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

SLIDE 1

Grammar Implementation with Lexicalized Tree Adjoining Grammars and Frame Semantics

Grammar implementation with XMG: Frames Laura Kallmeyer, Timm Lichte, Rainer Osswald & Simon Petitjean

University of Düsseldorf

DGfS Fall School, September 21, 2017

SFB 991

SLIDE 2

Yesterday: Syntax

Implementation of syntactic trees S NP↓ VP V⋄ =⇒

1

class alphanx0v

2

import VerbProjection[]

3

declare ?Subj

4

{

5

?Subj = Subject[];

6

?Subj.?VP = ?VP

7

}

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

SLIDE 3

Today: Frames!

Frame theories come with two components: atribute-value descriptions atribute-value constraints How to implement both with XMG?

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

SLIDE 4

Atribute-value descriptions (recap.)

Vocabulary / Signature Atr atributes (= dyadic functional relation symbols) Rel (proper) relation symbols Type type symbols (= monadic predicates) Nname node names (“nominals”) Nvar node variables

} Nlabel

node labels Primitive atribute-value descriptions (pAVDesc) t | p : t | p q | [p1, . . . , pn] : r | p k

(t ∈ Type, r ∈ Rel, p, q, pi ∈ Atr∗, k ∈ Nlabel)

Semantics

P∶t t

P [P [t] ]

P ≐ Q

P Q ⎡ ⎢ ⎢ ⎢ ⎢ ⎣

P

1

Q

1

⎤ ⎥ ⎥ ⎥ ⎥ ⎦

[P, Q]∶r

P Q

r

⎡ ⎢ ⎢ ⎢ ⎢ ⎣

P

1

Q

2

⎤ ⎥ ⎥ ⎥ ⎥ ⎦ r ( 1 , 2 ) P ≜ k

k

P [P k [ ] ]

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

SLIDE 5

Atribute-value formulas (recap.)

Primitive atribute-value formulas (pAVForm) k · p : t | k · p l · q | k1 · p1, . . . , kn · pn : r

(t ∈ Type, r ∈ Rel, p, q, pi ∈ Atr∗, k, l, ki ∈ Nlabel)

Semantics

k ⋅ P∶t

k

t

P k [P [t] ]

k ⋅ P ≜ l ⋅ Q

k l

P Q k [P

1 ]

l [Q

1 ]

⟨k ⋅ P,l ⋅ Q⟩∶r

k l

P Q

r

k [P

1 ]

l [Q

2 ]

r ( 1 , 2 )

Formal definitions (fairly standard) Set/universe of “nodes” V Interpretation function I : Atr → [V ⇀ V], Type → ℘(V), Rel →

n ℘(V n), Nname → V

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

SLIDE 6

Atribute-value formulas in XMG

p : t

t p

p
t
[p: t]

p q

p

1

q

1

[p: ?X1,

q: ?X1]

[p, q] : r p q r

p

1

q

2

r ( 1 , 2 )

NOT SUPPORTED YET

p k

k

p

p

k [ ]

[p: ?K[] ]

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

SLIDE 7

Atribute-value formulas in XMG

k · p : t

k

t p k

p
t
?K [p: t];

k · p l · q

k l

p q k

p

1

l
q

1

?K [p: ?X1];

?L [q: ?X1]

k · p, l · q : r

k l

p q r k

p

1

l
q

2

r ( 1 , 2 )

NOT SUPPORTED YET

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

SLIDE 8

Atribute-value formulas in XMG: Example

                    causation actor

1

theme

2

cause        activity actor

1

theme

2

       effect

4

mover

2

goal

3



                  

1

<frame>{

2

?0[causation,

3

actor:?1,

4

theme:?2,

5

cause:[activity,

6

actor:?1,

7

theme:?2],

8

effect:?4[mover:?2,

9

goal:?3]

10

]}

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

SLIDE 9

Atribute-value constraints (recap.)

Constraints (general format) ∀ϕ, ϕ ∈ AVDesc V, I, g ∀ϕ iff V, I, g , v ϕ for every v ∈ V Notation: ϕ ⇛ ψ for ∀(ϕ → ψ) Horn constraints: ϕ1 ∧ . . . ∧ ϕn ⇛ ψ

(ϕi ∈ pAVDesc ∪ {⊤}, ψ ∈ pAVDesc ∪ {⊥})

Examples activity ⇛ event

(every activity is an event)

causation ∧ activity ⇛ ⊥

(there is nothing which is both a causation and an activity)

agent : ⊤ ⇛ agent actor

(every agent is also an actor)

activity ⇛ actor : ⊤

(every activity has an actor)

activity ∧ motion ⇛ actor mover ...

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

SLIDE 10

Atribute-value constraints in XMG

activity ⇛ event

activity -> event

causation ∧ activity ⇛ ⊥

causation activity -> -

agent : ⊤ ⇛ agent actor

agent:+ -> agent=actor

activity ⇛ actor : ⊤

activity -> actor:+

activity ∧ motion ⇛ actor mover

activity motion-> actor=mover

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

SLIDE 11

Atribute-value constraints in XMG: Examples

1

frame-constraints = {

2

activity -> event, activity -> [actor:+],

3

motion -> event, motion -> [mover:+],

4

causation -> event, causation -> [cause:+,effect:+],

5

locomotion -> activity motion}

What is the graphical represention of this (“type hierarchy”)?

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

SLIDE 12

Atribute-value constraints in XMG: Examples

1

frame-constraints = {

2

activity -> event, activity -> [actor:+],

3

motion -> event, motion -> [mover:+],

4

causation -> event, causation -> [cause:+,effect:+],

5

locomotion -> activity motion}

What is the graphical represention of this (“type hierarchy”)?

1

frame-type-hierarchy = {

2

[event, [activity, actor:+, [locomotion]],

3

[motion, mover:+, [locomotion]],

4

[causation, cause:+, effect:+]]}

NOT YET SUPPORTED

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 12 11

SLIDE 13

Atribute-value constraints (recap.)

Graphical presentation of constraints

event activity actor ∶ ⊺ motion mover ∶ ⊺ causation cause ∶ ⊺ ∧ effect ∶ ⊺ activity ∧ motion actor ≐ mover translocation path ∶ ⊺

nset-causation

cause ∶ punctual-event extended- causation locomotion bounded-translocation goal ∶ ⊺ bounded-locomotion

Caveat: Reading convention required !

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 13 12

SLIDE 14

Implementation exercise with frames

1 implement the large type hierarchy 2 implement two frame descriptions e

bounded-translocation

goal

x



              causation actor

1

theme

2

cause        activity actor

1

theme

2

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

3 implement the unfication of these two frames ( e = 0 )

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 14 13

SLIDE 15

Case study: dative alternation (recap.)

Sketch

[→ Kallmeyer/Osswald 2013]

(1) a. John sent Mary the book. (double object construction) b. John sent the book to Mary. (prepositional object construction) a) S NP[i=x] VP[e=e] V◇[e=e] NP[i=z] NP[i=y]

e ⎡ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ causation cause [activity actor x] effect ⎡ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ change-of-possession theme y recipient z ⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ ⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦

b) S NP[i=x] VP[e=e] VP[e=e] PP[prep = to,i = z,e = e′] V◇[e=e] NP[i=y]

e ⎡ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ causation cause [activity actor x] effect e′ ⎡ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ bounded-translocation mover y goal z ⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ ⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 15 14

SLIDE 16

Implementation exercise with frames

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 16 15

SLIDE 17

Implementation exercise with frames

The trick: sharing of variables across dimensions!

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 17 15

SLIDE 18

Further applications

Morpho-semantic interface: modelling verbal prefixation in Russian (Zinova) modelling derivational morphology in English (Andreou & Petitjean) modelling root-patern morphology in Arabic (Petitjean, Samih & Lichte)

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 18 16

SLIDE 19

Tomorrow

Mon: introduction to grammar engineering and XMG Tue: implementing syntax with XMG Wed: implementing semantics with XMG Thu: parsing implemented grammars with TuLiPA Fri: conclusion

Kallmeyer, Lichte, Osswald & Petitjean (HHU Düsseldorf) 19 17

SLIDE 20

[1] Petitjean, Simon, Younes Samih & Timm Lichte. 2015. Une métagrammaire de l’interface morpho-sémantique dans les verbes en arabe. In Actes de la 22e conférence sur le Traitement Automatique des Langues Naturelles, 473–479. Caen, France.

http://www.atala.org/taln_archives/TALN/TALN-2015/taln-2015-court-024.

[2] Zinova, Yulia. 2016. Russian verbal prefixation: a frame semantic analysis. Düsseldorf, Germany: Heinrich-Heine-Universität Düsseldorf Dissertation.

https://user.phil-fak.uni-duesseldorf.de/~zinova/Thesis.pdf.