Design and the Prospects of a US Military Renaissance

OK, I'm frustrated with one of the doctrine clan, or at least the latest missive from the SWJ. Its this "Design" thing. I see in it the search for a pseudo-intellectual "killer Ap" to make a career. Someone needs to explain that the difference between Heraclitean ontology and Parmenidean reasoning is that the people who work off Heraclitean ontology can get a joke and the people who use Parmenidean reasoning will laugh at the joke but will have to figure it out latter.

This is because in the General Theory of Verbal Humor (GTVH) assumes that a joke is always related with two different scripts that are opposed to each other in a special way. The theory explains that the text of a joke is unambiguous up to the point of the punch-line. The punch-line triggers a switch (a "pivot") from one script to another and directs the attention of the audience to the realization that more interpretations of the text are possible from the beginning.

Then you realize, hey this looks a lot like the way we teach people to do stuff, to model a bunch of potential outcomes and then to put in decision points (the "pivot" points from GTVH) to decide how we thing the story will end.

Wow, what a concept, they way that people get jokes is the solution to the doctrine problem without having to twist out a few greek names of ancient dead guys who we don't really know what they said.

I would be a fan of this concept of dredging up old Greek ideas for several simple reasons, such as at least is steers away from the Buddhist mysticism that crept into a lot of ideas about leadership with vapid concepts like "mindfulness" or "mindlessness". It does show the appreciation that people have not changed and the solutions can be found in the western cultural heritage but the concept is so simple and so basic it is the same as either getting a joke or not getting a joke.

To get a joke, you have to have a model, the model has to provide a set of potential outcomes. Their has to be some indicators to show that the path to the outcome has shifted at decision points from a broad set of potential outcomes to a more narrow set of potential outcomes. You can call it what you will but the I believe the GTVH is a better way of saying it than calling it "Heraclitean."

As for the Parmenidean reasoning, that is that guy who when he reads a joke has to go through the following ARTEP steps

1. Read text and make predictions about the narrative outcome
2. Without perceiving a conflict with the expectations, continue reading.
3. If the narrative conflicts with the expectations:
a. If the conflict is placed in the body of the narrative; experience puzzlement
b. If the conflict is placed at the end of the narrative; try to resolve the conflict
c. If no cognitive rule is found to resolve the conflict; experiences puzzlement
d. If a cognitive rule is found to resolve the conflict; experience humor.

I leave it up for grabs who wants to build their career with the above detailed General Theory of Humor Based Operations (GTHBO)

Shiver me timbers! A thought-provoking piece indeed.

<em>At once as far as Angels kenn he views
The dismal Situation waste and wilde,
A Dungeon horrible, on all sides round
As one great Furnace flam'd, yet from those flames
No light, but rather darkness visible
Serv'd only to discover sights of woe,
Regions of sorrow, doleful shades, where peace
And rest can never dwell, hope never comes
That comes to all; but torture without end
Still urges, and a fiery Deluge, fed
With ever-burning Sulphur unconsum'd</em>

(John Milton, Paradise Lost, Bk I, Lines 59-69)

Humor Based Operations (HBO)

Why couldnt the Effects Based Operations (EBO) Bayesian Directed Acyclic Graph (DAG) get a joke?

Answer: Because, the DAG thing could not make sense of it.

Joking is an essential part of human behavior, to make system of analytic tools based on DAGs be able to do what people do will require them to be able to understand humor the way people understand humor.

Dr. Vannevar Bush developed the idea for what he called a "Memex" in a 1945 Atlantic Monthly article (Bush 1945). Dr. Bush described his idea of how a human brain navigates through information in the mind in what he called "trailblazing." Dr. Bush said, "It operates by association. With one item in its grasp, it snaps instantly to the next that is suggested by the association of thoughts, in accordance with some intricate web of trails carried by the cells of the brain." This can be called thought of as a process where labels are used to associate semantics (meaning) to data and enable information processing based on relationships. In the process of "trailblazing" the mind creates links and associations and creates cognitive models that are constantly being updated and tested. The models are either tested serially or simultaneously in groups and sets as they are retrieved from memory or created anew. The DAGS used in EBO do none of those things since the DAGS used in EBO are relatively static and most often there is only one model, one template, one process used. The "trailblazing" mind envisioned by Dr. Bush can get a joke, an EBO DAG can not.

There have been efforts to automate humor in computers. In order for an automated system to get a joke, there has to be a theory of humor. As humor plays such an important role in human reasoning and human-like cognition and behavior, there are ample reasons to investigate such a model can include reasoning (sense making) mechanisms (Nijholt 2003).

When Apple introduced OS 9 in 1991, it included a speech recognition system that could tell jokes (Hempelmann 2006). It was a very rudimentary system that reacted to the recognition of the spoken command "computer, tell me a joke."
When the computer recognizes the command, it starts a punning knock-knock joke:

You: Computer, tell me a joke.
Computer: Knock, knock.
You: Whos there?
Computer: Thelma.
You: Thelma who?
Computer: Thelma your soul. [Sell me your soul.]

Other than the voice interface this was a simple information retrieval (IR) approach. The user makes a request and the computer provides a result selected either by sequence or randomly without any customization. Simple template-based humor generation systems are not going to lead a better understanding of natural language engineering, and much less leading us to create theories that could tell us about the mechanisms involved in human sense making (Hempelmann 2006).

Joke structure and humor theory:

The basic parts of the way people tell jokes are the "build-up," which comprises the body of the joke, and the "punch-line" which structurally closes the joke. The punch-line semantically reverses the sense the audience of the joke expects from the build-up, and creates an unexpected focus for the audiences attention. This is dependant on the placement of a "pivot" in the narrative of the joke. The "pivot" points in the narrative contain dual meanings and ambiguity that allow for the unexpected "punch-line" (Norrick 2001).

The joke therefore depends on the "trailblazing" behavior described by Dr. Bush. There has to be a "pivot" that shifts from one frame of reference to another. The new frame of reference is either a model that was retrieved from memory or a model that was dynamically created. The reason an EBO DAG cant get a joke is it is stuck in one frame of reference and cannot dynamically create a new frame of reference.

A human comedian usually remembers only the "punch-line" and a bare skeleton of the body of the joke and uses this to blend the joke into the context of the situation where the joke is told. The jokes template is remembered by the comedian as a bare skeleton of the joke. In the semantics of joke telling this is the "primary sequence" composed of the narrative clauses of the joke. The comedian then uses the data from their surroundings to customize the joke to fit the circumstance of in which the joke is told (Norrick 2001). Thereby even telling the joke uses Dr. Bushs "trailblazing" strategy to retrieve, manipulate and associate information into the more complex narrative.

As with the model of "trailblazing" where the mind has "one item in its grasp" and "snaps instantly to the next that is suggested by the association of thoughts," a narrative jokes employ competing frames of reference. These encompass the perception of a single event "in two self-consistent but habitually incompatible frames of reference." Jokes therefore evoke a clash between opposed semantic parallel scripts. An example of this can be found in the narrative jokes referred to as "one-liners" (Norrick 2001).

The most approachable jokes are the classic one-liner such as: "A panhandler came up to me today and said he hadn't had a bite in weeks, so I bit him"

This joke contains three clauses relating the story of a panhandler and a passer-by. The humor in this joke is that the request for the panhandlers plea for a handout is misunderstood as a request to be bitten by the passer-by. The first frame of reference in the narrative is that of the panhandlers plea, the second is the frame of reference of the panhandler being attacked by the passer-by. The ambiguous phrase "have a bite" constitutes the pivot of the joke (Norrick 2001).

Assuming that the joke is being read by a human computer user, the sequences of responses that are produce should be (Mulder 2002):

1. A text is read, the user make predictions about the narrative outcome
2. While the user does not perceive a conflict with the users expectations, the user continues reading.
3. If the narrative conflicts with the users expectations:
a. If the conflict is placed in the body of the narrative, the user experiences puzzlement
b. If the conflict is placed at the end of the narrative, the user tries to resolve the conflict
c. If the user can not find a cognitive rule to resolve the conflict, the user experiences puzzlement
d. If the user finds a cognitive rule to resolve the conflict, the user experiences humor.

"Incongruity Theory" most closely fits jokes that require the resolution of a conflict within the narrative. There are other theories that apply as well to the panhandler joke, such as the "superiority theory" and the "relief theory" (Mulder 2002).

The superiority theory explains humor as a product of hidden aggression where there are victims and losers. The perception of humor reflects the appreciation of the victory of the winner over the loser (Mulder 2002). Therefore any analytical system that can appreciate a joke may also be able to appreciate winning and losing.

The relief theory explains humor as the result of the relief from stress produced by the situation. This can be the relief of the audience that the cognitive rule has been found to explain the conflict in the narrative (Mulder 2002). In the case of the panhandler joke, the relief vicariously enjoyed as seeing the passer-by being able to comply with the panhandlers request in an unexpected way. Therefore any analytical system that can appreciate a joke may need also be able to appreciate stress and the relief from stress.

The General Theory of Verbal Humor (GTVH) assumes that a joke is always related with two different scripts that are opposed to each other in a special way. The theory explains that the text of a joke is unambiguous up to the point of the punch-line. The punch-line triggers a switch (a "pivot") from one script to another and directs the attention of the audience to the realization that more interpretations of the text are possible from the beginning.

Humor as communication between humans and non-humans:

Fictional computers and robots are almost always portrayed as humorless, even if theyre skilled at natural language, graceful bipedal motion, and other human behaviors that current machines find challenging (Binsted 2006). It is not surprising therefore that the analytical systems currently proposed, using DAGS, are also constructed to be humorless.

The clichéd science fiction plot-line has a malevolent automated intelligence outwitted and deactivated simply by having the human protagonist use humor to overload the sinister machine with logical conflicts that the computer can not resolve (McMahon 1989). These science fiction story lines follow the same cycle as the ancient folktale of the clever villager who traps either death or the devil up a tree (Ashliman 1987). As such the story of the evil computer foiled by a joke may be simply a reformulation of the folktale of danger overcome by jokes.

Research in the language ability of apes reflects the tendency to believe that humor is a unique human feature. The artificial language "Yerkish" that was developed as a synthetic language did not have a word for "funny" or "joke" (Glasersfeld 1974). Apes that have been taught sign language are less restricted and have been proposed to have basic humor skills and to tell jokes (Gamble 2001). Interspecies communications are, however, at risk of the humans are not actually being aware of what is being communicated and instead reflect the hope of communication and the tendency of humans for self deception (Terrace 1979).

Humor is something that some would like to restrict purely to humans while others are very willing to extend to others. Would it really be desirable to have a computer capable of humor? Jokes such as the example of the panhandler being bitten by the passer-by, is also an example of "malicious compliance." Would if be beneficial to have a computer system that was capable of "malicious compliance?"

Future benefits of having humor built into automated systems:

The presence of humor affects attention and memory, facilitates social interaction, and helps reduce communication problems. Natural communications between humans and computers may require the ability of the computer to both understand and generate humor. Humor provides insight on how language is processed. To have humor requires more than just the retrieval of standard sentences, it requires that the sentences be composed with context and with a strategy to highlight key features of what is being communicated (Binsted 2006).

Psycholinguistic research suggests that human sentence processing is both incremental and predictive, because people integrate perceptual input with linguistic and conceptual information at multiple levels of representation (Binsted 2006). The processing of humor both for the creation and the understanding of jokes suggest that the processes of lexical access, semantic association, and contextual integration are simultaneous, and employ a cascade model (Binsted 2006). This is consistent with the GTVH and Dr. Bushs "trailblazing" theory.

Templates and the humor cycles:

The creation of humor and the perception of humor in the GTVH proceed in the two linked cyclic interactions between the joke teller and the joke audience. The joke teller uses situational information and applies the template of the "primary sequence" of the joke with the goal of focusing the audiences attention on an ambiguous/conflicting feature of the situation and creates the narrative. The audience receives the narrative, perceives the conflict in the narrative, identifies the ambiguous/conflicting feature, and uses problem solving to discover a cognitive rule to resolve the conflict. Humor is then the result of the shared cognitive process to resolve the conflict in the narrative. Therefore both the generation of the joke and the resolution of the joke as humor require sense-making. Sense making requires the situation to be composed in a simulation and for cognitive rules to be discovered to resolve any ambiguity in the models.

Humor offers a model for sense-making and provides a mechanism to resolve ambiguity. A system capable of generating humor would also have to be capable of anticipating a potential outcome and making sense of an unexpected outcome.

An automated joke generating systems must have some means of gaining situational context. Once sufficient context has been achieved templates can be used to select and retrieve the jokes "primary sequence." The use of templates is facilitated by the stylized format that is used for jokes as well as identifiable "joke words."

Beyond Apples "Computer, tell me a joke" from OS 9, the starting point for computer humor is the Joke Analysis and Production Engine (JAPE). JAPE use word substitution as the main mechanism. The program substitutes phonologically identical words into noun phrases. The resulting construct is used to generate a punch-line.

The JAPE algorithm proceeds as:
1. Choose a common noun phrase from the lexicon.
2. Choose an appropriate schema, into which the phrase can fit.
3. Fit the phrases constituent lexemes into the schema.
4. Instantiate the key lexemes into the schema.
5. Choose an appropriate template, thus instantiating the characteristic links.
6. Generate the near-surface form of the riddle, instantiating characteristic lexemes in the process.
7. Check that the generated joke is neither repetitive, nor a sensible question-answer pair.

The JAPE program was designed for use with disabled children using a childrens dictionary. The lexicon and the schemas are manually inputted. The computational model used by the JAPE system is in the form of schemas and templates with rules describing the linguistic structures of human puns (Mulder 2002).

An advance on the JAPE system STANDUP (System to Augment Non-speakers Dialogue Using Puns). STANDUP produces novel jokes. STANDUP links JAPEs core concepts and incorporates it into a joke recommender system that runs on user supplied data (Binsted 2006).

Several systems have been proposed that are simple recommender systems that simply perform information retrieval for existing jokes. Systems such as Eigentaste that works with the Jester website work in this manner (Goldberg 2001).

Bryan Anthony Hong and Ethel Ong created T-PEG (Template-Based Pun Extractor and Generator) (Hong 2009) which as the name suggests is a template based word substitution algorithm. T-PEG attempts to generated puns through the automatic identification, extraction and representation of the word relationships in a template, and then using these templates as patterns generate puns.

Similarly Rada Mihalcea and Carlo Strapparavas "Automatic Recognition of Humorous One-liners" (Mihalcea 2005) uses and compares the performance of templates as text classifiers (Naive Bayes and Support Vector Machines) to detect potential jokes in text.

A more advanced method along this path is employed by Jonas Sjobergh and Kenji Araki in "Recognizing Humor Without Recognizing Meaning" (Sjobergh 2007)

The true path to automating humor requires logical mechanisms to simulate an understanding of the meaning, the context and the strategy of the joke.

The Embodied Construction Grammar (ECG) language-understanding system begins to approximate a system that will allow the computer the "get" the joke. The ECG uses two main processes: analysis and simulation. The analysis system parses each input narrative into its constituent words and syntax, using a representation of the current communicative context as well as stored pairings of phonological and conceptual knowledge known as constructions. The simulation system runs dynamic simulations of the narratives content, producing inferences and updating beliefs about the world. The ECG simulation system uses a representational formalism based on stochastic Bayesian nets (BN). The BN allow dynamic and probabilistic modeling (Binsted 2006)

Simulation is the key to humor, templates and Bayesian nets can provide a method for recognizing and retrieving the "primary sequence" and punch-line for a joke and can test its appropriateness. Only multiple simultaneous simulations and the ability to identify the "pivot points" between the simulations will make it truly possible for the computer to get the joke. Only through the use of multiple simulations can alternate frames of reference be compared as required by the GTVH.

Conclusion:

To an automated analytical system able to even begin to compete with a human operator will require the automated system to be able to employ humor in a human way. The employment of humor in a human way requires more than just retrieving something labeled as a "joke." Employing a template to the narrative only allows the probable identification of what may be a "joke." Jokes are used to communicate an ambiguity and produce a common model between whom or what is telling the joke and the audience of the joke. To be able truly "get" the joke the computer will have to understand the joke. To understand the joke the computer will have to be able to process the narrative input, create multiple models based on the narrative and be able to determine how the models intersect.

Future Research:

No systems based on Bayesian networks or any other automated data processing system currently comes anywhere close to being able to truly tell or understand a joke. Before computers and other automated system can function as a system that can truly have situational awareness to allow them to create independent conclusions on even the simplest human environment they will have to be able to tell and understand jokes. The systems used in EBO and even the systems currently proposed can not tell or understand jokes; however their rigid simple templates may be configured to compute the probability that a joke may have occurred. The Network Centric Warfare (NCW) construct also requires future work in this field. The "Power to the Edge" theory only seems to promise that those on the periphery of the network will be able to get the joke. Work is therefore needed to ensure that only the people who should be in on the joke get the joke. Further that the idea of trying to create a joke without context, so that it can be appreciated by the entire network is pointless and research in that direction is not recommended because a joke without context is not funny.

References:
Ashliman, D. (1987) A Guide to Folktales in the English Language: Based on the Aarne-Thompson Classifcation System. Westport, CT: Greenwood Press

Binsted, K., Bergen, B., Coulson, S., Nijholt, A., Stock, O., Strapparava, C., Ritchie, G., Manurung, R., Pain, H., Waller, A., and O'mara, D. (2006) Trends & Controversies - Computational Humor IEEE Expert / IEEE intelligent systems. vol 21, no. 2, pp. 59-69

Bush, V.(1945) As We May Think. The Atlantic Monthly, vol 176, no.1, pp. 101-108

Gamble J. (2001) Humor in apes. International Journal of Humor Research. Vol 14, no 2, pp 163-179.

Glasersfeld, E. von (1974) The Yerkish language for non-human primates. American
Journal of Computational Linguistics 1, microfiche 12.

Goldberg, K., Roeder, T., Gupta, D., and Perkins, C. (2001) Eigentaste: A Constant Time Collaborative Filtering Algorithm, Information Retrieval J., vol. 4, no. 2, pp. 133-151

Hempelmann, C., Raskin, V. and Triezenberg, K. (2006) Computer, Tell Me a Joke ... but Please Make it Funny: Computational Humor with Ontological Semantics. Proceedings of the Nineteenth International Florida Artificial Intelligence Research Society Conference, Melbourne Beach, Florida, USA, May 11-13, AAAI Press, pp. 746-751

Hong, B and Ong, E. (2009) Automatically Extracting Word Relationships as Templates for Pun Generation. Proceedings of the NAACL HLT Workshop on Computational Approaches to Linguistic Creativity, pp 24-31

Mihalcea, R. and Strapparava, C. (2005) Computational Laughing: Automatic Recognition of Humorous One-liners, Proc. 27th Ann. Conf. Cognitive Science Soc. (CogSci 05), Lawrence Erlbaum Associates, pp. 1513-1518;

Mulder, M. and Nijholt, A. (2002) Humour research: State of the art, Enschede : Centre for Telematics and Information Technology, University of Twente, CTIT technical report series , no. 02-34 , pp. 1381-3625

Nijholt, A., Stock, O., Dix, A. & Morkes J. (2003). Humor modeling in the interface. In: Proceedings: New Frontiers, G. Cockton, P. Korhonen, E. Bergman, S. Bjork, P. Collings, A. Dey, S. Draper, J. Guliksen, T. Keinonen, J. Lazar, A. Lund, R. Malich, K. Nakakoji, L. Nigay, R. Oliveira Prates, J. Rieman & C. Snyder et al. (eds.), Association for Computing Machinery, New York, pp. 1050-1051

Norrick, N. (2001) On the conversational performance of narrative jokes: Toward an account of timing. Humor vol 14, no 3, pp255-274.

Sjobergh, J.and Araki, K. (2007) Recognizing Humor without Recognizing Meaning. In Proceedings of the Workshop on Cross-Language Information Processing. pp. 469-476.

Terrace, H., Petitto, L., Sanders, R. and Bever, T. (1979) Can an ape create a sentence? Science 206: pp 891-902. [ACMH]

Given that these ideas are about the philosophical foundation of officers, how can these ideas be integrated at the basic formative stage of officer training, eg, ROTC?

I ask as an advocate for ROTC at Columbia University. I think a new CU ROTC program would be a special opportunity to create an officer program that draws upon top-quality students, world-class university and global city resources to produce officers who are uniquely equipped to further COL Paparone's "renaissance" and carry forward the full-spectrum responsibilities of the modern American military officer.

What would such an innovative ROTC program at Columbia look like?