Proposition Bank: a resource of predicate-argument relations Martha Palmer University of Pennsylvania October 9, 2001 Columbia University 10/9/01 PropBank.

Download Report

Transcript Proposition Bank: a resource of predicate-argument relations Martha Palmer University of Pennsylvania October 9, 2001 Columbia University 10/9/01 PropBank.

Proposition Bank:
a resource of
predicate-argument relations
Martha Palmer
University of Pennsylvania
October 9, 2001
Columbia University
10/9/01
PropBank
1
Outline
 Overview (Ace consensus: BBN,NYU,MITRE,Penn)
 Motivation
 Approach
• Guidelines, lexical resources, frame sets
• Tagging process, hand correction of automatic
tagging
 Status: accuracy, progress
 Colleagues: Joseph Rosenzweig, Paul Kingsbury,
Hoa Dang, Karin Kipper, Scott Cotton, Laren
Delfs, Christiane Fellbaum
10/9/01
PropBank
2
Proposition Bank:
Generalizing from Sentences to Propositions
Powell met Zhu Rongji
battle
wrestle
join
debate
Powell and Zhu Rongji met
Powell met with Zhu Rongji
consult
Proposition: meet(Powell, Zhu Rongji)
Powell and Zhu Rongji had
a meeting
meet(Somebody1, Somebody2)
...
When Powell met Zhu Rongji on Thursday they discussed the return of the spy plane.
10/9/01
meet(Powell, Zhu)
discuss([Powell, Zhu], return(X, plane))
PropBank
3
Penn English Treebank






1.3 million words
Wall Street Journal and other sources
Tagged with Part-of-Speech
Syntactically Parsed
Widely used in NLP community
Available from Linguistic Data Consortium
10/9/01
PropBank
4
A TreeBanked Sentence
(S (NP-SBJ Analysts)
(VP have
(VP been
VP
(VP expecting
(NP (NP a GM-Jaguar pact)
VP
have
(SBAR (WHNP-1 that)
(S (NP-SBJ *T*-1)
NP-SBJ
been VP
(VP would
Analysts
(VP give
expecting NP
(NP the U.S. car maker)
SBAR
(NP (NP an eventual (ADJP 30 %) stake)
NP
S
(PP-LOC in (NP the British
a GM-Jaguar WHNP-1
company))))))))))))
VP
pact
that NP-SBJ
VP
*T*-1 would
NP
give
S
Analysts have been expecting a GM-Jaguar
pact that would give the U.S. car maker an
eventual 30% stake in the British company.
10/9/01
NP
the US car
maker
PropBank
PP-LOC
NP
an eventual
30% stake
in
NP
the British
company
5
The same sentence, PropBanked
(S Arg0 (NP-SBJ Analysts)
(VP have
(VP been
(VP expecting
Arg1
Arg1 (NP (NP a GM-Jaguar pact)
(SBAR (WHNP-1 that)
(S Arg0 (NP-SBJ *T*-1)
(VP would
(VP give
a GM-Jaguar
Arg2 (NP the U.S. car maker)
pact
Arg1 (NP (NP an eventual (ADJP 30 %) stake)
(PP-LOC in (NP the British
company))))))))))))
Arg0
that would give
have been expecting
Arg0
Analysts
Arg1
*T*-1
Arg2
the US car
maker
10/9/01
an eventual 30% stake in the
British company
expect(Analysts, GM-J pact)
give(GM-J pact, US car maker, 30% stake)
PropBank
6
Motivation
 Why do we need accurate predicate-argument relations?
 They have a major impact on Information Processing.
 Ex: Korean/English Machine Translation: ARL/SBIR
• CoGenTex, Penn, Systran (K/E Bilinugal Lexicon, 20K)
• 4K words ( < 500 words from Systran, military messages)
• Plug and play architecture based on DsyntS
(rich dependency structure)
• Converter bug led to random relabeling of predicate
arguments
• Correction of predicate argument labels alone led to tripling
of acceptable sentence output
10/9/01
PropBank
7
Focusing on Parser comparisons
 200 sentences hand selected to represent “good”
translations given a correct parse.
 Used to compare:
• Corrected DsyntS output
• Juntae’s parser output (off-the-shelf)
• Anoop’s parser output (Treebank trained, 95% F)
10/9/01
PropBank
8
Evaluating translation quality
 Compare DLI Human translation to system output (200)
 Criteria used by human judges (2 or more, not blind)
•
•
•
•
[g] = good, exactly right
[f1] = fairly good, but small grammatical mistakes
[f2] = Needs fixing, but vocabulary basically there
[f3] = Needs quite a bit of fixing, usually some
un-translated vocabulary, but most v. is right
• [m] = seems grammatical, but semantically wrong,
actually misleading
• [i] = irredeemable, really wrong, major problems
10/9/01
PropBank
9
Results Comparison = 200 sent.
Correct
Juntae
Anoop
0
20
40
60
80
100
Anoop
Juntae
Correct
Bad
5
9
3
Fixable
85
67
11
Good
10
24
85
10/9/01
PropBank
120
10
Plug and play?
 Converter used to map Parser outputs into MT
DsyntS format
• Bug in the converter affected both systems
• Predicate argument structure labels were being lost
in the conversion process, relabeled randomly
 The converter was also still tuned to Juntae’s
parse output, needed to be customized to
Anoop’s
10/9/01
PropBank
11
Anoop’s parse -> MTW DsyntS
–0010Target: Unit designations are normally transmitted in code.
–0010Corrected: Normally unit designations are notified in the code.
–0010Anoop: Normally it is notified unit designations in code.
P = Arg0
C = Arg1
designations
notified
normally
code
unit
10/9/01
PropBank
12
Anoop’s parse -> MTW DsyntS
0022Target: Under what circumstances does radio inteference occur?
0022Corrected: In what circumstances does the interference happen in the radio?
0022Anoop: Do in what circumstance happen interference in radio?
P = ArgM
C = Arg0
happen
P = Arg0
C = Arg1
circumstances radio
interference
what
10/9/01
PropBank
13
New and Old Results Comparison
Correct
J2
A2
0%
20%
40%
60%
80%
100%
A2
A1
J2
J1
Correct
Bad
4.5
5
4
9
3
Fixable
60.5
85
64.5
67
11
37
10
31
24
85
Good
10/9/01
PropBank
14
English PropBank
 1M words of Treebank over 2 years, May’01-03
 New semantic augmentations
• Predicate-argument relations for verbs
• label arguments: Arg0, Arg1, Arg2, …
• First subtask, 300K word financial subcorpus
(12K sentences, 35K+ predicates)
 Spin-off: Guidelines (necessary for annotators)
• English lexical resource
• 6000+ verbs with labeled examples, rich semantics
10/9/01
PropBank
15
Task: not just undoing passives
 The earthquake shook the building.
<arg0> <WN3>
<arg1>
 The walls shook; the building rocked.
<arg1> <WN3>; <arg1> <WN1>
 The guidelines = lexicon with examples:
Frames Files
10/9/01
PropBank
16
Guidelines: Frames Files
 Created manually – Paul Kingsbury
• working on semi-automatic expansion
 Refer to VerbNet, WordNet and Framenet
 Currently in place for 230 verbs
• Can expand to 2000+ using VerbNet
• Will need hand correction
 Use “semantic role glosses” unique to each verb
(map to Arg0, Arg1 labels appropriate to class)
10/9/01
PropBank
17
Frames Example: expect
Roles:
Arg0: expecter
Arg1: thing expected
Example: Transitive, active:
Portfolio managers expect further declines
in interest rates.
Arg0:
REL:
Arg1:
rates
10/9/01
Portfolio managers
expect
further declines in interest
PropBank
18
Frames File example: give
Roles:
Arg0: giver
Arg1: thing given
Arg2: entity given to
Example:
double object
The executives gave the chefs a standing ovation.
Arg0:
The executives
REL:
gave
Arg2:
the chefs
Arg1:
a standing ovation
10/9/01
PropBank
19
The same sentence, PropBanked
(S Arg0 (NP-SBJ Analysts)
(VP have
(VP been
(VP expecting
Arg1
Arg1 (NP (NP a GM-Jaguar pact)
(SBAR (WHNP-1 that)
(S Arg0 (NP-SBJ *T*-1)
(VP would
(VP give
a GM-Jaguar
Arg2 (NP the U.S. car maker)
pact
Arg1 (NP (NP an eventual (ADJP 30 %) stake)
(PP-LOC in (NP the British
company))))))))))))
Arg0
that would give
have been expecting
Arg0
Analysts
Arg1
*T*-1
Arg2
the US car
maker
10/9/01
an eventual 30% stake in the
British company
expect(Analysts, GM-J pact)
give(GM-J pact, US car maker, 30% stake)
PropBank
20
Complete Sentence
Analysts have been expecting a GM-Jaguar pact that
*T*-1 would give the U.S. car maker an eventual 30%
stake in the British company and create joint venture
that *T*-2 would produce an executive-model range
of cars.
10/9/01
PropBank
21
How are arguments numbered?
 Examination of example sentences
 Determination of required / highly preferred
elements
 Sequential numbering, Arg0 is typical first
argument, except
O ergative/unaccusative verbs (shake example)
O Arguments mapped for "synonymous" verbs
10/9/01
PropBank
22
Additional tags
(arguments or adjuncts?)
 Variety of ArgM’s (Arg#>4):
10/9/01
•
TMP - when?
•
LOC - where at?
•
DIR - where to?
•
MNR - how?
•
PRP -why?
•
REC - himself, themselves, each other
•
PRD -this argument refers to or modifies another
•
ADV -others
PropBank
23
Tense/aspect
 Verbs also marked for tense/aspect
O
O
O
O
Passive
Perfect
Progressive
Infinitival
 Modals and negation marked as ArgMs
10/9/01
PropBank
24
Ergative/Unaccusative Verbs: rise
Roles
Arg1 = Logical subject, patient, thing rising
Arg2 = EXT, amount risen
Arg3* = start point
Arg4 = end point
Sales rose 4% to $3.28 billion from $3.16 billion.
*Note: Have to mention prep explicitly, Arg3-from, Arg4-to, or could have
used ArgM-Source, ArgM-Goal. Arbitrary distinction.
10/9/01
PropBank
25
Synonymous Verbs: add in sense rise
Roles:
Arg1 = Logical subject, patient, thing
rising/gaining/being added to
Arg2 = EXT, amount risen
Arg4 = end point
The Nasdaq composite index added 1.01 to 456.6 on
paltry volume.
10/9/01
PropBank
26
Phrasal Verbs







Put together
Put in
Put off
Put on
Put out
Put up
...
10/9/01
PropBank
27
Frames: Multiple Rolesets
 Rolesets are not necessarily consistent between different
senses of the same verb
•
Verb with multiple senses can have multiple frames, but not
necessarily
 Roles and mappings onto argument labels are consistent
between different verbs that share similar argument
structures, Similar to Framenet
•
•
Levin / VerbNet classes
http://www.cis.upenn.edu/~dgildea/VerbNet/
 Out of the 179 most frequent verbs:
•
•
•
10/9/01
1 Roleset – 92
2 rolesets – 45
3+ rolesets – 42 (includes light verbs)
PropBank
28
Annotation procedure
 Extraction of all sentences with given verb
 First pass – automatic tagging
 Second pass: Double blind hand correction
• Variety of backgrounds
• less syntactic training than for treebanking
 Script to discover discrepancies
 Third pass: Solomonization (adjudication)
10/9/01
PropBank
29
Inter-annotator agreement
Quote 100
100
Comment 92
90
Compare 91
Earn 90
Announce 87
End 84
Seem 83
Result 83
80
Want 75Fall 76
Want 75
70
Result
82 82
Approve 81 Resign
Close 80
Elect 75
Change 84
Return 73
BeginBid
7070
Cost 67
Know 61
Call 59
60
KeepSell
52 52
Leave 50
50 Buy 48
Find 61
Work 63
Hit 57
Decline 53
Climb 62
Cause 55
Add 51
Base 46
Offer 43
Name 41
40
Bring 39
See 34
Gain 29
30
20
Tell 18
Believe 11
10
0
10/9/01
PropBank
30
Annotator Accuracy vs. Gold Standard
 One version of annotation chosen (sr. annotator)
 Solomon modifies => Gold Standard
Verb
Acquire
Add
Announce
Bid
Cost
Decline
Hit
Keep
Know
10/9/01
Darren
85%
86%
90%
Erwin
50%
78%
96%
PropBank
Kate Katherine
96%
93%
99%
95%
89%
61%
96%
60%
92%
53%
89%
69%
31
Status
 179 verbs framed (+ Senseval2 verbs)
 97 verbs first-passed
O 12,300+ predicates
O Does not include ~3000 predicates tagged for
Senseval
 54 verbs second-passed
O 6600+ predicates
 9 verbs solomonized
O 885 predicates
10/9/01
PropBank
32
Throughput
 Framing: approximately 2 verbs per hour
 Annotation: approximately 50 sentences per hour
 Solomonization: approximately 1 hour per verb
10/9/01
PropBank
33
Automatic Predicate Argument Tagger
 Predicate argument labels
• Uses TreeBank “cues”
• Consults lexical semantic KB
—Hierarchically organized verb subcategorization frames and
alternations associated with tree templates
—Ontology of noun-phrase referents
—Multi-word lexical items
• Matches annotated tree templates against parse in Treeadjoining Grammar style
• standoff annotation in external file referencing treenodes
 Preliminary accuracy rate of 83.7% (800+ predicates)
10/9/01
PropBank
34
Summary
 Predicate-argument structure labels are arbitrary to a
certain degree, but still consistent, and generic enough
to be mappable to particular theoretical frameworks
 Automatic tagging as a first pass makes the task feasible
 Agreement and accuracy figures are reassuring
10/9/01
PropBank
35
Solomonization
Source tree: Intel told analysts that the company will resume
shipments of the chips within two to three weeks .
*** kate said:
arg0 : Intel
arg1 : the company will resume shipments of the chips within
two to three weeks
arg2 : analysts
*** erwin said:
arg0 : Intel
arg1 : that the company will resume shipments of the chips
within two to three weeks
arg2 : analysts
10/9/01
PropBank
36
Solomonization
Such loans to Argentina also remain classified as non-accruing,
*TRACE*-1 costing the bank $ 10 million *TRACE*-*U* of
interest income in the third period.
*** kate said:
argM-TMP : in the third period
arg3 : the bank
arg2 : $ 10 million *TRACE*-*U* of interest income
arg1 : *TRACE*-1
*** erwin said:
argM-TMP : in the third period
arg3 : the bank
arg2 : $ 10 million *TRACE*-*U* of interest income
arg1 : *TRACE*-1
10/9/01
37
Such loans to Argentina PropBank
Solomonization
Also , substantially lower Dutch corporate tax rates helped the
company keep its tax outlay flat relative to earnings growth.
*** kate said:
argM-MNR : relative to earnings growth
arg3-PRD : flat
arg1 : its tax outlay
arg0 : the company
*** katherine said:
argM-ADV : relative to earnings growth
arg3-PRD : flat
arg1 : its tax outlay
arg0 : the company
10/9/01
PropBank
38