Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
Tree Adjoining Grammars
Syntax in LTAG
Laura Kallmeyer & Benjamin Burkhardt
HHU Düsseldorf
WS 2017/2018
1 / 23
Derivation trees: Example derivation
NP
Adam
VP
VP∗
Adv
always
S
VP
NP↓V
ate
NP↓ NP
NP∗
Det
the
NP
apple
3 / 23
Derivation trees
TAG derivations are uniquely described by derivation trees. The
derivation tree contains:
nodes for all elementary trees used in the derivation, and
edges for all adjunctions and substitutions performed
throughout the derivation, and
edge labels indicating the target node of the rewriting
operation.
Whenever an elementary tree γ rewrites the node at Gorn address pin the elementary tree γ′, there is an edge from γ′ to γ labeled with p.
Note that derivation trees are unordered trees.
Adjunction edges are sometimes depicted as dashed lines,
substitution edges as solid lines.
4 / 23
Derivation trees
For the node addresses of elementary trees, Gorn addresses are
used:
the root has address ε (or 0)
the ith daughter of the node with address p has address pi.
0
2
22
221
21
212211
1
1312
121
11
5 / 23
Derivation trees: example
Derived tree:
S
VP
VP
NP
NP
apple
Det
the
V
ate
Adv
always
NP
Adam
Derivation tree:
ate
adam apple
the
always
1 22
0
2
6 / 23
Linguistic analyses with LTAG
What is an elementary tree, and what is its shape?
elementary trees?⇐=
syntactic/semantic properties of
linguistic objects
⇒ Syntactic design principles from Frank (2002):
Lexicalization
Fundamental TAG Hypothesis (FTH)
Condition on Elementary Tree Minimality (CETM)
θ-Criterion for TAG
⇒ Semantic design principles [Abeillé & Rambow (2000)]
⇒ Design principle of economy
Further overviews are given in chapter 5 of Lichte (2015) (in
German) and chapter 4 of Kallmeyer (2010).
7 / 23
Syntactic design principles (1): Lexicalization
Each elementary tree has at least one non-empty lexical item, its
lexical anchor.
⇒ All widely used grammar formalisms support some kind of
lexicalization!
⇒ TAG → LTAG: Lexicalized Tree-Adjoining Grammar
[Schabes & Joshi (1990); Joshi & Schabes (1991)]
Recall: reasons for lexicalization
Formal properties: A �nite lexicalized grammar provides
�nitely many analyses for each string (�nitely ambiguous).
Linguistic properties: Syntactic properties of lexical items
can be accounted for more directly.
Parsing: The search space during parsing can be delimited
(grammar �ltering).
8 / 23
Syntactic design principles (2): FTH
Fundamental TAG Hypothesis (FTH); [Frank (2002)]
Every syntactic dependency is expressed locally within an
elementary tree.
“syntactic dependency”
valency/subcategorization
binding
�ller-gap constructions
. . .
“expressed within an elementary tree”
terminal leaf (i.e. lexical anchor)
nonterminal leaf (substitution node and footnode)
marking an inner node for obligatory adjunction
9 / 23
Syntactic design principles (2): FTH
Examples of ill-formed elementary trees:
S
VP
V
persuades
NP↓
S
VP
NP
ε
V
ate
NP↓
10 / 23
Complex primitives
Joshi (2004):
Complicate locally, simplify globally.
“[...] start with complex (more complicated) primitives, which capturedirectly some crucial linguistic properties and then introduce somegeneral operations for composing these complex structures (primitive orderived). What is the nature of these complex primitives? In theconventional approach the primitive structures (or rules) are kept assimple as possible. This has the consequence that information (e.g.,syntactic and semantic) about a lexical item (word) is distributed overmore than one primitive structure. Therefore, the informationassociated with a lexical item is not captured locally, i.e., within thedomain of a primitive structure.”[Joshi (2004)]
11 / 23
Syntactic design principles (3): CETM
Condition on Elementary Tree Minimality (CETM); ; [Frank (2002)]
The syntactic heads in an elementary tree and their projections must
form the extended projection of a single lexical head.
Examples of ill-formed elementary trees:
S
VP
VP
V
arrived
AP↓
NP↓
S
VP
S
VP
NP↓V
to eat
NP
ε
NP↓V
persuades
NP↓
12 / 23
Syntactic design principles (4): θ-Criterion for TAG
Thematic role (θ-role)
the semantic relationship of an argument with its predicate is
expressed through the assignment of a role by the predicate to the
argument. Di�erent theta-roles have di�erent labels, such as Agent,
Theme, Patient, Goal, Source, Experiencer etc.
example: Bart kicked the ball.kicked predicate
Bart Agent
ball Theme/Patient
The ball was kicked by Bart.kicked predicate
Bart Agent
ball Theme/Patient
13 / 23
Syntactic design principles (4): θ-Criterion for TAG
θ-Criterion (TAG version)
a. If H is the lexical head of an elementary tree T, H assigns all of
its θ-roles in T.
b. If A is a frontier non-terminal of elementary tree T, A must be
assigned a θ-role in T.
[Frank (2002)]
=⇒ Valency/subcategorization is expressed only with nonterminal
leaves!
S
VP
V
sleeps
NP↓
S
VP
S∗
V
try
NP↓
14 / 23
Further design principles
Semantic design principles
Predicate-argument co-occurrence:
Each elementary tree associated with a predicate contains a
non-terminal leaf for each of its arguments.
Semantic anchoring:
Elementary trees are not semantically void (to, that.)
Compositional principle:
An elementary tree corresponds to a single semantic unit.
Design principle of economy
The elementary trees are shaped in such a way, that the size of the
elementary trees and the size of the grammar is minimal.
15 / 23
Modi�cation and functional elements
How to insert modi�ers (e.g. easily) and functional elements(complementizers, determiners, do-auxiliaries, ...)?
either as co-anchor in the elementary tree of the lexical item
they are associated with
S
S
VP
sleeps
NP↓
Comp
that
S
VP
AP
easily
VP
sleeps
NP↓
or by separate auxiliary trees (e.g., XTAG grammar)
S
S∗
Comp
that
VP
AP
easily
VP∗
⇒ Footnodes/Adjunctions indicate both complementation and
modi�cation.
⇒ Enhancement of the CETM: [see Abeillé & Rambow (2000)]
16 / 23
Sample derivations: NP and PP complements
(1) Adam gave Eve the apple.
Elementary trees:
S
VP
NP↓NP↓V
gave
NP↓
NP
N
Adam
NP
N
Eve
NP
N
apple
NP
NP∗
Det
the
Derivation tree:gave
adam eve apple
the
1 22 23
0
17 / 23
Sample derivations: NP and PP complements
(2) Adam gave the apple to Eve.
Elementary trees:
S
VP
PP
NP↓P
to
NP↓V
gave
NP↓
NP
N
Adam
NP
N
Eve
NP
N
apple
NP
NP∗
Det
the
Derivation tree:gave
adam apple eve
the
1 22 232
0
18 / 23
Sample derivations: Sentential complements
(3) Adam hopes that Eve comes.
Elementary trees:
S
VP
S∗
V
hopes
NP↓
S
VP
V
comes
NP↓
NP
N
Adam
NP
N
Eve
S
S∗
Comp
that
Derivation tree: comes
adam
eve
hopes
that
0 1
0
1
19 / 23
Sample derivations: long-distance dependency
(4) Whati did Adam say (that) Eve ate _i?
NP
N
what
S
S
VP
S∗
V
say
NP↓
Aux
did
NP
N
Adam
NP
N
Eve
S
S
S
VP
NP
εi
V
ate
NP↓
COMP
ε
NPi↓
Derivation tree: ate
whateve did_say
adam
221 1 2
21
20 / 23
Sample derivations: Modi�ers
(5) The good student participated in every course during the semester.
Elementary trees:
S
VP
PP
NP↓P
in
V
part.
NP↓
NP
N
student
NP
N
course
NP
N
semester
NP
NP∗
Det
the
NP
NP∗
Det
every
N
N∗
AP
good
VP
PP
NP↓P
during
VP∗
21 / 23
Sample derivations: Modi�ers
Derivation tree:part_in
stud course during
the good every semester
the
1 222 2
0 1 0 22
0
22 / 23
References
Abeillé, Anne & Owen Rambow. 2000. Tree adjoining grammar: An overview. In Anne
Abeillé & Owen Rambow (eds.), Tree Adjoining Grammars: Formalisms, linguistic
analyses and processing (CSLI Lecture Notes 107), 1–68. Stanford, CA: CSLI Publications.
Frank, Robert. 2002. Phrase structure composition and syntactic dependencies. Cambridge,
Mass: MIT Press.
Joshi, Aravind K. 2004. Starting with complex primitives pays o�: complicate locally, simplify
globally. Cognitive Science 28. 637–668.
Joshi, Aravind K. & Yves Schabes. 1991. Tree-Adjoining Grammars and lexicalized grammars.
Tech. Rep. MS-CIS-91-22 Department of Computer and Information Science, University of
Pennsylvania. http://repository.upenn.edu/cis_reports/445/.
Kallmeyer, Laura. 2010. Parsing beyond context-free grammars Cognitive Technologies.
Heidelberg: Springer.
Lichte, Timm. 2015. Syntax und Valenz. Zur Modellierung kohärenter und elliptischer
Strukturen mit Baumadjunktionsgrammatiken (Empirically Oriented Theoretical
Morphology and Syntax 1). Berlin: Language Science Press.
Schabes, Yves & Aravind K. Joshi. 1990. Parsing with lexicalized tree adjoining grammar.
Tech. Rep. MS-CIS-90-11 Department of Computer and Information Science, University of
Pennsylvania. http://repository.upenn.edu/cis_reports/542/.