Exposing the Fallacy of Creole and Pidgin Genesis--Cornelius DuBois SpecGram Vol CXLIX, No 1 Contents

Adaptive Heuristic Caching in Name Recall1

by Trey Jones
Evil Linguist at Large2

This paper will present a brief case study and provide data pertaining to an apparently inconsistent linguistic behavior concerning name recall. This inconsistency will be resolved by means of a novel computational explanation for the phenomenon.

Let us consider the linguistic behavior of Marilyn, a female octogenarian with five children and nearly twenty grandchildren. In general, Marilyn is a feisty woman, with a quick wit and a sharp tongue, who keeps up on current events and has opinions on everything. While her body has begun to show some signs of aging, her mind has not, and her memory is generally good. However, she does exhibit one marked tendency that would seem to indicate a difficulty with her memory: when calling one of her children or grandchildren by name, Marilyn typically has to run through several names aloud before finding and stopping at the correct one.

Careful analysis of this behavior has led to the conclusion that it is not at all random (χ2 = 31416.61413, P < 0.000001). Rather, it seems that a collection of connected name-nodes are differentially stimulated based on a combination of factors, such as frequency of visits, age, familial associations, distance of relationship, and recent use. In the absence of other factors, children's names come before those of grand-children or spouses of children, and spouses of grandchildren come last; more frequent visitors' names come up first, and siblings of those whose names have been recently used precede others. During a visit from her granddaughter Becky, Marilyn ran through the names of all three of her daughters (Marilyn's youngest daughter is Becky's mother), followed by Becky's two older sisters before stopping at the right name, as in (1):

(1)   Mary-Sally-Julie-Sarah-Tina-Becky bring me a cup of water, please.
Later, after several similar name calling events, the list changed to include only one daughter, Becky's mother, as in (2).
(2)   Come here, Julie-Sarah-Tina-Becky, and get me down a can of peaches.
Finally the list stabilized, including only the granddaughter and her mother, as in (3):
(3)   Julie-Becky hand me my cane, dear.
Shortly after Becky left, Becky's cousin Alicia came to visit. Alicia's mother is Marilyn's second oldest daughter, and Alicia is the second oldest daughter in her immediate family. She was greeted thusly (4):
(4)   Well, goodness, if it isn't Julie-Mary-Sally-Beth-Alicia!
The order here is third daughter, first daughter, second daughter, first granddaughter, second granddaughter. This is clearly not the typical order, as it was never recorded outside of this one utterance. On several other occasions when Alicia visited first on a given day, however, a more expected order of first daughter, second daughter, first granddaughter, second granddaughter was observed (5):
(5)   Good morning, Mary-Sally-Beth-Alicia, you are here bright and early.
The unexpected ordering in (4) is clearly the result of Becky's earlier visit, which moved not Becky herself, but rather her mother up in the list of names called.

For the entire history of the field of Artificial Intelligence, researchers have looked to the natural world, and particularly to the human brain, as a source of inspiration. Often (perhaps typically) direct, faithful imitation is not possible due to the complexity involved in the natural system. Such simplifications can take on a life of their own, and evolve into complex systems different from those on which the original model was based. On such a foundation are many aspects of computer science, including AI research, based.

An interesting fact is that these newer, artificial systems often excel at tasks which the long-forgotten original model did not. Consider the modern day computer, with its incredible number-crunching power that far outstrips the arithmetical abilities of most if not all humans. Conversely, the vastly parallel processing power of the human brain makes short work of face-recognition, a task that until quite recently would bring a number-crunching super computer to its metaphorical knees. I see each of these situations as the mirror of the other: one high-powered system running a necessarily lower-powered emulation of another. Thus, a human's relatively poor number crunching abilities are in fact a testament to the power of a distributed fuzzy system to perform a sharply defined linear task.

In the case of Marilyn, we can draw a computational analogy that has a good chance at psychological validity. Marilyn has, with the rich experiences that come with age, literally filled up her brain. In order to compensate, she has begun using compression schemes and alternative computational models that require more time but less space. The data structures used to hold the information about her loved ones are no longer discrete entities with a large number of pairwise relations. In order to economize on space, Marilyn has collapsed several family members into a single structure, with the result that some individuals are only defined in terms of their relationship to (i.e., as a subclass of) more salient family members. Nodes are stored in a time-dependent most-recently-used order that reverts to the default in the absence of input. Clearly the equation used is simply:

(6)   Rn(ti)  =  Dn  +  Rn(ti-1)/KT  +  Σm[rn,m·Rm(ti)]  +  KLn·(ln(Vn) - 1)

         where Dn > max(Rm) for Gn > Gm

All variables have their traditional meanings, and constants are allowed to vary slowly over time according to the usual rules. In plain and unambiguous terms, the equation tells us that Marilyn remembers people through their family relations, with preferences for more recently seen family tree branches.

Final Thoughts:
While I harbor a fear that I have been overstating the obvious, especially with the equation in (6), I feel compelled to point out yet another obvious consequence of the hypothesis presented here: The savings in space that result from familial compression and group retrieval are more than sufficient to compensate for the costs for the fuzzy-neural emulation of the linear-digital caching algorithm. Clearly some fuzzy aspects still remain, decreasing the theoretic emulation costs, while adding to the potential psychological validity of the model.

More research (and hence more funding) will be required to uncover all the details of this fascinating and intricate system.

1 This paper is made possible by a grant from NECST, the National Endowment for Cool and Spiffy Things. "NECST looks for the future in the present."© ® SM TM "We know what will be the NECST cool thing."© ® SM TM "We have too many slogans."© ® SM TM "And you have to put them all in your acknowledgments."© ® SM TM

2 Trey Jones is currently linguistically unaffiliated. He was once a Roving Linguist for the United Nations, at which time he was U.N.-affiliated. Since then he has become a freelance linguist specializing in the burgeoning niche market of providing solutions for sociopathic super-villains, helping them leverage their intellectual capital to find solutions to pressing linguistics problems associated with attempts to take over the world and/or gain market share.

Exposing the Fallacy of Creole and Pidgin Genesis--Cornelius DuBois
SpecGram Vol CXLIX, No 1 Contents