Meaning Representations: A Practical Viewpoint

1
Meaning Representations A Practical Viewpoint

• CIS 530 Intro to NLP
• (Adapted from ongoing research efforts)

2
I. Ontonotes A corpus for literal semantics

• (Ongoing DARPA project)

3
Meaning representations should capture

• Entities
• Of some type Nation, Know-how
• Events and relations
• Predicates with arguments
• Recursively
• And More
• Quantifiers

The founder of Pakistan's nuclear program, Abdul
Qadeer Khan, has admitted he transferred nuclear
technology to Iran, Libya and North Korea.
4
Literal vs. Implicit Meaning

• A cognitive being automatically
• combines literal meaning
• with world knowledge
• to see implicit meaning
• Q Whose greed? Q Whose ambition?
• Understanding this involves infering implicit
  meaning

The founder of Pakistan's nuclear program, Abdul
Qadeer Khan, has admitted he transferred nuclear
technology to Iran, Libya and North Korea, a
Pakistani government official said Monday The
transfers were made during the late 1980s and in
the early and mid 1990s, and were motivated by
"personal greed and ambition," an official said.
5
Premise of what follows.

• Extracting literal meaning from free text is now
  possible
  
• Decoding methods now in use for syntax, etc. will
  suggest new decoding methods
  
• Different methods are necessary to replicate the
  human ability to extract implicit meaning
  
• Extracting literal meaning is a prerequisite for
  the extraction of implicit meaning
  

6
Levels of Representation
Ontonotes
7
(No Transcript)
8
Shallow Verb Semantics Propbank

• The
• founder
• of
• Pakistans
• nuclear department
• Abdul Qadeer Khan
• has
• admitted
• he
• transferred
• nuclear technology
• to
• Iran,
• Libya,
• and
• North Korea

NP
NP
PP
NP
S
NP
NP
VP
VP
SBAR
NP
S
VP
NP
PP
NP
NP
NP

• PropBank adds
• Lexical semantics of verbs

NP
9
(No Transcript)
10
Semantic Role Labeling Results (Pradhan)
ID Class
P R F1
Treebank
92 90 91
Automatic Parses
82 73 77
Chunks
72 67 70
Combined
83 75 79
11
Explicit Semantic Representation Ontobank

• The
• founder
• of
• Pakistans
• nuclear department
• Abdul Qadeer Khan
• has
• admitted
• he
• transferred
• nuclear technology
• to
• Iran,
• Libya,
• and
• North Korea

E1 Founder Names Abdul Qadeer Khan Descrip
tions The founder of Pakistans nuclear depar
tment, he
Establish Agent Org
E2 Agency Descriptions Pakistans nuclear
department
Subsidiary SubOrg SuperOrg
E3 Nation Names Pakistan
E5 Nation Names Iran
E6 Nation Names Libya
E7 Nation Names North Korea

• adds
• Representation of explicit entities, relations,
  and events
  
• Entity types and predicate frames from a formal
  hierarchy (an ontology)

12
(No Transcript)
13
Initial Wordsense Results (BBN)

• Five most polysemous verbs and nouns
• Verbs
• Nouns

14
An Ontology provides

• Key relations between concepts
• Superconcept/subconcept structure
• Links from words to concepts
• Different word senses of a word are represented
  by different concepts
  
• A Predicate/argument frame for each predicate

transfer-thing agent item
destination

15
The plan Annotators determine entities,
connect to ontology
E2 Agency Descriptions Pakistans nuclear
department
E3 Nation Names Pakistan
E4 Know-How Descriptions nuclear technology
E5 Nation Names Iran
E6 Nation Names Libya
E7 Nation Names North Korea
16
Annotators connect relations to ontology
Transfer Agent E1 Item E4 Dest E3,E6,E7
17
(No Transcript)
18
OntoNotes Integrated Shallow Semantic Annotation
and Word Sense
The founder of Pakistans nuclear department,
Abdul Qadeer Khan, has admitted he transferred
nuclear technology to Iran, Libya, and North
Korea.
19
Unified Relational DB Representation
20
Coming in OntoNotes 3.0
NW
BN
BC
550 K
Eng
200 K
200 K
Treebank
250 K
Chi
300 K
150 K
200 K
Ara
300 K
Eng
200 K
200 K
PropBank
150 K
250 K
Chi
300 K
Ara
?
300 K
200 K
200 K
Eng
Wordsense
250 K
300 K
150 K
Chi
200 K
Ara
Eng
300 K
200 K
200 K
Coreference
Chi
250 K
300 K
150 K
200 K
Ara
21
Semantic Representation
Jeff Bezos, Amazons founder
E1027 Person Names Jeff Bezos, Bezos Des
criptions Amazons founder
E13043 Corporation Names Amazon, Amazon.co
m
Descriptions The company
Bezos, was setting up Amazon.com
Jeff Bezos started the company in his garage
The semantic representation captures paraphrases
of what is explicitly stated
22
Recognizing Contradiction

• Barbara Jordan, a reporter and former news editor
  of The Chronicle of Willimantic, died Sunday of
  cancer. 1999-10-04
  
• Jordan would keep the extent of her health
  problems concealed until the day she died early
  in 1996.
  
• BARBARA JORDAN (1936-1996) Jordan was born in
  Houston.
  
• Jordan, 33, went on to become an all-conference
  outfielder at CSUN 1999-09-26
  
• But Jordan, from Agoura, realized a different
  goal last week

Birth-year(X, 1936)
Birth-place(X, Houston)
Birth-year(X, 1966)
Birth-place(X, Agoura)
Contradictions in argument values imply three
different individuals.
23
Ontobank is trivially mapped into logic
representation
A1 act Acknowledge agent E1 obj
ect Founder name Abdul Qadeer Khan descri
ption A3 act Establish agent
E1 organization E2 object Ag
ency description Pakistans nuclear depart
ment proposit
ion A2 act Transfer agent E1 theme
E4 object Know-How
description nuclear technology
destination (and E5 ob
ject Nation name Pakistan" E6
object Nation name Libya"
E7 object Nation name North Ko
rea" )
E1 Founder Names Abdul Qadeer Khan Descrip
tions The founder of Pakistans nuclear depar
tment, he
Establish Agent Org
E2 Agency Descriptions Pakistans nuclear
department
Subsidiary SubOrg SuperOrg
E3 Nation Names Pakistan
E4 Know-How Descriptions nuclear technology
Acknowledge Person Fact
E5 Nation Names Iran
E6 Nation Names Libya
Transfer Agent Item Dest
E7 Nation Names North Korea
24
Building decoders for literal meaning
Correct parse trees
Training Program
Training sentences
Current State of Art

Models
The founder of Pakistan's nuclear program, Abdul
Qadeer Khan, has admitted he transferred nuclear
technology to Iran, Libya and North Korea
Parser
25
II. Approaching Implicit Meaning

• (ARO MURI project)

26
II. SUBTLE Situation Understanding Bot Through
Language and Environment
27
Key Ideas

• We must move from robust sentence processing to
  robust utterance understanding.
  
• Must decode not only what sentence the speaker
  used, but also what the speakers intentions
  were.
  
• To achieve these goals, the communication system
  must exploit broad context.
  
• The linguistic specification should incorporate
  formal models of language
  
• which allow computationally efficient analysis
  methods
  
• so that we can study the habitability and
  effectiveness of the result
  

28
Context Urban Search and Rescue Example 1999
Worcester Cold Storage and Warehouse Fire
29
System Overview
30
Syntax TAGs!
31
Semantics Quantifier Scope

• Every boy read some book
• Not too hard to get
• ?z/boy'(z) ?y/book'(y) read'(z,y)
• Also possible
• ?y/book'(y) ?z/boy'(z) read'(z,y)

32
Quantifier Scope II

• Old Groucho Marx joke
• In this country a woman gives birth every 15 min.
  
  
• Our job is to find that woman and stop her.
• Two readings
• ? woman (? 15min gives-birth-during(woman,
  15min))
  
• ? 15min (? woman gives-birth-during(15min,
  woman))
  
• (Example from Jason Eisner)

33
Scope TAG Semantics Real example I
(Examples from Maribel Romero)
34
Scope TAG Semantics Real example II
35
Past Work Computing Presuppositions

• Example
• Q1 Which students got a grade of F in CS200 in
  Spring 2005?
  
• Nil
• Did anyone fail CS200 in Spring 2005?
• No
• How many people passes CS200 in Spring 2005?
• Zero
• Was CS200 given in Spring2005?
• No

36
Sample Dialogue (Kaplans COOP 75)

• Q Who advises projects in area 36?
• P Assuming that there are projects that are in
  area 36, who advises those projects?
  
• R I dont know of any area 36
• Q Which programmers from the ASD group are in
  superdivision 5000?
  
• P Which programmers are in superdivision 5000?
  Look for programmers that are from ASD groups.
  
• R I dont know of any ASD groups. I dont know
  of any programmers that are in superdivision
  5000.

37
Implicit Meaning

• Exchange between firefighters
• (Worcester Cold Storage and Warehouse Fire 1999)
• L-1/P2
• first answers question
• then appears to give command
• L-1/P1s implicit question How can we find you?
• Linguists call this the Question Under Discussion
  (QUD)
  
• L-1/P2 answers the QUD

38

• Central Hypothesis Effective communications
  with autonomous bots in real-time, high-stress
  situations requires that bots understand not only
  what is literally said, but also what is intended.

39
The most general QUDs go unasked

• They are inferred from the nature of the
  scenario
  
• Stretches of apparently incoherent discourse are
  linked by these QUDs

40
UMass Lowell SciFi Scenario

• Run 2
• The robot and the commander start the search in
  separate parts of the building. The business of
  the day, (standing orders) are to clear the
  building search for any bombs or weapons, find
  any injured persons and report them and their
  location so the medical team can find them, and
  tell any other persons to leave the building.
  
• 000 Commander Jr, I want you to search the room
  you are in first and when you go to leave head
  for the North end of the building, continue
  around East and we will meet up at the end of a
  long hallway.
• 032 Jr 10-4
• 313 Jr I have cleared the first room and am
  proceeding North down a hallway.
  
• 320 Commander Mark that room as clear and make
  certain to check all the rooms in the hall.
  
• 345 Jr Room is marked. Proceeding to the next
  room.
  
• 356 Commander Wait. Shouldnt you be going into
  the hallway?
  
• 402 Jr No, I went directly into the next room.
• 405 Commander Can you show me the map and mark
  your position.
  
• 415 Jr I am displaying the map, I am the yellow
  dot.
  
• 523 Commander Alright I see now, carry on.
• 529 Jr 10-4.

41
The PragBot Environment
42
Cooperation

• pragbot_chat_log_2007.11.01 AD at 19.48.20
  EDT.txt
  
• Player 2 i have 4H
• Player 1 I want it!
• Player 1 where is it?
• Player 2 should i leave it for you somewhere?
• Player 1 sure
• Player 1 where are you?
• Player 2 okay, where are you?
• Player 1 I'm near the top
• Player 2 i'm left side.
• Player 1 next to the gap near the middle
• Player 2 i'll leave the card in the upper left
  corner.
  
• Player 1 awesome

43
Where are you?

• Player 2 in the first box-like region in the
  center, 2nd row  
  
• Player 1 still at the right bottom corner  
• Player 2 at the right edge, 1/4 of the way
  down. 
  
• Player 1 I'm near the top next to the gap near
  the middle
  
• Player 2 i'm left side 
• Player 1 Pioneer valley! 
• Player 1 I'm sort of in the middle.  
• Player 2 I'm in that loop thing in the bottom
  right part of the Labyrinth Player 1 A loop? I
  am not sure what you mean. 
  
• Player 2 I am on the left at the second entrance
  from the top