Transcript Resources - IIT Bombay

CS626-449: NLP, Speech and
Web-Topics-in-AI
Pushpak Bhattacharyya
CSE Dept.,
IIT Bombay
Lecture 37: Semantic Role Extraction
(obtaining Dependency Parse)
Vaquious Triangle
Interlingua based
(do deep semantic process
Before entering the target language)
Transfer Based
(do deep semantic
process
Before entering the
target language)
Direct
(enter the target
Language immediately
Through a dictionary)
Vaquious: an eminent
French Machine
Translation ResearcherOriginally a Physicist
2
Universal Networking
Language




Universal Words (UWs)
Relations
Attributes
Knowledge Base
3
UNL Graph
He forwarded the mail to the minister.
forward(icl>send)
agt
@ entry @ past
gol
obj
He(icl>person)
minister(icl>person)
mail(icl>collection)
@def
@def
4
AGT / AOJ / OBJ

AGT (Agent)
Definition: Agt defines a thing which
initiates an action

AOJ (Thing with attribute)
Definition: Aoj defines a thing which is in a
state or has an attribute

OBJ (Affected thing)
Definition: Obj defines a thing in focus
which is directly affected by an event or state
5
Examples



John broke the window.
agt ( break.@entry.@past, John)
This flower is beautiful.
aoj ( beautiful.@entry, flower)
He blamed John for the accident.
obj ( blame.@entry.@past, John)
6
BEN

BEN (Beneficiary)
Definition: Ben defines a not directly
related beneficiary or victim of an event or
state

Can I do anything for you?
ben ( do.@entry.@interrogation.@politeness, you )
obj ( do.@entry.@interrogation.@politeness, anything )
agt (do.@entry.@interrogation.@politeness, I )
7
PUR

PUR (Purpose or objective)
Definition: Pur defines the purpose or
objectives of the agent of an event or the
purpose of a thing exist

This budget is for food.
pur ( food.@entry, budget )
mod ( budget, this )
8
RSN


RSN
(Reason)
Definition: Rsn defines a reason why
an event or a state happens
They selected him for his honesty.
agt(select(icl>choose).@entry, they)
obj(select(icl>choose) .@entry, he)
rsn (select(icl>choose).@entry, honesty)
9
TIM


TIM (Time)
Definition: Tim defines the time an
event occurs or a state is true
I wake up at noon.
agt ( wake up.@entry, I )
tim ( wake up.@entry, noon(icl>time))
10
TMF


TMF (Initial time)
Definition: Tmf defines a time an
event starts
The meeting started from morning.
obj ( start.@entry.@past, meeting.@def )
tmf ( start.@entry.@past, morning(icl>time) )
11
TMT


TMT (Final time)
Definition: Tmt defines a time an event
ends
The meeting continued till evening.
obj ( continue.@entry.@past, meeting.@def )
tmt ( continue.@entry.@past,evening(icl>time) )
12
PLC


PLC (Place)
Definition: Plc defines the place an
event occurs or a state is true or a thing
exists
He is very famous in India.
aoj ( famous.@entry, he )
man ( famous.@entry, very)
plc ( famous.@entry, India)
13
PLF


PLF (Initial place)
Definition: Plf defines the place an
event begins or a state becomes true
Participants come from the whole
world.
agt ( come.@entry, participant.@pl )
plf ( come.@entry, world )
mod ( world, whole)
14
PLT


PLT (Final place)
Definition: Plt defines the place an
event ends or a state becomes false
We will go to Delhi.
agt ( go.@entry.@future, we )
plt ( go.@entry.@future, Delhi)
15
INS

INS (Instrument)
Definition: Ins defines the instrument
to carry out an event

I solved it with computer
agt ( solve.@entry.@past, I )
ins ( solve.@entry.@past, computer )
obj ( solve.@entry.@past, it )
16
Attributes




Constitute syntax of UNL
Play the role of bridging the conceptual world and
the real world in the UNL expressions
Show how and when the speaker views what is said
and with what intention, feeling, and so on
Seven types:







Time with respect to the speaker
Aspects
Speaker’s view of reference
Speaker’s emphasis, focus, topic, etc.
Convention
Speaker’s attitudes
Speaker’s feelings and viewpoints
17
Tense: @past
He went there yesterday

The past tense is normally expressed by
@past
{unl}
agt(go.@entry.@past, he)
…
{/unl}
18
Aspects: @progress
It’s raining hard.
{unl}
man ( rain.@entry.@present.@progress,
hard )
{/unl}
19
Speaker’s view of reference


@def (Specific concept (already referred))
The house on the corner is for sale.
@indef (Non-specific class)
There is a book on the desk

@not is always attached to the UW which
is negated.
He didn’t come.
agt ( come.@entry.@past.@not, he )
20
Speaker’s emphasis

@emphasis
John his name is.
mod ( name, he )
aoj ( John.@emphasis.@entry, name )

@entry denotes the entry point or main
UW of an UNL expression
21
Subcategorization Frames

Specify the categorial class of the lexical item.
Specify the environment.

Examples:

kick: [V; _ NP]
cry: [V; _ ]
rely: [V; _PP]
put: [V; _ NP PP]
think: : [V; _ S` ]
22
Subcategorization Rules
Subcategorization Rule:
V
y /
_NP]
_]
_PP]
_NP PP]
_S`]
23
Subcategorization Rules
The boy relied on the friend.
1.
2.
3.
4.
5.
6.
7.
S  NP VP
VP  V (NP) (PP) (S`)…
NP  Det N
V  rely / _PP]
P  on / _NP]
Det  the
N  boy, friend
24
Semantically Odd
Constructions

Can we exclude these two ill-formed
structures ?



*The boy frightened sincerity.
*Sincerity kicked the boy.
Selectional Restrictions
25
Selectional Restrictions

Inherent Properties of Nouns:
[+/- ABSTRACT], [+/- ANIMATE]

E.g.,
Sincerity [+ ABSTRACT]
Boy [+ANIMATE]
26
Selectional Rules

A selectional rule specifies certain selectional
restrictions associated with a verb.
V
V
y /
frighten /
[+/-ABSTARCT] __
__ [+/-ANIMATE]
[+/-ABSTARCT] __
__ [+ANIMATE]
27
Subcategorization Frame
forward
V
__ NP PP
e.g., We will be forwarding our new
catalogue to you
invitation
N
__ PP
e.g.,
An invitation to the party
e.g.,
A program making science is more
accessible to young people
accessible
A
__ PP
28
Thematic Roles
The man forwarded the mail to the minister.
forward
V
__ NP PP
Event
FORWARD
([
Thing
THE MAN],
[Path
[Thing THE MAIL],
TO THE MINISTER]
)
29
How to define the UWs in UNL
Knowledge-Base?

Nominal concept



Verbal concept





Abstract
Concrete
Do
Occur
Be
Adjective concept
Adverbial concept
30
Nominal Concept: Abstract thing
abstract thing{(icl>thing)}
culture(icl>abstract thing)
civilization(icl>culture{>abstract thing})
direction(icl>abstract thing)
east(icl>direction{>abstract thing})
duty(icl>abstract thing)
mission(icl>duty{>abstract thing})
responsibility(icl>duty{>abstract thing})
accountability{(icl>responsibility>duty)}
event(icl>abstract thing{,icl>time>abstract thing})
meeting(icl>event{>abstract thing,icl>group>abstract thing})
conference(icl>meeting{>event})
TV conference{(icl>conference>meeting)}
31
Nominal Concept: Concrete thing
concrete thing{(icl>thing,icl>place>thing)}
building(icl>concrete thing)
factory(icl>building{>concrete thing})
house(icl>building{>concrete thing})
substance(icl>concrete thing)
cloth(icl>substance{>concrete thing})
cotton(icl>cloth{>substance})
fiber(icl>substance{>concrete thing})
synthetic fiber{(icl>fiber>substance)}
textile fiber{(icl>fiber>substance)}
liquid(icl>substance{>concrete thing})
beverage(icl>food,icl>liquid>substance})
coffee(icl>beverage{>food})
liquor(icl>beverage{>food})
beer(icl>liquor{>beverage})
32
Verbal concept: do
do({icl>do,}agt>thing,gol>thing,obj>thing)
express({icl>do(}agt>thing,gol>thing,obj>thing{)})
state(icl>express(agt>thing,gol>thing,obj>thing))
explain(icl>state(agt>thing,gol>thing,obj>thing))
add({icl>do(}agt>thing,gol>thing,obj>thing{)})
change({icl>do(}agt>thing,gol>thing,obj>thing{)})
convert(icl>change(agt>thing,gol>thing,obj>thing)
classify({icl>do(}agt>thing,gol>thing,obj>thing{)})
divide(icl>classify(agt>thing,gol>thing,obj>thing))
33
Verbal concept: occur and be


occur({icl>occur,}gol>thing,obj>thing)
melt({icl>occur(}gol>thing,obj>thing{)})
divide({icl>occur(}gol>thing,obj>thing{)})
arrive({icl>occur(}obj>thing{)})
be({icl>be,}aoj>thing{,^obj>thing})
exist({icl>be(}aoj>thing{)})
born({icl>be(}aoj>thing{)})
34
How to define the UWs in UNL Knowledge
Base?

In order to distinguish among the verb classes
headed by 'do', 'occur' and 'be', the following
features are used:
[ need an
object ]
English
UW
[ need an
agent ]
'do'
+
+
"to kill"
'occur'
-
+
"to fall"
'be'
-
-
"to know"
35
How to define the UWs in UNL KnowledgeBase?

The verbal UWs (do, occur, be) also take some
pre-defined semantic cases, as follows:
UW
PRE-DEFINED CASES
English
'do'
takes necessarily agt>thing
"to kill"
'occur'
takes necessarily obj>thing
"to fall"
'be'
takes necessarily aoj>thing "to know"
36
Complex
sentence
I want to watch this movie.
@entry.@past
want (icl>)
agt
:01
obj
watch (icl>do)@entry.@inf
agt
I (iof>person)
obj
movie(icl>)
I (iof>person)
@def
37
Approach to UNL
Generation
38
Problem Definition

Generate UNL expressions for English
sentences



in a robust and scalable manner,
using syntactic analysis and lexical resources
extensively.
This needs


detecting semantically relatable entities
and solving attachment problems
Semantically Relatable Sequences
(SRS)
Definition: A semantically relatable
Sequence (SRS) of a sentence is a
group of words in the sentence (not
necessarily consecutive) that appear in the
semantic graph of the sentence as linked
nodes or nodes with speech act labels
(This is motivated by UNL representation)
SRS as an intermediary to and
intermediary
Source
Language
Sentence
SRS
UNL
Target
Language
Sentence
Example to illustrate SRS
past tense
bought
“The man bought a
new car in June”
agent
time
object
man
car
the: definite
a: indefinite
modifier
new
June
in: modifier
Sequences from “the man bought a
new car in June”
e.
{man, bought}
{bought, car}
{bought, in, June}
{new, car}
{the, man}
f.
{a, car}
a.
b.
c.
d.
Basic questions

Which words can form semantic
constituents, which we call Semantically
Relatable Sequences (SRS)?


What after all are the SRSs of the given
sentence?
What semantic relations can link the
words in an SRS and the SRSs
themselves?
Postulate

A sentence needs to be broken into
Sequences of at most three forms



{CW, CW}
{CW, FW, CW}
{FW, CW}
where CW refers to content word or a
clause and FW to function word
SRS and Language
Phenomena
Movement: Preposition
Stranding

John, we laughed at.


(we , laughed.@entry)---------(CW, CW)
(laughed.@entry,at, John)---(CW, FW,
CW)
Movement: Topicalization

The problem, we solved.



(we , solved.@entry)------------(CW, CW)
(solved.@entry , problem)-----(CW,CW)
(the, problem)--------------------(CW,CW)
Movement: Relative
Clauses

John told a joke which we had already heard.
 (John, told.@entry) -------------------(CW, CW)
 (told.@entry, :01) ---------------------(CW,CW)
 SCOPE01(we,had,heard.@entry)-------(CW,
FW,CW)
 SCOPE01(already,heard.@entry)-------(CW,CW)
 SCOPE01(heard@entry,which,joke)---(CW,FW,CW)
 SCOPE01(a, joke)--------------------------(FW,CW)
Movement: Interrogatives

Who did you refer her to?
 (did , refer.@entry.@interrogative)-------(FW,CW)
 (you, refer.@entry.@interrogative)--------(CW,CW)
 (refer.@entry.@interrogative , her)-------(CW,CW)
 (refer.@entry.@interrogative , to,who)---(CW,FW,CW)
Empty Pronominals: toinfinitivals

Bill was wise to sell the piano.
 (wise.@entry , SCOPE01)---------------(CW,CW)
 SCOPE01(sell.@entry , piano)---------(CW,CW)
 (Bill, was, wise.@entry) -----------------(CW,
FW,CW)
 SCOPE01(Bill, to, sell.@entry)---------(CW,
FW,CW)
 SCOPE01(the, piano) --------------------(FW,CW)
Empty pronominal: Gerundial







The cat leapt down spotting a thrush on the lawn.
(The, cat) -------------------------------(FW, CW)
(cat, leapt.@entry) --------------------(CW, CW)
(leapt.@entry , down) ----------------(CW, CW)
(leapt.@entry , SCOPE01) -----------------(CW, CW)
SCOPE01(spotting.@entry,thrush)--------(CW,CW)
SCOPE01(spotting.@entry,on,lawn)---(CW,FW,CW)
PP Attachment

John cracked the glass with a stone.





(John, cracked.@entry)-------------(CW,CW)
(cracked.@entry, glass)-------------(CW,CW)
(cracked.@entry, with, stone)---(CW,FW,CW)
(a, stone)------------------------------(FW,CW)
(the,glass)-------------------------(FW,CW)
SRS and PP attachment
(Mohanty, Almeida, Bhattacharyya, 04)
Conditions
Sub-conditions
Attachment Point
[PP] is subcategorized
by the verb [V]
[NP2] is licensed
[NP2] is attached to the
by a preposition [P] verb [V] (e.g., He forwarded
the mail to the minister)
[PP] is subcategorized
by the noun in [NP1]
[NP2] is licensed
[NP2] is attached to the
by a preposition [P] noun in [NP1](e.g., John
published six articles on
machine translation )
[PP] is neither
subcategorized by the
verb [V] nor by the
noun in [NP1]
[NP2] refers to
[PLACE] / [TIME]
feature
[NP2] is attached to the
verb [V](e.g., I saw Mary in
her office; The girls met the
teacher on different days)
Linguistic Study to
Computation
Syntactic constituents to Semantic
constituents



A probabilistic parser (Charniak, 04) is used.
Other resources: Wordnet and Oxford
Advanced Learner’s Dictionary
In a parse tree, tags give indications of CW
and FW:
 NP, VP, ADJP and ADVP CW
 PP (prepositional phrase), IN (preposition)
and DT (determiner) FW
Observation:
Headwords of sibling nodes form
SRSs
(C) VP bought
“John has bought
a car.”
SRS:
{has, bought},
{a, car},
{bought, car}
(C) VP bought
(F) AUX has
(C) VBD bought
(C) NP car
(F) DT a
has
(C) NN car
bought
a
car
Need:
Resilience to wrong PP attachment



“John has published an
article on linguistics”
Use PP attachment
heuristics
Get
{article, on, linguistics}
(C)VP published
(C)VBD published
(F)DT an
(C)NP article
(F) PP on
(C)NNarticle
(F)IN on
published
an
(C)NPlinguistics
article
(C)NNS linguistics
on
linguistics
to-infinitival
“I forced him to watch this
movie”
Clause boundary is the VP
node, labeled with SCOPE
(C)VP forced
(C)VBD forced
Tag is modified to TO, a FW
tag, indicating that it heads
a to-infinitival clause,
The duplication and insertion
of the NP node with head him
(depicted by shaded nodes) as
a sibling of the VBD node with
head forced is done to bring out
the existence of a semantic
relation between force and him.
(C) S SCOPE
(C)NP him
(C)PRP him
(C)VP
(C)NP him
forced
him
(F)TO toto
(C)PRP him
him
to
(C)VP watch
Linking of clauses:
“John said that he was reading a novel”
Head of S node
marked as Scope SRS:
{said, that, SCOPE}.

Adverbial clauses
have similar parse
tree structures except
that the subordinating
conjunctions are
different from that.

(C) VP said
(C)VBD said
(F) SBAR that
(F) IN that
said
that
(C) S SCOPE
Implementation

Block Diagram of the system
Input Sentence
WordNet 2.0
Charniak Parser
Noun classification
Time and Place
features
Scope Handler
Parse Tree
Parse Tree modification and
augmentation with head and scope
information
THAT clause as Subcat property
Augmented
Parse Tree
Sub-categorization Database
Preposition as Subcat property
Attachment Resolver
Semantically Relatable Sequences
Generator
Semantically Related Sequences
Head determination
Uses a bottom-up strategy to determine the
headword for every node in the parse tree.
Crucial in obtaining the SRSs, since wrong head
information may end up getting propagated all the
way up the tree
Processes the children of every node starting from
the rightmost child and checks the head information
already specified against the node’s tag to
determine the head of the node
Some special cases are:







SBAR node
A VP node with PRO insertion, copula, Phrasal verbs etc.
NP nodes with of-PP cases and conjunctions under them,
which lead to scope creation.
Scope handler


Performs modification on the parse
trees by insertion of nodes in toinfinitival cases
Adjusts of the tag and head information
in case of SBAR nodes
Attachment resolver

Takes a (CW1, FW, CW2) as input and
checks




the time and place features of CW2,
the noun class of CW1 and
the subcategorization information for the CW1 and
FW pair
to decide the attachment.
If none of these yield any deterministic
results, take the attachment indicated by
the parser
SRS generator



Performs a breadth-first search on the
parse tree and performs detailed
processing at every node N1 of the tree.
S nodes which dominate entire clauses
(main or embedded) are treated as
CWs.
SBAR and TO nodes are treated as
FWs.
Algorithm
Algorithm
If the node N1 is a CW (new/JJ, published/VBD,
fact/NN, boy/NN, John/NNP) perform the
following checks:
If the sibling N2 of N1 is a CW (car/NN,
article/NN, SCOPE/S)
Then create {CW,CW} ({new, car}, {published,
article}, {boy, SCOPE})
If the sibling N2 is a FW (in/PP, that/SBAR,
and/CC)
Then, check if N2 has a child FW, N3 (in/IN,
that/IN) and a child CW, N4 (June/NN,
SCOPE/S)
If yes,
Then use attachment resolver to decide the CW
to which N3 and N4 attach.
Create{CW,FW,CW} ({published, in, June},
{fact, that, SCOPE})
If no,
Then check if next sibling N5 of N1 is a CW
(Mary/NN)

If yes,
Create {CW,FW,CW} ({John, and, Mary})
If the node N1 is a FW (the/DT, is/AUX, to/TO),
perform the following checks:
If the parent node is a CW (boy/NP,
famous/VP)
Check if sibling is an adjective.
i. If yes, (famous/JJ)
Then, create {CW,FW,CW} ({She, is, famous})
ii. If no,
(boy/NN)
Then, create {FW,CW} ({the, boy}, {has,
bought})
If the parent node N6 is a FW (to/TO) and the
sibling node N7 is a CW (learn/VB)
Use attachment resolver to decide on the
preceding CW to which N6 and N7 can
attach.
Create {CW,FW,CW} ({exciting, to, learn})
Evaluation



FrameNet corpus [Baker et. al., 1998], a
semantically annotated corpus, as the
testdata.
92310 sentences (call this the gold
standard)
Created automatically from the FrameNet
corpus taking verbs, nouns and adjectives
as the targets



Verbs as the target- 37,984 (i.e., semantic
frames of verbs)
Nouns as the target-37,240
Adjectives as the target-17,086
Score for high frequency verbs
Verb
Swim
Depend
Look
Roll
Rush
Phone162
Reproduce
Step
Urge
Avoid
Frequency
280
215
187
173
172
159
159
157
152
Score
0.695
0.709
0.804
0.835
0.7
0.775
0.797
0.795
0.765
0.789
Scores of 10 verb groups of
high frequency in the Gold
Standard
Scores of 10 noun groups of
high frequency in the Gold
Standard
An actual sentence

A. Sentence: A form of asbestos once
used to make Kent cigarette filters has
caused a high percentage of cancer
deaths among a group of workers
exposed to it more than 30 years ago,
researchers reported.
Relative performance on SRS
constructs
Parameters matched
(CW,CW)
(CW,FW,CW)
(FW,CW)
Total SRSs
Recall
Precision
0
20
40
60
Recall/Precision
80
100
Parameter
Results on sentence constructs
PP Resolution
Clause linkings
Complement-clause resolution
To-infinitival clause resolution
Recall
Precision
0
20
40
60
Recall/Precision
80
100
Rajat Mohanty, Anupama Dutta and Pushpak Bhattacharyya,
Semantically Relatable Sets: Building Blocks for Repesenting Semantics,
10th Machine Translation Summit ( MT Summit 05), Phuket, September, 2005.
Statistical Approach
Use SRL marked corpora

Daniel Gildea and Daniel Jurafsky. 2002. Automatic labeling of
semantic roles. Computational Linguistics, 28(3):245–288.

PropBank corpus


Role annotated WSJ part of Penn Treebank [10]
PropBank role-set [2,4]


Core roles: ARG0 (Proto-agent), ARG1 (Proto-patient) to ARG5
Adjunctive roles:
ARGM-LOC (for locatives),
ARGM-TMP (for temporals), etc.
SRL marked corpora contd…

PropBank roles: an example
[ARG0 It] operates] [ARG1 stores] [ARGM−LOC mostly in Iowa and Nebraska]
Fig.4: Parse tree output, Source: [5]

Preprocessing systems [2]




Part of speech tagger
Base Chunker
Full syntactic parser
Named entities recognizer
Probabilistic estimation [1]

Empirical probability estimation over candidate roles for each
constituent based upon extracted features
P(r | h, pt , gov, position , voice, t ) 
# (r , h, pt , gov, position , voice, t )
# (h, pt , gov, position , voice, t )
here,
t is the target word
r is a candidate role,
h , pt, gov, voice are features

Linear interpolation, with condition
i i  1
P(r | constituent )   1 P(r | t )   2 P(r | pt , t )   3 P(r | pt , gov, t )   4 P(r | h)   5 P(r | h, pt , t )
•
Geometric mean, with condition
r P(r | constituent )  1
1
P(r | constituent )  exp{ 1 P(r | t )   2 P(r | pt , t )   3 P(r | pt , gov, t )   4 P(r | h)   5 P(r | h, pt , t )}
z
A state-of-art SRL system: ASSERT
[4]

Main points [3,4]




Use of Support Vector Machine [13] as classifier
Similar to FrameNet “domains”, “Predicate Clusters” are introduced
Named Entities [14] is used as a new feature
Experiment I (Parser dependency testing)


Use of PropBank bracketed corpus
Use of Charniak parser trained on Penn Treebank corpus
Parse
Treebank
Charniak
Task
Precision (%)
Recall (%)
F-score (%)
Accuracy (%)
Id.
97.5
96.1
96.8
-
Class.
-
-
-
93.0
Id. + Class.
91.8
90.5
91.2
-
Id.
87.8
84.1
85.9
-
Class.
-
-
-
92.0
Id. + Class.
81.7
78.4
80.0
-
Table 1: Performance of ASSERT for Treebank and Charniak parser outputs.
Id. Stands for identification task and Class. stands for classification task. Data source: [4]
Experiments and Results

Experiment II (Cross genre testing)
1.
2.
Training on PropBanked WSJ data and testing on Brown Corpus
Charniak parser trained on first PropBank then Brown
Table 2: Performance of ASSERT for various experimental combinations
Date source: [4]