of 21
8/6/2019 NLP - Lecture1
1/21
CS 598 JH: Advanced NLP (Spring 09)
Ju lia Hockenmaier [email protected]
3324 Siebel CenterOfce Hours: Fri, 2:00-3:00pm
http://www.cs.uiuc.edu/~juliahmr/cs598
Review
http://www.cs.uiuc.edu/class/fa08/cs498jhhttp://www.cs.uiuc.edu/class/fa08/cs498jhhttp://www.cs.uiuc.edu/class/fa08/cs498jhhttp://www.cs.uiuc.edu/class/fa08/cs498jhhttp://www.cs.uiuc.edu/class/fa08/cs498jhhttp://www.cs.uiuc.edu/class/fa08/cs498jhhttp://www.cs.uiuc.edu/class/fa08/cs498jh8/6/2019 NLP - Lecture1
2/21
CS 598 JH: Advanced NLP (Spring 09)
What is the structure
of a sentence?Sentence structure is hierarchical:
A sentence consists of words ( I, eat, sushi, with, tuna )
..which form phrases or constituents: sushi with tuna
Sentence structure denes dependenciesbetween words or phrases:
I eat sushi with tuna
2
[ ][ ][ ][ ]
8/6/2019 NLP - Lecture1
3/21
CS 598 JH: Advanced NLP (Spring 09)
Strong vs. weak
generative capacityFormal language theory:- denes language as string sets- is only concerned with generating these strings
(weak generative capacity)
Formal/Theoretical syntax (in linguistics):- denes language as sets of strings with (hidden) structure-is also concerned with generating the right structures(strong generative capacity)
3
8/6/2019 NLP - Lecture1
4/21
CS 598 JH: Advanced NLP (Spring 09)
Context-free grammars (CFGs)
ca ture recursionLanguage has complex constituents
(the garden behind the house )
Syntactically, these constituents behavejust like simple ones.
(behind the house can always be omitted)
CFGs dene nonterminal categoriesto capture equivalent constituents.
4
8/6/2019 NLP - Lecture1
5/21
CS 598 JH: Advanced NLP (Spring 09)
Context-free grammars
A CFG is a 4-tuple N , ,R,SA set of nonterminals N(e.g. N = {S, NP, VP, PP, Noun, Verb, ....})
A set of terminals
(e.g. = {I, you, he, eat, drink, sushi, ball, })
A set of rules R R {A with left-hand-side (LHS) A N
and right-hand-side (RHS) (N )* }A start symbol S (sentence)
5
8/6/2019 NLP - Lecture1
6/21
An example
N {ball, garden, house, sushi } P {in, behind, with} NP N
NP NP PPPP P NP
N: nounP: prepositionNP: noun phrasePP: prepositional phrase
8/6/2019 NLP - Lecture1
7/21
CFGs dene parse trees
eat with tunasushiNPNP
VP
PPNPV P
N {sushi, tuna} P {with} V {eat} NP NNP NP PPPP P NPVP V NP
8/6/2019 NLP - Lecture1
8/21
CFGs are equivalent toPushdown automata PDAs
PDAs are FSAs with an additional stack:Emit a symbol and push/pop a symbol from the stack
This is equivalent to the following CFG:S a X bX a X bX a b
Push x on stack.Emit a
Pop x from stack.
Emit b
Accept ifstack
empty.
8/6/2019 NLP - Lecture1
9/21
The Chomsky HierarchyLanguage Automata Parsingcomplexity Dependencies
Type 3 Regular Finite-state linear adjacent words
Type 2 Context-Free Pushdown cubic nested
Type 1 Context-sensitiveLinear
Bounded exponential
Type 0 RecursivelyEnumerableTuring
machine
8/6/2019 NLP - Lecture1
10/21
CS 598 JH: Advanced NLP (Spring 09)
Constituents:
Heads and dependentsThere are different kinds of constituents:
Noun phrases : the man, a girl with glasses , Illinois Prepositional phrases : with glasses, in the garden Verb phrases : eat sushi, sleep, sleep soundly
Every phrase has a head:Noun phrases : the man, a girl with glasses , Illinois
Prepositional phrases : with glasses, in the garden Verb phrases : eat sushi, sleep, sleep soundly The other parts are its dependents .Dependents are either arguments or adjuncts
10
8/6/2019 NLP - Lecture1
11/21
Two ways to represent structure
eat with tunasushiNPNP
VP
PPNP
V P
sushieat with chopsticksNPNP
VP
PPVPV P
eat sushi with tuna
eat sushi with chopsticks
Phrase structure trees Dependency trees
8/6/2019 NLP - Lecture1
12/21
Structure (Syntax)corresponds to
Meaning (Semantics)Correct analysis
Incorrect analysis
eat with tunasushiNPNP
VP
PPNP
V P
sushieat with chopsticksNPNP
VP
PPVPV P
eat sushi with tuna
eat sushi with chopsticks
eat sushi with chopsticks
NPNP
NPVP
PPV P
eat with tunasushiNPNP
VP
PPVPV P
eat sushi with tuna
eat sushi with chopsticks
8/6/2019 NLP - Lecture1
13/21
CS 598 JH: Advanced NLP (Spring 09)
Dependency grammar
DGs describe the structure of sentences as graph.The nodes of the graph are the wordsThe edges of the graph are the dependencies.
The relationship between DG and CFGs:If a CFG phrase structure tree is translated into DG,the resulting dependency graph has no crossing edges.
13
8/6/2019 NLP - Lecture1
14/21
CS 598 JH: Advanced NLP (Spring 09)
CKY chart parsing algorithm
Bottom-up parsing:start with the words
Dynamic programming:save the results in a table/chart
re-use these results in nding larger constituents
Complexity: O(n 3|G|)n: length of string, |G|: size of grammar)
Presumes a CFG in Chomsky Normal Form:Rules are all either A B C or A a(with A,B,C nonterminals and a a terminal)
14
8/6/2019 NLP - Lecture1
15/21
we eat sushiwe eat
eat sushi
sushi
eat
we
S NP VPVP V NPV eat
NP
weNP sushi
We eat sushi
The CKY parsing algorithm
SNP
V
NP
VP
8/6/2019 NLP - Lecture1
16/21
Exercise: CKY parser
S NP VP
NP NP PP
NP Noun
VP VP PP
VP Verb NP
I eat sushi with chopsticks
8/6/2019 NLP - Lecture1
17/21
CS 598 JH: Advanced NLP (Spring 09)
Dealing with Ambiguity
A grammar might generate multiple trees for a sentence:
What s the most likely parse for sentence S ?
We need a model of P( | S)
17
eat with tunasushiNPNP
VP
PPNP
V P
sushieat with chopsticksNPNP
VP
PPVPV P
eat sushi with chopsticksNPNP
NP
VP
PPV P
eat with tunasushiNPNP
VP
PPVPV P
8/6/2019 NLP - Lecture1
18/21
CS 598 JH: Advanced NLP (Spring 09)
Using Bayes Rule:
The yield of a tree is the string of terminal symbols
that can be read off the leaf nodes
Computing P( | S)
arg max
P ( |S ) = arg max
P ( , S )P (S )
= arg max
P ( , S )
= arg max
P ( ) if S = yield( )
18
yield( )= eat sushi with tuna
eat with tunasushiNPNP
VP
PPNP
V P
8/6/2019 NLP - Lecture1
19/21
CS 598 JH: Advanced NLP (Spring 09)
T is the (innite) set of all trees in the language:
Weed to dene P( ) such that:
The set T is generated by a context-free grammar
Computing P( )
19
T : 0 P ( ) 1 T P ( ) = 1
L = {s | T : yield ( ) = s }
S NP VP VP Verb NP NP Det NounS S conj S VP VP PP NP NP PPS ..... VP ..... NP .....
8/6/2019 NLP - Lecture1
20/21
CS 598 JH: Advanced NLP (Spring 09)
Probabilistic Context-Free Grammars
For every nonterminal X, dene a probability distributionP(X | X) over all rules with the same LHS symbol X:
20
S NP VP 0.8S S conj S 0.2
NP
Noun 0.2NP Det Noun 0.4NP NP PP 0.2NP NP conj NP 0.2VP Verb 0.4
VP
Verb NP 0.3VP Verb NP NP 0.1VP VP PP 0.2PP P NP 1.0
8/6/2019 NLP - Lecture1
21/21
CS 598 JH: Advanced NLP (Spring 09)
Computing P( ) with a PCFG
The probability of a tree is the product of the probabilitiesof all its rules:
21
P( ) = 0.8 0.3 0.2 1.0= 0.00384
0.2 3
S
NP
Noun
John
VP
VP
Verb
eats
NP
Noun
pie
PP
P
with
NP
Noun
cream
S NP VP 0.8S S conj S 0.2NP Noun 0.2
NP
Det Noun 0.4NP NP PP 0.2NP NP conj NP 0.2VP Verb 0.4VP Verb NP 0.3
VP
Verb NP NP 0.1VP VP PP 0.2PP P NP 1.0