In this blog, I will be discussing the use of attribution models in relation to codified narrative. For this purpose, I will be referring to the plots of two films: the 1974 horror classic “The Texas Chainsaw Massacre”; and a 2014 dark comedy called “Tusk.” I have my own codification system called BERLIN: this is short for “Behavioural Event Reconstruction Linguistic Interface for Narratives.” An attribution model supports the inference of meaning from data. Imagine a student one day going to an introductory statistics class and noticing all of the other students clearing their desks. He asks nervously, “Hey, what’s going on? Is there something happening?” The student closest to him replies, “Yeah, there’s an exam.” “An exam - you mean right now! I can’t believe this. I didn’t know about the exam. I haven’t studied. When was this mentioned?” I was both stupefied and petrified. However, I passed that exam. Although I considered my score terrible, I later discovered that more than half of the students had failed. In comparison to them, I actually did quite well. Exam scores can be attributed to all sorts of things. Or, thinking about this in reverse, a researcher can be in possession of numerous attribution models to help make sense of exam scores. There are always a few students that are sadly unaware of the world around them who might enter exams unprepared. Some students “miss the bus” and start their exams late. Some exhibit tremendous apathy and put little effort into their studies. Such are the complex stories that might surround exam marks. If we put aside the issue of the marks, attribution models and narratives can be close partners to help explain many things.
On my database - a prototype mentioned in previous blogs - I can determine the extent to which particular stories fit specific profiles. (Although there is no firm timetable, I will eventually make the code open-source for those interested.) The profiles might include the following: “Bones,” “Buffy the Vampire Slayer,” and the “X-Files.” Episodes of Bones have certain shared attributes. If I encounter a story that contains some of these attributes, that story might be 1) an episode of Bones; 2) a type of case that might appear in an episode of Bones; or 3) a type of case that might be investigated by the characters found in Bones. Consequently for me, attribution models serve to rapidly give a sense of context based on the available lines of codified narrative. I once said that codified narrative provides context, which I consider entirely true. However, narrative cannot do so on its own - unless a human systematically and manually reviews all of the stories. On the other hand, for a computer to systematically skim through narrative, it is necessary to have attribution models engage narrative code. Below are some of the attributes that I find reasonable for Bones. The attributes in this case have been “defined” by me due to my authoritative familiarity - since I watch quite a lot of Bones. (Every time I mention a television series in my blogs, I probably have all of the DVDs.)
Some Attributes that I Associate with “Bones”
<testimony_comparison> comparison of different narratives
<person_heroine> “Bones” or Dr. Brennan
<person_hero> “Booth” the FBI agent
<facial_reconstruction> recreating a face from a person’s skull
(My terms for the linguistic interface don’t necessarily coincide with common usage. For example, “digital extraction” on my narrative database means removal of fingers. “Physical arrest” means preventing a person from moving his or her body. A mass linguistic converter can support conversion between different databases perhaps even if the codified narrative is written in another language. Consequently, the general idea is to use terms that make sense to me rather than attempt to impose authoritative terminologies. Organizations are free to maintain their own linguistic silos.)
Each item in < > brackets is an attribute that I refer to as a “hard attribute.” Hard attributes - at the risk of causing some confusion - are scanned by the search mechanism; and a “match” means that these attributes are present in the mass data object. However, the attributes in the search mechanism are called “soft attributes.” I describe only the attributes in the mass data object as hard. Hopefully that distinction isn’t too confusing. Why do I bother making it? Theoretically, the attributes in the search mechanism can be generated by the system itself based on everything that tends to occur in episodes of Bones. Such soft attributes need not cover only hard attributes. I use the term “soft attributes” to also include similarities in behaviour and setting; these aren’t really attributes in relation to the mass data objects. However, just to keep things simple here, think of “attribution modelling” as a way of identifying phenomena characterized by specific attributional constructs. If there is any confusion, ignore the hard and soft distinction. Maybe I am providing too much information - or too little. It’s hard for me to say - or maybe soft.
By the way, I call a soft attribute that exists specifically to support a coherent attribution model an “attributional conform” (pronounced CON-form): conforms confirm attributional congruence. Not all attributes are necessarily part of a coherent model. But many can be. For example, “digital extraction” can be part of an attribution model called “Dismemberment” that might also include limb detachment, organ harvesting, decapitation, and surgical tools. If any single attribute is present, dismemberment “might” be involved. If all of the attributes are present, dismemberment seems likely or at least reasonably inferred. Seems simple and really useful, right? Well, think of systematically sifting through a large number of files where dismemberment might not be explicitly stated but reasonably inferred. The inferential attributes might be obscure. For example, “blood painting” (painting with blood) might indicate some kind of dismemberment ritual or ceremony.
Constructing an Attribution Model
Before proceeding to the horror movies previously mentioned, I want to briefly discuss how an “attribution model” might be constructed. I provide an example below of something that can be used as the basis for an attribution model. I previously posted this survey that I took of a Canadian human rights tribunal case (below): it relates to a customer service grievance against the Toronto Transit Commission. This survey is designed to model faults in perception. Within the narrative of the tribunal case, specific linkages can be made to the survey to determine whether or not a fault occurred and in what respect. Attribution therefore is not necessarily to determine whether a narrative fits an episode of Bones. One could ponder, for example, “I wonder what people (reporters, agents, or terrorists) might be drawn to a place like this (or this type of situation)?” This sort of question can be addressed using attribution models - perhaps a different model for each type of person.
If a case contains attributes that fit the survey - and the survey is designed to model faults in perception - then the attributes of a case together with the survey can be used to identify faults in perception. However, the attributional constructs of the survey (soft) and the case (hard) have to coincide - by which I mean that the processing system must be able to confirm congruence through pattern matching. Such a task is straightforward using a linguistic interface like BERLIN - straightforward although not necessarily simple, just to clarify.
There was a television series that aired from 1974 to 1975 called “The Night Stalker.” I remember watching this series as a child - although many of the details are understandably foggy at this point. Some people feel that the Night Stalker inspired the X-Files. I believe that the lead investigator was a reporter. Consider the opportunities that would now be available if this reporter had at the time the ability to store information about peculiar creatures as little “packets” of case attributes (soft). The mass data objects that my prototype uses can store the following: settings, behaviours, and attributes (hard). It is therefore possible to compare conforms against perpetrator behaviours, surroundings, and attributes. What makes this useful? A serial killer might persist for many decades - sometimes outlasting the careers of those most familiar with available cases. The killer might also travel to many different countries. It is worthwhile to have a systemic means of detection irrespective of the people present at the time in the intelligence organizations.
I want to emphasize that attribution modelling doesn’t necessarily have to involve an episode of Bones, a human rights tribunal case, or a serial killer. There are many different applications. Thinking of a business as a system where the players move in and out while performing their individual tasks, it is easy to appreciate the level of disconnection between people. This disconnection exists among those in the organization today for example between different departments; it also exists between people occupying different time periods. For example, a company encountering a severe snow storm might make a number of mistakes causing enormous loss in public confidence. A few winters later, another major snow storm might cause the same errors involving the exact same departments as if nobody had learned a thing. Such an organization is impaired from a structural-informational standpoint - unable to learn from its mistakes.
The Texas Chainsaw Massacre (Dark Sky Films, 1974)
I call the attribution model that I prepared for this film “Idol Capitalism.” I designed it specifically for this movie. Some young people off on a long drive find themselves interacting with a family of hillbilly cannibals. The largest and apparently most intellectually challenged of these cannibals chases people with a chainsaw. (A series of movies called “Wrong Turn,” the first released in 2003, exhibits a number of similarities to Texas Chainsaw.) Although I am no fitness expert, I don’t believe that a big heavy guy carrying a chainsaw would in real life be able to keep up with sprightly young girl. I noticed that this towering serial killer was wearing a suit, tie, and a skin-mask made of human face. A fabricated identity is asserted through the use of a human mask. Some attempt is made for the killer to blend in, pose, and pass as an ordinary person. I personally consider the film full of interesting symbolism. It is “abnormal” for uneducated hillbillies to kill people, dismember their bodies, and eat their flesh; yet it is “normal” for privileged people to exploit other members of society and in some respects treat them like livestock. Through Idol Capitalism, Texas Chainsaw might be regarded as a political comedy.
Attributions are added to BERLIN codified narrative using the tag “_as” at any given line of code. (By the way, “_as” may appear more than once on a line.) Consequently, one might find “_as *idol_capitalismconformance*” associated with a behaviour - and by convention <idol_capitalismconformance> among the hard attributes. The individual attributes are listed below each with a short explanation.
<idol_capitalismhumanface>: Capitalism is humanized and often portrayed using human faces. In recent decades, these faces have belonged to the “successful middle class.” The “face” is an important aspect of Chainsaw since the main killer wears a mask made of human skin. By no means am I implying that a person with facial deformities is any less human than the next person. The underlying idea here is that identity is being concealed.
<idol_capitalismconformance>: The main killer can be found wearing a suit and tie.
<idol_capitalismbymachine>: The main killer uses a chainsaw to kill people. There are some narrative parallels in Chainsaw about the slaughtering of livestock. Indeed, people are “eaten” by cannibalistic hillbillies in the film. The “machine” therefore reduces people to commodities, implying that people are born and bred to feed others.
<idol_capitalismincessant>: The main killer chases the heroine of the story and simply refuses to stop chasing.
<idol_capitalismfeerequired>: The young victims in the film were targeted only after they refused to make payment for something.
The above soft attributes would appear in a “test” file among many hundreds or thousands of other attributional tests that are systematically compared against all available narratives (the latter containing hard attributes). In some cases, some hard attributes might not be articulated as part of a coherent model at the “front-end” of processing - i.e. they do not serve as conforms. The user can create a “pro-attributional” - an attribution model that has been deconstructed to make use of vernacular descriptors: using “human_face_mask” rather than “idol_capitalismhumanface” - keeping in mind that the former provides no assurance of the latter. An attribution model might therefore be described as a system of indicators rather than a single authoritative test. There is “back-end” burden on the user after the algorithmic comparisons have been completed.
Tusk (Sony Entertainment, 2014)
Tusk contains certain aspects of a comedy although I think many people would describe it as a horror story. I find the premise so peculiar that insisting on it use to me makes the film comedic. The story involves the surgical transformation of an unwilling human participant into a creature resembling a walrus. Careful not to give too much of the plot, I will only mention that there is an epic battle among walruses. I consider the issue of “disability” and “disablement” really relevant in this film. I call the attribution model that seems to fit Tusk the “General Disablement Model v 2015” (GDMv2015) which can be applied to many different types of stories. It is the first attribution model that I designed for BERLIN - given my preoccupation with certain types of narratives.
The GDM can be broadly applied different forms of disablement including challenging environments, colonialism, bullying, stalking, predation, and abductions. In Tusk, the main forms of disablement involve the following: luring through false portrayal; drugging; forced confinement; physical injury; and this horrific process of surgical transformation into a walrus. The perpetrator exerts enormous control over another human being and, through surgical transformation, literally robs that individual of his humanity. It’s a fantastic example of disablement. The hard conforms are configured in the following: Scale value & Projection value & Association value. There are also six major “forms” of disablement that I routinely use for codified narratives: [spa_] vocal; [spa_] sensory; [spa_] mobility; [spa_] physical; [spa_] input; and [spa_] output. Just to get to the important parts, in Tusk there is a single perpetrator (S = 0); a victim but not necessarily the only one (P = 1) this being the case of serial-killer; and essentially a stranger as the perpetrator or victim (A = 4). Conforms include the following: S0P1A4_vocal since the victim could not speak afterwards; S0P1A4_mobility since the victim was trapped; S0P1A4_physical since the victim could barely move; S0P1A4_input since the victim was deprived of human necessities; and S0P1A4_output since the victim was permanently impaired in his ability to exercise autonomy. Recall that the symbol or tag “_as” is used for attributions. Consequently, one or more lines of code can be attached to attributional conforms noted below.
_as *S0P1A4_vocal*_ as *S0P1A4_mobility*_ as *S0P1A4_physical*_ as *S0P1A4_input*_ as *S0P1A4_output*
The above line is quite difficult to read; but then again it is meant for a computer more than a human analyst. Also - and I think this is important to note as far as attributional models go - the codification tends to result in ontological precision exceeding the capabilities of the language in use. The above represents a very specific shade of disablement. By the way, I have used S0P1A4 in other instances. It has been used for “possibly alien aggressor or killer.” I currently have two other cases on the database previously reviewed containing this form of disablement: the film “Europa Report” (Magnet 2013); and a case (read it here) of a lady (Dynel Lane) luring a pregnant woman (Michele Wilkins) in order to extract her baby apparently using a crude cutting device. The attack in both of these stories (the former fictional - the latter true) is on or to the body although the perpetrator doesn’t have any real attachment to the victim. The perpetrator has a need or desire that can be fulfilled by various victims. This is precisely the situation with Tusk, also. There was actually “another” walrus - a dead one - in the movie representing one of the previous victims. The disablement model implemented as an attributional model for codified narrative accomplishes its mission in exactly the intended manner. It’s pretty awesome, right. I think so.
Detecting Bodies in the Quotidian
That’s me above as a ghostly halo thanks to the ability to make bitmap modifications through my own programming. I just want to wrap things up. Imagine having global databases of codified narrative where queries can be posed through pattern matching against attribution models. Why not match against the codified narrative directly? No reason. It is possible to pattern match directly against the code. However, attribution models represent an intellectual asset that can build up in sophistication over time. Not only this, but as in my Night Stalker example some investigators might know the nature of their cases extremely well while they might have limited knowledge of the narratives that build up in national databases over many decades. The attribution model is a completely different asset than the codified narrative although they function together. I regard myself as a theoretical ontologist primarily because of my use of attribution models, which influences my recognition of phenomena in the codified narrative.
I might ask myself a question such as, “How is it possible to identify cases of child abuse in narrative that does not explicitly state that it occurred?” “How can a terrorist be found in a narrative that isn’t actually about terrorism?” These are engaging questions. In this blog, I have described an approach that is non-mathematical in nature but nonetheless highly database-oriented. As I mentioned earlier, I hope to have the opportunity to eventually offer the source-code openly. But having source-code is not quite the same as having an arsenal of attribution models and stockpile of codified narratives. This type of research probably requires a lot of humans. An attribution model is the product of creativity. Arguably, codified narrative is the product of intimate familiarity in terms of language and context. It seems unlikely that organizations would be able to “import services” from cheap sources of labour abroad. For those that would like to wield these assets, the source-code won’t cost anything - for those that accept mine. But it would be reasonable to expect investments in people for investigation, coding, and analysis. I use the term “investigation” rather than “research” to emphasize how narrative databases require the collection of stories. It is a coherent task more like investigative journalism and less like academic studies. Please continue reading on the subject, of course; this being where I fit into the scheme of things.
Pace of Development
My source-code will take some time to release (10+ years) since it is still a work in progress. This does not stop others from creating their own code, of course. I work on the code myself when I am free as the circumstances dictate. There is no timetable. Nor do I have any specific objectives. This is development by osmosis. I remember trying to get into a PhD program a number of years ago. I reminded myself that once I pass the intersection, there is little chance of backtracking. Young people are free to do all sorts of things - to experiment, fail, and try again. I have found the situation radically different for a person like myself - turning 50 this year. There is often little room for maneuvering. It is within this rather confined context that I develop attribution modelling for codified narrative - or at least my take on it. In a different world, if people were free to follow their pursuits, the pace of development would likely be accelerated. In fact, the source-code may have been released in my 30s. I therefore want to portray this situation not as a matter of personal unwillingness or inability but rather a logical outcome of the current environment. Researchers in other countries might release their own versions of this research long before I do - if they are inclined to share it.