Chapter Evolving Connectionist and Fuzzy Connectionist Systems: Theory and Applications for Adaptive, On-line Intelligent Systems

6.5. ECOS for evolving language systems
The areas of the human brain that are responsible for the speech and the language
abilities of humans evolve through the whole development of an individual
[2,52,56]. Computer modelling of this process, before its biological, physiological
and psychological aspects are made completely known, is an extremely difficult
task. It requires flexible techniques for adaptive learning trough an active
interaction with a teaching environment.
It can be assumed that in a modular spoken language evolving system, the
language (languages) modules evolve through using both domain text data and
spoken information data fed from the speech recognition part. The language
module produces final results as well as a feedback for the adaptation in the
previous modules. This idea is currently being elaborated with the use of ECOS.

---

131
Fig.6a. The average normalised values of each of the 26 
mel scale coefficients of the
reference data and the new speaker data on the phoneme /I/.
Fig. 6b. Adaptation of an already evolved EFuNN from NZEnglish phoneme /I/ data to a
new accent data on the same phoneme. After a single pass of and additional evolving
(adaptation) of the /I/ EFuNN on the new accent data, 9 out of 10 frames from the new
accent data were correctly recognised (none of the 10 speech frames from the new accent
were recognised before the adaptation took place). The y-axis shows the output activation
value of the adapted /I/ phoneme EFuNN to the new accent data (10 vectors).
7. ECOS and 
EFuNNs for on-line, adaptive, 
multi-modal
(speech, image, text) information processing
Several methods for multi-modal information processing that involve images (e.g.,
lip movement) to enhance speech recognition have been developed [23,55,73].
Other methods use speech to enhance image recognition. But when the
multimodal -based recognition (or identification) process has to be performed in a
real time, on-line, adaptive mode, most of the above methods would fail to achieve
satisfactory results. That is because of the speed of the processing needed and a
method of adaptation needed that can deal with fast adaptation to new data, some
of them presented only for a very short period of time, in a noisy environment.
Here, a brief reference to a framework AVIS for integrated auditory and visual
information processing published in [43 ], is made. In sub-section two the use of
ECOS and EFuNNs for the implementation of AVIS is discussed and directions
for further implementations are given.