Objectifying the Subjective Self

parsing and generation (considered from a competence view) are non-deterministic processes, i.e., without any further (non-grammatical) information both processes have inherent degrees of freedom; cf. also Shieber (1993). Furthermore, the mentioned properties (reversibility, implicit search space, and modularity) are also important in the context of the “recursion-only hypothesis” discussed by Hauser, Chomsky & Fitch (2002), who claim that recursion (i.e., providing the capacity to generate an infinite range of expressions from a finite set of elements) is the only uniquely human component of the faculty of language. However, at least from a language usage perspective, concrete language utterances seem to be deterministic, i.e., at some point some decisions are made. What is the nature of these decisions, if they are not grammatical? At least two possibilities can be considered: either the decisions are based on preferences (which are learned through past experience) or through control, i.e., explicit strategies that are used to interpret the results of other processes in order to provide feedback. I consider preferences as “un-intelligent” in the sense that they are merely applied (blindly) and control-strategies as “intelligent” because they are applied purposely. Of course, both aspects are somewhat integrated, i.e., language processing is both preferencedirected and controlled. It seems that the interleaved approach and the anticipatory drive at least have important properties that classify them as an explicit control-strategy. As already said at the beginning, the proposed computational model is mainly rooted in mathematics and computational linguistics and does not claim any cognitive “realism”. However, the realization of the underlying ideas (i.e., grammar reversibility, uniform parsing and generation, self-monitoring) on such a technical algorithmic level requires fine-grained details. Furthermore, the idea of the interleaved approach of parsing and generation is also strongly motivated by the assumption that complex anticipation feedback loops are necessary for the modelling of highly self-adaptive natural language systems. And as such, it might be of interest for the further outline of the model of the anticipatory drive proposed by Butz, especially concerning the aspects of language and modularity. Clearly, the computational model in its current form is realized on a high symbolic level. Probably, it is too high to integrate it directly on a neuronal level. Seen as such, it could be of scientific interest to explore a) how to integrate sub-symbolic approaches into such computational models as I have outlined, and b) how to integrate such complex symbolic interactions into the model of the anticipatory drive.