Folding

Currently, we stream the words of a document into a grid (a list of string cells), perform small local changes that do not rely on inheritance of properties (15 psi, $4,000, etc.), and then build individual sentences into parse chains.

There are several problems with this:

· It is hard to see across parse chains, and it is often necessary to run back a long way to see how a nested list started, for example, so that a decision can be made between a noun and a verb.

· We introduce too much detail into the parsing process, detail we than have to abstract out again.

· The network parsing process is slow, relying on the building of structure.

We can do a lot more at the grid level.
We can turn noun phrases into noun groups.

There is a lot of resource expended in building a noun phrase.

nounphraseParse.JPG (255826 bytes)

If we build a NounPhraseGroup instead, we get

nounphraseGroup.JPG (170845 bytes)

The reduction in parsing effort is obvious. If we extend this to compound (nouns joined by conjunctions) and composite (nouns having a prepositional tail) noun phrases, we will greatly reduce the parsing effort at the network level.

Design Notes