 3 SPEC)=INDEF, ( NUM)=SG, ( )= studet, ( NUM)=SG, { =, = } { = } ( )= fll<subj>, ( )=PAST NP T ( SPEC)=INDEF, ( )= studet, ( NUM)=SG } { ( NUM)=SG } DET N V ( )= fll<subj>, { ( } )=PAST 6 studet 7 ( SUBJ)=, =, =, =, {( SUBJ)=, = } =, = ( 3 SPEC)=INDEF, ( NUM)=SG, ( )= studet, ( NUM)=SG, { =, = } { = } ( )= fll<subj>, ( )=PAST ( SPEC)=INDEF, ( )= studet, { ( NUM)=SG } { ( NUM)=SG } ( )= fll<subj>, { ( } )=PAST SUBJ 3 4 b c d PAST fll<subj> SPEC NUM e f g studet INDEF SG miiml model M 4. Stisfibility test d costructio of the miiml model 4. Stisfibility test d costructio of the miiml model 17 18 3

The LFG Derivtio Reltio SUBJ 3 4 b c d PAST fll<subj> SPEC NUM e f g I. Abstrctio from the odes: M L AV studet INDEF SG II. Abstrctio from the uiverses: [M L AV ] Give rbitrry LFG grmmr G Δ G = { (s,f ) G ssigs to strig s the f-structure f } Prser: Pr G (s) = { f Δ G (s,f ) } 4. Stisfibility test d costructio of the miiml model Geertor: Ge G (f) = { s Δ G (s,f ) } 19 20 Forml Properties Pr G (s) is fiite set, for every G d s. Proscriptio of obrchig domice chis i c- structure (off-lie prsbility) gurtees fiite umber of trees for ech strig, d ech tree supports fiite umber of f-structures Ge G (f) is ifiite, for some G d f. Loger d loger strigs with the sme f-structure: G: f: [X Y] S S A Ge G (f) = + S A A ( X)=Y S A S A S A Reductio of the f-descriptio Spce Give: -ry LFG rule r Ist(r,(t,t 1.. t )) Uio of the isttited ottios of r i.e. - ll occurreces of re substituted by t, - ll occurreces of i the ottio of the j-th dughter re substituted by t j r = ( SUBJ)= Ist(r,(, ( SUBJ) )) = {( SUBJ)=( SUBJ), =} 21 22 Reductio of the f-descriptio Spce Reductio of the f-descriptio Spce Give: -ry LFG rule r j is m-defible i Ist(r,(t,.. )) iff Ist(r,(t,.. )) I _ j = (t σ) S ( SUBJ) NP 1 = ( SUBJ) = ( SUBJ)= ( SUBJ) = 2 joh V ADV ( SUBJ ( )= joh = 4 ( ( ADJ)=( ELE) = fell 6 quickly 7 ( ( )= fll<subj> ( ) = quickly ( ( )=PAST 23 24 4

Reductio of the f-descriptio Spce Reductio of the f-descriptio Spce Two descriptios D d D with miiml models M d M re equivlet iff M L AV M L AV Remiig odes: d ll dughters which re ot m- defible However: their deottio is biuique T j ( σ) ( j σ ) if j is ot m-defible the from the descriptio it cot follow: j = for Τ Suppose: j = the j = ( j σ ) d σ = 0 thus ( σ) = j 2 26 Reductio of the f-descriptio Spce Reductio of the f-descriptio Spce F c e g SUBJ ADJ b c d e fll<subj> PAST ELE C F Biuiquely reme ll dughters which re ot m-defible Reducig substitutio = compositio of the defiitios d the remig substitutio g f joh h quickly ( SUBJ) Itroduce coicl costts for ech elemet g tht is ot deoted by tomic feture vlue Nodes ot occurrig i the f-descriptio 27 28 Reductio of the f-descriptio Spce Determitio of the f-descriptio Spce F c e g SUBJ ADJ b fll<subj> f c joh d PAST e g h ELE quickly C F ll terms which deote i the expsio (but do ot refer to the deottio of tomic feture vlue) C F d ( SUBJ) ( ADJ) ( g ELE) Appropritely isttited rules Give m-ry LFG rule r d terms t, t 1.. t m of T F (r, (t, t 1.. t m )) is ppropritely isttited iff (i) if t j = the j does ot occur i Ist(r,(t,.. m )) (ii) if j is m-defible i Ist(r,(t,.. m )) the Ist(r,(t,.. m )) I _ j = t j (iii) otherwise t j C F d t j = / t d t j = / t i for ll t i, i = / j plus ll terms tht re obtied from them by substitutig the ew costts by ( SUBJ) ( ADJ) ( ELE) plus T F 29 30

Determitio of the f-descriptio Spce Comptible isttitios IR F = set of ll ppropritely isttited rules fiite Two ppropritely isttited rules (r, (t, t 1.. t m )), (r, (t, t 1.. t l )) re comptible iff ll dughters which re ot m-defible re pir-wise distict isttited IRD F = set of ll subsets IR of IR F such tht (i) the isttited descriptios of the rules i IR together describe f fiite i.e. they hve miiml model M with M L AV f (ii) ll rules i IR re pir-wise comptible 31 32 Give: LFG G = (N,T,S,R) d iput f-structure [F] Costruct: CFG grmmr G F = (N F,T F,S F,R F ) S F = ew strt symbol (ot i N T) N F = {S F } {A:t:IR A N, t T F, IR is subset of set i IRD F } R F strt rules for every IR i IRD F, R F cotis rule S F S::IR T F = {:t: T, t T F } 33 34 R F The other rules ( SUBJ) For every LFG rule r = A X 1.. X m R F cotis rules of the form D 1 D m S: S ( SUBJ)= NP Joh NP:( SUBJ) NP VP: 2 (, ) ( SUBJ)= = 2 (, ) (, ) SUBJ PAST joh fll<subj> A:t:IR X 1 :t 1 :IR 1... X m :t m :IR m such tht Joh: V: V 3 ( )= joh = (, ) (i) IR = {(r,(t,t 1.. t m ))} IR 1.. IR m (ii) ech isttited rule occurrig i {(r,(t,t 1.. t m ))} IR i or IR i IR j ist selfcomptible fell: ( )= fll<subj> ( )=PAST 3 36 6

( SUBJ) ( SUBJ) (, ( ) SUBJ) ) ( SUBJ)= S: NP Joh (((, SUBJ) 3 ), ) NP:( SUBJ) VP: 2 (,, ) 4 ) (, ( SUBJ) ) ( SUBJ)= IR 1 IR 2 S::IR NP Joh (( SUBJ), ) IR 3 NP:( SUBJ):IR 1 VP::IR 2 1 2 (, ) IR 4 ( ( 4,, ) ) Joh: V: 4 fell: 1. Replce isttitig odes by the vlues of the reducig substitutio Joh::IR 3 V::IR (, ) 4 fell::ir IR 37 38 S:: (, ( SUBJ) ) ( SUBJ)= NP Joh (( SUBJ), ) (, ) (, ) S:: (, ( SUBJ) ) ( SUBJ)= NP Joh (( SUBJ), ) (, ) (, ) NP Joh (( SUBJ), ) NP:( SUBJ): (, ) 1 VP:: (, ) Joh:: (, ) term compoet simultes reducig substitutio V:: ( SUBJ) fell:: NP:( SUBJ): NP Joh (( SUBJ), ) Joh:: rule compoet cotis ll rules licesig the subderivtio (with the licesed odes replced by (, ) VP:: (, ) (, ) V:: fell:: 39 40 (, ( SUBJ) ) ( SUBJ)= NP Joh (( SUBJ), ) (, ( SUBJ) ) ( SUBJ)= NP Joh (( SUBJ), ) S:: (, ) (, ) S:: (, ) (, ) NP:( SUBJ): NP Joh (( SUBJ), ) Joh:: (, ) VP:: (, ) NP:( SUBJ): NP Joh (( SUBJ), ) Joh:: (, ) VP:: (, ) rule compoet cotis ll rules licesig the subderivtio (with the licesed odes replced by (, ) V:: fell:: rule compoet cotis ll rules licesig the subderivtio (with the licesed odes replced by (, ) V:: fell:: 41 42 7

S:: (, ( SUBJ) ) ( SUBJ)= NP Joh (( SUBJ), ) (, ) (, ) S:: (, ( SUBJ) ) ( SUBJ)= NP Joh (( SUBJ), ) (, ) (, ) NP:( SUBJ): NP Joh (( SUBJ), ) Joh:: (, ) VP:: (, ) NP:( SUBJ): NP Joh (( SUBJ), ) Joh:: (, ) VP:: (, ) rule compoet cotis ll rules licesig the subderivtio (with the licesed odes replced by (, ) V:: fell:: rule compoet cotis ll rules licesig the subderivtio (with the licesed odes replced by (, ) V:: fell:: 43 44 Observtios d Cosequeces Observtios d Cosequeces Two phses of LFG geertio Costructio of the CFG grmmr G F for [F] Well-uderstood, fiite represettio of (ifiitely) my strigs Apply CF geertio lgorithm to produce strigs Use pumpig lemm to select short strigs Use PCFG techology to select most probble strig Select by other heuristics Decidble properties of Ge G (f) It is decidble whether Ge G (f) is empty, fiite, or ifiite. Decidbility of geertio does ot deped o proscribig obrchig domice chis (off-lie prsbility), ulike prsig. 4 46 Observtios d Cosequeces Observtios d Cosequeces Exteds to other forml devices: sets, fuctiol ucertity, fuctiol precedece, etc. (but ot to the restrictio opertor) Applies to other cotext-free bsed uifictio formlisms: PATR, ALE, perhps some versios of HPSG 47 48 8