The Social Organization of the Domestic Dog;
A Longitudinal Study of Domestic Canine Behavior and the Ontogeny of Canine Social Systems
By Alexandra Semyonova, copyright 2006, All Rights Reserved

Abstract

   The theory that a hierarchy based on dominance relationships is the organizing principle in social
groups of the sort canis lupus is a human projection that needs replacing.  Furthermore, the model has
unjustifiably been transferred from its original place in the discussion of the behavior of wolves to the
discussion of the behavior of domestic dogs (canis familiaris).  This paper presents a new, more
adequate model of how familiaris organizes itself when in groups.  This paper is based on a longitudinal
study of a permanent group of five randomly acquired dogs living in their natural habitat, as they interact
with each other within the group, with newcomers of various species who joined the group, and with
fleetingly met individuals of various species in their outside environment.  This study shows that the
existence of the phenomenon "dominance" is questionable, but that in any case "dominance" does not
operate as a principle in the social organization of domestic dogs.  Dominance hierarchies do not exist
and are in fact impossible to construct without entering the realm of human projection and fantasy.  The
hypotheses were tested by repeatedly starting systems at chaos and observing whether the model
predicted the evolution of each new system.  The study shows that domestic canine social groups must be
viewed as complex autopoietic systems, whose primary systemic behavior is to gravitate as quickly as
possible to a stable division of the fitness landscape so that each animal present is sitting on a fitness hill
unchallenged by other group members.  Aggression is not used in the division of the fitness landscape.  It
is not possible for an observer to measure the height of respective hills.  There is no hierarchy between or
among the animals.  The organization of the system is based on binary relationships, which are converted
by the agents as quickly as possible from competitive to complementary or cooperative binaries, through
the creation of domains of consensus.  The production processes by which this is done are twofold. The
first is an elegant and clear, but learned, system of communicative gestures which enables the animals to
orient themselves adequately to each other and emit appropriate responses in order to maintain or restore
the stability of their fitness hills and the larger social landscape.  The second is learning.  It is the learning
history of each animal, which determines how adequately the animal can operate within the system and
what the components of its individual fitness hill will be, and which, in the end, is more crucial to the animal’
s survival than even presumed genetic factors or some human-constructed “dominance” position.  


This resource was first published as:

Semyonova, A., 2003, The social organization of the domestic dog; blowing up the dominance myth, The
Carriage House Foundation, The Hague.

This resource should be cited as:

Semyonova, A. 2003, The social organization of the domestic dog; a longitudinal study of domestic canine
behavior and the ontogeny of domestic canine social systems, The Carriage House Foundation, The
Hague, www.nonlineardogs.com , version 2006.


Introduction

   The theory of a linear hierarchy based on dominance relations, originally developed from observations of
ants, was one of the first models used in ethology to describe or account for the behavior and the social
structure of wolves and the groups they live in (Mech 1995, 2000; Sax 1997).  The dominance hierarchy
model was adopted by others to explain the behavior of canis familiaris, and is still broadly in use today
among both scientists and laymen who deal with domestic canine behavior.  
   This model as applied to wolves was from the beginning, based on dubious evidence (Mech 2000).  
Furthermore, it has been shown that throughout history humans have modeled the animal kingdom in
ways analogous to the societies humans themselves were living in (Dahles 1993; Darnton 1985; Evans
1994), and that perceptions of scientists are influenced by their belief systems and the need to protect
various kinds of investment (Kuhn 1962; Pernick 1985; Phillips 1993; Rollin 1989).  The dominance
hierarchy model was developed in a period in which many human societies were struggling with
authoritarian forms of government and the culture and ideologies that form of government propagated
(Deichmann 1996; Sax 1997).  Its spread continued in a post-war world in which a competitive market
economy and its ideologies, based on a selective and flawed interpretation of Darwin's theory of natural
selection, shaped a new generation of humans' perceptions of natural reality.  A final factor is that the
model was developed in a period in which there were very few women involved in scientific research. This
means that a limited group of existential repertoires and paradigms was used as background in the
search for explanations of observed animal behavior.  For example, it is now widely known that human
males and females differ from a very early age, with males displaying largely competitive behavior in
groups even before they develop verbal skills, while females tend at the same age to show cooperative
and appeasing strategies in dealing with group membership.  This raises the question of whether
observations were not biased in advance toward perceiving mostly the competitive elements of any
observed social system.  It is also a widely investigated psychological fact that the first thing human males
do when two or more of them have to share a physical space is investigate and order their relative power
relations within the (fleeting) group.  The conclusion that dogs are equally preoccupied with establishing
“dominance” in their social interactions is most likely a failure of imagination.  Unable to conceive of any
other way of organizing a group, scientists seem to have projected their own existential paradigm onto the
animals they were observing.  Secondly, most of the legal and illegal violence in human societies is
committed by males.  This raises the question of a third observer bias, namely the possible tendency to
give more weight than is really justified to seemingly violent encounters between observed animals, and
thus the model's focus on what it calls aggression.  Finally, it has also been shown that women make the
scientific effort in a different way from men, less career oriented, more interested in fundamentally sound
and thorough research (Kollantaj 1982; Holton 1998; Sonnert 1998a, b).  This, combined with the proven
tendency of humans, including scientists, to impose their own existential paradigms in modeling the world
around them, suggests that the model contains, besides a cultural bias, a gender bias in its perceptions
and models of the behavior of other species as well as the methods by which models are developed.  
The use of statistical analysis to prove the existence of and unravel dominance hierarchies does not
provide a solution to any of the above mentioned biases.  These analyses begin based on definitions
derived from the dominance hierarchy model, counting only behaviors that already fit the model, and are
therefore unable to do anything but recirculate and affirm the model on which they are based.  A reanalysis
of data often shows no hierarchy, or a different hierarchy emerging when group members  are isolated and
then reassembled in the same group; shifting patterns over time remain unpredictable (Dickey 2002;
Dickey et al 2002a; Mesterton-Gibbons 1999).  Patterns found also shift when delineations of behavior to
be counted are changed (e.g. resource holding rather than "wins" in exclusively aggressive agonistic
conflicts).  The definitions themselves have been sloppy.  What is a "win"?  What is a resource?  How are
resources designated as such, and by whom?  This point has been missed, and this hiatus seems to
have led to a search for other, presumably better, statistical approaches.  However, the measure of a
"better" statistical approach again derives from the model itself, since a "good" technique is one that turns
out to predict dominance relations as perceived and defined by researchers (e.g., Dickey, et al 2002b).  
Where no stable hierarchy is found, the statistical method and not the model itself is presumed to have
failed.
   In addition to this closed loop problem, superstition seems to play a role in the use of statistical
analysis.  Experimenters use statistics to assert that results obtained and conclusions drawn from
observing a group of ten to forty animals have the same validity as a wide population study would.  Even
where the groups observed are larger, the animals are treated as static beings living in some permanent
observable state.  The statistical approach seems intended to sidestep instead of facing the complex
problem of the individual learning histories of the organisms studied in order to draw broad conclusions
from small group studies.  However, the ability to learn is such a critical part of the manifestation of life in
vertebrates that any study which tries to exclude the effects of this ability invalidates itself a priori as relevant
to understanding the life of these organisms.  A second superstition is that pulling data through a
statistical program will, all by itself, solve the problem that the observer is inevitably interpreting what s/he
sees (see below: The super-observer).
   But it was, in the first place, not correct to apply a model developed watching wolves to the behavior of
domestic dogs.  Though it has become clear that domestic dogs share a common ancestor with the wolf,
the genetic similarity is a weak basis for assuming that models derived from the studies of wolves are
applicable to the behavior of domestic dogs or adequate to understand, explain or predict the behavior of
the latter.  It is clear that wolves and domestic dogs have occupied strongly differing natural habitats at
least since the human agricultural revolution and probably long before that.  Therefore, factors relevant to
the survival of both individuals and the species themselves will strongly differ.  In light of this, it seems
reasonable to propose that the behavior of wolves and domestic dogs may differ as much as the behavior
of chimpanzees and humans do.  Furthermore, the relationship between genotype and phenotype is not
well understood, let alone the role the genotype plays in determining discrete behaviors of an organism
during the course of its life.  Finally, selection operates on phenotypes, not on genotypes.  Thus, we must
at least for the moment assume that the observation of phenotypes and their behavioral ontology in the
course of their lives in the particular environment they inhabit is critical in understanding any species.  The
dominance hierarchy model and many conclusions about wolf behavior have nevertheless been
transferred to domestic dogs.  At the same time, domestic dogs, like wolves, have rarely been studied in
their natural habitat.  In fact, the domestic dog's natural habitat is near or among humans and all the
species humans live with, yet most studies take place in laboratories or in animal shelters, and none
encompass the entire life-span of multiple animals in their natural habitat.      
   The transfer of the dominance hierarchy model as it has been derived from observations of wolves to
domestic dogs, and attempts to preserve the model in the face of all contradictory evidence, have led to the
model becoming an inelegant chaos in its new application, one full of contorted appendages and internal
contradictions.  To name a few:
  • Dogs are, like other domestic animals, described as neotenic.  The fact that wolf pups do not
    organize themselves around a leader is then difficult to reconcile with the assertion that domestic
    dogs do.  
  • It has, for example, been shown that domestic dogs exhibit different play behavior when interacting
    with humans as opposed to conspecifics (Rooney 2000), yet the assumption is still made
    throughout the literature that a domestic dog sees its human owner as some sort of dog which it
    will try to “dominate”.  This ignores the fact that domestic dogs are able to organize their social
    groups to include other species despite vast differences between the species involved, and does
    not address the question of how they do that.  
  • the fact that most social conflict does not involve the presumably dominant animal, and that in fact a
    presumably dominant domestic dog often will defer to a presumably subordinate domestic dog in a
    conflict, has led to the hypothesis that “dominance” (also referred to as a “high rank” in the
    “dominance hierarchy”) brings some sort of zen tolerance with it.  This ignores the dominance
    hierarchy model's own designation of the “dominant” animal as, by definition, the animal that can
    most use aggression with impunity and/or the animal that "wins" the most aggressive encounters.  
    In an effort to solve this problem without abandoning the model, some researchers have shifted the
    definition of dominance to mean the animal most successful in resource holding.  What is meant
    by "success" has never been satisfactorily defined, partly because what is meant by "resources"
    has been too narrowly limited and too observer dependent.  
  • Wolves spend most of their lives in stable family groups which normally do not incorporate
    outsiders (Mech 1995, 2000).  Domestic dogs form loose, temporary groups (Beck 1973, 1975;
    Rubin 1982; Scott 1973) and/or interact fleetingly with each other during outings with a human
    caretaker. It seems ridiculous to talk about a linear dominance hierarchy in such a group of dogs,
    the composition of which may change from minute to minute as the dogs interact on a city field,
    since rank, which is a statistical construct, can only have meaning within in a group that stays
    together long enough for a statistical pattern to emerge.  
  • This construct has been reified.  It assumed that when two dogs meet and execute the ritual
    greeting and “bluffing” ceremony, they are establishing "rank" with respect to one another.  It is
    furthermore assumed that the animals themselves have some sort of consciousness of their
    respective ranks as such (Askew 1996; Overall 1997, 2002; Voith 1982; Voith & Borchelt 1982;
    Borchelt & Voith 1996; Fisher 1998, 1991; Neville 1993; Trumler 1972; Lorenz 1995), despite the
    fact that there is still much dispute regarding the possibility of cognitive content in animal
    consciousness.  This assumption is probably another attempt to salvage the model, which, as it is
    applied to domestic dogs, can only be maintained if various assumptions are made about
    consciousness in the species and the contents of that consciousness.  
  • The treatment of domestic dogs with presumably dominant aggression problems is always based
    on the principles of respondent and operant conditioning (Askew 1996; Overall 1997, 2002; Voith
    1982; Voith & Borchelt 1982; Borchelt & Voith 1996) and often includes the use of anxiolytic
    medication (Overall 1997, 2002; Borchelt & Voith 1996).  Yet the causes of this aggression are still
    sought in presumably instinctive behavior as dictated by the dominance hierarchy model and the
    owner's failure to establish "dominance" over the dog (Askew 1996; Overall 1997, 2002; Voith 1982;
    Voith & Borchelt 1982; Borchelt & Voith 1996).  The contradiction arises that high status is
    presumed to create the self-assurance and tolerant calm needed to explain the lack of aggression
    in a "high ranking" animal, yet a dog whose aggressive behavior requires the use anxiolytics is also
    diagnosed as "dominant aggressive", i.e., that the animal perceives its “rank” as high with respect
    to humans.  
  • The dominance hierarchy model violates the rules of parsimony.  No broad, comparative study has
    been done, for example,  comparing the incidence of "dominant" aggression between domestic
    dogs raised and trained using positive reinforcement and those raised and trained using negative
    reinforcement and punishment techniques designed to elicit avoidance behavior.  This should have
    been done before behavior was attributed to an internal, inherited need to operate within a
    “dominance hierarchy”, presuming the animal is thinking about "rank" in its interactions with
    humans, etc.
  • Statistical analysis has shown that food guarding behavior in domestic dogs correlates with the
    development of "dominant" behavior toward humans (Overall 1997).  Thus, it is assumed that food
    guarding is an early sign of a "dominant personality" in a dog (Overall 1997).  In fact, this correlation
    is a result of operant conditioning.  In guarding food, aggression may be reinforced by the other
    (human) animal's withdrawing to a greater distance.  The reinforced behavior emancipates itself
    and, reinforced in other situations, becomes a generalized behavior.  This is an example of the way
    in which statistical analysis can produce trivial information and serve to mask rather than reveal the
    mechanisms which are, in fact, operating.  It is also an example of how a model, once adopted as a
    persistent belief, can act as a filter distorting perceptions to the point that observations lose all
    value and enter the realm of fantasy.   

   These are just a few examples of the problems that arise when an attempt is made to understand
domestic canine behavior and social structure by applying the dominance hierarchy model to domestic
dogs.  In short, the model creates more conceptual and practical problems than it solves, forces its users
into the realm of fantasy and projection, and has required the addition of inelegant appendages and
various contortions of thought – yet the problems and contradictions remain.  I assert that the only thing we
can do with this model is discard it.  Dogs are neither wolves nor male humans.  If you want to know about
dogs, you must observe dogs, not wolves.  Not only that, but you must also study the animals in their
natural habitat (near or among humans and all the species humans live with), and your study must
encompass at least one entire life-span of multiple animals in their natural habitat.  This paper is the result
of exactly such a study.
   In this paper I present a model based on non-linear dynamics and work on self-organizing autopoietic
systems as these theories have been developed to deal with systems in which competition takes place.  L.
David Mech has already presented a model in which he describes wolves as role-oriented rather than
dominance- or status-oriented (Mech 2000).   I will contend that this model is also relevant to canis
familiaris.  Mech’s model describes, however, only one of several mechanisms for part production in a
larger, autopoietic system, since domestic dogs, unlike wolves, live in two essentially different types of
group during their lives: family groups and stranger groups.  This autopoietic system and its parts gravitate
toward stability and predictability, choosing at any given moment between multiple optima in the fitness
landscape.  The learning history of  each animal is essential to the emergence and successful operation
of the social system.  This model is adequate to explain all observed interactions of canis familiaris, with
conspecifics and with other species.  I will also describe the mechanism by which discrete events on the
ground generate a social structure and by which they lead to predictability in the social behavior and
interactions of individual dogs within their social environment.  


Theoretical framework


Complex self organizing systems and autopoiesis

    In order to deal with the complex interaction of multiple variables at many levels or organization
simultaneously, I refer to theories about complex self-organizing systems (Beckerman 1999; Lucas 2002
a&b), and to H.R. Maturana and V. Varela’s work on autopoietic systems (Maturana 1975; Maturana &
Varela 1980; Varela 1981).  A brief review follows here.  Those who are not familiar with these scientific
paradigms can click on the links in the title of this section to get to pages in which extensive explanation is
given.  
      When we look at a dog, what we are really observing is a creature that is a discrete and complex living
system in itself, composed of many smaller systems (e.g., cells and their cellular organs, tissues, organs
such as brain and heart).  All of these smaller systems have an effect on the behavior of the whole that we
are observing under the name “dog”.  A dog’s perceptions and reactions will vary according to what we call
its inner state, depending on all kinds of ever changing factors within its own internal system.  The balance
is always shifting, and a dog will be juggling many internal variables as it tries to maintain some kind of
equilibrium inside itself as a system.  At the same time, this system we call “dog” is situated in an external
environment.  Events in the external world can trigger changes inside the dog as a system.  The dog will try
to restore some kind of internal equilibrium, but as it does this it will have to take the outside world into
account.  External factors may limit the choices a dog has.  Interaction with the external world is not a one
way street: the dog’s behavior will, in many cases, trigger a change in the outside world: output returns to
the dog as input.  As a dog seeks internal balance, it will often simultaneously have to control how its
output affects the external world and find ways not to disturb an equilibrium in parts of that external world.  
Thus, behavior is not a result of static traits, but of a complex interaction of internal and external variables
and processes, and of several levels of organization changing and having to be managed all at the same
time.  In other words, in studying the organism called “dog”, we are focusing on only one level of
organization in a multi-level system, while the dog deals with all those levels at once.  All these levels of
organization affect and are constantly being affected by all other levels (cellular<-> organic <-> dog <->
social system <-> habitat).  While changes at one level neither cause nor specify the changes at another
level, they constitute perturbation on other levels; the system may, at any level, seek a new attractor to
accommodate the change.  A description of any one level of a system (e.g., the dog) must try to find and
include at least the relevant perturbations that originate at other levels (e.g., a social system) as the
organism attempts to juggle optimization on various levels simultaneously.  
In order to understand how this works, we must first clarify a number of theoretical terms as they are used
in reference to complex self-organizing systems.
   In this context, “fitness” has a different meaning than in the theory of evolution.  “Fitness” refers only to
criteria contained inside the system we are talking about.  A fitness landscape is a distribution of optima
the system can choose from: we call these optima “fitness hills”.  These are positions in the system’s
state space that result in best meeting criteria internal to the system (in this case, the dog), given the
options available at a particular moment.  Moving between these optima (or fitness hills) involves the
balancing of many variables and is relevant to a current niche rather than to any imposed static function or
criterion.  There may be several more or less equally preferable optima available at any one time, again
depending on the combination of conditions prevailing at that moment.  The system itself constrains the
options available and chooses between them referring only to internal factors; fitness is determined with
respect to internal criteria.  In complexity theory, fitness can include factors affecting the “quality of life” as
perceived by the organism.  Where an animal can’t optimize all factors at once since they affect each other,
a compromise between factors may be the best route to the best practically possible solution: this is
epistasis (Lucas 2002c).
   “Selection” refers to the system choosing between equilibriums in reference to external factors.  
Selection may then operate on various self-organizing systems according to their emerging phenotypic
states (Lucas 2002b).  External selection pressure operates as a force on the system to perturb it to
migrate to a different attractor, but it does not determine how the system, once perturbed, will behave, nor
which attractor will be chosen.  The system will move through its state space referring to internal rules and
according to internal processes.  Although a self-organizing system refers to internal processes and rules,
the system may relate to its environment and must move within that environment.  This is always the case
with living organisms.  This is called situated self-organization (Lucas 2002b).  The internal structure of a
system may become coupled to relevant features of that environment.  This is structural coupling.  In a
system that is structurally coupled with its environment, perturbations can come from within the system or
from outside the system.  The environment becomes a factor in selecting which of the attractors available
to the system will be chosen at a given moment, although the choices are still made according to internal
rules and constraints.  Structural coupling can operate at various levels.  In systems in which the parts are
living organisms, it can operate on the level of the organism as a system in itself.  Structural coupling can
also operate on the level of the behaving, whole organism; it can operate on the level of a larger system of
which the parts are themselves living systems.
   A self-organizing system consists of parts that may themselves be autopoietic systems.  The parts are
non-equivalent and may obey different laws.  System parts can be interconnected in various ways.  An
interconnection means that system parts are positioned in space and time in such a way that a transfer of
energy, matter or signals takes place.  A signal is defined here as any change in or output from one part of
a system that would allow other parts of the system to change their orientation or output in an interaction.  
Interactions between these parts refers to the influences parts have on each other due to their
interconnections.  This influence can trigger a change in one or both parts’ internal state, and a signal or
perturbation can percolate in some way to other parts.  Interactions between the system’s parts must be
net positive sum to be sustainable (Lucas 2002b).
   A final note: the author uses the term “binary” to indicate a pair of interacting organisms, whether the
interaction is fleeting or results in a two-part system that continues to exist over a long period of time.  The
term is common in the literature concerning self-organizing systems.  The word “dyad” is specifically
rejected due to the unavoidably observer-dependent definition of what a “significant” sociological
relationship would be.


Learning

   Each dog has an internal fitness landscape, a sort of map of all the possible system states, some of
which provide more opportunity for success than others with respect to various parameters and system
functions.  This fitness landscape shows the rating of each option in terms of some attribute or
achievement that pertains to the system's optimal condition according to its own internal criteria.  Higher
hills represent preferred states and lower hills the less preferred states.  An organism or any other system
will migrate between fitness hills, trying to optimize its position on this fitness landscape.  In this paper we
are dealing with organisms capable of learning.  This means that on the level of the organism, the fitness
landscape and the migration across it will be affected by the dog's history.  This history will play a role in
the animal's mapping of the heights of various fitness hills, and in determining the choice whether or not to
attempt an adaptive walk to a higher peak, the tactics an animal will use to achieve this migration, and the
routes it may choose from.  If it does decide to migrate, a dog must have a plan.  A plan is any attempt to
predetermine the trajectory of an adaptive walk across the fitness landscape.  This involves predictions
about the animal's own actions, about the actions of others in the system, as well as how these actions
will affect the landscape and the system as a whole.  Lack of precise knowledge of starting conditions,
input and/or other agents in the system can be responsible for lack of precision in predictions.  New
information may effect the plan and the trajectory, as well as changing the fitness landscape.  Beliefs (in
behavioristic terms: learned associations, conclusions about prevailing contingencies, superstitious
learning) can also effect strategy.  (Beckerman 1991, passim)  Thus, a dog’s movement within its state
space is a result of learning as an ongoing production process.  
   In a complex self-maintaining systems at points of instability and at points far from equilibrium, new
forms of order are generated, which lead to higher levels of organization and increased diversity (Capra,
1996).  The system moves away from chaos and towards one (or more) attractor(s).  The processes which
lead complex systems to and from various stable states, or provide for maintenance of self-maintaining
systems and their parts, are called production processes.  These can be purely mechanical, chemical or
physical, involving the movement of matter, energy or information, or can, in the case of living organisms
include mental processes such as self-generated thoughts, perception, learning and cognition.  Learning
is a critical production process in the autopoietic system “dog”.  A domestic dog is a complex, self-
organizing, structurally plastic system.  Learning is a production process by which the structure of the
animal as a system changes through time (e.g., changes in brain structure and the organization of
electrical patterns within in the brain, in available internal representations of the environment, in
metabolism, in behavior of both subsystems within the dog and of the dog as a whole system).  This
affects both the animal’s ability to maintain itself as a living system and its ability to conceive plans that
enable it to participate in the organization at the next level, the social system.  A number of learning
processes are well understood.  Before we engage in superstitious behavior and posit some mysterious
internal property (e.g., “dominance”) of the animal as a cause of behavior, we should make a rigorous
attempt to understand what learning process and contingencies could result in behavior we observe.  In
fact, these learning processes are adequate to explain and predict many changes that take place over time
in an organism as a system and in its relation to its environment.  
   Learning is also included in the model presented in this paper as a crucial production process in the
larger social systems domestic dogs occupy.  It is through learning that these systems are generated and
structured.  This inclusion of learning allows us to model dogs’ behavior and the organization of their
social systems without making any presumptions about genetic determination of their discrete behaviors,
about the content of their consciousness or about how dogs learn beyond those processes that
behaviorists have unraveled, yet without excluding the possibility of cognitive processes which may in time
be proven to take place.  
   Although the various aspects of consciousness and cognition in animals are still being researched, it is
sufficient here to include only discoveries made by Pavlov, Skinner and Sidman and some of the insights
further experimental research has provided based on their work (Skinner 1938, 1953, 1969; Sidman
1989).  In particular the following behavioral phenomena have been mistakenly interpreted as “dominant”
and/or “submissive” behavior in domestic dogs: spontaneous recovery, the extinction burst, extinction
aggression (Azrin et al 1966; Kelly & Hake 1970; Hutchinson et al 1968), learned aggression (Baisinger &
Roberts 1972; Powell et al 1972), avoidance behavior (Sidman 1989), behavior produced by a variable
reinforcement schedule (Skinner 1938), behavioral depression (Sidman 1989), and superstitious behavior
(Skinner 1938, 1953, 1969; Sidman 1989).  Though an internal change may operate as a perturbation that
triggers movement of the dog as a system within its state space, it is learning that will determine which
behavioral choices the dogs perceives as available, its valuation of the various choices and outcomes, and
the strategies the dog perceives as available for moving along a certain trajectory.  In other words, it is
learning that determines which path a particular dog moves upon as it seeks a new internal equilibrium in
response to a perturbation.  Not only do the concepts of “dominance” and “submission” assume that the
contents of a dog’s consciousness are similar to the contents of human consciousness, they are also
unparsimonious and inelegant, and are entirely superfluous in explaining the behavior of the domestic
dog.  Respondent and operant conditioning are sufficient to explain the behavioral responses on the
organizational level of the system called “dog”, and to explain the emergence and maintenance of a larger
canine social system.  


Competition

   At the level of the social group, we are dealing with discrete living systems that are parts in one or more
larger systems consisting of one or more other animals.  Not only will each individual animal's actions
change its own fitness landscape, but it will also affect the fitness landscapes of other agents in any larger
system it is part of.  This means that competition between animals may occur, since no two dogs can have
the same ball or eat the same bit of food simultaneously.  In a competitive situation, stability can be
described as a state where each agent is better off just sitting on its present attractor so long as the other
does, or others do, the same (Kauffman 1996; Beckerman 1999).  In this case, stability also means
resistance to change.  Each agent will prefer to stay where it is, neither improving its position nor allowing it
to deteriorate.  Each agent may, if necessary, somehow defend or try to preserve its position on its own
fitness landscape in the face of perturbation.  But the animals are also participants in a single larger
system together, and they are all structurally plastic.  If the agents in a system selfishly optimize their
position on the fitness landscape with little regard for other units, their lack of coordination will result in a
wildly fluctuating, uncoordinated fitness landscape for the group as a whole (Beckerman 1999).  This
means that organisms that are part of a system may choose stability of the system they occupy or some
other form of epistasis above the maximization of their own positions.  Many factors may operate as
variables in the determination of an animal’s fitness landscape and in determining its choices as to
whether, and if so, how, to migrate across the landscape.  Competition on one level of organization may be
limited by the organization of the system on another level, as agents seek optimization on more than one
level at once.  Thus, competition is not as important in determining the system’s structure as is generally
assumed.  Epistasis may be chosen and achieved within a frame of reference that includes variables from
several levels of organization of the system that an animal occupies, as well as from its own states as a
system.  The choices may seem paradoxical or illogical to an observer, but that is only because a super-
observer does not exist.  

The super-observer

   As Whitaker points out, "the precise form(s) and function(s) by which systems are distinguished are
unavoidably imposed by whatever observer is addressing them" (Whitaker 1996, p 5).  A mammal
engages the environment via disturbances in its nervous system.  It is limited to the internal
representations of the literal external environment, which result from these disturbances.  These internal
representations of the external environment are called descriptions (Maturana 1970).  The organism in
which these descriptions reside, and which operates within the realm of these descriptions, is the
observer (Whitaker 1996).  All observers operate within a cognitive domain.  That is, a domain that
circumscribes "…all the descriptions which [the observer] can possibly make" (Varela 1979, p 46).  
Observation is fundamentally based on making distinctions.  That is, "the pointing to a unity by performing
an operation which defines its boundaries and separates it from a background" (Maturana 1975, p 325 ).  
This enables the observer to behave as if he were "external to (distinct from) the circumstances in which he
finds himself" (Maturana 1975, p 315).  However, the observer operates only within the domain of his/her
closed nervous system, and is not actually external to the circumstances in which s/he finds her/himself
(Whitaker 1996): her/his nervous system is one of those circumstances, as is the attractor upon which that
nervous system sits at the instant of observation.  This nervous system is partly physically determined, and
partly determined by the learning history the observer as an organism has undergone.  Thus, observations
are always relative to the person and history of the observer.  Finally, where descriptions of experience are
negotiated or shared among multiple observers, observations are qualified by the interactions of the
observers, their persons, their personal histories and their histories of interaction (Whitaker 1996; Kuhn
1962).  What we call knowledge is "inseparable from our bodies, our language, and our social history"
(Varela et al 1991, p 149).  
     The super-observer, one who does stand apart from the circumstances in which s/he finds her/himself,
whose descriptions are something other than internal representations, and whose observations are not
dependent on her/his personal and social history and context, does not exist.  All (scientific) statements are
observer-dependent.  This phenomenon is studied by the historians and anthropologists of science as
they observe the scientific observer, but scientists themselves have been too unaware that they, too, are
animals whose behavior must be studied, and that they are subjects operating within a cognitive and
social domain upon which the very questions they ask, their observations and their conclusions are
dependent.  The use of statistics is not adequate to solve this problem.  In addition, though science is
practiced with the aim of eliminating as much superstitious learning as it can from its models and belief
systems, it is inevitable that some superstition always remains – though the content can shift with time.  
This has all had a huge effect on the way animal societies are studied, the models, which have been
proposed, and the conclusions, which have been drawn throughout history.  


Linguistic and other consensual domains

   An animal is, as a living system, constantly behaving.  It is interacting with its environment at all times.  
Adequate behavior is that behavior which enables an interaction to take place without disintegration of the
system's unity (Maturana & Varela 1980).  On the level of the organism, this means behavior which enables
the animal to interact with its environment without undergoing grave bodily damage.  On the next level,
adequate behavior is behavior that will not lead to disintegration of the system occupied with one or more
other animals.  Adequate behavior is, on this level, not determined by the choice of a single animal, but of
more than one, all of whom are learning and behaving as parts of each other’s environments.  In order to
interact adequately, the animals involved must do this within a consensual domain.  
   A consensual domain is a range of interlocked, intercalated and mutually triggering sequences of
possible states with respect to each other, determined through the ontogenic interactions between
structurally plastic state-determined systems (Maturana 1975), which can arise when two or more living
organisms interact.  The animals generate, through time, "…a history of recurrent interactions leading to
the structural congruence between two (or more) systems" (ibid., 1975, p75), in a "…historical process
leading to the spatio-temporal coincidence between the changes of states."(ibid., p321).   
   One of the areas of consensus that dogs develop in interactions is a linguistic domain.  That is, "…a
consensual domain of communicative interactions in which the behaviorally coupled organisms orient
each other with modes of behavior whose internal determination has become specified during their
coupled ontogenies" (Maturana & Varela 1980, p 120).  In this paper, a signal refers to these
communicative modes of behavior.  Signals in a domestic dog include a positioning of the ears, tail, lips,
eyes, head, or body, vocalizations, regulation of distance from other animals, manner of approach, and any
other sounds, movements or gestures which allow other animals to orient themselves to interact with the
subject in an adequate way.  After they are born, domestic dogs go through a learning process in their
interactions with conspecifics, through which a basic consensual domain is achieved with respect to the
significations assigned to various physical signals the species is capable of emitting.  This system of
signaling is loosely called "body language", and constitutes a linguistic domain.  The linguistic domain is
part of a larger consensual domain, which is necessary for adequate interaction to take place.


Mood and emotion

   Signals give an indication of the respective, changing internal states of agents, enabling them to
mutually orient and seek to generate adequate behavior in an interaction.  The model presented in this
paper requires a re-definition of some of the signals used by domestic dogs, since the old terms really
only describe human emotions.  But this will be of no use unless we replace the old terms with well-
defined new terms.  In order to define the meaning of various signals accurately, we will first consider what
we mean by "mood" and "emotion".  Where an association results in physiological changes when a
stimulus is presented, the changes can be such things as levels of salivation, secretion of gastric juices, a
change in hormone levels, muscle tension, rate of the heartbeat.  The changes in an internal response are
often referred to as "mood" or "emotion".  These "moods" and "emotions" can only be deduced, by
measuring, for example, stress or sexual hormone levels, by observing responses such as flight or
approach, or by observing signaling responses in species which have signals available.  The names we
give these responses are analogies we make with our own responses.  It is common in the literature
describing canine social systems to refer, for example, to fearful aggression vs. dominant aggression, or a
submissive approach vs. a dominant approach.  We are not then speaking of the behavioral expressions,
but using analogies of our own feelings to describe the internal state we presume to underlie the
behavior.  Because of the inexactness of terms referring to moods and emotions, I will use (and precisely
define) only two of them here: fear and anxiety.  Fear as used in this paper refers to a classically
conditioned internal response in an animal to a stimulus, where behavior includes signals that the animal
is perceiving or anticipating a threat to its physical integrity as a system and that the behavioral response
with the highest probability is flight.  If flight is not possible, an animal may attack (defined below).  Anxiety
will refer to an internal response to or in anticipation of an aversive (Skinner 1938, 1969), though not
necessarily integrity-threatening, event, to which flight is not necessarily the response with the highest
probability.  When anxious, the dog can show stress signals such as panting or displacement behavior, it
can resort to snapping or inhibited or uninhibited biting, it can emit sounds, and so forth.


Aggression

   Scientists have had huge trouble defining what they mean by aggression in animals.  This is largely
because scientists have projected how they, themselves, would feel in a given context rather than
describing a behavioral event.  This has resulted in the definition of aggression shifting as contexts and
observers shift.  Sometimes mere approach is called aggression, sometimes it’s not, depending on how
the observer thinks the approached animal must feel.  In such a case, the observer is not describing a
behavioral event in the animals, but rather describing his/her own rules and emotions about social space.   
Sometimes a grab that does not damage the other animal’s skin is called aggression, sometimes it’s not,
again dependent on observer bias.  There is argument about what we will and won’t call a bite.  
Sometimes the killing of prey is considered aggression, sometimes it’s not.  Where a term does not fall
within a consensual linguistic domain shared by those who are attempting to communicate with each
other, discussion using the term is pointless.  Despite the illusion that all are discussing a single
phenomenon, in fact each participant is remarking on a different (internally defined) phenomenon than
each of the others.  Such a discussion may feel satisfying to each participant, but in reality no participant
really has any idea what the other is taking about as they all refer to their differing internal descriptions.  
Unless this term is clearly defined, we will never gain clarity about the role aggression plays in the lives
and social organizations of any animals.   
   In this paper, aggression by a domestic dog is defined, quite simply, as the delivery of an uninhibited bit
to another organism, with the exception of the uninhibited biting that takes place as a part of the activity of
eating.  Where a human is the aggressor, aggression refers to actions (as opposed to omissions) by the
human that could damage a dog’s integrity as a functioning living system (e.g., hitting, kicking, jerking on a
choke chain, delivering electric shocks, hanging a dog by the neck, and many of the other things people do
to dogs).  Attack refers to initiation of aggression by an organism towards something in its surroundings.  


Resources

   A similar problem has turned up in the definition of what a resource is, as scientists tried to save the
dominance hierarchy model by claiming the “dominant” animal was the one who could hold and control the
most resources.  Inevitably, these resources ended up being defined as things male humans value most:
food, sleeping spots, sex, and some went so far as to claim that even an abstract status within the group
was a resource that would be defended.  Obviously, this is all mistaken.  It is irrelevant to attempt to
quantify which animal is capable of keeping the most of one or more resources a human author values
and think you are learning anything about the animals or how their social system works.  There was an
attempt made in other quarters to define resources as  related to survival and evolutionary chances, but for
some reason this again led back to food, sleeping spots, sex and status.  As we will see later, observer
bias also led scientists to overlook many of the resources dogs are in fact trying to conserve.
I will use the definition of resource as it is given in an ordinary dictionary: “1a : a source of supply or support
: an available means – usu. Used in pl.  b : a natural source of wealth or revenue – usu. Used in pl.  c :
computable wealth – usu. Used in pl.  d : a source of information or expertise  2 : something to which one
has recourse in difficulty : EXPEDIENT  3 : a possibility of relief or recovery  4 : a means of spending one’s
leisure time  5 : an ability to meet and handle a situation : RESOURCEFULNESS” (Mish, et al, 1983).  
Because dogs can’t count money, “source of wealth” means, in this context, anything that increases the
quality of a dog’s life as perceived by the dog itself.  During this study, I did not, as most researchers do,
define "resources" in advance.  Rather, I observed the dogs until their own behavior made clear what they
were treating as resources.  Resource-holding is recognized as a subjective activity whose subjectivity
resides in the dog, not the human, and whose subjectivity therefore not only does not have to be
discounted, but is a critical part of how they canine system works (at all levels).  No attempt is made to
define resources that would supposedly be valid for the whole species, because it turned out that among
dogs the definition and valuation of a resource varied from one individual to another, and in each individual
according to time and circumstances, just as subjective as it is from one human to another, and just as
dependent on many individual factors.    


The signification of signals

   As will be shown below, dogs are also capable of adjusting the signification they attribute to various
signals so that a smaller consensual domain is achieved with a single other dog in a binary interaction,
separate from the consensual domain in which the entire species or group operates.  For example, a dog
can learn that a certain tail position means something very specific and different in this one other dog, but  
not in any other dog.  Through a mutual process of learning during interactions with members of other
species, domestic dogs are also capable of generating a consensual linguistic domain with those non-
conspecifics.  This domain again exists outside the domain that is achieved with conspecifics and,
perhaps, yet other species.  However, with humans, there is often failure to achieve a consensual domain.  
This failure is at least in part caused by the cognitive domain in which humans operate, and the low
plasticity of this domain's structure.  Once a human has adopted a particular model by which it organizes
its perceptions, it may take a long time for this model to change (Kuhn 1962).  This model, in turn, may limit
the descriptions which it is possible for the human organism to make.  This can be due to a filtering effect
– the human fails to perceive that which does not fall within the model structuring her/his cognition (e.g.,
Phillips 1993).  It can also be due to the threat of loss of system integrity if the structure of cognition is
changed, that is, another model adopted.  When an observer moves out of the consensual domain s/he
shares with other humans, group membership or even the very existence of the observer can be
threatened (Kuhn 1962).  
   The dominance hierarchy model has become widely popular both among laymen with one or more dogs
in the household and among scientists studying canine behavior.  One of the areas of study is the
linguistic domain.  Human observers have attempted to attribute signification to the signals domestic dogs
use in their social interactions.  The significations "dominant" and "submissive", are derived from the
dominance hierarchy model and have been assigned to certain ear, tail, eye and bodily positions, and this
consensus has led to a limitation of the ability to perceive any other possible signification for these
gestures.  That is, the model structures the cognitive domain of human observers such that this cognitive
domain does not (for the moment) include any other possible descriptions of these signals.  In this regard,
human cognition is possibly less plastic than that of a dog with all its neurological and cognitive
limitations, since the dog is not subject to some of the complex group pressures humans operate under,
and may be more able to incorporate new descriptions than a human at a particular point in history.  
I assert that the dominance hierarchy model and its concepts of “dominance” and “submission” are
examples of superstition as defined by Skinner. Rather than curing the problem, extensive dependence on
statistics has operated as a typical feedback loop, actually resulting in an increase in superstitious
learning in many areas of study.  This dependence is also an expression of the limitations of the shared,
historically shaped cognitive domain within which scientists operate.  
   In attempting to describe the linguistic domain of domestic dogs, it is therefore necessary to abandon
the use of the terms "dominant" and "submissive", since these terms keep our perceptions locked inside
the cognitive domain defined by a single model.  These terms in themselves tie us to a pre-existing human
consensus, thus themselves structuring our perceptions and preventing us from considering or,
considering, being able to discover another model of domestic dog behavior and group organization.  This,
in turn, prevents us from arriving at some other consensual domain among ourselves.  It is therefore
impossible to develop a new model while continuing to use these terms.  Furthermore, pharmacological
evidence suggests that the underlying system states related to these signals are not what humans have
so far thought them to be (Borchelt & Voith 1996; Simpson & Simpson 1996). This implies that some other
signification of these gestures may help us more accurately describe the consensual domain within which
the animals interact and to find a consensual domain with the animals as we interact with domestic dogs
ourselves.  
    The following gestures are generally described as "dominant" in domestic dogs: ignoring the other,
erect ears, staring, holding the tail higher than horizontal, retracting the lips to show incisors and fangs,
standing on straight legs so that the body shows a high posture, placing the head or a forepaw on the
shoulders of another animal, mounting another animal in a non-sexual context (Askew 1997; Borchelt &
Voith 1996; Dunbar & Bohnenkamp 1986; Fisher 1998; Gaus 1995; Morris 1994; Overall 1997, 2002;
Schenkel 1967 – to name only a few).  While most researchers have included standing over in the list of
dominant gestures, one researcher does not consider this a dominant gesture (Mech 2002).  Other
gestures are generally described as "submissive":  ears held low on the skull and folded back in the neck,
averted gaze and/or head, holding the tail lower than horizontal, bending the legs so that the body
assumes a lower posture, licking at the corners of the other animal's mouth, licking at the air during an
approach, lifting a forepaw in a tapping motion at a distance, lifting a hind leg to expose the groin, lying on
the back, urinating without assuming the normal urinating posture (Askew 1997; Borchelt & Voith 1996;
Dunbar & Bohnenkamp 1986; Fisher 1998; Gaus 1995; Morris 1994; Overall 1997, 2002; Schenkel 1967 –
again, among many others).  Some authors have referred to these gestures as "pacification" or
"appeasement" gestures (Abrantes 1997).  While this is seems to imply a departure from the dominance
hierarchy model and an abandonment of the concepts of “dominance” and “submission”, this terminology
does not go far enough.  A state of war is implied in which aggression is always imminent and must be
pacified or appeased in order to interact at all.  Retracting the lips in a sort of smile to show the incisors,
fangs and molars is generally described as a gesture motivated by fear, though one researcher maintains
that this gesture is motivated by "submission" (Abrantes 1997).  When two domestic dogs meet each other
for the first time, their gesturing at each other is generally described as an attempt to establish respective
“ranks”.  Ritual fighting can take place, but in the end one animal is described as reverting to "submissive"
gestures, at which point “rank” is supposedly established between the two. This categorization of and
assignment of signification to the various gestures does not do justice to the complexity of the interactions
nor to the animals as plastic systems.  Finally, it has apparently led to the ascription of faulty significations
to many signals (e.g., the "dominant" aggression which in fact can be treated with the help of anxiolytic
medication [Overall 1997, 2002]).
   In this paper, another system of signification will be used.  The concepts of dominance and submission
are abandoned as irrelevant and unverifiable.  Instead, gestures will be categorized as indicating a threat
or the absence of a threat to another animal.  Further discussion will follow later in this paper, but for now it
will suffice to define the new, more accurate significations.  

Threat gestures
    A threat is any signal indicating that the gesturing animal's internal state is such that it perceives or
anticipates a perturbation to its position on a fitness hill or to the larger system.  A threatening dog is not
being “dominant”.  It can be indicating that it feels threatened by or anxious about the anticipated
perturbation of its position in its fitness landscape, that it wants more information before conceiving a plan
as defined above, and/or that it wants the other to back off (i.e., maintain or increase physical distance) until
enough information is exchanged so that its internal state returns to (or migrates to) an equilibrium.  
The following gestures will be referred to in this paper as threat gestures:
  • staring
  • growling
  • wrinkling or lifting an upper lip (sometimes for a split second)
  • retracting the lips so that teeth are bared
  • standing on straight legs so that the body shows a high posture
  • lifting the tail to a position higher than horizontal
  • freezing up
  • approach in a straight line (direct approach) unless non-threat signals are given during approach
  • barking in a low tone
  • opening the jaws wide, teeth exposed, while moving the jaws around the face/neck of the other
  • the delivery of an inhibited bite

Non-threat gestures
    A non-threat gesture is any signal by which the signaling animal indicates its explicit intention not to
perturb the fitness hill the receiving animal is sitting on.  Some of these signals are referred to as
“dominant” and others as “submissive” in the old theory.  
The following gestures will be referred to in this paper as non-threat gestures:
  • ignoring the other
  • ears held low on the skull and folded back in the neck
  • ears turned (erect) so that the openings point sideways/outwards
  • averted gaze and/or head, or otherwise avoiding direct eye contact
  • bending the legs so that the body assumes a lower posture
  • licking at the corners of the other animal's mouth
  • holding the tail in a position lower than horizontal
  • licking at the air during an approach
  • lifting a fore paw in a tapping motion at a distance
  • sitting
  • freezing up
  • lifting a hind leg to expose the groin
  • approach in a curved line (indirect approach)
  • maintaining or increasing distance from the other
  • not pursuing/approaching when the other increases the distance
  • orienting the side or back of the body towards the other
  • emitting high tones (yelping, squealing, barking in a high tone)
  • lying on the back
  • urinating without assuming a normal urinating posture
  • gestures which are recognized as creating the consensual domain of play in an interaction
  • so-called displacement behaviors (e.g., yawning, scratching, sniffing the ground)

   Turid Rugaas (1997) was the first to recognize many of these non-threat gestures as “calming signals”
rather than any expression of “rank”, and was thus also the first to explicitly acknowledge threat signals as
expressions of anxiety.  In this paper, I expand on her pioneering work, explaining mechanisms and
placing it all in a theoretical context, showing how these signals fit into a complex self-organizing system
and how they are used to achieve stability on multiple levels of organization.    


Methodology

   When dealing with complex self-organizing systems, we are dealing with systems in which an
enormous number of variables operate, and where feedback loops mean the results of the system's
output return to it in the form of input, so that the system's operation itself ends up supplying variables
which affect the determination of the system's future behavior.  The whole is more than the sum of the
parts, and the sum of the variables taken altogether in their interaction is more than and qualitatively
different from the sum of each of their effects separately.  In fact, each variable is, in isolation, meaningless
since the relevance of a variable is derived from its interaction with all the other variables.  We are therefore
not studying separate variables, but the interaction of all the variables simultaneously.  
   One of the characteristics of a self-organizing system is the fact that it can choose between various
attractors in its state space.  A complex self-organizing system is, therefore, deterministic, but it is not
necessarily predictable.  The general direction the system will take is determined by the driving function,
but the path it may take to any one of various fitness hills is broad with many side roads or forks
(bifurcations), and the system may switch midway to a path toward another attractor.  Thus, we cannot
predict the system's trajectory through state space.  In the of study self-organization, it is necessary to
observe a system operating.  In the case of artificial intelligence, artificial life, cellular automata, genetic
algorithms, neural networks, and such, the researchers create a set of rules, set the system in motion,
note any stable patterns that emerge, and then repeat the sequence many times.  Generalizations with
some statistical probability can eventually be attempted (Lucas 2002b, passim).  With living organisms,
however, we do not know the rules in advance and must begin by observing an operating system without
any assumptions as to which rules will turn out to apply.  The aim of research is to discover general
properties that apply to topologically equivalent systems.  Once these rules are discovered, they can be
tested by repeatedly starting a system at some point in state space and observing whether they adequately
describe the evolution of the system through time.  These are the methods used in this study.  
    We are not super-observers.  In observing a system, any criteria we use in plotting the height of hills in
the fitness landscape are our own criteria and the map which results is our own map.  When the system
parts are living organisms, they will be mapping their fitness hills according to their own internal criteria,
which may remain invisible to us.  Domestic dogs are complex organisms subject to many kinds of input,
able to produce many kinds of feedback-generating output, and capable of learning.  It is therefore
impossible for an observer to determine all the criteria the animal itself is using in attributing heights to the
various fitness hills in the landscape.  We cannot say that a hill occupied by the use of threat signals is
higher than some other hill unless we specify that this is a criteria we have imposed ourselves which may
be irrelevant for the animal(s) involved.  We cannot determine that possession of some resource will
always lead to a heightening of a fitness hill, nor that the relinquishing of a resource has led to a lowering
of the height of a fitness hill, since factors as the learning history, valuation of system membership, and
such internal metabolic states as hunger or sickness may influence an animal's valuation of a resource at
any given moment or in a given interaction.  We can, to some extent, predict that a condition such as famine
might lead to a higher valuation of food and result in changes in interactions involving the access to or
possession of food.  However, in such cases we are observing an example of what is called an Error
Catastrophe: too high a rate of innovation is not controllable by selection and leads to information loss,
chaos and breakdown of the system (Lucas 2002b).  This will not tell us how the system operates, but how
it disintegrates.  The author did not, therefore, attempt to rank fitness hills in any sort of hierarchy.  
If we want to understand social organization, we must study an organism in it natural habitat.  Such a study
must encompass the entire life span of at least one organism.  The author argues that the observations
made in such a context will have more validity and be better related to the behavioral realities than many
studies that have been done or are being done in  laboratory or academic settings, for many of the same
reasons that studies of wolves have recently been criticized (Mech, 2002).  Though a laboratory or
academic setting succeeds in reducing the variables involved when very specific aspects of behavior are
being studied, they are unable to provide a context for studying the full range of natural behavior or for
discovering the principles which govern the social organization of domestic dogs.      
   The model presented here is derived from information gained during a longitudinal study of a group of
domestic dogs living in their natural habitat: a) in groups of two or more and b) among humans.  The study
began in 1994, and is still in progress.  I refer to the group studied as “the home dogs”.  Though the
membership of the group changed over time, the defining factor was that the dogs shared a home and
were never separated for more than a few hours once they’d joined the group.  These dogs live in my
home, in an urban setting.  The dogs in the home group were observed 24  hours a day, seven days a
week, which meant the evolution of their behavior could be followed without missing events that triggered
changes.  The group has been built up by a process of accident.  Human preferences or expectations did
not play a role in selection of types or sizes of dogs.  This accidentalness also provides a more accurate
microcosm of the natural habitat domestic dogs live in than would a group based on human selection,
since dogs interact during their lives with many conspecifics that were not selected by the human they live
with.  Because a defining characteristic of an autopoietic system is that it produces its own functioning
system parts, reproduction and the raising of offspring must be understood if we are to understand the
functioning of a system composed of living organisms.  To deduce system principles, we must observe the
system functioning as a whole for at least the time span of an entire life cycle of at least one system part.  
The system formed by the home dogs has, at the time of writing, been observed for a period of twelve years
(probably more than three quarters of the life span of the oldest animal present in the permanent group),
and has included several rounds of introduction of offspring,  However, domestic dogs are not limited to
the groups in which they are born and raised.  When two dogs meet, they are usually capable of quickly
arriving at a point where they can interact while both maintain their own system integrity.  They are also
capable of doing this with other species.  This means that other species can participate in the system.  
Part of the data for this study comes from watching the permanent group of dogs absorb new members
into their group and watching them interact with strangers of various species inside and outside of the
home.  This allowed observation of how the larger social group moves across its landscape in the face of
a perturbation, as well as the establishment and evolution of new binaries between hundreds of dogs (and
tens of other animals) that had never met.  
   A second part of the data for this study comes from data arrived at through participant observation
engaged in with domestic dogs in an animal shelter.  In this case the author chose to work with animals
showing fear or aggression toward humans.  This was partly to test the proposed system rules by
repeatedly starting systems at or near chaos and testing whether the proposed rules predicted the
evolution of these systems.  The underlying proposition was also that much could be learned about
system principles by entering an interaction in chaos and participating in moving the interaction into a non-
chaotic region of state space, so that I could amend my model if need be.  That is, that by repeatedly
undergoing production processes and allowing myself to be shaped into an adequately functioning system
part, I would discover some of the principles governing the system's organization, determination of
structure, and movement through state space.  Besides, participant observation was unavoidable as the
only way to solve the problem of the super-observer.  You can’t send a dog into an interaction and know he
is following the rules you propose.  You obviously can’t solve this problem by giving a dog a set of
instructions to follow in an interaction with another dog, asking the dog to test your rules for you.  The only
way to be sure my proposed rules were being tested was to enter the interactions myself and behave
according to those rules.  This also provided an arena for testing the signification I had hypothetically
assigned to dogs’ signals: responding to a dog’s signal as if it meant what I proposed it meant, signaling
back (within the human limitations) with signals from my own dictionary of significations, watching the dog’
s response to see whether its internal state as indicated by its own signaling changed as I predicted it
would.  My work in the shelter was, thus, a testing both of my proposed model and of the new significations
I propose for dogs’ signals.  Did we generate, through time, Maturana’s "…history of recurrent interactions
leading to the structural congruence between [our] systems" (Maturana, 1975, p75), in a "…historical
process leading to the spatio-temporal coincidence between the changes of [our] states."(ibid., p321), as I
predicted we would?  Could the dog and I construct a stable binary social system in which were able to
display adequate behavior while sharing a physical space and a social landscape?
   The first dog present in the permanent group was a Basenji crossing, male, born in 1979, abandoned to
me by a neighbor in 1984.  This study started In 1994 when I found a male puppy on the street, a Scottish
Collie/Labrador mix (hereafter mCL).  This puppy was approximately eight weeks old when I introduced
him to the Basenji.  He had been bought at a market in Antwerp at the age of seven weeks, spent a week in
a household with two adults and two children (both male, eight and 11 years old), then was dumped onto
the street when Dad got tired of cleaning up pee.  mCL had been alone on the sidewalk for about five
minutes when I came along.  When mCL reached the age of 30 months, in May 1997, a ten-week-old male
Border Collie (hereafter mBC) puppy was acquired from a farmer who was planning to drown him.  MBC
had spent his first eight weeks alone in a small shed with his mother and siblings, and his eighth to tenth
week alone with the mother.  He saw humans when food was brought, when people came to choose a
pup, or when neighborhood children stopped by to look at the pups.  The mother was a well-socialized
working dog.  When mBC reached the age of 31 months in the autumn of 1999, two new puppies were
added to the group, a female German Shepherd (hereafter fGS) and a female Labrador (hereafter fLab),
both eight weeks at the time of introduction to the group.  These pups were referred to the author by the
local SPCA.  In May, 2001 a castrated male Jack Russell (hereafter mJR), five years old, was added to the
group when another person moved into the house.  This dog had been acquired in a shelter at the age of a
year.  In June, 2002, a female Canadian shepherd (hereafter fCan) was added to the group.  This dog was
18 months old.  Her original owner had bought her from a breeder (conditions unknown) at seven weeks
old.  She had lived until her 18th month with this man, his wife and three small children (two male, three
and five years old, one female, seven years old).  All of them were often severely beaten by the man, the
dog from her eighth week of life.  FCan had spent much time in dog parks.  She was extremely fearful of
human males, but she had no trouble interacting with dogs or other animals.  I had fCan sterilized as soon
as she joined my household.  In August, 2002, an already sterilized female Friese stabij (fStab) was added
to the group.  She had lived eleven years with a woman, until she was re-homed due to the birth of a
handicapped child.  FStab had had the life of an ordinary household dog, living with humans, spending
time in dog parks.    
   This permanent group of dogs interacts with visitors, both canine and human, in their home.  This
includes both puppies and adult dogs of both sexes, and children and adult humans of both sexes.  Visitor
dogs sometimes stay a few hours, most come two to four days every week (thus allowing repeated
observation of the resumption of interrupted relationships), and sometimes they stay for several weeks.  
The dogs are fed in one space, about half the time with one or more visitor dogs present, which are also
fed.  In addition, the group interacts in city parks with randomly encountered dogs of all ages and both
sexes.  When traveling to a city park, the entire home group is present, including any visitor dogs that are
staying with us.  At no time did the researcher interfere in the interactions between the dogs in the home.  
In parks, most interactions were left to the dogs. The first exception was in cases where the owner of a non-
group dog panicked during an interaction between the dogs.  In six cases, a dog was observed to be
delivering uninhibited bites to a second dog.  The interaction was stopped by human intervention, as it
would be unethical to allow a dog to be damaged in the name of research.  This did provide an opportunity
to observe the later reactions of dogs to a conspecific that had exhibited aggressive behavior.  The third
exception was when intervention was necessary to save the life of a dog from an attack by one of the
breeds humans have developed for the purpose of killing other animals and dogs, or to avoid such an
attack from even starting.  
   For ethical reasons, I prevented the adult females in the group from reproducing.  The evolution of
Interactions with (and between) puppies younger than seven weeks were observed only twice. Both
periods of intermittently observing neonates during their first seven weeks took place in the shelter, when
carelessness had led to the birth of litters there.  This is a gap in the present study.  However, as research
continues the gap may be filled, or perhaps other observers will be able to present data, which will
compensate this hiatus and correct any flaws it may have led to.    
   The testing of rules in systems starting at chaos was done in an animal shelter in The Hague, two days
a week, from May 1997 through June 2000.  This shelter places approximately 800 dogs per year, giving
the author a chance to observe a wide population of city dogs.  It is important that this population was not
socialized by the author or her own dogs, since this provided a chance to test the proposed model against
observations of a population of variously raised and socialized dogs, without further knowledge of their
past experiences and in a context where the dogs were dealing with strangers of all species they
encountered.  I entered binary interactions with each of 53 dogs.  These dogs were not randomly chosen;
focus was on dogs with extreme fear problems or aggressive behavior toward humans.  An attempt was
made to arrive at a consensual domain with each animal, such that the participants could, finally, interact
on an open field without flight, threat or aggression occurring.  The proposed rules were tested to see
whether they would move the binary through state space to a stable attractor, with each participant
undisturbed on a fitness hill while also constituting part of each other's fitness hill.  To further test the rules,
guard against omissions and exclude purely personal biases, a research assistant also participated in
forming binary relationships with some of these dogs using the same set of instructions.  In the shelter, I
also had the chance to observe the reactions of dogs to personnel who continued to believe in and behave
according to the old dominance model, including the interpretation of the dogs’ signals.  It was interesting
to watch personnel respond to threat signals with their own threat signals, and to watch relations quickly
deteriorate (i.e., move further toward chaos, often to the point where personnel became afraid that attempts
to remove a dog from its cage would result in aggression as defined in this paper).  This provided a
chance to validate my own model.  I was able to contrast the results of its application in interactions with
many specific dogs to the results of applying the old model in interactions with the same dogs.  
   One group of dogs was excluded from this study:  the fighting breeds (the pit bull, the American and
English Staffordshire bull terriers, the English bull terrier, the akita inu, the tosa inu, the American bull dog,
the Presa Canaria, and all the other breeds that humans have artificially selected for killing behavior).
There are anomalies in the behavior of the fighting breeds which make it too dangerous to allow them to
interact with other dogs merely for the purpose of observation.  An exploration or explanation of these
anomalies is beyond the scope of this paper.  This does, however, constitute the one exception to the
aforementioned randomness of the interactions between observed dogs, and must be mentioned here.



Summary of Results


Aggression: Dog to dog

   Dog-to-dog aggression turned out to be extremely rare.  In twelve years, I observed six cases in parks in
which a dog delivered one or more uninhibited bites to another dog.  (Remember here that bred by
humans based specifically on artificial selection for aggressive behavior are excluded from this study.)  
Aggression was never in the context of a dispute about a resource.  Attacks occurred each time outside the
context of any ongoing interaction, coming as it were out of nowhere, not embedded in any social context.  A
seventh dog had been brought to the shelter with an owner-reported history of killing other dogs.  In the
shelter, this dog was kept isolated from other dogs, so this behavior was not actually observed.    
When aggression did occur, dogs reacted to it as abnormal behavior.  Attacked dogs showed fear (if
possible, flight behavior with all attending fear signals), uttered screams, did not attempt to bite back but
rather whipped head and body around trying to avoid or fend off bites and (it seemed) attempting to get a
look at their attacker (i.e., visually orient), looked for an opening to flee.  After an attack was stopped by
human intervention, the attacked dog would stand there trembling, often barking in a high tone at the dog
that had attacked it – highly emotional, whatever those emotions may be, but we can presume full of
adrenaline in any case, yet not attempting to re-engage physically.  The experience must be called
traumatic, because an immediate change in behavior in later social interactions was always observed.  
Some dogs became fearful even of familiar conspecifics; some of all unknown conspecifics, others of
dogs that shared some feature (size, color, breed) with the attacking dog.  They all showed highly
increased anxiety when they saw their attacker later (at a distance), growling, bristling, lifting the tail.  They
showed avoidance behavior, increasing distance or attempting to leave the site altogether while the
attacker was still far away.  The reactions remained very emotional (i.e., they made clear that the inner state
of the dog was highly perturbed at seeing a stimulus associated with aggression).  These changes
sometimes lasted for years.   The heavy emotional reactions of the dogs, the fact that an attack always
resulted in long lasting fear or anxiety, the fact that it damaged a dog’s ability to engage in social
interactions, and the fact that the attacker was avoided, that all led to the following conclusion.  
Aggression is not relevant to the social organization of the domestic dog.  Dogs do not use aggression in
organizing either binary or larger social systems.  Aggression leads to the complete disintegration of
interactions.  Furthermore, it damages the attacked dog as a functioning part of any social system.  Thus,
the first and most basic, important rule governing relations in the domestic dog’s binary interactions and in
the emergence of social systems is: aggression will not be used.   


Aggression: Dog to human

   Dog to human aggression turned out also to be rare.  Dog to human aggression was observed only in
the shelter.  In two of these cases, owner-reported original conditioning was sufficient to explain the
response.  A third case involved an adult Doberman, male, who had been in the shelter for a time without
showing any sort of fear or threats toward humans.  On the contrary, he was friendly, sought human
companionship, allowed himself to be touched all over his body, and was generally considered by staff to
be an ideal companion animal.  The dog was eventually re-homed, but he was brought back after six
weeks.  Upon return, he showed marked fear behavior when an unfamiliar human approached his kennel
(e.g., he fled to the outdoor run, maximizing distance to whatever human stood before his cage).  Humans
he had know before the re-homing were able to take the dog out of the kennel to walk him.  The attack
occurred when a female staff member took the dog out of his kennel, leashed him, then (standing to the
left of the dog) abruptly bent over the dog’s back to grasp and examine a wound on the dog’s right front
foot.  The Doberman did not attack the staff member who was bending over him and handling his wounded
foot.  Rather, he attached a second caretaker who was standing in front of him three yards away.  This
woman was directly in the dog’s line of vision, and she later reported that she had been looking him
directly in the eyes while the other woman tried to examine his foot.  We were unable to work extensively
with these three dogs because they were executed within four weeks of attacking a human.  However, and
in particular as illustrated by the case of the Doberman, learning remains sufficient to explain the attack
behavior without reference to mysterious internal traits in any of these dogs.  
   With fearful dogs and threatening dogs, we followed a procedure of focused respondent conditioning
and at the same time developing consensual domains with each dog.  We succeeded in establishing
stable binary systems with each and every dog that initially showed fear or threat behavior.  The dogs all
showed signs that our presence heightened their fitness hills long before we reached the stage of taking
them out of their cages onto a field.  Here, the importance of learning as a production process was clearly
demonstrated.  Once we had established a linguistic domain with these dogs, the dogs were able to
generalize this domain to other humans who used the same signals.  In the end, various shelter
volunteers were able to enter into stable binaries with these dogs, some of whom shelter personnel hadn’t
dared take out of their cages for as long as a year.  Aggression did not occur with any of the participants
who cooperated in establishing consensual domains with these dogs.  
   Threat behavior did not decrease, and sometimes aggression did occur, toward shelter personnel who
continued to believe in and behave according to the idea that you must “dominate” a dog (they did not do
this to fearful dogs).  These humans were not willing to try using other signals. All of them expressed
indignation at the idea of  “submitting” to these dogs.  What was the use of establishing peaceful
interactions if these were not the result of winning a power struggle?  Here we were again witnessing the
low plasticity of the human cognitive domain.  These humans were defending various investments and
resisting having to make new ones.  They had invested much time in learning the dominance hierarchy
theory.  There seemed to be psychological investment in the image of the self as already having expertise,
and reluctance to let go of this ego image.  Learning new ideas would have meant leaving a negotiated
and shared consensual domain, thus also possible loss of position within a peer group.  In addition, we
were watching persistent projection of the rules governing human systems onto the dogs’ system: not only
are all relations power struggles, but one can only count a “win”  where one has prevailed by dominating
rather than by compromising.  None of these humans ever removed these dogs from their cages, so
personnel’s hypothesis that these dogs would attack them was not tested.  Aggression consisted of dogs
biting any object, including hands, that were inserted into the cage by these people (e.g., to feed the dogs).  
It remains unclear whether these were really just high level threats (reflecting a high level of anxiety about
what these humans would do next), or whether these dogs (or some of them) were really determined to
drive the human out of the shared physical space (and thus out the dogs’ fitness landscapes altogether) by
the use of aggression.  
   This contrast in the dogs’ behavior with each set of humans shows that the success of my model and
the failure of the old one is not due to something in the dog.  It also shows that my proposition is correct
that relationships are binary, always established between two participants and applicable only to them.  
Though the linguistic domain becomes available to others once it is learned and established, each of
these others must use it to establish his/her own binary system with a dog.  
Aggression towards humans was so rare that it must be considered an anomaly.   Aggression is clearly
not relevant to organization in the systems dogs form with humans – at least not from the dogs’ side of the
interactions.  Rather, it is related to the loss of the consensual domains that are necessary to maintain a
system containing one or more human agents.  This confirmed the rule stated above: aggression will not
be used.  Where aggression does occur, it is not used in organizing a system, but is an indication of total
system breakdown.  


Resources

   Learning was clearly of great importance in determining what a dog considered a resource.  Interest in
toys was entirely dependent on learning history.  Other resources were also defined by the dog’s learning
history.  I will cite one example at length here to give clarity about the method of analysis, and to emphasize
how important it is not to be sloppy in such analysis.  One of the dogs in the home group had been
severely beaten by a previous owner.  She initially responded with fear (flight behavior) to any sign that
human attention was directed toward her (e.g., looking directly at her).  I responded by ignoring her and not
decreasing distance.  Eventually fCan began to approach carefully, emitting all kinds of emphatic non-
threat signals.  I responded by also emitting non-threat signals.  Fcan was, in fact, testing whether I would
obey the rule that aggression will not be used: she was learning about me in a process of mutual
orientation as described above.  Within eight weeks fCan began to seek frequent interaction with me, not
only giving an orienting response to receive a food reward, but also soliciting attention and stroking.  This
behavior allows us to conclude that attention and stroking were now operating as reinforcers.  We must,
thus, conclude that fCan now perceived attention and stroking as increasing the quality of her life.  She
learned to experience human attention as a resource, whereas it had first been a sign of danger.  FCan
had also learned that certain signals of hers would be followed by certain signals from me, some of which
predicted the appearance of these reinforcers – a new linguistic domain was established.  Once these
consensual domains been opened up, fCan was able to transfer it to other humans.  She became
interested in investigating whether she could safely establish social interactions with new humans.  This
indicates that there was a second resource involved.  fCan got plenty of stroking from me, so there was no
reason to seek more of this in and of itself.  The more so since seeking this resource from another human
meant voluntarily undergoing a strong perturbation of her inner state (a high level of anxiety) and – in her
eyes – the risk of being attacked.  I conclude that fCan actively sought to repeat experiences of non-violent
interactions with humans because each new peaceful interaction led to decreased anxiety that a human
meant, by definition, an extreme perturbation of the shared fitness landscape at the least, and a danger to
herself as a functioning living system at the worst.  Because fCan was willing initiate interactions in which
she had to overcome fear and risk attack to test this stability, we must conclude she perceived stability in
the social landscape as a resource in itself.  fCan was attempting to find our how frequently this resource
was available by actively seeking repetition of the experience she’d had first with me.  If other humans
turned out to share the linguistic domain (responding to non-threat signals with non-threat themselves),
and if humans could be generally regarded as following the non-aggression rule, a great increase in the
quality of fCan’s life would be reached.  She would be able to move within a state space in which violent
perturbations by humans did not occur on the binary level, and internal perturbations were decreased by
her own changed perceptions of humans.  Thus, generalized reduction of fear, increase of inner stability
and increased stability in the general social landscape were clearly all resources fCan was willing to face
fear to gain.  This was, however, a binary process.  The responses of the other dogs or myself to a new
person had no influence whatsoever on fCan’s responses.  Each new human had to establish its own
predictability with, and thus value as a resource to, this dog.  The other dogs in the home group had
different learning histories.  They considered all human attention to be a resource and no human to be a
threat to stability on any level.  (The analysis followed in all cases is the same as that used with fCan: the
other dogs behaved as if human attention was a reinforcer for all kinds of soliciting behavior; from their
behavior we can conclude that human attention was perceived as increasing the quality of their lives and
that the non-aggression was assumed to operate; approach was not careful, non-threat signals were
subtle; the dogs behaved as if the consensual domains necessary for social interaction could be taken for
granted as pre-existent.)   
   But learning does not only determine what a dog considers a resource.  Learning was also important to
behavior around resources.  Dogs whose owners constantly took things away from them showed intense
threat behavior if approached by a human while the dog had what it regarded as a resource in its
possession.  Dogs whose owners did not constantly take things away from them did not show this
behavior around humans and resources.  A dog’s behavior toward other dogs was dependent on both
experience with other dogs in general and knowledge of whatever dog was approaching.  Dogs were able
to learn to predict the behavior of another individuals as individuals and to adjust their behavior accordingly,
with responses varying according to experience of a particular individual in the past.  In interactions around
resources between dogs who did not know each other, the resource-holding dog would often emit high
level threat signals.  Threat signals always decreased, finally, in response to non-threat signals emitted by
a dog not holding a resource.  After several interactions in which a particular dog had consistently emitted
non-threat signals, threat behavior by the other dog ceased to occur at all toward this particular dog in the
presence of resources.  For example, some of the home dogs initially (in the first days after joining the
group) showed threat behavior toward the other dogs while treats were being handed out.  These dogs all
quickly learned that some other dog receiving a treat was a reliable indicator that a treat for her/himself
would follow.  Threat behavior ceased and was replaced by strong orienting responses toward the human
at the sight of anot