8/12/2019 Computational Humour
1/18
Why Computers need to take jokes seriously? - Anintroduction to Computational Humour
KVS Dileep
September 8, 2011
1/18
8/12/2019 Computational Humour
2/18
Outline
Motivation - So what if computer does not have a funnybone?
Surface Approaches - Scratch the surface to make the
computer tickle itself.Failures - There are some jokes computers cant laugh.
Theory of Humour - Yes, Some people take humor veryseriously
Deeper Approach - Deeper Processing Better Humour
2/18
8/12/2019 Computational Humour
3/18
Motivation
Some of the applications of Computational Humour are as follows
Natural Language Systems should be able to handle humourImprove user-friendliness.Improve human-computer interaction.Develop better intelligent systems.Improve second language training.
Computational Advertising - Humourous ads are much betterreceived than normal ones.
Information Retrieval - Discard unnecessary or irrelevantinformation.
Automatic Counselling - Humour to counsel drug addicts, andpeople with suicidal tendencies.
Search for unintended humour in serious text. PreventBushisms.
3/18
8/12/2019 Computational Humour
4/18
Surface Approaches[1]
In these approaches we try to seperate the jokes fromstatements by trying to exploit features that are present injokes.
Here we try to classify the joke without making a effort tocompletely understand them.
Essentially employ machine learning approaches to classify thejokes. Now we shall look at features that can be used for thetask.
Linguists have suggested stylistic features that characterize
humorous text. They areAlliterationAntonymy
4/18
8/12/2019 Computational Humour
5/18
Alliteration based jokes
Structural and phonetic properties are also important along withcontent.(??)
Veni,Vidi,Visa. I came. I saw. I shopped.Beauty is in the eye of the beer holder.(Beholder)
Identify and count the alliteration chains in the data set.
5/18
8/12/2019 Computational Humour
6/18
Antonymy based jokes
Humour always relies on some form of incongruity, opposition orcontradiction.Identify the location of antonyms in the sentence and determinethe antonymy relation through WordNet.
Acleandesk is a sign of acluttereddesk drawer.
Always bemodestand try to beproudof it.
Money cant buyfriends, but it ensures a better class ofenemies.
6/18
8/12/2019 Computational Humour
7/18
A little long joke...
A person once was asked by his friend...How many idlies can youeat on an empty stomach? He replied,Five. No, you can eatonly one, for after one idly your stomach is no longer empty.Impressed, he went on to another friend and asked the samequestion. The other friend replied, Three. Oh! No... If you hadtold five, I would have told you a good joke.
7/18
8/12/2019 Computational Humour
8/18
A little long joke...
A person once was asked by his friend...How many idlies can youeat on an empty stomach? He replied,Five. No, you can eatonly one, for after one idly your stomach is no longer empty.Impressed, he went on to another friend and asked the samequestion. The other friend replied, Three. Oh! No... If you hadtold five, I would have told you a good joke.The point is regarding thefailure to generalize. Sometimes anobvious point is completely missed. We will now try to come at
another fundamental nature of a joke.
8/18
8/12/2019 Computational Humour
9/18
Another joke to arrive to next point
Teacher- We need to discuss about your son, Rams appearance inclass.Mr. Dasaradh: Why? What is wrong with his appearance?
Teacher- He has not made a single one in class till date.
9/18
A h j k i i
8/12/2019 Computational Humour
10/18
Another joke to arrive to next point
Teacher- We need to discuss about your son, Rams appearance inclass.Mr. Dasaradh: Why? What is wrong with his appearance?
Teacher- He has not made a single one in class till date.The joke is due to ambiguity in the meaning ofappearance.So ambiguity in natural language is what makes the humourin the sentence. So you cant escape understanding the joke.
10/18
C l J k Thi i l f ld
8/12/2019 Computational Humour
11/18
Contextual Joke- This one requires a lot of worldunderstanding
Rahul Dravid to make debut as opening bowler, clarifies MS
DhoniClearing the air over the seemingly desperate decision to recallRahul Dravid for the T20 match and the ODI series againstEngland, Indian captain Mahendra Singh Dhoni clarified in a pressconference here today that Rahul had been picked to open thebowling, rather than batting, in the absence of a pace lead. WithZaheer Khan down with a strain caused by excessive stress on theskin surrounding the abdominal region, and Praveen Kumarafflicted by a facial strain brought on by all the intense scowling
that he does while bowling, the Indian team needed an experiencedcampaigner who could add some much needed fighting spirit to theattack.
11/18
C t t l J k Thi i l t f ld
8/12/2019 Computational Humour
12/18
Contextual Joke- This one requires a lot of worldunderstanding
Rahul Dravid to make debut as opening bowler, clarifies MS
DhoniClearing the air over the seemingly desperate decision to recallRahul Dravid for the T20 match and the ODI series againstEngland, Indian captain Mahendra Singh Dhoni clarified in a pressconference here today that Rahul had been picked to open thebowling, rather than batting, in the absence of a pace lead. WithZaheer Khan down with a strain caused by excessive stress on theskin surrounding the abdominal region, and Praveen Kumarafflicted by a facial strain brought on by all the intense scowling
that he does while bowling, the Indian team needed an experiencedcampaigner who could add some much needed fighting spirit to theattack.
Can computers recognize this joke? Requires a lot ofmodeling.
12/18
I id J k Th j k i
8/12/2019 Computational Humour
13/18
Insider Joke - The joke is on us...
Figure: Courtesy: phdcomics.com13/18
Th i s f H [2]
8/12/2019 Computational Humour
14/18
Theories of Humour[2]
Cognition based theories depend on humorous stimulus.
Incongruity - Humour arises from something that violates anexpectation.
Social approaches are based on interpersonal effects.
Superiority - people laugh at each others infirmities.
Psychoanalytical approaches are based on the audience reaction.
Relief - explain the link between humour and laughter.Humour provides relief from mental, nervous and psychic
energy.
14/18
Script Based Semantic Theory of Humour[3]
8/12/2019 Computational Humour
15/18
Script Based Semantic Theory of Humour[3]
A script can be defined as a stereotypical sequence of events.
Two conditions for text to be humourous:
A text has to be compatible, fully, or in part with two differentscripts.
The two scripts with which the text is compatible are opposite.
A man walks into a bar. Ouch!!
15/18
Script Based Semantic Theory of Humour[3]
8/12/2019 Computational Humour
16/18
Script Based Semantic Theory of Humour[3]
A script can be defined as a stereotypical sequence of events.
Two conditions for text to be humourous:
A text has to be compatible, fully, or in part with two differentscripts.
The two scripts with which the text is compatible are opposite.
A man walks into a bar. Ouch!!
Customer to shop assistant, Can I try out the dress in thewindow? Assistant, Sure, but it would be more comfortable
in the changing room . Script modeling more difficult.
16/18
Conclusion is simply the place where you get tired of
8/12/2019 Computational Humour
17/18
Conclusion - is simply the place where you get tired ofthinking
Humour Detection is an interesting problem.
Lot of issues yet to be resolved - efficient models for
representing world knowledge, etc.Extend to speech and vision - funny pictures and funny way oftelling.
Humour Generation is also a good problem.
Social media humour
17/18
References
8/12/2019 Computational Humour
18/18
References
Rada Mihalcea and Carlo Strapparava.
Learning to laugh (automatically): Computational models forhumor recognition.Computational Intelligence, 22(2):126142, 2006.
Julia Taylor.Towards Informal Computer Human Communication:Detecting Humor In Restricted Domain.PhD thesis, University of Cincinnati, 2008.
Julia Taylor.Ontology-based view of natural language meaning: the case of
humor detection.Journal of Ambient Intelligence and Humanized Computing,1:221234, 2010.10.1007/s12652-010-0014-2.
18/18