Sentence Analyzer - enabling business and enterprise applications to handle sentences and text

While working on the document-extraction demo at text2data.net, the repititive nature of sentences in most officialese documents got me thinking on a simple way to handle sentences.

Sentences seem to form structural as well as content boundaries in a document. Any information extraction process becomes doubly powerful once sentence structure in included among other things.

My approach to sentencs is essentially document-centric, the identification of a document, the parsing of it using structural or linguistic boundaries etc.

Existing NLP practices seemed to me too inflexible, incapable of consuming a wide range of sentences, and of outputting the results in a way that business users would find friendly enough.

I also felt that any software using sentences ought not to inflict upon the user any other learning curve except other sentences.

The basic rule I have followed throughout is to go wherever grammar (i.e. whatever little I learned at high school) leads, and not depend on maths and statistics. So that if needed, the real grammar experts could improve the whole thing simply by using better grammar rules.

That way, I hope to include stronger and better grammar rules in the next version, as soon as I can get hold of an expert.

The key design principle is KISS, do the least minimum, keep many things configurable etc.

More reading for those interested...