disclaimer : the jokes during the seminar were generated either by ai (artificial intelligence) or...
TRANSCRIPT
Disclaimer :
The jokes during the seminar
were generated either by AI (Artificial Intelligence)
or by AI (Aaditya’s Intelligence).
The bottomline, AI is good.
Humour & AI
Devshree D Sane
08305059
devshreedsane@cse
Aditya M Joshi
08305908
adityaj@cse
Under the guidance of
Dr. Pushpak Bhattacharyya
Why Humour?
Why AI?
Why Humour & AI?
• Trust
• Interpersonal Attraction
• Stress Release
• Use Existing Intelligent Systems - humans
• Model Intelligent systems as close as possible to them
Motivation
• Computers As Social Actors• Cognitive science studies
“Humour is a powerful weapon - you can even break ice with it.”
Humour & AI
Humour theoryComputational
HumourJAPE-1 HAHAcronym
Humour Recognition
Applications of Computational
humour
Humour theory
Components
Humour Research
Challenges
• Wit
• Mirth
• Laughter
• Manner
• Humour theory• Sociological Research• Gelotology (Health effects)• Computational humour
What is humour?
• Different to different people• Different at different times
Superiority theory
Relief theory
Incongruity theory•Focus on feelings necessaryfor humour.
• Mixture of pleasure and painat the base of amusement
• Focus on effect of humour• Release of nervous energy
Theories of humour
• Gives a necessary conditionFor humour – a ‘twist’. • Humour arises from showingsomething absurd fromsomething that is not.• Based on contradiction of some sort.
Dry humour is a form of humour which is narrated as if it is not a joke at all (i.e. narrated in a serious tone, perhaps.)
Examples of jokes
Incongruity theory: "Some people can tell what time it is by looking at the sun. But I have never been able to make out the numbers."
Superiority theory: All the “blonde” or “Sardarji” jokes that are cracked.
Relief theory: The “battle-of-the-sexes” jokes
A pun in Hindi:
Sawaal: Shahrukh Khan ne ek sansthaa ko Rs.10000 ka chandaa diya. Us chande ko kya kehte hain?
Jawab: “KHAN-DAAN”.
Humour & AI
Humour theoryComputational
HumourJAPE-1 HAHAcronym
Humour Recognition
Applications of Computational
humour
ComputationalHumour
Definition
Areas
Our Focus
• Using computers in humourresearch.
• Modelling humour in a computationally tractable way.
• Humour Generation
• Humour Recognition
What is *computational* humour?
• Out of all forms, text-based / Verbal Humour• Humour in one-liners
Phonological
Morphological
Syntactic
• Same sounds, different meaning.
• Three ways:•Syllable substitutionE.g. What do short-sighted ghosts wear? Spooktacles. •Word substitutionE.g. How do u make gold soup? Put 14 carrots in it.•Metathesis (Reversal of sounds)
•Words with same surfacestructure.
E.g. : The book is read / red.
Computational Humour – Linguistic Ambiguity
• As a result of structure orsyntax of sentence.• Example: “Squad helps dogbite victims.”
A word is ambiguous if it has more than one meaning. (‘Ambiguous’ is thankfully not ambiguous. )
Humour & AI
Humour theoryComputational
HumourJAPE-1 HAHAcronym
Humour Recognition
Applications of Computational
humour
JAPE-1
JAPE-1
• Generates question-answer style puns using phonological similarities
•For example,What do you give an elephant that’s exhausted?Trunkquillizers.
Lexicon
Schemata
Template
•A set of lexemes.
•Lexeme is an abstract entity,roughly corresponding to ameaning or a phrase.
•In addition, a homonym base.
A set of relationships whichmust hold between the lexemes
JAPE-1 : Units
To produce the surface form of a joke from the lexemes
and relationships specifiedin an instantiated schema.
JAPE-1: Example
Lexicon
Schemata
Template
Lexeme : jumperSynonym : SweaterCategory: NounCountable: YesSpecifying adjective : Warm
“What will you get if you cross____ and ____?”Answer: _______
Humour & AI
Humour theoryComputational
HumourJAPE-1 HAHAcronym
Humour Recognition
Applications of Computational
humour
HAHAcronym
HAHAcronym
http://www.haha.itc.it
About HAHAcronym
Features
Examples
• European project• Humorous Agent for Humourous Acronyms.• Acronym Ironic Re-analyzer and generator
• Makes fun of existing acronyms.
• Produces new acronyms based on concepts provided by the user.
ACM :• We say: Association for Computing machinery• HAHA says: Association forConfusing machinery
Synset
WordNet
WordNet Domains
• group of data elements that are considered semantically equivalent for the purposes of information retrieval.• Eg. Person, Human, Individual
•A large database of English.
•Words are grouped into sets of synonyms (synsets), each expressing a distinct concept. •Synsets are interlinked by meansof semantic and lexical relations.
HAHAcronym : Concepts
•Augment WordNet withdomain labels.
•Example, the word ‘bank’ has two labels – Economy and Geology.
HAHAcronym : Acronym modification
Acronym parsing and
construction of logical form
Recognizes individual constituents such as NP, VP, etc. using acronym grammar.
Choice of whatto modify
and what to keep unchanged
Substitutions
1. Using semantic field oppositions.2. Reproducing rhyme and rhythm.3. Adjectives: antonym clustering and semantic relations in WordNet.
HAHAcronym : Examples of Acr. Modification
CCTT
CHI •Close Combat Tactical Trainer•Close Combat Theological Trainer
Two changes: antonym strategy for first adjective and semantic
opposition found in ‘religion' domain for tactical to theological.
Computer Human Interface.
Computer Harry_Truman Interface.
Unexpected result: due to rhyming of "human"
to "harry_truman"
HAHAcronym : Acronym generation
Input/Output
Example Input: Main concept +
Attribute
Output: A new funny acronym.
Concept: WritingAttribute: Creative
CAUSTIC – Creative Activity forUnconvincingly Sporadically
Talkative IndividualisticCommercials
Humour & AI
Humour theoryComputational
HumourJAPE-1 HAHAcronym
Humour Recognition
Applications of Computational
humour
HumourRecognition
Experiment
Strategies • Humorous data : One-liners• Non-humorous data :
• Sentences from BNC• Sentences from Reuters• Proverbs
• Stylistic features
• Content-based features
• Combination of the two
Humour Recognition
Stylistic Features
Content-based features• Antonymy( A clean desk is a sign ofcluttered drawer.)
• Alliteration( Infants don’t enjoy infancyas much as adults do adultery.)
• Naïve Bayes
• Uses joint probability of words and documents.
Humour Recognition : Strategies
Humour Recognition : N-grams
Naïve Bayes assumes independent words.
N-gram :
• Allows a computer to process ‘word sequences’.
• Uses conditional probability to predict word based on N-1 previous words.
• Allows processing of ‘multi-words’.
• Example: wild card, freak out, etc.
Humour & AI
Humour theoryComputational
HumourJAPE-1 HAHAcronym
Humour Recognition
Applications of Computational
humour
Applications ofComputational
humour
• Advertisements
• Human-Computer Interaction– Text-based Interface– Embodied Agents (Helpdesks)– Chat bots (A.L.I.C.E.)
(Artificial Linguistic Internet Computer entity)
PLEASE visit: www.alicebot.org
Applications
Human: It’s a funny thing!
ALICE: I thought it was too.
Human: Then why didn’t you laugh?
ALICE: Perhaps it is not part of my programming.
Human: So you are not programmed to laugh at all?
ALICE: Interesting deduction. Ha Ha.
Applications : ALICE
• Humour is a combination of intellect and emotions. It is subject to time, individual experience and perception.
• Humour generation produces instances of humour. Textual humour is generated in systems like JAPE and HAHA.
• Humour recognition takes help of machine learning techniques to understand the ‘humour’ content of a situation/statement.
Conclusion
A conclusion is simply the place where you got tired of thinking.
References
Humour Theory and Computational Humour: www.dcs.gla.ac.uk/~kimb/dai_version/dai_version.html
JAPE-1: • Kim Binsted and Graeme Ritchie. An implemented model of punning riddles. In Twelfth National Conference on Artificial Intelligence (AAAI-94), pages 1-6, 1994.
HAHAcronym:• An Experiment in Automated Humorous Output Production. Oliviero Stock and Carlo Strapparava. In IUT 2003, pages 1-3, 2003.
Humour Recognition:• Making Computers Laugh. Rada Mihalcea and Carlo Strapparava. In Proceedings of HLT/EMNLP, pages 531-538, 2005.
• www.wikipedia.org