Multi Context-Free Tree Grammars and Multi-component Tree Adjoining - - PowerPoint PPT Presentation

multi context free tree grammars and multi component tree
SMART_READER_LITE
LIVE PREVIEW

Multi Context-Free Tree Grammars and Multi-component Tree Adjoining - - PowerPoint PPT Presentation

Multi Context-Free Tree Grammars and Multi-component Tree Adjoining Grammars Joost Engelfriet 1 Andreas Maletti 2 1 LIACS, , Leiden, The Netherlands 2 Institute of Computer Science, , Leipzig, Germany maletti@informatik.uni-leipzig.de


slide-1
SLIDE 1

Multi Context-Free Tree Grammars and Multi-component Tree Adjoining Grammars

Joost Engelfriet1 Andreas Maletti2

1 LIACS,

, Leiden, The Netherlands

2 Institute of Computer Science,

, Leipzig, Germany

maletti@informatik.uni-leipzig.de

Altenberg, Germany

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 1

slide-2
SLIDE 2

Motivation

Definition

Context-free grammar (N, Σ, S, R) is in Greibach normal form if each rule ρ ∈ R \ {S → ε} is of the form ρ = A → σA1 · · · An with σ ∈ Σ and A, A1, . . . , An ∈ N

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 2

slide-3
SLIDE 3

Motivation

Definition

Context-free grammar (N, Σ, S, R) is in Greibach normal form if each rule ρ ∈ R \ {S → ε} is of the form ρ = A → σA1 · · · An with σ ∈ Σ and A, A1, . . . , An ∈ N

Theorem [Greibach 1965]

Every CFG can be turned into an equivalent CFG in Greibach normal form

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 2

slide-4
SLIDE 4

Motivation

Definition

CFG (N, Σ, S, R) is lexicalized if occΣ(r) = ∅ for each rule (A → r) ∈ R \ {S → ε}

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 3

slide-5
SLIDE 5

Motivation

Definition

CFG (N, Σ, S, R) is lexicalized if occΣ(r) = ∅ for each rule (A → r) ∈ R \ {S → ε} CFG in Greibach normal form is lexicalized

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 3

slide-6
SLIDE 6

Motivation

Definition

CFG (N, Σ, S, R) is lexicalized if occΣ(r) = ∅ for each rule (A → r) ∈ R \ {S → ε} CFG in Greibach normal form is lexicalized lexicographers (linguists) love lexicalized grammars

  • ccurrence of lexical element in a rule is called anchor

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 3

slide-7
SLIDE 7

Motivation

S NP PRP We VP MD must VP VB bear PP IN in NP NN mind NP NP DT the NN Community PP IN as NP DT a NN whole

linguists nowadays care more about the parse tree than the membership of its yield in the (string) language modern grammar formalisms generate tree and string languages

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 4

slide-8
SLIDE 8

Motivation

Definition

For two tree grammars G and G′, of which G′ is lexicalized, G′ weakly lexicalizes G if yield(L(G′)) = yield(L(G)) G′ strongly lexicalizes G if L(G′) = L(G)

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 5

slide-9
SLIDE 9

Motivation

Definition

For two tree grammars G and G′, of which G′ is lexicalized, G′ weakly lexicalizes G if yield(L(G′)) = yield(L(G)) G′ strongly lexicalizes G if L(G′) = L(G) tree language preserved under strong lexicalization string language preserved under weak lexicalization

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 5

slide-10
SLIDE 10

Motivation

Definition

For two tree grammars G and G′, of which G′ is lexicalized, G′ weakly lexicalizes G if yield(L(G′)) = yield(L(G)) G′ strongly lexicalizes G if L(G′) = L(G) tree language preserved under strong lexicalization string language preserved under weak lexicalization lifed to classes C and C′ as usual C′-grammars strongly lexicalize C-grammars if for every G ∈ C there exists a lexicalized G′ ∈ C′ such that L(G′) = L(G)

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 5

slide-11
SLIDE 11

Motivation

Some results: CFGs (local tree grammars) weakly lexicalize themselves [Greibach 1965] Tree adjoining grammars (TAGs) strongly lexicalize CFGs [Joshi, Schabes 1997]

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 6

slide-12
SLIDE 12

Motivation

Some results: CFGs (local tree grammars) weakly lexicalize themselves [Greibach 1965] Tree adjoining grammars (TAGs) strongly lexicalize CFGs [Joshi, Schabes 1997] TAGs strongly lexicalize themselves [Joshi, Schabes 1997]

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 6

slide-13
SLIDE 13

Motivation

Some results: CFGs (local tree grammars) weakly lexicalize themselves [Greibach 1965] Tree adjoining grammars (TAGs) strongly lexicalize CFGs [Joshi, Schabes 1997] TAGs strongly lexicalize themselves [Joshi, Schabes 1997] TAGs do not strongly lexicalize themselves [Kuhlmann, Satta 2012]

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 6

slide-14
SLIDE 14

Motivation

Some results: CFGs (local tree grammars) weakly lexicalize themselves [Greibach 1965] Tree adjoining grammars (TAGs) strongly lexicalize CFGs [Joshi, Schabes 1997] TAGs strongly lexicalize themselves [Joshi, Schabes 1997] TAGs do not strongly lexicalize themselves [Kuhlmann, Satta 2012] Context-free tree grammars (CFTGs) strongly lexicalize TAGs and themselves [Maletti, Engelfriet 2013]

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 6

slide-15
SLIDE 15

Contents

1

Motivation

2

Main notion

3

Lexicalization

4

Expressive Power

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 7

slide-16
SLIDE 16

Main notion

Definition [Engelfriet, Maneth 1998; Kanazawa 2010]

Multiple context-free tree grammar (MCFTG) G = (N, B, Σ, S, R) finite totally ordered ranked alphabet N (nonterminals) partition B ⊆ P(N) of N (big nonterminals) finite ranked alphabet Σ (terminals) S ∈ N(0) with {S} ∈ B (initial big nonterminal) finite set R of rules of the form A → r with A ∈ B and N-linear forest r ∈ CN∪Σ(X)+ such that rk+(r) = rk+(A) and B saturates occN(r)

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 8

slide-17
SLIDE 17

Main notion

Definition [Engelfriet, Maneth 1998; Kanazawa 2010]

Multiple context-free tree grammar (MCFTG) G = (N, B, Σ, S, R) finite totally ordered ranked alphabet N (nonterminals) partition B ⊆ P(N) of N (big nonterminals) finite ranked alphabet Σ (terminals) S ∈ N(0) with {S} ∈ B (initial big nonterminal) finite set R of rules of the form A → r with A ∈ B and N-linear forest r ∈ CN∪Σ(X)+ such that rk+(r) = rk+(A) and B saturates occN(r) MCFTGs generalize (linear, nondeleting) CFTGs to multiple components multiple components synchronously applied to “synchronized” nonterminal occurrences

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 8

slide-18
SLIDE 18

Main notion

Nonterminals S, A, C, C′, T1, T2, T3:

A → T1 σ C T2 T3 C x1 → σ C x1 C′ A C′ x1 → σ C x1 C′ A T1 x1 T2 T3 → γ T1 τ x1 σ T2 α ν T3 S → γ A C x1 → x1 C′ x1 → x1 T1 x1 T2 T3 → x1 α β

(nonterminals that constitute a big nonterminal connected by splines)

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 9

slide-19
SLIDE 19

Main notion

T1 x1 T2 T3 → γ T1 τ x1 σ T2 α ν T3 nonterminals T1, T2, T3 with T1 < T2 < T3, terminals {γ, τ, σ, α, ν} big nonterminal in lhs and rhs: {T1, T2, T3} of ranks 1, 0, 0 3 corresponding rhs contexts with 1, 0, 0 variables

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 10

slide-20
SLIDE 20

Main notion

A → T1 σ C T2 T3 C x1 → σ C x1 C′ A C′ x1 → σ C x1 C′ A T1 x1 T2 T3 → γ T1 τ x1 σ T2 α ν T3 S → γ A C x1 → x1 C′ x1 → x1 T1 x1 T2 T3 → x1 α β

Derivation:

A ⇒ T1 σ C T2 T3 ⇒ T1 σ σ C T2 C′ A T3 ⇒ T1 σ σ T2 C′ A T3 ⇒ γ T1 τ σ σ σ T2 α C′ A ν T3 ⇒ γ τ σ σ σ α α C′ A ν β

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 11

slide-21
SLIDE 21

Main notion

A → T1 σ C T2 T3 C x1 → σ C x1 C′ A C′ x1 → σ C x1 C′ A T1 x1 T2 T3 → γ T1 τ x1 σ T2 α ν T3 S → γ A C x1 → x1 C′ x1 → x1 T1 x1 T2 T3 → x1 α β

Derivation:

A ⇒ T1 σ C T2 T3 ⇒ T1 σ σ C T2 C′ A T3 ⇒ T1 σ σ T2 C′ A T3 ⇒ γ T1 τ σ σ σ T2 α C′ A ν T3 ⇒ γ τ σ σ σ α α C′ A ν β

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 11

slide-22
SLIDE 22

Main notion

A → T1 σ C T2 T3 C x1 → σ C x1 C′ A C′ x1 → σ C x1 C′ A T1 x1 T2 T3 → γ T1 τ x1 σ T2 α ν T3 S → γ A C x1 → x1 C′ x1 → x1 T1 x1 T2 T3 → x1 α β

Derivation:

A ⇒ T1 σ C T2 T3 ⇒ T1 σ σ C T2 C′ A T3 ⇒ T1 σ σ T2 C′ A T3 ⇒ γ T1 τ σ σ σ T2 α C′ A ν T3 ⇒ γ τ σ σ σ α α C′ A ν β

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 11

slide-23
SLIDE 23

Main notion

A → T1 σ C T2 T3 C x1 → σ C x1 C′ A C′ x1 → σ C x1 C′ A T1 x1 T2 T3 → γ T1 τ x1 σ T2 α ν T3 S → γ A C x1 → x1 C′ x1 → x1 T1 x1 T2 T3 → x1 α β

Derivation:

A ⇒ T1 σ C T2 T3 ⇒ T1 σ σ C T2 C′ A T3 ⇒ T1 σ σ T2 C′ A T3 ⇒ γ T1 τ σ σ σ T2 α C′ A ν T3 ⇒ γ τ σ σ σ α α C′ A ν β

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 11

slide-24
SLIDE 24

Main notion

A → T1 σ C T2 T3 C x1 → σ C x1 C′ A C′ x1 → σ C x1 C′ A T1 x1 T2 T3 → γ T1 τ x1 σ T2 α ν T3 S → γ A C x1 → x1 C′ x1 → x1 T1 x1 T2 T3 → x1 α β

Derivation:

A ⇒ T1 σ C T2 T3 ⇒ T1 σ σ C T2 C′ A T3 ⇒ T1 σ σ T2 C′ A T3 ⇒ γ T1 τ σ σ σ T2 α C′ A ν T3 ⇒ γ τ σ σ σ α α C′ A ν β

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 11

slide-25
SLIDE 25

Main notion

A → T1 σ C T2 T3 C x1 → σ C x1 C′ A C′ x1 → σ C x1 C′ A T1 x1 T2 T3 → γ T1 τ x1 σ T2 α ν T3 S → γ A C x1 → x1 C′ x1 → x1 T1 x1 T2 T3 → x1 α β

Derivation:

A ⇒ T1 σ C T2 T3 ⇒ T1 σ σ C T2 C′ A T3 ⇒ T1 σ σ T2 C′ A T3 ⇒ γ T1 τ σ σ σ T2 α C′ A ν T3 ⇒ γ τ σ σ σ α α C′ A ν β

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 11

slide-26
SLIDE 26

Main notion

A → T1 σ C T2 T3 C x1 → σ C x1 C′ A C′ x1 → σ C x1 C′ A T1 x1 T2 T3 → γ T1 τ x1 σ T2 α ν T3 S → γ A C x1 → x1 C′ x1 → x1 T1 x1 T2 T3 → x1 α β

Derivation:

A ⇒ T1 σ C T2 T3 ⇒ T1 σ σ C T2 C′ A T3 ⇒ T1 σ σ T2 C′ A T3 ⇒ γ T1 τ σ σ σ T2 α C′ A ν T3 ⇒ γ τ σ σ σ α α C′ A ν β

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 11

slide-27
SLIDE 27

Main notion

A → T1 σ C T2 T3 C x1 → σ C x1 C′ A C′ x1 → σ C x1 C′ A T1 x1 T2 T3 → γ T1 τ x1 σ T2 α ν T3 S → γ A C x1 → x1 C′ x1 → x1 T1 x1 T2 T3 → x1 α β

Derivation:

A ⇒ T1 σ C T2 T3 ⇒ T1 σ σ C T2 C′ A T3 ⇒ T1 σ σ T2 C′ A T3 ⇒ γ T1 τ σ σ σ T2 α C′ A ν T3 ⇒ γ τ σ σ σ α α C′ A ν β

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 11

slide-28
SLIDE 28

Main notion

A → T1 σ C T2 T3 C x1 → σ C x1 C′ A C′ x1 → σ C x1 C′ A T1 x1 T2 T3 → γ T1 τ x1 σ T2 α ν T3 S → γ A C x1 → x1 C′ x1 → x1 T1 x1 T2 T3 → x1 α β

Derivation:

A ⇒ T1 σ C T2 T3 ⇒ T1 σ σ C T2 C′ A T3 ⇒ T1 σ σ T2 C′ A T3 ⇒ γ T1 τ σ σ σ T2 α C′ A ν T3 ⇒ γ τ σ σ σ α α C′ A ν β

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 11

slide-29
SLIDE 29

Main notion

A → T1 σ C T2 T3 C x1 → σ C x1 C′ A C′ x1 → σ C x1 C′ A T1 x1 T2 T3 → γ T1 τ x1 σ T2 α ν T3 S → γ A C x1 → x1 C′ x1 → x1 T1 x1 T2 T3 → x1 α β

Derivation:

A ⇒ T1 σ C T2 T3 ⇒ T1 σ σ C T2 C′ A T3 ⇒ T1 σ σ T2 C′ A T3 ⇒ γ T1 τ σ σ σ T2 α C′ A ν T3 ⇒ γ τ σ σ σ α α C′ A ν β

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 11

slide-30
SLIDE 30

Main notion

Definition

The tree language generated by the MCFTG G = (N, B, Σ, S, R) is L(G) = {t ∈ TΣ | S ⇒∗ t}

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 12

slide-31
SLIDE 31

Contents

1

Motivation

2

Main notion

3

Lexicalization

4

Expressive Power

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 13

slide-32
SLIDE 32

Lexicalization

Definition

Tree language L ⊆ TΣ has finite ambiguity if for every w ∈ (Σ(0))∗ {t ∈ L | yield(t) = w} is finite

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 14

slide-33
SLIDE 33

Lexicalization

Definition

Tree language L ⊆ TΣ has finite ambiguity if for every w ∈ (Σ(0))∗ {t ∈ L | yield(t) = w} is finite every string w has finitely many “parses” in L (i.e., finitely many tree representations that have w as yield) property of the language, not the grammar (not to be confused with the similarly named notions for grammars)

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 14

slide-34
SLIDE 34

Lexicalization

MCFTG G:

A → T1 σ C T2 T3 C x1 → σ C x1 C′ A C′ x1 → σ C x1 C′ A T1 x1 T2 T3 → γ T1 τ x1 σ T2 α ν T3 S → γ A C x1 → x1 C′ x1 → x1 T1 x1 T2 T3 → x1 α β

L(G) has finite ambiguity

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 15

slide-35
SLIDE 35

Lexicalization

Definition

MCFTG (N, B, Σ, S, R) is lexicalized if occΣ(0)(r) = ∅ for every A → r ∈ R

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 16

slide-36
SLIDE 36

Lexicalization

Definition

MCFTG (N, B, Σ, S, R) is lexicalized if occΣ(0)(r) = ∅ for every A → r ∈ R each rule contains an anchor (from Σ(0)) lexicalized MCFTGs generate tree languages with finite ambiguity

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 16

slide-37
SLIDE 37

Lexicalization

Theorem [MCFTGs strongly lexicalize themselves]

For every MCFTG G it is decidable whether L(G) has finite ambiguity

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 17

slide-38
SLIDE 38

Lexicalization

Theorem [MCFTGs strongly lexicalize themselves]

For every MCFTG G it is decidable whether L(G) has finite ambiguity and if so an equivalent lexicalized MCFTG can be constructed.

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 17

slide-39
SLIDE 39

Lexicalization

Theorem [MCFTGs strongly lexicalize themselves]

For every MCFTG G it is decidable whether L(G) has finite ambiguity and if so an equivalent lexicalized MCFTG can be constructed. multiplicity remains the same (multiplicity = maximal cardinality of big nonterminals) width increases at most by 1 (width = maximal rank of nonterminals) derivation trees are even related by means of linear deterministic top-down tree transducers with regular look-ahead

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 17

slide-40
SLIDE 40

Lexicalization

Lexicalization approach: normalize terminal rules to contain at least 2 anchors

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 18

slide-41
SLIDE 41

Lexicalization

Lexicalization approach: normalize terminal rules to contain at least 2 anchors normalize unary rules to contain at least 1 anchor

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 18

slide-42
SLIDE 42

Lexicalization

Lexicalization approach: normalize terminal rules to contain at least 2 anchors normalize unary rules to contain at least 1 anchor guess-and-verify strategy for remaining rules Derivation tree (of another MCFTG):

ρ1 ρ2 ρ4 ρ6 ρ0 ρ′

4

ρ4 ρ5 ρ8 ρ′

5

ρ8 ρ0 ρ9 ρ′

4

ρ6 ρ′

6

ρ0 ρ9 ρ9 ρ2 ρ6 ρ8 ρ7 ρ0 ρ8 β β α α α α

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 18

slide-43
SLIDE 43

Lexicalization

Extraction (verification) of lexical symbol Original rule: T1 x1 T2 T3 → x1 α β Constructed rule: T1 x1 T α

2

x1 T3 → x1 x1 β

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 19

slide-44
SLIDE 44

Lexicalization

Extraction (verification) of lexical symbol Original rule: T1 x1 T2 T3 → x1 α β Constructed rule: T1 x1 T α

2

x1 T3 → x1 x1 β Guess of lexical symbol (lexicalizing the rule) Original rule: Constructed rule: A → T1 σ C T2 T3 A → T1 σ C T α

2

α T3

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 19

slide-45
SLIDE 45

Contents

1

Motivation

2

Main notion

3

Lexicalization

4

Expressive Power

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 20

slide-46
SLIDE 46

Expressive Power

Definition

Context c ∈ CN∪Σ(Xk) with k variables is footed if k = 0 or there is a subtree of the form σ(x1, . . . , xk)

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 21

slide-47
SLIDE 47

Expressive Power

Definition

Context c ∈ CN∪Σ(Xk) with k variables is footed if k = 0 or there is a subtree of the form σ(x1, . . . , xk) Rule A → r is footed if all contexts in r are footed

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 21

slide-48
SLIDE 48

Expressive Power

Definition

Context c ∈ CN∪Σ(Xk) with k variables is footed if k = 0 or there is a subtree of the form σ(x1, . . . , xk) Rule A → r is footed if all contexts in r are footed MCFTG (N, B, Σ, S, R) is a multi-component tree adjoining grammar (MC-TAG) if all the rules of R are footed. Non-footed rule: Footed rule: A x1 x2 → σ γ σ x1 α x2 A x1 x2 → σ γ σ x1 x2 α

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 21

slide-49
SLIDE 49

Expressive Power

Theorem

For every MCFTG G there exists an equivalent MC-TAG G′ footed normal form for MCFTGs footed CFTGs as expressive as TAGs [Kepser, Rogers 2011]

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 22

slide-50
SLIDE 50

Expressive Power

Theorem

For every MCFTG G there exists an equivalent MC-TAG G′ footed normal form for MCFTGs footed CFTGs as expressive as TAGs [Kepser, Rogers 2011] result also true for strict MC-TAG (our notion of MC-TAG is essentially “non-strict MC-TAG”) if MCFTG G lexicalized, then so is MC-TAG G′

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 22

slide-51
SLIDE 51

Expressive Power

Proof idea: Decompose context into footed contexts:

Original rule: C x1 x2 → γ γ σ α x1 τ τ x2 Cα Cγ Cσ Cτ Constructed rule: Cγ x1 Cσ x1 x2 x3 Cα Cτ x1 ↓ γ γ x1 σ x1 x2 x3 α τ τ x1

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 23

slide-52
SLIDE 52

Expressive Power

Proof idea: Adjust “calls” appropriately:

Original rhs of rule: Constructed rhs of rule: γ A C β τ x1 γ A Cγ Cσ Cα β Cτ τ x1

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 24

slide-53
SLIDE 53

Expressive Power

Corollary [MC-TAGs strongly lexicalize themselves]

For every MC-TAG G it is decidable whether L(G) has finite ambiguity and if so an equivalent lexicalized MC-TAG can be constructed.

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 25

slide-54
SLIDE 54

Key points: MCFTGs and MC-TAGs equally expressive both allow strong lexicalization

September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 26

slide-55
SLIDE 55

Key points: MCFTGs and MC-TAGs equally expressive both allow strong lexicalization

Thank you for your attention.

Full version available on

arXiv September 21, 2017 MCFTG and MC-TAG

  • J. Engelfriet, A. Maletti

· 26