Grammar Implementation with Lexicalized Tree Adjoining Grammars and - - PowerPoint PPT Presentation

▶
grammar implementation with lexicalized tree adjoining
SMART_READER_LITE
LIVE PREVIEW

Grammar Implementation with Lexicalized Tree Adjoining Grammars and - - PowerPoint PPT Presentation

Sep 23, 2023 456 likes •1.29k views

Grammar Implementation with Lexicalized Tree Adjoining Grammars and Frame Semantics Further linguistic analyses Laura Kallmeyer, Timm Lichte, Rainer Osswald & Simon Petitjean University of Dsseldorf DGfS CL Fall School, September 13, 2017

slide-1
SLIDE 1

Grammar Implementation with Lexicalized Tree Adjoining Grammars and Frame Semantics

Further linguistic analyses Laura Kallmeyer, Timm Lichte, Rainer Osswald & Simon Petitjean

University of Düsseldorf

DGfS CL Fall School, September 13, 2017

SFB 991

slide-2
SLIDE 2

Outline of today’s course

1

Extraction phenomena in LTAG

2

Generalization and factorization within the elementary trees Tree families LTAG & metagrammar specification

3

LTAG semantics Synchronous TAGs for semantics Unification-based LTAG semantics with predicate logic Unification-based LTAG semantics with frames

4

Summary & outlook

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 2 2

slide-3
SLIDE 3

Outline of today’s course

1

Extraction phenomena in LTAG

2

Generalization and factorization within the elementary trees Tree families LTAG & metagrammar specification

3

LTAG semantics Synchronous TAGs for semantics Unification-based LTAG semantics with predicate logic Unification-based LTAG semantics with frames

4

Summary & outlook

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 3 3

slide-4
SLIDE 4

Extraction: some examples

certain constructions permit an element in one position to fill the grammatical role associated with another position

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 4 4

slide-5
SLIDE 5

Extraction: some examples

certain constructions permit an element in one position to fill the grammatical role associated with another position the positions can be arbitrarily far apart

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 5 4

slide-6
SLIDE 6

Extraction: some examples

certain constructions permit an element in one position to fill the grammatical role associated with another position the positions can be arbitrarily far apart filler – gap constructions

topicalization wh-movement relative clause

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 6 4

slide-7
SLIDE 7

Extraction: some examples

certain constructions permit an element in one position to fill the grammatical role associated with another position the positions can be arbitrarily far apart filler – gap constructions

topicalization wh-movement relative clause

long-distance dependencies

subject extraction

  • bject extraction

preposition stranding AP complement extraction

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 7 4

slide-8
SLIDE 8

Topicalization

Topicalization Placing a constituent (subject, object, ...) into a sentence-initial position.

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 8 5

slide-9
SLIDE 9

Topicalization

Topicalization Placing a constituent (subject, object, ...) into a sentence-initial position. (1) a. Adam gave an apple to Eve. (base configuration)

  • b. an applei, Adam gave _i to Eve.

(object NP) c. Evei, Adam gave an apple to _i. (NP from PP)

  • d. To Evei, Adam gave an apple _i.

(PP)

  • e. *Adam, _i gave an apple to Eve.

(no subject topicalization!)

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 9 5

slide-10
SLIDE 10

Topicalization

Unbounded dependency The dependency between an extracted constituent and its trace may extend across more clause boundaries.

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 10 6

slide-11
SLIDE 11

Topicalization

Unbounded dependency The dependency between an extracted constituent and its trace may extend across more clause boundaries. (2) a. The applei, Adam ate _i.

  • b. Applesi, Eve knows (that) Adam loves _i.

c. The applei, Adam believes (that) Eve knows (that) the snake ate _i.

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 11 6

slide-12
SLIDE 12

Wh-constructions

Wh-questions wh-questions involve a (possibly long-distance) extraction of a con- stituent as a wh-phrase.

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 12 7

slide-13
SLIDE 13

Wh-constructions

Wh-questions wh-questions involve a (possibly long-distance) extraction of a con- stituent as a wh-phrase. (3) a. [Who]i _i ate my apple?

  • b. [What]i did Eve eat _i?

c. [Which apple]i did Adam say Eve had eaten _i?

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 13 7

slide-14
SLIDE 14

Wh-constructions

Wh-questions wh-questions involve a (possibly long-distance) extraction of a con- stituent as a wh-phrase. (3) a. [Who]i _i ate my apple?

  • b. [What]i did Eve eat _i?

c. [Which apple]i did Adam say Eve had eaten _i? Subject-auxiliary inversion wh-questions involve subject-auxiliary inversion: The auxiliary verb (‘do’, ‘have’, ‘be’, ...) precedes the subject.

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 14 7

slide-15
SLIDE 15

Subject-auxiliary inversion

Obligatory subject-auxiliary inversion in direct questions with object extraction: (4) a. Whati does Adam eat _i?

  • b. *Whati Adam does eat _i?
  • c. *Whati Adam eats _i?

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 15 8

slide-16
SLIDE 16

Subject-auxiliary inversion

Obligatory subject-auxiliary inversion in direct questions with object extraction: (4) a. Whati does Adam eat _i?

  • b. *Whati Adam does eat _i?
  • c. *Whati Adam eats _i?

No subject-auxiliary inversion in embedded wh-questions: (5) a. Eve wonders [whati Adam eats _i].

  • b. *Eve wonders [whati does Adam eat _i].

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 16 8

slide-17
SLIDE 17

Subject-auxiliary inversion

Obligatory subject-auxiliary inversion in direct questions with object extraction: (4) a. Whati does Adam eat _i?

  • b. *Whati Adam does eat _i?
  • c. *Whati Adam eats _i?

No subject-auxiliary inversion in embedded wh-questions: (5) a. Eve wonders [whati Adam eats _i].

  • b. *Eve wonders [whati does Adam eat _i].

No subject-auxiliary inversion in topicalization: (6) a. *[The apple]i, has Adam eaten _i.

  • b. [The apple]i Adam has eaten _i.

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 17 8

slide-18
SLIDE 18

Extraction: elementary trees

subject extraction

  • bject extraction

preposition stranding

Whoi _i ate the apple? Whati did Adam eat _i? Whati does Adam dream of_i?

S S VP NP↓ V ate NPi ϵ NPi↓ S S VP NPi ϵ V eat NP↓ NPi↓ S S VP PP NPi ϵ P⋄ V dream NP↓ NPi↓

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 18 9

slide-19
SLIDE 19

Extraction: features

Features for extraction, taken from the XTAG grammar (XTAG Research Group 2001) extracted := + | – indicates extraction in the S-node wh := + | – indicates the presence of a wh-pronoun inv := + | – indicates inversion

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 19 10

slide-20
SLIDE 20

Extraction: features

Features for extraction, taken from the XTAG grammar (XTAG Research Group 2001) extracted := + | – indicates extraction in the S-node wh := + | – indicates the presence of a wh-pronoun inv := + | – indicates inversion Handling: no inversion with topicalization

(Booksi, people read _i.)

no topicalized subject

(*Peoplei, _i read books.)

no inversion with subject wh-extraction

(Whoi _i read books?)

inversion with object wh-extraction

(Whati do people read _i?)

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 20 10

slide-21
SLIDE 21

Extraction: elementary trees with features

Elementary tree for subject extraction: (7) Whoi _i arrived?

S

       inv

4

wh

3

extr +         S       inv

4

wh

3

            inv – agr

2

      VP

  • agr

2

  • V

arrived NP

  • trace

5

  • ϵ

NP↓         agr

2

wh

3 +

trace

5

       

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 21 11

slide-22
SLIDE 22

Inversion with object extraction

in case of object extraction

topicalization → no inversion wh-questions → inversion

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 22 12

slide-23
SLIDE 23

Inversion with object extraction

in case of object extraction

topicalization → no inversion wh-questions → inversion

⇒ equation of the wh feature of the extracted NP and the upper inv feature of the lower S node:

S

       inv

3

wh

3

extr +         S

  • inv

3

  • inv

–

  • NP↓
  • wh

3

  • Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf)

23 12

slide-24
SLIDE 24

Analyses

(8) Apples, Adam ate.

S

       inv

3

wh

3

extr +         S       inv

3

agr

2

            inv – agr

1

  • 3sg

+

     VP NP ϵ V ate NP↓

  • agr

1

  • NP↓
  • wh

3

  • NP
  • wh

–

  • apples

NP

  • agr
  • 3sg

+

  • Adam

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 24 13

slide-25
SLIDE 25

Analyses

Derived tree with top and botom feature structures:

S

       inv

3

wh

3

extr +         S       inv

3

agr

2

            inv – agr

1

  • 3sg

+

     VP NP ϵ V ate NP

  • agr

1

  • agr
  • 3sg

+

  • Adam

NP

  • wh

3

  • wh

–

  • apples

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 25 14

slide-26
SLIDE 26

Analyses

Final derived tree afer top-botom unification:

S         inv – wh – extr +         S       inv – agr

1

  • 3sg

+

     VP NP ϵ V ate NP

  • agr
  • 3sg

+

  • Adam

NP

  • wh

–

  • apples

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 26 15

slide-27
SLIDE 27

Analyses

(9) What does Adam eat?

S

       inv

3

wh

3

extr +         S       inv

3

agr

2

            inv – agr

1

      VP NP ϵ V eat NP↓

  • agr

1

  • NP↓
  • wh

3

  • NP
  • wh

+

  • what

NP

  • agr
  • 3sg

+

  • Adam

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 27 16

slide-28
SLIDE 28

Analyses

S

       inv

3

wh

3

extr +         S       inv

3

agr

2

            inv – agr

1

      VP NP ϵ V eat NP

  • agr

1

  • agr
  • 3sg

+

  • Adam

NP

  • wh

3

  • wh

+

  • what

S

     inv + agr

  • 3sg

+

     S∗

  • agr
  • 3sg

+

  • inv

–

  • V

does

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 28 17

slide-29
SLIDE 29

Analyses

S

       inv

3

wh

3

extr +         S       inv

3

agr

2

            inv + agr

  • 3sg

+

     S

  • agr
  • 3sg

+

     inv – agr

1

      VP NP ϵ V eat NP

  • agr

1

  • agr
  • 3sg

+

  • Adam

V does NP

  • wh

3

  • wh

+

  • what

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 29 18

slide-30
SLIDE 30

Analyses

S         inv + wh + extr +         S       inv + agr

  • 3sg

+

     S       inv – agr

  • 3sg

+

     VP NP ϵ V eat NP

  • agr
  • 3sg

+

  • Adam

V does NP

  • wh

+

  • what

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 30 18

slide-31
SLIDE 31

Outline of today’s course

1

Extraction phenomena in LTAG

2

Generalization and factorization within the elementary trees Tree families LTAG & metagrammar specification

3

LTAG semantics Synchronous TAGs for semantics Unification-based LTAG semantics with predicate logic Unification-based LTAG semantics with frames

4

Summary & outlook

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 31 19

slide-32
SLIDE 32

Tree families

Lexical anchoring Unanchored trees and their lexical anchors are specified separately. ⇒ the contribution of the lexical anchor is factored out.

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 32 20

slide-33
SLIDE 33

Tree families

Lexical anchoring Unanchored trees and their lexical anchors are specified separately. ⇒ the contribution of the lexical anchor is factored out. Lexical anchoring example:

S VP NP↓ V⋄

  • agr

1

  • NP↓
  • agr

1

  • Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf)

33 20

slide-34
SLIDE 34

Tree families

Lexical anchoring Unanchored trees and their lexical anchors are specified separately. ⇒ the contribution of the lexical anchor is factored out. Lexical anchoring example:

S VP NP↓ V⋄

  • agr

1

  • NP↓
  • agr

1

  • V
  • agr
  • 3sg

+

  • eats

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 34 20

slide-35
SLIDE 35

Tree families

Lexical anchoring Unanchored trees and their lexical anchors are specified separately. ⇒ the contribution of the lexical anchor is factored out. Lexical anchoring example:

S VP NP↓ V⋄

  • agr
  • 3sg

+

  • eats

NP↓

  • agr
  • 3sg

+

  • Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf)

35 20

slide-36
SLIDE 36

Tree families

Unanchored elementary trees are organized in tree families, which capture variations in the (syntactic) subcategorization frames.

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 36 21

slide-37
SLIDE 37

Tree families

Unanchored elementary trees are organized in tree families, which capture variations in the (syntactic) subcategorization frames. Example: (10) a. Adam eats apples.

  • b. Who eats apples?

c. Apples are eaten by Adam.

  • d. What was eaten by Adam?

e. Apples, Adam eats. f. ...

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 37 21

slide-38
SLIDE 38

Tree families

Unanchored tree family for transitive verbs:

S NP VP V◇ NP S NP S NP VP ε V◇ NP S NP VP V◇ PP P NP by S NP S NP VP ε V◇ PP P NP by S NP S NP VP V◇ NP ε ...

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 38 22

slide-39
SLIDE 39

Tree families

Unanchored tree family for transitive verbs:

S NP VP V◇ NP S NP S NP VP ε V◇ NP S NP VP V◇ PP P NP by S NP S NP VP ε V◇ PP P NP by S NP S NP VP V◇ NP ε ...

Options for the specification/generation of tree families: Transformation rules applied to base trees (e.g., metarules in XTAG) Classes of tree constraints (“metagrammar”, XMG system)

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 39 22

slide-40
SLIDE 40

LTAG & metagrammar specification

eXtensible MetaGrammar (XMG, Crabbé et al. 2013) A framework for specifying (the elementary structures of) tree based grammars by means of a declarative language (e.g., by dominance and precedence constraints) The specifications are organized into classes that can be reused (“imported”) by other classes. Classes may contain descriptions from different dimensions, and the XMG system can be extended in this respect, e.g., by a dimension of frame descriptions. An XMG compiler generates the elementary structures of a grammar from a metagrammar.

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 40 23

slide-41
SLIDE 41

LTAG & metagrammar specification: Example

Class CanSubj S NP ≺ VP V◇ Class ExtrSubj S NP[wh=yes] ≺∗ S NP ≺ VP ε V◇ Class Subj CanSubj ∨ ExtSubj Class DirObj VP V◇ ≺∗ NP Class ByObj VP[voice=passive] V◇ ≺∗ PP P ≺ NP by Class ActV VP[voice=active] V◇ Class PassV VP[voice=passive] V◇ Class Transitive ((Subj ∧ ActV ) ∨ ByObj ∨ PassV ) ∧ ((DirObj ∧ ActV ) ∨ (Subj ∧ PassV ))

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 41 24

slide-42
SLIDE 42

LTAG & metagrammar specification

Overall architecture metagrammar classes compilation unanchored tree families lexical entries lexical selection LTAG Next step: Add (frame) semantics to all components and link syntax to semantics.

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 42 25

slide-43
SLIDE 43

Outline of today’s course

1

Extraction phenomena in LTAG

2

Generalization and factorization within the elementary trees Tree families LTAG & metagrammar specification

3

LTAG semantics Synchronous TAGs for semantics Unification-based LTAG semantics with predicate logic Unification-based LTAG semantics with frames

4

Summary & outlook

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 43 26

slide-44
SLIDE 44

LTAG semantics: overview

Goal: an LTAG architecture of the syntax-semantics interface that is compositional: the meaning of a complex expression can be computed from the meaning of its subparts and its composi- tion operation. pairs entire elementary trees with meaning components.

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 44 27

slide-45
SLIDE 45

LTAG semantics: overview

Three principal approaches:

1 LTAG semantics with synchronous TAG (STAG)

(Shieber 1994; Nesson & Shieber 2006; 2008)

2 Unification based LTAG semantics with predicate logic

(Kallmeyer & Joshi 2003; Gardent & Kallmeyer 2003; Kallmeyer & Romero 2008)

3 Unification based LTAG semantics with frames

(Kallmeyer & Osswald 2013; Kallmeyer et al. 2016) We will use the third approach in this course and only briefly present the other two.

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 45 28

slide-46
SLIDE 46

LTAG semantics: STAG

Idea: pair two TAGs, one for syntax and one for L(ogical) F(orm) (= typed predicate logic), and do derivations in parallel. Formalism used for this: synchronous TAG (STAG) Shieber & Schabes (1990); Shieber (1994). STAG = two TAGs G1, G2 whose trees are related to each other. More precisely, it contains pairs γ1,γ2,link where γ1 is an elementary tree from G1, γ2 an elementary tree from G2, and link is a set of pairs of node addresses from γ1 and γ2 respectively.

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 46 29

slide-47
SLIDE 47

LTAG semantics: STAG

  • S 1

VP 3 V ‘laughed’ NP 2 t 1 , 3 e 2 e,t laugh

  • (The links are depicted with boxed numbers.)

The non-terminals of the semantic TAG are types t,e,e,t,. . . . The semantic TAG describes the syntactic structure of typed predicate logical formulas. The links in this example tell us, for instance, that the subject NP corresponds to the e argument of laugh.

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 47 30

slide-48
SLIDE 48

LTAG semantics: STAG

STAG derivation proceeds as in TAG, except that all operations must be paired: In every derivation step: A new elementary tree pair γ1,γ2 is picked. γ1 is atached (substituted or adjoined) to the syntactic tree while γ2 is atached to the semantic tree. The nodes that the two trees atach to must be linked. The link that is used in this derivation step disappears while all

  • ther links involving the atachment sites are inherited by the

root of the ataching tree.

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 48 31

slide-49
SLIDE 49

LTAG semantics: STAG

S 1 VP 3 V ‘laughed’ NP 2 t 1 , 3 e 2 e,t laugh NP ‘John’ e john

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 49 32

slide-50
SLIDE 50

LTAG semantics: STAG

S 1 VP 3 V ‘laughed’ NP ‘John’ t 1 , 3 e john e,t laugh

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 50 32

slide-51
SLIDE 51

LTAG semantics: STAG

S 1 VP 3 V ‘laughed’ NP ‘John’ t 1 , 3 e john e,t laugh VP VP∗ Adv ‘sometimes’ t t∗ t,t sometimes

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 51 32

slide-52
SLIDE 52

LTAG semantics: STAG

S 1 VP VP V ‘laughed’ Adv ‘sometimes’ NP ‘John’ t 1 t e john e,t laugh t,t sometimes

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 52 32

slide-53
SLIDE 53

LTAG semantics: STAG

S 1 VP VP V ‘laughed’ Adv ‘sometimes’ NP ‘John’ t 1 t e john e,t laugh t,t sometimes Logical form: sometimes(laugh(john))

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 53 32

slide-54
SLIDE 54

Unification-based LTAG semantics with predicate logic

Kallmeyer & Romero (2008), Gardent & Kallmeyer (2003): Syntax-Semantics Interface for LTAG Idea: Each elementary tree is paired with

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 54 33

slide-55
SLIDE 55

Unification-based LTAG semantics with predicate logic

Kallmeyer & Romero (2008), Gardent & Kallmeyer (2003): Syntax-Semantics Interface for LTAG Idea: Each elementary tree is paired with A set of typed predicate logic expressions and of scope con- straints (i.e., constraints on sub-term relations)

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 55 33

slide-56
SLIDE 56

Unification-based LTAG semantics with predicate logic

Kallmeyer & Romero (2008), Gardent & Kallmeyer (2003): Syntax-Semantics Interface for LTAG Idea: Each elementary tree is paired with A set of typed predicate logic expressions and of scope con- straints (i.e., constraints on sub-term relations) interface features that characterizes a) which arguments need to be filled, b) which elements are available as arguments for

  • ther elementary trees and c) the scope behaviour.

The features are linked to positions in the elementary tree.

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 56 33

slide-57
SLIDE 57

Unification-based LTAG semantics with predicate logic

S VP[p= 2 ]

[p=l1]

V ‘laughed’ NP[i= 1 ] l1 : laugh( 1 )

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 57 34

slide-58
SLIDE 58

Unification-based LTAG semantics with predicate logic

NP[I=x] ‘John’ l3 : john(x) S VP[p= 2 ]

[p=l1]

V ‘laughed’ NP[i= 1 ] l1 : laugh( 1 )

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 58 34

slide-59
SLIDE 59

Unification-based LTAG semantics with predicate logic

NP[I=x] ‘John’ l3 : john(x) S VP[p= 2 ]

[p=l1]

V ‘laughed’ NP[i= 1 ] l1 : laugh( 1 )

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 59 34

slide-60
SLIDE 60

Unification-based LTAG semantics with predicate logic

S VP[p= 2 ]

[p=l1]

V ‘laughed’ NP[i=x] ‘John’ l1 : laugh(x), l3 : john(x)

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 60 34

slide-61
SLIDE 61

Unification-based LTAG semantics with predicate logic

S VP[p= 2 ]

[p=l1]

V ‘laughed’ NP[i=x] ‘John’ l1 : laugh(x), l3 : john(x) VP[p=l2] VP∗[p= 7 ] Adv ‘sometimes’ l2 : sometimes( 6 ),

6 ⊳∗ 7

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 61 34

slide-62
SLIDE 62

Unification-based LTAG semantics with predicate logic

S VP[p= 2 ]

[p=l1]

V ‘laughed’ NP[i=x] ‘John’ l1 : laugh(x), l3 : john(x) VP[p=l2] VP∗[p= 7 ] Adv ‘sometimes’ l2 : sometimes( 6 ),

6 ⊳∗ 7

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 62 34

slide-63
SLIDE 63

Unification-based LTAG semantics with predicate logic

S VP[p= 2 ]

[p=l2]

VP[p= 7 ]

[p=l1]

V ‘laughed’ Adv ‘sometimes’ NP[i=x] ‘John’ l1 : laugh(x), l3 : john(x), l2 : sometimes( 6 ),

6 ⊳∗ 7

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 63 34

slide-64
SLIDE 64

Unification-based LTAG semantics with predicate logic

S VP[p = l2] VP[p = l1] V ‘laughed’ Adv ‘sometimes’ NP[i = x] ‘John’ l1 : laugh(x), l3 : john(x), l2 : sometimes( 6 ),

6 ⊳∗ l1

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 64 34

slide-65
SLIDE 65

Unification-based LTAG semantics with predicate logic

S VP[p = l2] VP[p = l1] V ‘laughed’ Adv ‘sometimes’ NP[i = x] ‘John’ l1 : laugh(x), l3 : john(x), l2 : sometimes( 6 ),

6 ⊳∗ l1

6 ⊳∗ l1 signifies that the formula labeled l1 is a subformula of the

formula that has to be placed in the hole 6 . Disambiguation leads to john(x) ∧ sometimes(laugh(x))

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 65 34

slide-66
SLIDE 66

Unification-based LTAG semantics with frames

Semantic representations are linked to entire elementary trees (as in the previous approaches). Semantic representations: frames, expressed as typed feature structures. Interface features relate nodes in the syntactic tree to nodes in the frame graph. Frame composition by unification, triggered by the unifica- tions on the interface features that are in turn triggered by substitution, adjunction and final top-botom unification on the derived tree.

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 66 35

slide-67
SLIDE 67

Unification-based LTAG semantics with frames

Reminder: (11) Adam ate an apple.

S VP[I=e] NP[I=y] V ‘ate’ NP[I=x] e          eating actor x theme y         

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 67 36

slide-68
SLIDE 68

Unification-based LTAG semantics with frames

Reminder: (11) Adam ate an apple.

NP[I=u] ‘Adam’ u       person name ‘Adam’       S VP[I=e] NP[I=y] V ‘ate’ NP[I=x] e          eating actor x theme y         

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 68 36

slide-69
SLIDE 69

Unification-based LTAG semantics with frames

Reminder: (11) Adam ate an apple.

NP[I=u] ‘Adam’ u       person name ‘Adam’       S VP[I=e] NP[I=y] V ‘ate’ NP[I=x] e          eating actor x theme y          x u

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 69 36

slide-70
SLIDE 70

Unification-based LTAG semantics with frames

Reminder: (11) Adam ate an apple.

NP[I=u] ‘Adam’ u       person name ‘Adam’       S VP[I=e] NP[I=y] V ‘ate’ NP[I=x] e          eating actor x theme y          NP[I=v] ‘an apple’ v

  • apple
  • x u

y v

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 70 36

slide-71
SLIDE 71

Unification-based LTAG semantics with frames

Reminder: (11) Adam ate an apple.

NP[I=u] ‘Adam’ u       person name ‘Adam’       S VP[I=e] NP[I=y] V ‘ate’ NP[I=x] e          eating actor x theme y          NP[I=v] ‘an apple’ v

  • apple
  • x u

y v

S VP[I=e] NP[I=y] ‘an apple’ V ‘ate’ NP[I=x] ‘Adam’ e              eating actor x       person name ‘Adam’       theme y

  • apple

           

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 71 36

slide-72
SLIDE 72

Unification-based LTAG semantics with frames

A more complex example that requires structure sharing in the frame: (12) Adam persuaded Eve to eat an apple. persuade is an object control verb: The object is not only an argument

  • f persuade but also an implicit argument of the embedded infinitive.

The semantic frame should (roughly) look like this:

                           persuasion actor       person name ‘Adam’       addressee x2       person name ‘Eve’       content           eating actor x2 theme

  • apple

                                   

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 72 37

slide-73
SLIDE 73

Unification-based LTAG semantics with frames

NP[i=x5] ‘Adam’ x5       person name ‘Adam’       S VP[i=e1] S∗[i=x2,e=e2] NP[i=x2] V ‘persuaded’ NP[i=x1] e1             persuasion actor x1 addressee x2 content e2             S[i=x3,e=e3] VP NP[i=x4] V ‘to eat’ NP ε e3          eating actor x3 theme x4          NP[i=x6] ‘Eve’ x6       person name ‘Eve       NP[i=x7] ‘an apple’ x7

  • apple
  • Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf)

73 38

slide-74
SLIDE 74

Unification-based LTAG semantics with frames

NP[i=x5] ‘Adam’ x5       person name ‘Adam’       S VP[i=e1] S∗[i=x2,e=e2] NP[i=x2] V ‘persuaded’ NP[i=x1] e1             persuasion actor x1 addressee x2 content e2             S[i=x3,e=e3] VP NP[i=x4] V ‘to eat’ NP ε e3          eating actor x3 theme x4          NP[i=x6] ‘Eve’ x6       person name ‘Eve       NP[i=x7] ‘an apple’ x7

  • apple
  • Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf)

74 38

slide-75
SLIDE 75

Unification-based LTAG semantics with frames

S VP[i=e1] S∗[i=x2,e=e2] NP[i=x2] ‘Eve’ V ‘persuaded’ NP[i=x1] ‘Adam’ e1                     persuasion actor x1       person name ‘Adam’       addressee x2       person name ‘Eve’       content e2                     S[i=x3,e=e3] VP NP[i=x4] ‘an apple’ V ‘to eat’ NP ε e3           eating actor x3 theme x4

  • apple

        

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 75 39

slide-76
SLIDE 76

Unification-based LTAG semantics with frames

S VP[i=e1] S∗[i=x2,e=e2] NP[i=x2] ‘Eve’ V ‘persuaded’ NP[i=x1] ‘Adam’ e1                     persuasion actor x1       person name ‘Adam’       addressee x2       person name ‘Eve’       content e2                     S[i=x3,e=e3] VP NP[i=x4] ‘an apple’ V ‘to eat’ NP ε e3           eating actor x3 theme x4

  • apple

        

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 76 39

slide-77
SLIDE 77

Unification-based LTAG semantics with frames

S VP[i=e1] S[i=x2,e=e2] VP NP[i=x4] ‘an apple’ V ‘to eat’ NP ε NP[i=x2] ‘Eve’ V ‘persuaded’ NP[i=x1] ‘Adam’ e1                            persuasion actor x1       person name ‘Adam’       addressee x2       person name ‘Eve’       content e2           eating actor x2 theme x4

  • apple

                                   

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 77 40

slide-78
SLIDE 78

Outline of today’s course

1

Extraction phenomena in LTAG

2

Generalization and factorization within the elementary trees Tree families LTAG & metagrammar specification

3

LTAG semantics Synchronous TAGs for semantics Unification-based LTAG semantics with predicate logic Unification-based LTAG semantics with frames

4

Summary & outlook

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 78 41

slide-79
SLIDE 79

Summary & outlook

Summary LTAG features for extraction phenomena grammar factorization LTAG approaches for the syntax-semantics interface

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 79 42

slide-80
SLIDE 80

Summary & outlook

Summary LTAG features for extraction phenomena grammar factorization LTAG approaches for the syntax-semantics interface Tomorrow introduction to frame semantics formalization using atribute value logic type constraints, type hierarchy extensions

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 80 42

slide-81
SLIDE 81

References

Crabbé, Benoit, Denys Duchier, Claire Gardent, Joseph Le Roux & Yannick Parmentier. 2013. XMG: eXtensible MetaGrammar. Computational Linguistics 39(3). 1–66.

http://hal.archives-ouvertes.fr/hal-00768224/en/.

Gardent, Claire & Laura Kallmeyer. 2003. Semantic Construction in FTAG. In Proceedings

  • f eacl 2003, 123–130. Budapest.

Kallmeyer, Laura & Aravind K. Joshi. 2003. Factoring Predicate Argument and Scope Semantics: Underspecified Semantics with LTAG. Research on Language and Computation 1(1–2). 3–58. Kallmeyer, Laura & Rainer Osswald. 2013. Syntax-driven semantic frame composition in Lexicalized Tree Adjoining Grammars. Journal of Language Modelling 1(2). 267–330. Kallmeyer, Laura, Rainer Osswald & Sylvain Pogodalla. 2016. For-adverbials and aspectual interpretation: An LTAG analysis using hybrid logic and frame semantics. In Christopher Pi˜ nón (ed.), Empirical issues in syntax and semantics EISS, vol. 11, . Kallmeyer, Laura & Maribel Romero. 2008. Scope and situation binding in LTAG using semantic unification. Research on Language and Computation 6(1). 3–52. Nesson, Rebecca & Stuart M. Shieber. 2006. Simpler TAG semantics through synchronization. In Proceedings of the 11th conference on formal grammar, Malaga, Spain.

http://www.eecs.harvard.edu/~shieber/Biblio/Papers/Nesson-2006-SSS.pdf.

Nesson, Rebecca & Stuart M. Shieber. 2008. Synchronous vector tag for syntax and semantics: Control verbs, relative clauses, and inverse linking. In Proceedings of the ninth international workshop on tree adjoining grammars and related formalisms (tag+ 9), Tübingen, Germany.

slide-82
SLIDE 82

References (cont.)

Shieber, Stuart M. 1994. Restricting the weak-generative capacity of synchronous Tree-Adjoining Grammars. Computational Intelligence 10(4). 271–385. Shieber, Stuart M. & Yves Schabes. 1990. Synchronous Tree-Adjoining Grammars. In Proceedings of coling, 253–258. XTAG Research Group. 2001. A Lexicalized Tree Adjoining Grammar for English. Tech. rep. Institute for Research in Cognitive Science, University of Pennsylvania Philadelphia, PA.