8/3/2019 Rambow Slides Leysin Tag
1/142
Tree Adjoining Grammar:
Introduction
Owen RambowCCLS, Columbia University
2008-03-12
8/3/2019 Rambow Slides Leysin Tag
2/142
Overview
Terminology (ppt)
What is Tree Adjoining Grammar?
Some Syntactic Analyses
TAG and Syntactic Theory
Problem: German
8/3/2019 Rambow Slides Leysin Tag
3/142
Three Main Points About Tree Adjoining
Grammar (TAG):
TAG is a constrained mathematical
formalism
TAG supports the development of lexicalized
grammars for natural languages
TAG is not a linguistic theory
8/3/2019 Rambow Slides Leysin Tag
4/142
Constrained mathematical formalisms
Mathematical device for specifying sets of
structures
Constrained: mathematical device cannot
specify all possible sets
8/3/2019 Rambow Slides Leysin Tag
5/142
Why constrained mathematical formalisms?
Linguistically appealing because scope oflinguistic theory is restricted
Computationally appealing because of
efficient processing models
8/3/2019 Rambow Slides Leysin Tag
6/142
Lexicalization: central role of lexicon in syntax
Many syntactic phenomena are idiosyncraticto specific lexical items or classes of lexical
items
Natural language processing: use of corpora
pervasive, corpora (typically) consist of words
8/3/2019 Rambow Slides Leysin Tag
7/142
Reminder: Lexical Idiosyncracies in Syntax
(Subcategorization)
(1) a. He told a secret to Mary
b. He told Mary a secret
c. He divulged a secret to Mary
d. *He divulged Mary a secret
8/3/2019 Rambow Slides Leysin Tag
8/142
Lexicalization
A formalism is lexicalized if every elementary
structure contains at least one terminal symbol (=
lexical item)
8/3/2019 Rambow Slides Leysin Tag
9/142
Which Constrained Mathematical Formalism?
Context-Free Grammar (CFG) (Chomsky
1957), used all over NLP, BUT
8/3/2019 Rambow Slides Leysin Tag
10/142
Which Constrained Mathematical Formalism?
Context-Free Grammar (CFG) (Chomsky
1957), used all over NLP, BUT Wrong descriptive level (issue of lexicalization)
Not powerful enough formally (Shieber 1985
on Swiss German)
8/3/2019 Rambow Slides Leysin Tag
11/142
Which Constrained Mathematical Formalism?
Context-Free Grammar (CFG) (Chomsky
1957), used all over NLP, BUT Wrong descriptive level (issue of lexicalization)
Not powerful enough formally (Shieber 1985
on Swiss German)
Most current linguistic formalisms go beyond
CFG (TAG, HPSG, LFG)
8/3/2019 Rambow Slides Leysin Tag
12/142
Which Constrained Mathematical Formalism?
Context-Free Grammar (CFG) (Chomsky
1957), used all over NLP, BUT Wrong descriptive level (issue of lexicalization)
Not powerful enough formally (Shieber 1985
on Swiss German)
Most current linguistic formalisms go beyond
CFG (TAG, HPSG, LFG)
Most of these lose constraint on formal power
EXCEPT TAG
8/3/2019 Rambow Slides Leysin Tag
13/142
Lets Start with a CFG
(2) a. S NP VP
b. VP really VP
c. VP V NP
d. V likes
e. NP John
f. NP Lyn
Elementary structures of this grammar: the rules
8/3/2019 Rambow Slides Leysin Tag
14/142
Derivation in a CFG
Rule 1: S NP VP
Rule 2: VP really VP
Rule 3: VP V NP
Rule 4: V likes
Rule 5: NP John
Rule 6: NP Lyn
8/3/2019 Rambow Slides Leysin Tag
15/142
Derivation in a CFG
Rule 1: S NP VP
Rule 2: VP really VP
Rule 3: VP V NP
Rule 4: V likes
Rule 5: NP John
Rule 6: NP Lyn
S = NP VP (Rule 1, then Rule 5)
8/3/2019 Rambow Slides Leysin Tag
16/142
Derivation in a CFG
Rule 1: S NP VP
Rule 2: VP really VP
Rule 3: VP V NP
Rule 4: V likes
Rule 5: NP John
Rule 6: NP Lyn
S = NP VP (Rule 1, then Rule 5)
= John VP (next Rule 2)
8/3/2019 Rambow Slides Leysin Tag
17/142
Derivation in a CFG
Rule 1: S NP VP
Rule 2: VP really VP
Rule 3: VP V NP
Rule 4: V likes
Rule 5: NP John
Rule 6: NP Lyn
S = NP VP (Rule 1, then Rule 5)
= John VP (next Rule 2)
=
John really VP (next Rule 3)
8/3/2019 Rambow Slides Leysin Tag
18/142
Derivation in a CFG
Rule 1: S NP VP
Rule 2: VP really VP
Rule 3: VP V NP
Rule 4: V likes
Rule 5: NP John
Rule 6: NP Lyn
S = NP VP (Rule 1, then Rule 5)
= John VP (next Rule 2)
=
John really VP (next Rule 3)= John really V NP (next Rule 6)
8/3/2019 Rambow Slides Leysin Tag
19/142
Derivation in a CFG
Rule 1: S NP VP
Rule 2: VP really VP
Rule 3: VP V NP
Rule 4: V likes
Rule 5: NP John
Rule 6: NP Lyn
S = NP VP (Rule 1, then Rule 5)
= John VP (next Rule 2)
=
John really VP (next Rule 3)= John really V NP (next Rule 6)
= John really V Lyn (next Rule 4)
8/3/2019 Rambow Slides Leysin Tag
20/142
Derivation in a CFG
Rule 1: S NP VP
Rule 2: VP really VP
Rule 3: VP V NP
Rule 4: V likes
Rule 5: NP John
Rule 6: NP Lyn
S = NP VP (Rule 1, then Rule 5)
= John VP (next Rule 2)
= John really VP (next Rule 3)
= John really V NP (next Rule 6)
= John really V Lyn (next Rule 4)
8/3/2019 Rambow Slides Leysin Tag
21/142
= John really likes Lyn
8/3/2019 Rambow Slides Leysin Tag
22/142
CFG is a string-rewriting system
Elementary structure: context-free rewrite rule
Operation: rewrite
Record of derivation is a phrase-structure tree,
called derivation tree
8/3/2019 Rambow Slides Leysin Tag
23/142
Derivation Tree:
really
VP
John
S
NP
V
VP
likes Lyn
NP
S=
NP VP (Rule 1)= John VP (Rule 5)
= John really VP (Rule 2)
= John really V NP (Rule 3)
= John really V Lyn (Rule 6)
= John really likes Lyn (Rule 4)
8/3/2019 Rambow Slides Leysin Tag
24/142
Lexicalization (reminder)
A formalism is lexicalized if every elementary
structure contains at least one terminal symbol (=
lexical item)
Rule 1: S NP VP
Rule 2: VP really VPRule 3: VP V NP
Rule 4: V likes
Rule 5: NP John
Rule 6: NP Lyn
Is this grammar lexicalized?
8/3/2019 Rambow Slides Leysin Tag
25/142
Lexicalization (reminder)
A formalism is lexicalized if every elementary
structure contains at least one terminal symbol (=lexical item)
Rule 1: S NP VP
Rule 2: VP really VP
Rule 3: VP V NP
Rule 4: V likes
Rule 5: NP John
Rule 6: NP Lyn
Is this grammar lexicalized?
No
8/3/2019 Rambow Slides Leysin Tag
26/142
Lexicalizing a CFG (ctd)
Can we lexicallize this CFG?
8/3/2019 Rambow Slides Leysin Tag
27/142
Lexicalizing a CFG (ctd)
Can we lexicallize this CFG?
Greibach Normal Form: changes rules,
linguistically unappealing
8/3/2019 Rambow Slides Leysin Tag
28/142
Lexicalizing a CFG (ctd)
Can we lexicallize this CFG?
Greibach Normal Form: changes rules,
linguistically unappealing
S NP likes NP
8/3/2019 Rambow Slides Leysin Tag
29/142
Lexicalizing a CFG (ctd)
Can we lexicallize this CFG?
Greibach Normal Form: changes rules,
linguistically unappealing
S NP likes NP
What about really?
Go from string rewriting to tree rewriting!
8/3/2019 Rambow Slides Leysin Tag
30/142
Why not make trees the elementary structures of
a grammar?
NP
V
VP
S
likes
NP
NP
John Lyn
NP
1
2 3
Need operations to combine these trees!
8/3/2019 Rambow Slides Leysin Tag
31/142
Substitution
A
=>
S SA
A
8/3/2019 Rambow Slides Leysin Tag
32/142
Substitution
32
41
NP
V
VP
S
likes
NP
Lyn
NPNP
John
VP
NP
Lynlikes
VJohn
S
NP
8/3/2019 Rambow Slides Leysin Tag
33/142
New formalism: Tree Substitution Grammar
(TSG)
Elementary structures: phrase-structure trees
Operations: substitution
Comparison with CFG
8/3/2019 Rambow Slides Leysin Tag
34/142
Comparison with CFG
Formal power?
Comparison with CFG
8/3/2019 Rambow Slides Leysin Tag
35/142
Comparison with CFG
Formal power?
Equivalent to CFG
Linguistic expressive power (what range of
theories can we formulate)?
8/3/2019 Rambow Slides Leysin Tag
36/142
Comparison with CFG
Formal power?
Equivalent to CFG
Linguistic expressive power (what range of
theories can we formulate)?
Greater than CFG; examples: agreement,
subcategorization; extended domain oflocality
8/3/2019 Rambow Slides Leysin Tag
37/142
Agreement
[number=pl]NP
NP
like [number=pl]
S
VP
V
1
[number=sg, person=3]NP
NP
likes [number=sg,person=3]
S
VP
V
1
8/3/2019 Rambow Slides Leysin Tag
38/142
Subcategorization
V
VPNP0
S
tell
V
VP
tell
NP1NP2
NP0
to
NP1 PP
P NP2
S
Comparison with CFG (ctd)
8/3/2019 Rambow Slides Leysin Tag
39/142
Comparison with CFG (ctd)
Linguistic adequacy issue:
G ( )
8/3/2019 Rambow Slides Leysin Tag
40/142
Comparison with CFG (ctd)
Linguistic adequacy issue:
But what about really?
Adj ti
8/3/2019 Rambow Slides Leysin Tag
41/142
Adjunction
=>A
S SA
A*
A
A
Adjunction
8/3/2019 Rambow Slides Leysin Tag
42/142
Adjunction
1 54
John
S
NPNP
V
VP
S
John
likes Lyn
NP
V
likes Lyn
NP
really VP*
VP
VP
VPreally
N f m lism Tree Adjoining Grammar (TAG)
8/3/2019 Rambow Slides Leysin Tag
43/142
New formalism: Tree Adjoining Grammar (TAG)
Elementary structures: phrase-structure trees
Operations: substitution, adjunction
Note: trees that can be adjoined are called
auxiliary trees, trees that can be substituted
initial trees
Comparison with CFG and TSG
8/3/2019 Rambow Slides Leysin Tag
44/142
p
Formal power?
Comparison with CFG and TSG
8/3/2019 Rambow Slides Leysin Tag
45/142
Formal power?
Greater than CFG and TSG
Linguistic expressive power (what range of
theories can we formulate)?
Comparison with CFG and TSG
8/3/2019 Rambow Slides Leysin Tag
46/142
Comparison with CFG and TSG
Formal power?
Greater than CFG and TSG
Linguistic expressive power (what range of
theories can we formulate)?
Greater than CFG, but same as TSG;
examples: agreement, subcategorization;extended domain of locality
Derivation of our sentence
8/3/2019 Rambow Slides Leysin Tag
47/142
Derivation of our sentence
NP
V
VP
S
likes
NP
really VP
VP
John
NP
Lyn
NP
really VP
VP
NP
Lynlikes
V
NP
S
John
Derivation structure?
8/3/2019 Rambow Slides Leysin Tag
48/142
Derivation structure?
NP
V
VP
S
likes
NP
really VP
VP
John
NP
Lyn
NP
really VP
VP
NP
Lynlikes
V
NP
S
John
like
Lyn reallyJohn
2
John
ADJUNCT
OBJ
SUBJ
2.2
1
like
Lyn really
A note on elementary structures and rewriting
8/3/2019 Rambow Slides Leysin Tag
49/142
CFG: elementary structures are rewrite rulesS NP VP
TAG: elementary structures are PS trees
A note on elementary structures and rewriting
8/3/2019 Rambow Slides Leysin Tag
50/142
A note on elementary structures and rewriting
CFG: elementary structures are rewrite rulesS NP VP
TAG: elementary structures are PS treesTAG trees are really tree rewrite rules:
VP
VP
NPV
really
NP
S
likes
really VP
VP
VPS
likes
NP
S
VP
V
NP
Summary of comparison CFG, TSG, TAG
8/3/2019 Rambow Slides Leysin Tag
51/142
Summary of comparison CFG, TSG, TAG
CFG TSG TAG
Elementary Struc. string PS tree PS tree
Derived Structure string PS tree PS tree
Derivation Struc. PS tree Dep tree Dep tree
Can lexicalize Engl? no no yes
Overview
8/3/2019 Rambow Slides Leysin Tag
52/142
Terminology (ppt)
What is Tree Adjoining Grammar?
Some Syntactic Analyses
TAG and Syntactic Theory
Problem: German
Why TAGs are useful in Linguistics
8/3/2019 Rambow Slides Leysin Tag
53/142
y g
Constrained mathematical formalism
Extended domain of locality
Extended Domain of Locality
8/3/2019 Rambow Slides Leysin Tag
54/142
Elementary structure is a tree, not a single layerof a tree
CFG: VP V NP[obj]TAG:
S
8/3/2019 Rambow Slides Leysin Tag
55/142
S
rr
rrr
NP
[case:nom]
[agr: 1 ]
VP
rrr
V
[agr: 1 ]
/eat/
NP
[case:obj]
We can use Extended Domain of Locality for:
8/3/2019 Rambow Slides Leysin Tag
56/142
Case requirements imposed on complements(example: German objects)
Agreement through co-referential feature
structures between different parts of the tree
(e.g., subject-verb agreement or object
clitic-participle agreement in F) Strongly governed prepositions
Subcatgorization frame of lexical item (basicsubcat and variations)
Syntax of lexical item (basic and variation)
Idioms
8/3/2019 Rambow Slides Leysin Tag
57/142
Note: lexical item = single word or word+prepor idiom
Strongly governed preposition
8/3/2019 Rambow Slides Leysin Tag
58/142
S
8/3/2019 Rambow Slides Leysin Tag
59/142
S
rr
rr
rrr
NP
[nom,agr: 1 ]
VP
rr
rr
r
V[agr: 1 ]
/discriminate/
PP
rrr
P
against
NP
[obj]
Lois discrminates against linguists
Subcategorization frame:
8/3/2019 Rambow Slides Leysin Tag
60/142
S
rr
r
NP VP
rr
V
/eat/
NP
S
8/3/2019 Rambow Slides Leysin Tag
61/142
rr
rrr
NP VP
rr
rrr
V
/give/
NP PP
rr
P
to
NP
Substitution nodes (or foot nodes) designate
required arguments
Note: adjuncts handled through adjunction
8/3/2019 Rambow Slides Leysin Tag
62/142
S
rr
r
NP VP
rr
V
/eat/
NP
VP
8/3/2019 Rambow Slides Leysin Tag
63/142
rr
AdvP
often
VP*
Mary often eats beans
Adjuncts are not required, but the adjunct can
only modify certain trees
English Passive:
8/3/2019 Rambow Slides Leysin Tag
64/142
S
rr
rr
rr
NP[nom,role:theme
agr: 1 ]
VP
rr
r
Aux
[agr: 1 ]
/be/
VP
V
[past-part]
eaten
Apples were eaten
Fancy English syntax (topicalization):
8/3/2019 Rambow Slides Leysin Tag
65/142
S
rr
rrr
NP[obj]
S
rr
rr
NP
[nom,agr: 1 ]
VP
V
[agr: 1 ]
/eat/
Apples John eats
Note: trace optional
8/3/2019 Rambow Slides Leysin Tag
66/142
Idiom
8/3/2019 Rambow Slides Leysin Tag
67/142
S
rr
rr
rr
NP
[nom,agr: 1 ]
VP
rr
rr
V[agr: 1 ]
/kick/
NP
rr
the N
bucket
Bernice kicks the bucket (= Bernice dies)
Examples of Derivations
8/3/2019 Rambow Slides Leysin Tag
68/142
(1) Subcategroization for clauses
(2) Long-distance topicalization
(3) Control verbs
(4) Raising
(5) Extraction from picture-NPs
Subcategorization for clauses
8/3/2019 Rambow Slides Leysin Tag
69/142
Approach 1: Substitution
S
rr
rrr
NP
[nom,agr: 1 ]
VP
r
rr
V
[agr: 1 ]
/think/
S
S
8/3/2019 Rambow Slides Leysin Tag
70/142
rr
rrr
r
COMP
that
S
rr
rrr
NP
[nom,agr: 1 ]
VP
rrr
V
[agr: 1 ]
/like/
NP
S
r
8/3/2019 Rambow Slides Leysin Tag
71/142
rrr
rr
NP
[nom,agr: 1 ]
VP
rr
rr
V
[agr: 1 ]
/think/
S
rr
rr
COMP
that
S
rrrr
NP
[nom,agr: 2 ]
VP
rr
V
[agr: 2 ]
/like/
NP
Ivan thinks that Natasha likes pears
Subcategorization for clauses
8/3/2019 Rambow Slides Leysin Tag
72/142
Approach 2: AdjunctionS
S
rr
rr
VP
rrr
V
/think/
S*
S
r
rrr
COMP
that
S
rr
r
NP VP
rr
V
/like/
NP
S
rr
8/3/2019 Rambow Slides Leysin Tag
73/142
rrr
rr
NP
[nom,agr: 1 ]
VP
rr
rr
V
[agr: 1 ]
/think/
S
rr
rr
COMP
that
S
rrrr
NP
[nom,agr: 2 ]
VP
rr
V
[agr: 2 ]
/like/
NP
Ivan thinks that Natasha likes pears
Subcategorization for clauses
8/3/2019 Rambow Slides Leysin Tag
74/142
Approach 2: Adjunction (ctd)
Exactly the same derived tree (except for S root
node of matrix clause)
Advantage of adjunction: extraction from
embedded clause (coming up!)
Long-distance topicalization and wh-movement
8/3/2019 Rambow Slides Leysin Tag
75/142
S
rr
rr
NP VP
rrr
V
/think/
S*
S
rr
r
NP S
rr
NP VP
V
/eat/
S
rrr
8/3/2019 Rambow Slides Leysin Tag
76/142
r
NP S
rrr
NP VP
rrr
V
/think/
S
rr
NP VP
V
/eat/
Cherries Belinda thinks Jo eats
S
S
S
rrr
8/3/2019 Rambow Slides Leysin Tag
77/142
S
rrr
NP VP
rr
V
/think/
S*
r
NP S
rrr
COMP
that
S
rr
NP VP
V
/eat/
S
rr
r
8/3/2019 Rambow Slides Leysin Tag
78/142
r
NP S
S
rr
r
NP VP
rr
r
V
/think/
S
rrr
COMP
that
S
rr
NP VP
V
/eat/
Cherries Belinda thinks that Jo eats
Long-distance wh-movement (ctd)
8/3/2019 Rambow Slides Leysin Tag
79/142
S
rr
r
8/3/2019 Rambow Slides Leysin Tag
80/142
NP S
S
rr
rr
NP VP
rr
rr
V
/think/
S
rr
rr
COMP
that
S
rr
rr
NP VP
rr
r
V
/suspect/
S
rrr
COMP
that
S
rr
NP VP
Cherries Belinda thinks that Mark suspects that
Jo eats
8/3/2019 Rambow Slides Leysin Tag
81/142
Control verbs
8/3/2019 Rambow Slides Leysin Tag
82/142
S
rr
rrr
NP
[ref: 1 ]
VP
rr
rr
V
wishes
S*
[mood:inf-to]
[control: 1 ]
S
8/3/2019 Rambow Slides Leysin Tag
83/142
[mood:inf-to]
[control: 1 ]
rr
rr
NP
PRO
[index: 1 ]
VP
rrr
V
to sing
NP
Emily wishes to sing songs
Control verbs
8/3/2019 Rambow Slides Leysin Tag
84/142
S
rr
rr
rr
NP VP
rr
rr
rr
r
V
asks
NP
[ref: 1 ]
S*
[mood:inf-to]
[control: 1 ]
S
8/3/2019 Rambow Slides Leysin Tag
85/142
[mood:inf-to]
[control: 1 ]
rr
rr
NP
PRO
[index: 1 ]
VP
rrr
V
to sing
NP
Emily asks Alfred to sing songs
Raising verbs
VP S
8/3/2019 Rambow Slides Leysin Tag
86/142
[agr: 1 ]
rrr
V
[agr: 1 ]
/seem/
VP*
rr
r
NP
[agr: 1 ]
[agr: 1 ]
VP
[agr:-]
V
to sing
S
r
rr
8/3/2019 Rambow Slides Leysin Tag
87/142
rr
NP
[agr: 1 ]
VP
[agr: 1 ]
rrr
V
[agr: 1 ]
/seem/
VP
[agr:-]
V
to sing
Emily seems to sing
Raising verbs (ctd)
S
r
8/3/2019 Rambow Slides Leysin Tag
88/142
rr
rr
NP
[agr: 1 ]
VP
[agr: 1 ]
rrr
V
[agr: 1 ]
/seem/
VP
rr
V
to appear
VP
V
to sing
Emily seems to appear to sing
Picture-NP extraction
S
8/3/2019 Rambow Slides Leysin Tag
89/142
S
rr
rr
NP VP
rrr
V
/paint/
NP
NP
rrr
8/3/2019 Rambow Slides Leysin Tag
90/142
r
NP
pictures
PP
rr
P
of
NP
John paints pictures of bridges
How do we get Bridgesi John paints pictures of ti?
Picture-NP extraction (ctd)
S
rr
8/3/2019 Rambow Slides Leysin Tag
91/142
rrr
NP VP
rr
V
/paint/
NP
{ S
rr
NP
[index= 1 ]
S*
, NP
rrr
NP
pictures
PP
rr
P
of
NP
[index= 1 ]
}
S
rr
rr
NP S
8/3/2019 Rambow Slides Leysin Tag
92/142
NP
[index= 1 ]
S
rr
r
NP VP
rr
r
V
/paint/
NP
rrr
NP
pictures
PP
rr
P
of
NP
[index= 1 ]
Bridgesi John paints pictures of ti
Overview
Terminology (ppt)
8/3/2019 Rambow Slides Leysin Tag
93/142
Terminology (ppt)
What is Tree Adjoining Grammar?
Some Syntactic Analyses
TAG and Syntactic Theory
Problem: German
The Lexicon as Grammar
Q: Where do all these trees come from? Arent
generalizations being missed (wh movement the
8/3/2019 Rambow Slides Leysin Tag
94/142
generalizations being missed (wh-movement the
same for like and dislike)?
A: We can generalize across classes of lexicalitems (eg, Transitive-Verb), and associate lexical
item with class (eg, like is a Transitive-Verb)
The Lexicon as Grammar
Q: Where do all these trees come from? Arent
8/3/2019 Rambow Slides Leysin Tag
95/142
Q: Where do all these trees come from? Aren tgeneralizations being missed (wh-movement the
same for like and dislike)?
A: We can generalize across classes of lexical
items (eg, Transitive-Verb), and associate lexical
item with class (eg, like is a Transitive-Verb)
Q: Still, wh-movement from the subject position is
the same in intransitive, transitive, and ditransitive
verbs isnt a generalization being missed?
A: Generating all possible trees is the goal of a
TAG-based theory of syntax
The Lexicon as Grammar (ctd)
Q: How big are the elementary trees?
8/3/2019 Rambow Slides Leysin Tag
96/142
Q: How big are the elementary trees?A: A tree is a lexical item and its syntactic
projection.
Option 1: DP
rr
Det
the
NP
NP
tree
Option 2: DP
rr
D NP
Det
8/3/2019 Rambow Slides Leysin Tag
97/142
Det NP
tree
the
Option 3: NP
tree
NP
r
r
the NP*
The exact definition is subject to the particular
linguistic theory.
TAG is not a linguistic theory (nor a linguistic
framework)
Lik CFG TAG i th ti l f li
8/3/2019 Rambow Slides Leysin Tag
98/142
Like CFG, TAG is a mathematical formalism
Unlike GPSG, HPSG, LFG, CCG, and others, TAG
is not a combination theory-and-formal-framework
This means: a linguistic theory (or a linguistic
framework) must be added to TAG to use it for
linguistic description and/or theorizing
We speak of TAG-based linguistic theories/
approaches/frameworks
What is a TAG-based linguistic theory?
Model: GB
8/3/2019 Rambow Slides Leysin Tag
99/142
Model: GBLexicon
andParameters
D-Structure S-Structuremove-
Principles
constrains
constrain constrain constrain
Model: Minimalism
project
(Spellout)
8/3/2019 Rambow Slides Leysin Tag
100/142
move-
Multiclausal
LFmove-
move-
andParameters
Principles
Lexicon
project
extend-target
insert
Multiclausal
PF
constrainconstrain constrain
Model: TAG-Based Syntax
Lexicon D-Structure S-Structureconstrains
Clausal Clausal
move- formal
derivation
S-Structure
Multiclausal
8/3/2019 Rambow Slides Leysin Tag
101/142
andParameters
Principles
constrain constrain constrain
Model: GBLexicon
andParameters
D-Structure S-Structuremove-
Principles
constrains
constrain constrain constrain
TAG-Based Syntactic Theory
Syntactic theory explains variation on domainsf
8/3/2019 Rambow Slides Leysin Tag
102/142
Syntactic theory explains variation on domainsof extended projections only
Combination of extended projections only
through the formal operations of substitution
and adjunction
Scope of theory greatly reduced!
See Frank (2000) for sample account
Overview
Terminology (ppt)
8/3/2019 Rambow Slides Leysin Tag
103/142
gy (pp )
What is Tree Adjoining Grammar?
Some Syntactic Analyses
TAG and Syntactic Theory
Problem: German
German: long scrambling in Mittelfeld
(3) a. dass es Hans zu reparieren versucht
8/3/2019 Rambow Slides Leysin Tag
104/142
( )that it Hans to
prepair tries
that Hans tries to repair it
b. * dassthat
esit
HansHans
repariertrepair
zuto
habenhave
bereutregrets
Intended meaning: that Hans regrets
having repaired it
Long scrambling is iterable and combinable with
extraposition
8/3/2019 Rambow Slides Leysin Tag
105/142
(4) a. dassthat
esitACC
HansHans
den Kindern[the children]DAT
zuto
gebengive
versuchttries
that Hans tries to give it to the children
b. dass es Hans den Kindern versucht zu
geben
dass es Hans den Kindern versucht zu geben
=>A
S SA
A
8/3/2019 Rambow Slides Leysin Tag
106/142
=>A
A*
A
A
TAG adjunction produces only 5 segments, we
need 6!
Solution: V-TAG, UVG-DL, DSG (Rambow 1994,
Rambow et al 2001)
TAG: grammar = ready-made trees; trees are
8/3/2019 Rambow Slides Leysin Tag
107/142
g y ;
combined
Solution: V-TAG, UVG-DL, DSG (Rambow 1994,
Rambow et al 2001)
TAG: grammar ready made trees; trees are
8/3/2019 Rambow Slides Leysin Tag
108/142
TAG: grammar = ready-made trees; trees are
combined
DSG: grammar = trees in parts, assembly
required; assembled trees are combined
Example:arg: +
NP
VP arg: +
VPNP
VP
VP
arg: +VP
VP
VP
VP gebenNP
head: +
8/3/2019 Rambow Slides Leysin Tag
109/142
path constraint:no [ complete:+ ]
NP VPNP VPVPnom
VP gebenNP
VP
V
arg: dat acc
path constraint:
no [ head:+ ]
head:+ arg:+
head:+ arg:+VPNP
VP
acc
8/3/2019 Rambow Slides Leysin Tag
110/142
arg: +
path constraint:
no [ complete:+ ]
NP
VP arg: +
VPNP
VP
VP
arg: +VP
VPnom
VP
VP gebenNP
VP
V
head: +
arg: dat acc
path constraint:
no [ head:+ ]
head:+ arg:+
VP
V
geben
VP
VP
nomNP
NPdat
acc
head: arg:
head:+ arg:+head:+ arg:
Example with Long Scrambling:
NP
VP
VP
arg: +
NP
VP
VP
arg: +arg: +VP
VPNP
8/3/2019 Rambow Slides Leysin Tag
111/142
arg: arg:
accdatnom
path constraint:
no [ complete:+ ]
arg: +
head: + VP
VP geben
V
versuchenVP
VP head: +
V
VP
Example with Long Scrambling:
VP arg: +d t
NP
VP arg: +
8/3/2019 Rambow Slides Leysin Tag
112/142
arg:
arg:
arg: +VP
nomNP
acc
NP
dat
versuchenVP
VP head: +
V
VP arg: +
head: + VP
VP geben
V
New Model: V-TAG-Based SyntaxClausal
formalderivation
Lexicon
constrain
Structureformalderivation
S-Structure
Multiclausal
8/3/2019 Rambow Slides Leysin Tag
113/142
and Parameters
Principles
Old Model
Lexicon
andParameters
D-Structure S-Structure
Principles
constrains
Clausal Clausal
move- formalderivation
S-Structure
Multiclausal
constrain constrain constrain
What about LF? Use Synchronous TAG (Shieber
1994, 2000)ClausalStructure
MulticlausalLF
8/3/2019 Rambow Slides Leysin Tag
114/142
PF
derivation
synchronizedformalderivation
synchronizedLexicon
and
Parameters
Principles
constrain
Clausal
Structure
Multiclausal
formal
Conclusion
Tree Adjoining Grammar is a good formalismfor describing natural language syntax
8/3/2019 Rambow Slides Leysin Tag
115/142
g g g y
Exact type of TAG, and exact type of
linguistic theory, are active areas of research
Backup Slides
8/3/2019 Rambow Slides Leysin Tag
116/142
Desiderata for a formal system for NL syntax
(based on mildly context-sensitive, Joshi 1987):
Constrained weak generative capacity (at
8/3/2019 Rambow Slides Leysin Tag
117/142
most context-sensitive)
Sentences made up incrementally from lexicalitems: constant-growth property
Can form a new sentence using conjunction:closure under Kleene-star
Sentential subjects: closure under iterated
substitution
Polynomially parsable
Adjunction constraints on nonterminal nodes in
elementary trees:
8/3/2019 Rambow Slides Leysin Tag
118/142
Null adjoining constraint (NA) (=terminalnode for rewriting)
Obligatory adjunction constraint (OA)
(=nonterminal node for rewriting)
Selective adjoining constraint (SA)
Increases formal power
Example of adjoining constraint
1 2
8/3/2019 Rambow Slides Leysin Tag
119/142
**
S
NP VP
V
seen
OA( , ) 1 2
VP
VP VP
VP
Aux Aux
has is
1 2
TAG with Feature Structures
Represent adjoining constraints with fixed-sized
feature structures
8/3/2019 Rambow Slides Leysin Tag
120/142
b3
t1 v t2
t3A
t2
A
A
b1 v b3
t3
b2
t1
b2A
b1A
Example
tense: +
S
VP
1
tense: +
tense: +
tense: -
tense: +
tense:
VP
2
tense: +
tense:
t
8/3/2019 Rambow Slides Leysin Tag
121/142
**
*
tense: -
tense: +
NP VP
V
seen
VPAux
has
tense:
tense:
VPAux
is
tense:
tense: -
VP
VP
Aux
been
3
tense: -
tense:
tense:
tense: -
What about questions?
S
tense: [1]
tense: +
tense: [1]
VP
tense: -
tense: +
tense:
VP tense: +
tense:
tense:
1 2
8/3/2019 Rambow Slides Leysin Tag
122/142
*
**
*VP
V
seen
tense: -
[ ]
VPAux
has
tense:VPAux
is
tense:
tense: -
Aux
has
tense:
tense: -
tense: +
tense:
Aux
is
tense: +
tense:
tense:
tense: -
4S
S
S
S
3
NP
The Lexicon as Grammar (or the Grammar as
Lexicon)
We have many choices how to define elementary
8/3/2019 Rambow Slides Leysin Tag
123/142
trees in a linguistic grammar
S
rrr
NP
Betty
VP[agr:3sg]
VP
[agr:3sg]
V
blushes
S
rr
rr
NP VP
8/3/2019 Rambow Slides Leysin Tag
124/142
Betty
rrr
V
[agr:3sg]
NP
bats
V
[agr:3sg]
likes
(These are stupid elementary trees.)
8/3/2019 Rambow Slides Leysin Tag
125/142
The Lexicon as Grammar (ctd)
Almost all linguistic uses of TAG follow certain
basic assumptions we will discuss those basic
8/3/2019 Rambow Slides Leysin Tag
126/142
assumptions here
But the basic assumptions are assumptions
linguists bring to TAG, not assumptions that TAG
imposes on the linguists!
For example, we can use EDL for capturing
linguistic properties as just discussed, but TAG
does not impose that choice
The Lexicon as Grammar (ctd)
It seems that in language:
Properties (morphological, syntactic,
ti ) i t d ith l i l it
8/3/2019 Rambow Slides Leysin Tag
127/142
semantic) are associated with lexical items
and with classes of lexical itemsTAG: Given the extended domain of locality,
we can state all relevant properties in an
elementary tree
The properties of larger structures are
compositionally derivable from the propertiesof the composing lexical items
TAG: Grammar is set of trees each
associated with a single lexical item
(exploiting lexicalization)
8/3/2019 Rambow Slides Leysin Tag
128/142
The Lexicon as Grammar (ctd)
A better set of elementary trees:
Srr
Sr
8/3/2019 Rambow Slides Leysin Tag
129/142
rr
rr
NP
[agr:3sg]
VP
V
[agr:3sg]
blushes
rr
rr
NP
[agr:3sg]
VP
rr
V
likes
NP
NP
Betty
NP
bats
8/3/2019 Rambow Slides Leysin Tag
130/142
The Lexicon as Grammar (ctd)
Q: In a TAG grammar, is there one tree per lexical
item?
A: No
8/3/2019 Rambow Slides Leysin Tag
131/142
A: No
Example: verb
Voice alternations (active/passive)
Syntactic alternations (wh-movement,topicalization, dative shift, . . . )
. . . Every lexical item is associated with a set of
trees (family) which represents the set of
syntactic (and morphological) variations
8/3/2019 Rambow Slides Leysin Tag
132/142
The Lexicon as Grammar (ctd)
Conclusion:
A TAG grammar (in the formal sense) is
8/3/2019 Rambow Slides Leysin Tag
133/142
an enumeration of all lexical items in the
language, in all possible syntactic
variations
The formal TAG grammar is a lexicon of thelanguage
English Passive:
8/3/2019 Rambow Slides Leysin Tag
134/142
S
rrr
rrr
NP VP
8/3/2019 Rambow Slides Leysin Tag
135/142
NP
[nom,role:themeagr: 1 ]
VP
rr
rr
Aux
[agr: 1 ]
/be/
VP
V
[past-part]
eaten
VP
rrr
VP* PP rr
8/3/2019 Rambow Slides Leysin Tag
136/142
r
P
by
NP
[obj]
Apples were eaten by Joey
English Passive with agent:
8/3/2019 Rambow Slides Leysin Tag
137/142
S
rrr
rrr
NP VP
8/3/2019 Rambow Slides Leysin Tag
138/142
[nom,role:theme
agr: 1 ]
rr
rr
r
Aux
[agr: 1 ]
/be/
VP
rrr
r
V
[past-part]
eaten
PP
r
rr
P
by
NP
[obj,
role:agent]
Apples were eaten by Joey
8/3/2019 Rambow Slides Leysin Tag
139/142
Using features to force adjunction (English
agentless Passive):
8/3/2019 Rambow Slides Leysin Tag
140/142
S
rr
rrr
[case:nom] [agr: 1 ]
8/3/2019 Rambow Slides Leysin Tag
141/142
[role:theme]
[agr: 1 ]
NP
[tense:+]
VP
[tense:-]
[mood:past-part]
V[past-part]
eaten
[agr: 1 ]
[tense:+]
VPrr
8/3/2019 Rambow Slides Leysin Tag
142/142
r
rr
rr
[agr: 1 ]
Aux
/be/
VP
[tense:-]
[mood:past-part]
Apples were eaten