Transcript PP attachment and - Indian Institute of Technology Bombay

Syntactic Disambiguation
through
Lexicon Enrichment
Second Stage Project Presentation
Guide: Pushpak Bhattacharyya
Ashish Almeida
03M05601
Overview
•
•
•
•
•
•
•
Motivation
Problem definition
Linguistic theory
Lexical enrichment
Design and implementation
Results
Future work
July 16, 2015
2
Motivation
• Robust and scalable UNL generation required
• English analysis for extracting meaning
• Correct analysis  correct meaning
representation
• Identification of correct syntactic
representation
• Identification of correct semantic relation
July 16, 2015
3
Preposition Phrase Attachment Problem
• John covered the baby with a blanket.
covered
John
the baby
covered
with
a blanket
John
the baby
with
a blanket
Verb attachment
Noun attachment
July 16, 2015
4
Universal Networking Language
• He forwarded the mail to the minister.
forward(icl>send)
agt
@ entry @ past
gol
obj
He(icl>person)
minister(icl>person)
mail(icl>collection
)
July 16, 2015
@def
@def
5
Linguistic Insights
• Syntactic level
– Syntactic Frame
– Subcategorization
• Semantic level
– Selectional restrictions
– Thematic/theta roles
July 16, 2015
6
Syntactic Frame
• Sequence of words as they appear in
sentences
– [V-ART-N]
… handed a book
– [NP-to-NP]
... the mail to the minister
– [V-NP-P-NP] … forwarded the mail to the minister
July 16, 2015
7
Subcategorization
• Verbs
– He put the book on the table.
– *He put the book.
– *He put.
• put: [ _NP PP-on]
• Nouns
– his reliance on/*at/*with her help.
– *his reliance.
• reliance: [ _PP-on]
• Adjectives
– He is fond of reading.
• fond: [ _ PP-of]
July 16, 2015
8
Selectional Restrictions
• The boy murdered John.
• *The boy murdered the tree.
– Thus the verb ‘murder’ needs a human as object.
• murder: [HUMAN _ HUMAN]
July 16, 2015
9
Thematic Roles
• Each argument of verb has a unique role
associated with it.
• Each role is assigned to a single argument.
E.g.
– The boy murdered John.
• The boy
• John
- agent
- patient/theme
• Other thematic roles : Instrument, locative,
goal
• UNL relations: analogous to thematic roles.
July 16, 2015
10
Lexicon Enrichment
• Idiosyncratic information
– Subcategorization
– Thematic roles in terms of UNL relations
• How to get this information ?
– Subcategorization
• Oxford advanced learner’s dictionary, WordNet
– UNL relations
• Beth Levin, manual effort
July 16, 2015
11
An Example Dictionary Entry
• forward
• E.g. he forwarded the mail to the minister
•
[forward]{}“forward(icl>sent)”
(VRB,VOA,VOA-PHSL, #_TO_A2,#_TO_A2_gol)<E,0,0>;
headword
Universal Word
July 16, 2015
Attributes
12
Issues
• The work focuses on
– The [V-NP-P-NP] frame
– Commonly used prepositions
• In, on, of, with, from, to, for
– Disambiguating to
– Active voice
July 16, 2015
13
Design
• Fill the valency of the nearest element first.
• If in the frame [V-N1-P-N2]
– both V and N1 have #P in their subcategorization
frames, then satisfy the demand of the nearest
element to P, i.e., the noun first.
• Else, give priority to that element which
subcategorizes the preposition P
• Else, give priority to the events and actions (can
be verb or noun)
– destroyV, destructionN etc.
July 16, 2015
14
Summarization of Algorithm
Conditions
Action
Attributes
of
Attributes
of
Attachment
of
Examples
1 #<P>
#<P>
_
N1
…paid a visit to the museum.
2 #<P>
Not
#<P>
V
...passed the ball to Bill.
…imposed heavy penalties on fuel
dealers.
3 Not
a #<P>
Not
#<P>
3 Not
b #<P>
Not
#<P>,
EVENT
Attributes
of
V
5
Not
#<P>
NP1
#<P>
July 16, 2015
NP2
_
NP2
PLACE
V
TIME
…met him in his office.
…met him in the afternoon.
PLACE
TIME
N1
…cancelled a meeting with his friends.
N1
…supplied plans for projects.
15
Problems with to
• Infinitival to
– Do not allow onion to brown
• Preposition to
– The lights changed from green to brown
Problem:
Detect if the lexical element is to-preposition
or to-infinitive
July 16, 2015
16
Heuristics to Detect to-preposition
Criterion
Preposition
to
- to is followed by a determiner

- to is followed by a word followed by a plural marker

- to is followed by an adjective

- to is followed by a proper noun

- to is followed by a pronoun

- the matrix verb specifies that it needs a topreposition complement.

- to is preceded by a noun which specifies that it
needs a to-preposition complement.

Infinitival
to
- the matrix verb specifies that it needs a toinfinitival complement.

- to is preceded by a noun which specifies that it
needs a to-infinitival complement.

- to is followed by a base verb

July 16, 2015
17
Implementation
• Creating new dictionary with extra attributes
• Writing new rules to use these new attributes
– Rules to use subcategorization information
– Rules for processing events (nouns and verbs)
July 16, 2015
18
Analysis Engine (Enconverter)
sentence
Word1 Word2 Word3 Word4
LCW
LAW
RAW RCW
… Wordn
windows
• Analysis windows
– Left Analysis Window (LAW)
– Right Analysis Window (RAW)
• Condition windows
– Many in number
– LCWs, RCWs
July 16, 2015
19
Operations in Analysis
•
•
•
•
•
Movement of heads
Addition of two nodes
Deletion of a node
Creating relation between two nodes
Adding dynamically inferred attributes to
node
July 16, 2015
20
Rules
; Right shift to affect noun attachment
R{VRB,#_FOR_AR2:::}{N,#_FOR:::}(PRE,#FOR)P60;
This states that
IF
The left analysis window is on a verb which takes a for-pp as
the second argument (indicated by #_FOR_AR2)
AND
The right analysis window is on a noun which takes a for-pp
as an argument (indicated by #_FOR)
AND
The preposition for follows the noun (indicated by
(PRE,#FOR) )
THEN
Shift right (indicated by R at the start of the rule) anticipating
noun attachment for the PP.
July 16, 2015
21
Other Rules
; Create relation between V and N2, after resolving the preposition
preceding N2
<{VRB,#_FOR_AR2,#_FOR_AR2_rsn:::}
{N,FORRES,PRERES::rsn:}P25;
;Delete the preposition ON
>(VRB,EVENT,VOA){PRE,#ON:::}
{N,UNIT,TIME,DAY:+ONRES,+PRERES::}P27;
;Create the relation tim between verb and noun
<{VRB,VOA:::}
{N,TIME,UNIT,ONRES,PRERES::tim:}P20;
July 16, 2015
22
Testing
• Resources:
– British National Corpus
– WordNet
– Brown corpus
• Filtered out
– Phrasal verbs
– Compound nouns
– Longer sentences
• Semantically different types of constructs
tested in [V-N-P-N] frame.
July 16, 2015
23
Cases of with
Different semantic Roles in different syntactic and semantic environments
Attachment
Semantic
relation
Example
Noun
obj
He cancelled [a meeting with his students].
Noun
and
She wore [a green skirt with a blouse].
Verb
ins
He [covered the baby with a blanket].
Verb
gol
That [provides him with a living].
Verb
ptn
He [is playing chess with his friend].
July 16, 2015
24
Results for of-preposition
The results of testing for solving PP attachment and generating UNL
Corpus
Frames
BNC
WSJ
V-N1-of-N2
V-N1-of-N2
July 16, 2015
Total no. of
Sentences
1000
661
No. of Correct
attachments & UNL
relations
Accuracy
%
886
597
88
90
25
Conclusion
• Lexical enrichment originating from key
linguistic principles makes the analysis more
correct
• Rule-base design simplified due to distinction
made between complements and adjuncts
during analysis
July 16, 2015
26
Future Work
• Handling the alternation patterns of verbs
• Applying the algorithm on all prepositions
• Extracting the information through various
resources
– such as dictionaries and annotated corpus
July 16, 2015
27
References
•
UNDL Foundation: The Universal Networking Language (UNL) specifications
version 3.2. (2003) http://www.unlc.undl.org
•
Grimshaw, Jane.: Argument Structure. The MIT Press, Cambridge, Mass. (1990)
•
Brill, E. and Resnik, R.: A Rule based approach to Prepositional Phrase
Attachment disambiguation. Proc. of the fifteenth International conference on
computational linguistics. Kyoto. (1994)
•
Levin, Beth.: English verb Classes and Alternation. The University of Chicago
Press, Chicago. (1993)
•
Hornby, A. S.: Oxford Advanced Learner’s Dictionary of Current English. Oxford
University Press, Oxford.(2000)
July 16, 2015
28
Thank you
July 16, 2015
29
Example UNL
In
I deposited my money in my bank account.
{unl}
gol(deposit(icl>put):02.@entry.@past,
account(icl>statement):0W)
obj(deposit(icl>put):02.@entry.@past,
money(icl>currency):0F)
agt(deposit(icl>fasten):02.@entry.@past, I:0C)
mod(money(icl>currency):0F, I:0C)
mod(account(icl> statement):0W, bank(icl>possession):0R)
mod(account(icl> statement):0W, I:0O)
{/unl}
July 16, 2015
30
Example UNL
On
I put the book on the table.
{unl}
gol(put(icl>move):02.@present.@entry,
table(icl>object):0M.@def)
obj(put(icl>move):02.@present.@entry,
book(icl>publication):0A.@def)
agt(put(icl>move):02.@present.@entry, I:00)
{/unl}
July 16, 2015
31
Example UNL
To
They served a wonderful meal to fifty delegates.
{unl}
gol(serve(icl>provide):05.@entry.@past,
delegate(icl>person):12.@pl)
obj(serve(icl>provide):05.@entry.@past,
meal(icl>food):0O.@indef)
agt(serve(icl>provide):05.@entry.@past,
they(icl>thing):00)
mod(meal(icl>food):0O.@indef, wonderful(mod<thing):0E)
qua(delegate(icl>person):12.@pl, fifty(icl>number):0W)
{/unl}
July 16, 2015
32