Shallow Parsing

1
GETTING SERIOUS ABOUT ? HUMOUR ?
R. Ambareesha Harin Vadodaria Mitesh Khapra Ajay
Sarda
1
2
INTRODUCTION Computational Humour

• What is Humour?
• A message that causes amusement.
• Varies from person to person.
• Depends on mood and other factors.
• Sophisticated form of human intelligence.
• AI-Complete.
• Modelling humour in all its facets is as
  difficult as solving the toughest AI problems.
  

Question Which channel does the brain
watch? Ans The Neural Network
3
INTRODUCTION Computational Humour

• Artificial Intelligence
• Successfully tackled well-defined aspects of
  human intelligence like logical reasoning.
• Focus has shifted to intelligent and even
  creative behaviour.
• Modelling humour is a challenge.
• Stages in processing humour
• Understanding the joke.
• Producing emotions of amusement.
• Moving the muscles.

4
APPLICATIONS

• Better user interfaces which are more natural and
  effective.
• Creating robots with a sense of humour
  
• Advertising.
• Helps getting and keeping people's attention.
  
  
• Therapy for children with Complex Communication
  Needs (CCN).
• Developing a full and detailed theory of how
  humour works, would yield interesting insights
  into human behaviour and thinking.

5
ROADMAP

• Cognitive View
• Theories of Humour
• Models
• Computational View
• HAHAcronym
• JAPE
• Jokes for the Challenged
• Concluding remarks
• References

6
Humour A Cognitive View

• Theories of Humour
• Models

7
BACKGROUND Theories of Humour

• Superiority Theory
• Based on the observation that people laugh at
  other people's infirmities, especially if they
  are enemies.
• Dates back to Aristotle and Plato.
  
• Release-based
• Humour is a response to repression.
  
• Incongruity-Resolution Theory
• Script-based Semantic Theory
• General Theory of Verbal Humour
  
• They are considering different aspects of humour

8
Incongruity-Resolution Theory

• Fat Ethel sat down at the counter and ordered a
  whole fruit cake.
• Shall I cut it into four or eight pieces? asked
  the waitress.
• Four ,said Ethel, I'm on a diet.

9
Incongruity-Resolution Theory

• Humour is created by a multi-stage process in
  which an initial incongruity is created, and then
  some further information causes that incongruity
  to be resolved.
• Heuristic that accounts for vast samples of
  humour.
• A joke is analysed as being in 2 main parts
• The set-up
• The punch line
• Two main variants of the IR theory
• Surprise Disambiguation Model
• Two-stage Model

10
Incongruity-Resolution Theory -Surprise
Disambiguation Model

• The setup has two different interpretations, one
  much more obvious to the audience, who does not
  become aware of the other meaning.
  
• The meaning of the punch line conflicts with the
  obvious interpretation, but evokes the other,
  hitherto hidden meaning.
• E.g.
• Postmaster Here's your five-cent stamp
• Shopper (with arms full of bundles) Do I
  have to stick it on myself?
• Postmaster Nope. On the envelope.
  

11
Incongruity-Resolution Theory -Surprise
Disambiguation Model

• Entities centrally involved in the SD account
• M1 the first (more obvious) interpretation of
  the set-up
• M2 the second (hidden) interpretation of the
  set-up
• M3 the meaning of the punch line.
• Relationships of interest
• OBVIOUSNESS M1 is more likely than M2 to be
  noticed by the reader.
• CONFLICT M3 does not make sense with M1
• COMPATIBILITY M3 does make sense with M2
• COMPARISON There is some contrastive
  relationship, even a clash, between M1 and M2.
• INAPPROPRIATENESS M2 is inherently odd,
  eccentric or preposterous, or is taboo.
• These relationships are the ingredients of IR
  humour.
  

12
Incongruity-Resolution Theory -Two-Stage Model

• The punch line creates incongruity, and then a
  cognitive rule must be found which enables the
  content of the punch line to follow naturally
  from the information established in the setup.
  
  
• E.g.
• Question What is grey, has four legs and a
  trunk?
• Ans A mouse on vacation.

• Falsified Expectations
• Incongruity of the joke's ending refers to how
  much the punch line violates the recipient's
  expectations.
  
  
• E.g.
• O'Riley was on trial for armed robbery.
• The jury came out and announced, Not
  guilty.
• Wonderful, said O'Riley, does that mean
  that I can keep the money?

13
Humour A Computational View

• HAHAcronym
• JAPE
• Jokes for Challenged

14

• FBI
• Fantastic Bureau of Intimidation!
• Federal Bureau of Investigation

15
Humorous Agent for Humorous Acronym - HAHAcronym

• Acronym ironic re-analyzer
• Reuse WordNet, Incongruity Detector/Generator
• WordNet Domains Extension to simple WordNet.
  Synsets are annotated with Domain e.g. sports,
  art, religion, technology etc..

16
Domain Hierarchy
17
HAHAcronym

• Concept of Domain Opposition e.g. Religion vs.
  Technology
• Adjectives Plays a crucial part in Acronyms
• Acronym Tweak some words - HAHAcronym
• Parse Acronym (ATN Parser) Identify head
  (remains unchanged) Identify parts to be
  modified look for substitution retain
  phonological analogy

18
HAHAcronym contd..

• ACM
• Association for Computing Machinery
• Association for Confusing Machinery!
• MIT
• Massachusetts Institute of Technology
• Mythical Institute of Theology!

19

• NLP
• Notorious Linguistic Pranks!!

20
A Punning Riddle

• Why is a river lazy?
• Because it seldom gets out of its bed.

20
21
EnglishA very p(h)unny language

• A pun is a deliberated confusion of similar
  sounding words or phrases for rhetorical effect,
  whether humorous or serious. Oxford Dictionary
• Completely Confusable text segments
• What do you get when you cross a rabbit with a
  lawn sprinkler?
• Hare spray ? (hair and hare)
• Partially Confusable text segments
• Rhyming and alliterating texts
• What does a near sighted ghost wear?
• spooktacles ?

21
22
Strategies to generate punning riddle

• Juxtaposition
• Place confusable text segments near each other
  and treat them as a normal construction.
• What do you get if you cross a dog with a
  kangaroo?
• A pooch with a pouch.
• Substitution
• Substitute one confusable segment for another
• What to you call a depressed train?
• A low-comotive (Rhyming text segments)
• Comparison
• Compare two confusable texts
• Whats the difference between money and a
  bottom?
• One you spare and bank and the other you bare and
  spank. (Rhyming and alliterating text segments)

22
23
JAPE Architecture

• Construct a non-lexicalized word/phrase - cereal
  killer
• Construct a plausible meaning for that segment -
  A crunchy murderer
• Construct a question answer pair using the
  word/phrase and its meaning
• What do you call a crunchy murderer?
• A cereal killer

Schemata Contain and assert relations between
lexical items and constructed items.
Small Adequate Description Generator Construct a
plausible description of the concept .
Templates Translate relations into Question
Answer Pairs
23
24
Knowledge Resources Hand-built Lexicon/WordNet

• Phonological information
• To determine phonological similarities or
  differences between two texts
• Semantic information
• To evoke particular concepts or words in the
  listener
• Class (lemon, fruit) --gt Hypernymy
• Specifier (lemon, yellow)
• Adjective (lemon, sour)
• Has (lemon, seed) --gt Meronymy
• act_verb (grows)
• inact_verb (squeeze)
• Syntactic information
• To use the words grammatically in the final joke

24
25
Example 1
Schema
An instance of the schema

• Lexical preconditions
• written_form (serial, serial)
• homophone (serial, cereal)
• written_form ( cereal, cereal)
• noun_phrase (serial_killer)
• component_lexemes (serial_killer,serial,
  killer)
• written_form (cereal, killer, cereal killer)
• SAD constraints
• described_by(cereal, serial_killer,
• class(Lex, murderer), spec(Lex,crunchy))
• Relationships
• describes(cereal killer,
• class(Lex, murderer), spec(Lex,crunchy))

• Lexical preconditions
• written_form (LexA, WordA)
• homophone (WordA, HomWord)
• written_form ( HomLex,HomWord)
• noun_phrase (NPLex)
• component_lexemes (NPLex, LexA, LexB)
• written_form (HomLex, LexB, NPWF)
• SAD constraints
• described_by(HomLex, NPLex, Desc)
• Relationships
• describes(NPWF, Desc)

Template spec_class (Specifier, Noun) What do
you call a Specifier Noun?
25
What do you call a crunchy murderer? A cereal
killer
26
Example 2
Schema
An instance of the schema
Lexical preconditions rhyme (WordA,
WordB) rhyme (WordC, WordD) alliterate (WordA,
WordB) alliterate (WordC, WordD) written_form
(WordC, LexC) written_form (WordD, LexD) noun
(LexC) noun (LexD) written_form (LexA, LexD,
WFAD) written_form (LexA, LexD, WFBC) SAD
constraints decribed_by (LexA, LexD,
SAD1) decribed_by (LexB, LexC,
SAD2) Relations describes (WFAD, SAD1) describes
(WFBC, SAD2)
Lexical preconditions rhyme (cute,
mute) rhyme (kitten, mitten) alliterate (cute,
kitten) alliterate (mute, mitten) written_form
(kitten, kitten) written_form (mitten,
mitten) noun (kitten) noun (mitten) written_fo
rm (cute, mitten, cute mitten) written_form
(mute, kitten, mute kitten) SAD
constraints decribed_by (cute, mitten, class
(Lex1, glove), specifier(Lex1, pretty)) decribed_
by (mute, kitten, class (Lex1, cat),
specifier(Lex1, silent)) Relations describes
(cute mitten, class (Lex1, glove),
specifier(Lex1, pretty)) describes (mute kitten,
class (Lex1, cat), specifier(Lex1, silent))
Template negcompare (Spec1, Class1, Spec2,
Class2, NP1, NP2) What is the difference between
NP1 and NP2?
26
What is the difference between a pretty glove and
a silent cat? Ones a cute mitten and the others
a mute kitten.
27
Jokes for the Challenged

• Verbal wordplay is a critical part of language
  development in children.
• They provide a structure within which words and
  sounds can be experienced.
• They offer an opportunity to practise language,
  conversation and interaction skills.
  
• Children with CCN do not always have language
  play opportunities.
• Providing access to verbal humour play for
  children with CCN will narrow the gap of language
  experience opportunities between children of all
  abilities.

28
Jokes for the Challenged -Keyword Manipulation
Task

• Designed especially for children with expressive
  language difficulties.
• Involves identifying and changing a keyword in a
  humorous item.
• E.g.
• Policeman to little boy We are looking for
  a thief with a bicycle.
• Little boy Wouldn't you be better using
  your eyes?
• Three distracters (ears, nose, hat) and an
  appropriate alternative (glasses) are
  presented as options to replace the keyword eyes.
• A correct choice from the participant indicates
  some underlying knowledge of the joke technique.

29
Jokes for the Challenged -Keyword Manipulation
Task

• A complete test consists of 10 examples of verbal
  humour, including riddles and jokes of different
  types.
  
• The questions vary in difficulty by decreasing
  the semantic differences in keyword alternates.
  
  
• Results indicate that it is possible to identify
  humour recognition in children with CCN,
• These skills probably evolve more slowly than
  with typically developing children.

30
Jokes for the Challenged -The Standup Project

• Interactive software package developed to allow
  users to play with language.
• Emphasis is on allowing the user to build their
  own novel jokes.
  
• The goal was to provide users with the means to
  construct jokes on topics, using familiar
  vocabulary, enabling them to experiment with
  different forms of jokes.
  
• Is based on the JAPE software and has an
  interface suitable for children with CNN.

31
Critique and Conclusion

• Incongruity theory and its variants form the
  foundation for computational humour
• HAHAcronym
• Rhyming texts An unnecessary constraint.
• Sometimes fails to generate humour.
• CHI Computer Human Interface
• CHI Computer Harry_Truman Interface
• JAPE
• Successful (to a great extent) in generating
  punning riddles.
• Does not generate jokes using higher level
  ambiguities like
• Contextual ambiguity
• Pragmatic ambiguity
• A long way to go
• Recognizing humour is much more difficult than
  generating humour.

32
References

• Kim Binsted, Machine Humor An implemented model
  of puns, In Proceedings of the Third Conference
  on Applied Natural Language Processing, Trento,
  Italy, 1996.
• O. Stock C. Strapparava, HAHAcronym Humorous
  Agent for Humorous Acronym, In Proceedings of ACL
  Interactive Poster and Demonstration Session, Ann
  Arbor , June 2005
• Ritchie, G., Developing the incongruity-resolution
  theory. In Proceedings of the AISB Symposium on
  Creative Language Stories and Humour, pages 78
  85, Edinburgh, Scotland, 1999.
• Ritchie, Graeme. Current Directions in
  Computational Humour. In Artificial Intelligence
  Review, 16(2), pp.119-135, 2001.
• Julia Michelle Taylor. Computational Recognition
  of Humor in a Focused Domain, Ph.D Thesis, 2004
• O. Stock and C. Strapparava, Getting Serious
  about the Development of Computational Humor,
  Proc. 18th Intl Joint Conf. Artificial
  Intelligence (IJCAI 03), Morgan Kaufmann, 2003,
  pp. 5964.
• M. P. Mulder, A. Nijholt, Humour Research State
  of Art, Technical Report, CTIT, University of
  Twente, 2002.
• OMara,D., Waller, A., Manurung, R., Ritchie, G.,
  Pain, H., Black, R. (2006) Designing and
  evaluating joke-building software for AAC users.
  In Proceedings of ISAAC, 2006.

33
THANK YOU
33
34
EXTRA SLIDES
34
35
HAHAcronym

• Resources
• WordNet Domains
• Rule Database for Domain Opposition
• Parser, Acronym Grammar
• Morphological Analyzer
• Rhyming Dictionary
• Proper Noun Database
• Opposition of Semantic Field
• Incongruity Theory
• Apparent Contradiction
• Absurdity

36
HAHAcronym

• Adjectives
• Descriptive Beautiful, Interesting etc..
• Relational Computational, Machanical etc..
• Relational adjectives Derived from Nouns

37
HAHAcronym

• Rhymes
• CMU Pronouncing Dictionary
• ATN (Augmented Transition Network) Parser
• Grammar in form of Network
• Interpreter for traversal
• A dictionary