Hypothesis-Test Cycles and Symbol Grounding

Abstract

Language acquisition is a process of symbol grounding, which is construction of sym-
bol systems adapted to environment. Environmental adaptation defines the values
which cognitive agents pursue primarily by means of hypothesis-test cycles encompass-
ing both the inside and outside of their bodies. In addition to these directly grounded
cycles, there are also hypothesis-test cycles within cognitive agents. Cognitive processes
are combinations of these cycles, where cycles embody typical cognitive phenomena such
as navigation and language use.
Cycles are essentially countable, so that systems comprising cycles necessarily have
discrete structures. A cognitive agent is hence formulated as a discrete system consist-
ing of cycles including both directly grounded cycles and symbols (indirectly grounded
cycles), where each cycle embodies some value or meaning directly or indirectly associ-
ated with environmental adaptation. Computational models of cognition as combina-
tion of such cycles (values = meanings) are far more efficient (simpler and less prone to
overdesign) than traditional models stipulating possibly non-cyclic information flows.
Environmental-adaptation cycles operate at multiple spatiotemporal scales, including
real-time adaptive behavior, middle-term learning, and evolution across generations. It
is vitally important to address real-time adaptation behavior in terms of cycles, which
will raise the efficiency of the computational model not only at the level of real-time
adaptation but also accordingly at higher levels. Cycle-based (meaning-based) com-
putational models are necessary also because cycles derive meta-level constraints such
as symmetry bias and naming insight, which are indispensable for abductive reasoning
and language acquisition.
Existing technologies including deep learning fail to reflect such a value-based
(meaning-based) architecture of cognition. For the sake of thorough symbol grounding,
novel approaches are necessary which should integrate environmental-adaptation cycles
in the entire computational model at multiple levels of meaning and value.