markus egg, alexander koller, joachim niehren the constraint language for lambda structures

49
Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures Ventsislav Zhechev SfS, Universität Tübingen e-mail: [email protected]

Upload: tayten

Post on 20-Mar-2016

21 views

Category:

Documents


0 download

DESCRIPTION

Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures. Ventsislav Zhechev SfS, Universität Tübingen e-mail: [email protected]. Introduction Motivation Basic Terms. Elements of CLLS -terms -structures Discussed Phenomena. Agenda. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

Markus Egg, Alexander Koller, Joachim NiehrenThe Constraint Language

for Lambda StructuresVentsislav Zhechev

SfS, Universität Tübingene-mail: [email protected]

Page 2: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 2

Agenda• Introduction• Motivation• Basic Terms

• Elements of CLLS• -terms• -structures• Discussed Phenomena

• Introduction• Motivation• Basic Terms

• Elements of CLLS• -terms• -structures• Discussed Phenomena

Page 3: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 3

Agenda (continued)• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

• Introduction• Motivation• Basic Terms

• Elements of CLLS• -terms• -structures• Discussed Phenomena

Page 4: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 4

Agenda (continued)• Interaction of Quantifiers, Anaphora and

Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

• Introduction• Motivation• Basic Terms

• Elements of CLLS• -terms• -structures• Discussed Phenomena

Page 5: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 5

Agenda (continued)• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Computational Aspects• Conclusion

• Introduction• Motivation• Basic Terms

• Elements of CLLS• -terms• -structures• Discussed Phenomena

Page 6: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 6

• Introduction• Motivation• Basic Terms

• Introduction• Motivation• Basic Terms

• Introduction• Motivation• Basic Terms

• Linguistic Phenomena• Scope Ambiguities• Anaphora• VP Ellipsis

• Underspecification

• Elements of CLLS• -terms• -structures• Discussed Phenomena

IntroductionMotivation

Page 7: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 7

• Introduction• Motivation• Basic Terms

• Trees• Underspecification• Constraint Language for Lambda Structures:

A Combination of Constraints• Dominance Constraints• Anaphoric Binding Constraints• Parallelism Constraints• -binding Constraints

• Introduction• Motivation• Basic Terms

• Elements of CLLS• -terms• -structures• Discussed Phenomena

Basic Terms

Page 8: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 8

• Introduction• Motivation• Basic Terms

• Elements of CLLS• -terms• -structures• Discussed Phenomena

• Every linguist attends a workshop.• (a workshop)(x

(every linguist)(y(attend x) y))

• Types• e individuals• t truth values (0 or 1 / true or false)• <e,t> one-place predicates• <e,<e,t>> two-place predicates• etc.

• Elements of CLLS• -terms• -structures• Discussed Phenomena

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

• Elements of CLLS• -terms• -structures• Discussed Phenomena

Elements of CLLS-terms

Page 9: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 9

• (a workshop)(x(every linguist)(y(attend x) y))

• lam -abstraction• @ functional application• var bound variable• variable binding

• Elements of CLLS• -terms• -structures• Discussed Phenomena

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

• Elements of CLLS• -terms• -structures• Discussed Phenomena

-structuresattend

Page 10: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 10

• Scope Ambiguity• Every linguist attends a workshop.• (a workshop)(x

(every linguist)(y(attend x) y))

• (every linguist)(y (a workshop)(x(attend x) y))

• dominance

• Elements of CLLS• -terms• -structures• Discussed Phenomena

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

• Elements of CLLS• -terms• -structures• Discussed Phenomena

Discussed Phenomena

Page 11: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 11

• Elements of CLLS• -terms• -structures• Discussed Phenomena

• VP Ellipsis• Every man sleeps, and so does Mary.•

• Parallelism Constraint:X1/X2~Y1/Y2

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

Page 12: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 12

• Elements of CLLS• -terms• -structures• Discussed Phenomena

• Anaphora• Johni said hei

j liked hisj mother.

• ana anaphora• anaphoric link

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

Page 13: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 13

• Elements of CLLS• -terms• -structures• Discussed Phenomena

• The Capturing Problem• Variable binding in -terms is usually indicated by

using variable names, i.e. x binds all occurrences of x in its scope

• Possible Problems:• -calculus has to exclude the capturing of free

variables by unintended binders• Problems with constraints used for scope ambiguities

• Problems in the presence of parallelism constraints

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

Page 14: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 14

• Elements of CLLS• -terms• -structures• Discussed Phenomena

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

• Trees and Tree Structures• The Algebra of Trees

• Let be a set of function symbols, f, g, a, b• Each function symbol f has fixed arity• We write fk for a function symbol f with arity k ≥ 0• We define tree as a ground term built from a set of

function symbols• We define path as a word over ℕ (the natural numbers)• We identify each node in a tree with the path from the

root to this node• The empty word, , identifies the root• Concatenation is written as • A word is a prefix of , iff there is a word 1

such that =1

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

Syntax and Semantics of CLLS

Tree Structures

Page 15: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 15

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

• Tree Structures• tree domain is a finite nonempty set of nodes,

which is prefix closed ( ) and closed under left siblings (i j for all 1≤j<i)

• tree structure is defined as follows:

• Given nodes 0,..., n, we write 0:f(1,..., n) for(0, 1,..., n) :f

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

Page 16: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 16

• Tree Structures and -structures• Formalization

• For -structures we assume:{var0, ana0, lam1, @2}

• We define -structures as follows:

• We draw -structures as tree-like graphs

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

-structures

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

Page 17: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 17

• Dominance• Let be a -structure and , two of its nodes.

We say that dominates (⊲*), if lies above ,i.e. is a prefix of

• Dominance is a partial order on the domain of and it is reflexive, transitive and antisymmetric

• Parallelism• We call any pair / of nodes , in with ⊲* a

segment of , where is called the root and the hole of the segment

• We define:

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

Dominance and Parallelism

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

Page 18: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 18

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

• Correspondence Functions between segments:

• Parallelism Relation:

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

Page 19: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 19

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

Page 20: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 20

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

• We assume an infinite set of node variables, ranged over X, Xi, Y, etc.

• We pick relation symbols for all relations defined so far

• Finally we define CLLS with the following abstract syntax:

• The Semantics of CLLS is defined by interpretation of constraints over the class of-structures:• A pair of a -structure and a variable assignment into

the domain of satisfies a constraint , iff it satisfies each atomic conjunct of it

• We call (, ) a solution of in this case

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

The CLLS

Page 21: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 21

• CLLS Constraints are usually hard to read in the standard syntax. That is why we will use constraint graphs for presenting the constraints

• For Example:

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

Constraint Graphs

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

Page 22: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 22

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Computational Aspects• Conclusion

• Syntax and Semantics of CLLS• Tree Structures• -structures• Dominance and Parallelism• The CLLS• Constraint Graphs

• Throughout the chapter we assume a fixed signature:={@2, lam1, var0, ana0, before2, mary0, read0, ...}

• We follow the convention that proper nouns are always analyzed as constants of type e, except as contrasting elements in ellipses where the other contrasting element is a quantifier

Interaction of Quantifiers, Anaphora and Ellipsis

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

Page 23: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 23

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Computational Aspects• Conclusion

• Every linguist attends a workshop.• Every computer scientist does, too.• The pair of sentences has three possible

readings, although it may seem that there are four

• The CLLS constraint for the two sentences looks like this:

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

Quantifier Parallelism

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

Page 24: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 24

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Computational Aspects• Conclusion

• John likes his mother, and Bill does too.• The sentence has two readings: strict (Bill

likes John’s mother) and sloppy (Bill likes Bill’s mother)

• We describe the meaning of the sentence using parallelism and anaphoric linking constraints:

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

Strict/Sloppy Ambiguities

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

Page 25: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 25

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Computational Aspects• Conclusion

• According to the parallelism constraint the tree part of the -structure below Xt is the same as the one below Xs, except for the contrasting elements, as follows:

• This is yet not a complete -structure, because the anaphor at Xa’ doesn’t have an antecedent

Page 26: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 26

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Computational Aspects• Conclusion

• John revised his paper before the teacher did, and so did Bill.

• This sentence comprises nested ellipsis: the source clause of the ellipsis is elliptical itself

• The sentence is further complicated by the presence of the anaphor, which induces a complex strict/sloppy ambiguity

• We follow Dalrymple et al. (1991) in assuming five readings for the sentence

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

Nested Ellipses

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

Page 27: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 27

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Computational Aspects• Conclusion

• All five readings are represented by the following constraint:

Page 28: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 28

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Computational Aspects• Conclusion

• Mary read a book she liked before Sue did.• The sentence has three readings

• In the first reading the indefinite NP a book she liked outscopes both clauses

• The second and the third reading arise from a strict/sloppy ambiguity that occurs if the operator before outscopes the indefinite

• Here is a constraint describing the readings:

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

A Complex Interaction

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

Page 29: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 29

• Schematic representations of the solutions:• First reading:

• Second reading:

• Third reading:

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Computational Aspects• Conclusion

Page 30: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 30

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Computational Aspects• Conclusion

• John greeted every person that Max did.• The problem is that the ellipsis is contained in

the VP it refers to• In CLLS the meaning of the sentence is

described as follows:

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

Antecedent-Contained Ellipsis

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

Page 31: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 31

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Computational Aspects• Conclusion

• There is one problem with this analysis:the notion of binding equivalence as defined is too strong a restriction for ACD

• The redefinition is given as:

Page 32: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 32

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Computational Aspects• Conclusion

• The preconditions for the two branches of the definitions are given here as (a) and (b) respectively:

• This analysis also accounts for the difference between the following two sentences (the first one lacking one of the two readings of the second sentence):• John wants Bill to read everything that Max does.• John wants Bill to read everything Max wants him to

read.

Page 33: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 33

• Interaction of Quantifiers, Anaphora and Ellipsis• Quantifier Parallelism• Strict/Sloppy Ambiguities• Nested Ellipses• A Complex Interaction• Antecedent-Contained Ellipsis

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Phrase Structure Rules:

• The Lexicon is defined by a relation Lex, which relates words W and lexical categories{Det, N, IV, TV, SV, RP, ...}. Terminal productions (a13) expand lexical categories to words of this category

The Syntax-Semantics InterfaceGrammar

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Computational Aspects• Conclusion

• Computational Aspects• Conclusion

Page 34: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 34

• The Syntax-Semantics Interface should factor out as much of the constraint construction as possible into the interface rules

• Most of the lexical entries introduce just one labeling constraint

• For each node in the syntax tree a constraint is generated; the constraint of the whole tree is the conjunction of these subconstraints

• Each node ℕ* in the syntax tree is associated with two variables, X

s (the local scope domain of ) and X

r (the root of the subconstraint for )

Semantic Construction

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Computational Aspects• Conclusion

Page 35: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 35

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• We add a constraint Xs⊲*X

r for each determiner which is not an indefinite. We also add this constraint whenever is a verb

• We associate with each NP an index i that is used in the syntactic tree for coindexation with a variable Xi

• The variables associated with syntactic nodes are related by the following rules:

• Computational Aspects• Conclusion

Page 36: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 36

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Computational Aspects• Conclusion

Page 37: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 37

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Computational Aspects• Conclusion

Page 38: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 38

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Computational Aspects• Conclusion

Page 39: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 39

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• The complete constraint which the interface produces is the conjunction of all the local constraints we just mentioned, plus the labeling constraints for the lexical entries, of the typeX

r:sleep• Exceptions to this rule:

• The elliptic does (too) does not add a labeling constraint; its semantics is determined via a parallelism constraint

• Whenever coindexation signifies a relation between an anaphor ’ and its antecedent , we add the constraintX

r=Xi, when we process and the constraint X’r:ana

ante(X’r)=Xi when we process ’

• Computational Aspects• Conclusion

Page 40: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 40

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• The constraint for a relative pronoun with index i at isX

r=Xi Xi:var; and the constraint for the corresponding trace (say, at ’) is X’

r=Xi. This, together with rule (b11), enforces correct binding of the trace

• The constraints for possessive pronouns, such as his, are as follows:

• Computational Aspects• Conclusion

Page 41: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 41

• Every linguist attends a workshop.•

• First the lexical elements introduce several labeling constraints:X11

r:every, X121r:linguist, X21

r:attend,X221

r:a, X2221r:workshop

An Example

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Computational Aspects• Conclusion

Page 42: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 42

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• The constraints for the NPs are built from the above by the rules (b8) and (b9):•

• A1:@(XA2r, XL1

r) A2:@(X11r, X121

r) L1:lam(XL2r)

X1r:var XL2

r⊲*X1r (X1

r)=XL1r XL1

r≠X1r X1

s=X11s=X121

s

• A3:@(XA4r, XL3

r) A4:@(X221r, X2221

r) L3:lam(XL4r)

X22r:var XL4

2⊲*X22r (X22

r)=XL3r XL3

r≠X22r

X22s=X221

s=X2221s

• Then rule (b3) combines the transitive verb and its object• X2

r:@(X21r, X22

r) X2s=X21

s=X22s

• Computational Aspects• Conclusion

A2

A1

L1

L2

Page 43: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 43

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Rule (b1) analogously combines the subject and the VP• X

r:@(X2r, X1

r) Xs=X2

s=X1s

• So far we have the following constraint:•

• X11r:every X121

r:linguist X21r:attend X221

r:a X2221

r:workshop A1:@(XA2r, XL1

r) A2:@(X11r, X121

r) L1:lam(XL2

r) X1r:var XL2

r⊲*X1r (X1

r)=XL1r XL1

r≠X1r

A3:@(XA4r, XL3

r) A4:@(X221r, X2221

r) L3:lam(XL4r) X22

r:var XL4

2⊲*X22r (X22

r)=XL3r XL3

r≠X22r X2

r:@(X21r, X22

r) X

r:@(X2r, X1

r) X1s=X11

s=X121s=X22

s=X221s=X2221

s=X21s=X

s=X2s

• Computational Aspects• Conclusion

A4A3 A2

A1

L4

L3

L2

L1

Xr

X2r

Page 44: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 44

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Finally we add the relevant scope island constraints:• The complete sentence is associated with the

variable Xs, and all other Xs variables are forced to

be equal to this one by other constraints• Node 11 is a determiner and node 21 is a verb; so

we add the constraint X11s ⊲* X11

r X21s ⊲* X21

r

• Computational Aspects• Conclusion

Page 45: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 45

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• The final constraint we get is the following: •

• X11r:every X121

r:linguist X21r:attend X221

r:a X2221

r:workshop A1:@(XA2r, XL1

r) A2:@(X11r, X121

r) L1:lam(XL2

r) X1r:var XL2

r⊲*X1r (X1

r)=XL1r XL1

r≠X1r

A3:@(XA4r, XL3

r) A4:@(X221r, X2221

r) L3:lam(XL4r) X22

r:var XL4

2⊲*X22r (X22

r)=XL3r XL3

r≠X22r X2

r:@(X21r, X22

r) X

r:@(X2r, X1

r) X1s=X11

s=X121s=X22

s=X221s=X2221

s=X21s=X

s=X2s

Xs ⊲* X11

r Xs ⊲* X21

r

• Computational Aspects• Conclusion

A4

A3A2

A1

L4

L3

L2

L1

Xr

X2r

X221r X2221

rX121

r

X22r

X1r

Page 46: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 46

• The Syntax-Semantics Interface• Grammar• Semantic Construction• An Example

• Disambiguation of arbitrary CLLS description is very complex: it has been shown that even CLLS without binding is equivalent to context unification, whose decidability is an open problem in theoretical computer science

• There are, however, semi-decision procedures which will eventually enumerate all solved forms of a constraint

• For the sublanguage of dominance constraints it was shown, that the satisfiability problem is decidable, but NP-complete

Computational Aspects

• Computational Aspects• Conclusion• Conclusion• Computational Aspects

Page 47: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 47

• An implementation of a solver for dominance constraints can be obtained by employing constraint programming with finite sets. The constraints can be solved by always performing deterministic propagation steps to eliminate hopeless choices before making case distinctions.

• It can be shown that all dominance constraints that are needed for the linguistic application belong to a fragment called normal dominance constraints. Satisfiability of a normal constraint can be checked by a graph algorithm of polynomial runtime; each reading can be enumerated in polynomial time as well. However the graph algorithm is not a complete solver for all dominance constraints.

• Conclusion• Computational Aspects

Page 48: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 48

• Computational Aspects

• CLLS allows the representation of scope ambiguities, anaphora and ellipsis in simple underspecified structures that are transparent and suitable for processing.

• We have shown that CLLS correctly represents many notorious problems from the literature involving scope, anaphora, ellipses and their interactions.

• Furthermore CLLS can be used to model reinterpretation (meaning shift) of aspect and NPs in an underspecified way.

• Nevertheless the linguistic coverage of CLLS still has to be extended.

• Various more formal aspects can also be pursued in the future.

Conclusion

• Conclusion• Conclusion

Page 49: Markus Egg, Alexander Koller, Joachim Niehren The Constraint Language for Lambda Structures

19.01.2005 49

Thank you!

• Conclusion