ORB Visualization
(soon)
National Knowledge
Project
comments
In regards to Dr. Ballard’s remarks, everyone has an opinion, but few have an implementation, and even fewer have one that is bottom up and generalizable.
Some scholars are stuck in the all-or-nothing model of meaning -- which is indeed intractable. For these processes (meaning and understanding) to have physical computable realization, they need to reduce to simpler operations whose combined results produce the aggregate outcomes.
The problem of reference (specifically internal reference) is the key issue. No progress can be made without a bootstrap decomposition of reference -- and you will see when you pay attention, few are even aware of this central point let alone prepared to bring forward an implementation and corresponding theoretical approach (which must include an account of semantic perception).
Nobody will be able to render a useful opinion unless they already at least almost understand. But almost understanding is not enough, one has to also be aware of the limits to single coherent viewpoints.
Consequently, it will be useful to include a demo of the working system answering questions, parsing text, generating sentences. When we look at how the system performs these operations, then the picture can become much clearer to those who almost already understand a little bit.
People interested in the approach to meaning might benefit from reading on hermeneutics.
Regarding this notion of aligning semantic (entity-relationship) networks with natural language syntax patterns. How does one justify a belief that grammatical relations from syntax are not projected into semantics?
It is true that language communicates agreements between individuals whom have already a shared understanding of the concept description. On the horizon of experience we rely on these shared agreements, but when there is separation having an ontological status, then we move back to the problem where phenomenology is separated from the interpretation of signs and symbols.
Language is dependent upon shared levels of education and successful mutual assumptions of purpose and intent. But perhaps these observations form evolutionary psychology is at the wrong level of analysis.
Type-consistent description logics can't do very much, especially with the important parts of language like tropes (metaphor) and belief contexts (inherent in strategic language, competition, cooperation).
But, machine parsable text (and understandable) is a useful concept that is part of the research program.