Abstract:
A set of inter-related theories and research studies is presented that fit together to produce a large-scale, coarse-grained, integrative theory of how human language is likely to have evolved. The goal is to develop an overall framework and methodology that is useful for further studies. An important objective is to handle the continuity of evolution, with incremental benefits for selection, and without major discontinuities and gaps.
Theories:
First, that human language is a communication component of coordinated complementary skills that require external communication for coordination.
Second, that skills are internal information processing specifications (like software programs).
Third, that skills have evolved both through the evolution of the information processing specifications and through the evolution of the requisite information processing capabilities.
Fourth, that evolution is the outcome of an information processing mechanism based on copying information such as skill specifications, and where the number of copies (zero plus) depends indirectly on variations in the information content (selection).
Fifth, that there are alternative information processing mechanisms that result in evolution. In specific, DNA, mimicry, experimentation, and some of the uses of human language (both oral and written), are alternate copying mechanisms that can lead to an evolution of the skills being copied.
Methodology:
First, investigate specific hypotheses arising from the theories with analyses of information processing requirements and constraints.
Second, architect and design simplified information processing systems that address the requirements and constraints.
Third, conduct feasibility studies by constructing simplified simulation models of the proposed information processing systems.
Fourth, investigate similarities differences between observed skill-based action patterns and the behaviour of the simulations.
Falsification can be done in two ways. We can find a skill evolution sequence that cannot be handled by the framework. We could also develop an alternate framework that is better and more complete in explaining the evolution sequence.
Research projects on skilled action (theories 1 & 2 above):
Identifying components and processes that are critical in converting skills to action: An investigation of the information processing required for basic physical action skills
For simple action such as walking, rotation about the joints must be controlled with coordinated timing. To do so, information must be sent to at least a pair of muscles in each plane of rotation, to control relative tension. This information must be renewed at roughly 10msec intervals. To keep all the muscles at appropriate tension, a high rate to data to the muscles is required. For our analysis, we simplify the joint angle geometry and focus on the angles. The first challenge is that we have to provide angle information relative to local limbs, but not relative to gravity, body orientation and the environment. For instance, to pick up a full cup of coffee, we have to keep the hand at right angles to gravity as we are changing angles at the waist, the shoulder, at the elbow, and at the wrist. The changes in angle must be complementary to keep the cup steady. There is a lot of processing to calculate these angles. Furthermore, there are a lot of time-related calculations to get the appropriate rate of change in the angles.
Picking up the cup of coffee without spilling is a skill that we can apply repeatedly. From an information processing perspective, the skill consists in the internal 'program' (to use the information processing analogy), that does the geometric and timing calculations. (For simplicity, we shall ignore forces and momentum.) All vertebrates must have such skills for locomotion etc. For innate skills, such 'programs' must be stored in the DNA and copied into the brain.
Early computers stored such programs in machine language, at a very fine level of detail. That worked for a few simple programs. It was found that larger and more complex programs could be compressed by using higher level languages. DNA also has a limited storage capacity. To allow for the storage of more, and more complex skills, we propose that there is an internal 'programming language' somewhat like ballet choreography (e.g. Labanotation) to allow the DNA representation of more, and more complex skills. Part of the evolutionary advantage of such a 'programming language' is that complex skills can draw on and reuse skill components of other skills. Such compression and reuse allows DNA to store a larger set of skills. Designing and simulating such a language is the focus of some of the initial feasibility models.
To continue with the coffee cup example, to keep it level it helps to look at the cup as we lift it. We therefore need to integrate information from visual perception into the action skill. We therefore bend the head and move the eyes as we bend at the waist etc. to lift the cup. We see that moving the head and eyes is itself an action that is coordinated with the cup lifting action. If we follow the head motion with a video camera, we see that the image and the perspective keeps changing. We therefore have another geometry problem, related to interpreting the 2D image. Integrating perception into the skill requires further programming. The geometry for the internal image must match the internal geometry for the action, so that we can adjust the action. We are currently exploring a 3D geometry linked to vision and the internal perception of gravity, to bring visual perception in line with controlling action.
Lifting coffee cups, while a nice illustration, is hardly shared across vertebrates. We use behavioural mimicry and experimentation as target skills for further investigation, since they illustrate the integration of perception and action (and since they are candidates for alternate mechanisms leading to the further evolution of skills - below).
To show evolutionary benefits for inter-individual communication (external language), we need to find coordinated complementary skills (division of labour) that require external communication for coordination. Because of sexual reproduction and differentiated roles in raising offspring, division of labour is common to vertebrates. For many innate skills, one-way communication with very simple messages is sufficient. Advertising for mates is the most commonly known.
For two-way communication, we speculate on more complex cooperative hunting strategies, or interactive herd-based defensive strategies. Hunting in packs using both flushers and ambushers is an example of complementary skills, where two-way communication may provide evolutionary benefits through increased probability of success. We are still looking for good examples that require two-way real-time communication, but are simple enough to be modelled.
Research projects on the evolution of skills (theories 3, 4, & 5 above):
Identifying processes that are critical to the evolutionary progression: An investigation into the information processing mechanisms that are likely to result in a process of evolution
DNA-based copying has an extremely wide information scope but a very narrow time scope. DNA conveys almost all attributes and functions of the new offspring, including innate skills. Variation is provided primarily through the two-way mixing while copying. The copying only occurs once, at conception. Selection is through survival and through competition for access to mates.
We propose that behavioural mimicry provides a second mechanism for information copying that is likely to lead to the evolution of skills. Copying involves an individual perceiving the actions of another, and then producing similar behaviour. The behaviour pattern must be abstracted into a language representation that it can be stored as a skill that can lead to similar action later. It is perception plus an internal stored representation that makes mimicry more than just a one-time action copy of a behaviour pattern. Variation occurs through inexact copying and through experimentation (below). If the mimicked behaviour pattern is more successful, the skill can supplant other skills (selection).
We hypothesize that experimentation provides a third mechanism for information copying that is likely to lead to the evolution of skills. By experimentation we mean a spontaneously generated behaviour pattern. One such pattern is curiosity, where the animal explores the surroundings. If the pattern is rewarded, it is more likely to be repeated. Once it is repeated consistently in similar circumstances then it fits the definition of a skill. Like mimicry, the information scope is narrow with an extended time range.
We propose that human language itself provides a fourth mechanism for information copying that is likely to lead to the evolution of skills. Much of schooling can be seen as copying language-based information and reproducing it for examinations and in other contexts. A lot of creativity goes into producing new language-based output, from poetry to scientific research to new financial instruments. Most of this is based on variations and combinations of earlier language productions. Successful earlier productions are more frequently copied and used as base for new variations.
Identifying processes that are critical to the development of the language skill, since a language like English clearly is not innate: An investigation into the within-human-individual evolutionary stages that are likely to result in language use
We see babbling as an innate skill that produces a form of experimentation, where some sounds receive more parental reward, and thus eventually predominate, assuming that each allophone cluster in the language is a separate target for selection. (Seemingly random arm and leg movements likely result from another innate form of experimentation that leads to coordinated action.) Turn, point, and name is another childhood behaviour that reflects innate skills at mimicry, and that also contributes to eventual mastery of spoken language. Schooling tends to use rudimentary language skills to produce more advanced skills. Mimicry is an important component of schooling, probably best illustrated by traditional school drills of learning by heart, reciting, and testing. Experimentation such as trial and error learning may be encouraged as creativity.
Identifying processes that are critical to the evolution of languages themselves, since only a few scholars still speak classic Greek or old Aramaic: An investigation into the within-human-species evolutionary stages that are likely to have resulted in both oral and written language, and the semantic and usage progression from Homer to the theory of relativity.
We start from the assumption that there was no or very little DNA-based evolution since the beginning of Homo Sapiens Sapiens. We therefore cannot use the DNA-based evolution of skills to trace the beginning of oral language, and its development through written language to the present. We therefore have to look at the other three forms of evolution for the progression of modern language usage and the skills associated with it.
In the oral tradition, stories were passed on, with more attention and better survival for the more successful ones. For scientific papers we look at the reference count, and how much new research is generated. Law uses precedence, business uses case studies, and medicine uses cases. All of these can be seen as knowledge and skills that are selectively passed forward.
To speculate on the evolution of communication we look at selection factors that would favour more complex communication. The most relevant evidence is based on the requirement to coordinate complex multi-individual tasks. Another factor is the timing and speed of the adaptation that is required. A human example of one-way communication is an instruction booklet that encodes the skill required to put together Ikea furniture, or to wire up a home theatre. Some of these instructions are difficult, and it helps to have someone available to answer questions. Some tasks take four hands or more. Operating room skills may be good examples.
Results
The theory supports Noam Chomsky's universal grammar, assuming it is based on the components of the inner language with its action components, geometric objects, and the timing. It also fits the observation that two engineers from different language can communicate better that the engineer and a poet in the same language group. The theory suggests that semantic spaces and 'shared reality' is linked to professional groupings with similar skill sets and action domains. The overlapping part of semantic spaces in a common languages may be limited to common tasks and common schooling.
Present status
Most of the effort to this date has gone into modelling basic action skills and integrating perception. The action component works quite well, based on a macro-expansion approach based on ballet choreography. Earlier versions were evaluated by displaying action with OpenGL or with animation software such as Maya. The new version is based on SVG and SMIL to be used interactively over the Web (http://www.konigslow.com - allows access to some of the work in progress).
Action versions with integrated perception are still under development. The challenge is to extract the central principles without being swamped by clever but irrelevant programming. Neuroscience has made excellent headway on the perception side, so we focus more on the integration of the output of perception with action and skills (planning, prediction, etc.)
Basic models of evolution based on the information-processing approach are running. Work presently focuses on the chunking of skills, i.e. on variability in skill representation that don't destroy the basic action.
Models abstracting skills from action or from experimentation are still in early planning - rough prototyping stages.