The Power of Feedback
Charts depicting topics and situations for each paragraph can enhance human-parser
interactions in several ways. The data flow diagram
shows two important ones which take the form of feedback loops:
-
The upward
arrow at far right shows that MORPHOLOGY can heuristically rank possible incoming word
senses by using semantic models still being built and/or topics recently
accessed. This negative feedback loop impacts
short term memory as words are interpreted, directs attention to senses which relate, and makes parsing simpler by
reducing lexical ambiguity.
The arrow pointing upwards to LEXICON shows vocabulary
improvements arising from semantic models of sentences.
Their assertions, if properly verified, can
expand and refine models of word meanings within long
term system memory. This important knowledge
acquisition pattern is the core of human education - sometimes called
learning by being told.
Effect #1 benefits the parser itself. It takes a tight coupling between components,
such that adding a meaning for word N to the model of a partly analyzed phrase can
reduce lexical ambiguity for word N+1. This effect can help out any parser, but
it may be especially vital to improvements in continuous speech recognition.
And What of Effect #2?
This effect is more global, and it can be gained flexibly by post-processing charts.
Such processing can be part of an NLP application in model Q&A or voice I/O, or it can
be done much later, asynchronously, to help software learn from what users said.
Both alternatives
should reduce the costs of developing smart application software.
Our national power to exploit intelligent agents is currently much impeded
by our inability to communicate with them.
Charting software can ease such limits, and enhance translation,
multimedia indexing and retrieval, message routing, content management,
voice synthesis, knowledge representation, and many related fields.
Topic Map and RDF communities in particular will benefit from charts,
which can be accessed under either metadata paradigm.
This new form of robust bulk input will stimulate new R&D in several ways,
as chart-processing becomes a major new mechanism for training agents
in knowledge-based applications as well as driving them at run time.
All of this forms a positive feedback loop. As engineers know, they often
cause explosive growth - a nice effect we predict for chart-related sales.
|
