For a system to be a communication system, it must be composed of two complementary informative processes. If we are to study a communication system as composed of more than one function, that is, we choose to treat it as something other than an indivisible whole, it is necessary to break the larger process into smaller processes. Central to breaking a process into parts is determining which descriptions of processes are most natural and best help us describe, predict, and understand the communication system. In some decompositions, the resulting sub-processes may appear to be natural, having a reasonable explanation, while in other cases, the division into sub-processes will appear very arbitrary in nature with unnatural sub-processes. Obviously, a decomposition that produces reasonable explanations is superior to one where the boundaries between processes are arbitrarily chosen.
Communication processes may be delimited so that they represent an entire receiver or an entire sender, or processes may be viewed as being much smaller, so that many of these smaller processes together constitute the processing capability of an individual. If the most natural form of a hierarchy is with a large number of small processes, it may be easiest for an organism to learn the details of each smaller function instead of learning the nature of a single, large function. On the other hand, in some cases it may be beneficial to define a process larger than a single individual, with, for example, a group of people producing a document, or an individual reading a document, including the document and its production, in a single large process.
There are a number of processes that occur frequently in higher level species, including a number of components to natural language processing, such as the movement of vocal cords, the reception of sound and its translation into neural signals, syntactic processing, and phonological processing. Hockett has suggested that there are at least 13 different features (and thus functions) in human communication [Hau96, p. 47]. No other species has all 13 features, although many species have several. These may serve as a basis for an understanding of the human processes in natural language communication hierarchies.
Those processes that exist and are observable are more likely to have contributed to the survival of the greater systems within which the process exists than are less adaptable processes that did not contribute as much. A process may contribute to its own survival, or it may contribute to the survival of similar or identical processes in other systems, when contributing to the survival of others increases the chance for similar processes to appear in future systems. For example, someone who can communicate the nature of a vaccine for a childhood disease that can save the lives of millions of children may not increase their own chance of survival but will increase the number of communicative humans with intellectual abilities similar in some respects to those of the vaccine's developer.
We refer to systems that take on increasingly sophisticated and non-random structures as they evolve as self-organizing systems. Functions may be learned or may evolve through self-organization over time. This often occurs because increased organization or structure itself leads to increased adaptability and survival, although we should not make the mistake of assuming that increased complexity necessarily implies increased survivability [Gou97]. Through learning processes, a function may infer the characteristics and parameters of a function of interest through supervised learning, where labeled cases are learned when inferring a generalizing principle. The effect of learning also may be produced through evolution, with randomly generated variations either surviving with increased or decreased frequency over time. This has the effect of learning adaptive characteristics.
Learning communicative functions through evolutionary processes requires the presence of a higher level layer that both produces and receives information. In the case of bees said to be communicating through dancing, there is a process above the dancing process (which has evolved) that allows the dancing process to be be evolutionarily supported by increasing the survival rate of bees with dancing (or understanding) skills.
Any process located above another process may be a goal process that rewards, allowing the function before it to evolve. We refer to a function that maximizes a goal function as maximally useful. For an informative function to be learned genetically, it must support a goal which is relatively constant in the species, with evolution leading toward it becoming maximally useful. Goal processes often contribute toward reproductive behavior or are survival related. Communication processes produce results with the systems in which they occur such that similar processes are as likely or more likely to survive and reproduce than they were before. This is because introducing communicative capabilities essentially makes a system larger, which can lead to increased self-organization, than would be the case with individual, smaller, non-communicative systems.
Given our model of communication, communication systems are likely to develop in self-organizing systems because of their support for what are conceptually larger systems, leading to more self-organization and increased adaptability. We assume that it is almost as likely that if one process evolves then its inverse will also evolve. We also assume that information about another system may be beneficial to the system in question under many circumstances and thus communication systems are evolutionarily adaptive.