Database Reference
In-Depth Information
But it does not hinder the current work on the computational reconstruction of
natural language communication in DBS. The reason is an important differ-
ence between natural and artificial cognitive agents regarding what we call the
auto-channel and the service channel . 25
A natural agent has only an auto-channel for recognition and action. The
auto-channel evolves naturally during child development and comprises ev-
erything a natural agent can see, hear, feel, taste, etc., as well as consciously
do. The auto-channel also includes natural speech recognition and production.
In a standard computer, e.g., a notebook or a desktop computer, in contrast,
there is no auto-channel. Instead, there is only a service channel consisting of
the keyboard and the screen. It allows users and scientists alike to access the
hardware and software of the computer directly.
Building an artificial cognitive agent consists largely of reconstructing an
artificial auto-channel, with external interfaces for vision, audio, locomotion,
manipulation, etc. In higher cognitive agents, the auto-channel is connected to
the agent's autonomous control (Chap. 5).
The computational reconstruction of the auto-channel is an incremental pro-
cess which relies heavily on the keyboard and the screen. They function as the
service channel for direct access to, and manipulation of, the hardware and
software of the robot under development. After completion, the artificial agent
will be able to interact autonomously with the external world via its auto-
channel, including communication with the human user. However, in contrast
to a natural agent, an artificial agent will not only have an auto-channel, but
also a service channel as a remnant of the process of its construction.
The essential role of the service channel in bootstrapping the reconstruction
of cognition is especially clear in the area of natural language communication.
This is because it does not really matter whether a language surface gets into
the computer via the auto-channel or via the service channel. All that matters
for modeling natural language understanding in computational linguistics is
that the word form surfaces get into the computer, and for this typing them at
the keyboard of today's standard computers is sufficient.
Similarly for language production in the speak mode: all that matters for
the user's communication with the computer is that the surfaces derived from
prior cognitive processing be realized externally, and for this optical character
synthesis on the screen of today's standard computers is sufficient - at least
for users who can see and have learned to read.
The possibility to reconstruct natural language processing in artificial agents
via the service channel is good news for developing capable automatic speech
Search WWH ::




Custom Search