Background

ReVerb takes sentences and breaks them up into triples with confidence levels. The ReVerb example code passes in a simple sentence, which it parses into an accurate triple.  This test is based on a compound sentence, in which there are two sentences formed, one around a belief, and one formed around a causal claim; the belief depends on the causal claim.  Here is what ReVerb did:

Given

"Many engineers believe that CO2 is the cause of climate change."

Results

{(Many engineers, is the cause of, climate change)=0.7707088312566096,
(Many engineers, believe, that CO2)=0.15512383345636682}

Comments

In an ideal situation, a triple would be nested like this:

{Many engineers, believe that, {CO2, is the cause of, climate change}}

To implement that, each triple should really be a quad, with identity as one of the cells, in which case, the result might look like this:

{101, CO2, is the cause of, climate change}
{102, Many engineers, believe that, 101}

In that case, to render functional tuples, we would need ObjectType, and, for that matter, SubjectType cells in each tuple; that supports inferences and search.

While we are on the subject, each tuple should have a cell which carries ACL (access control list) codes.

Tuple creation along the lines just suggested were part of my thesis research and will be documented elsewhere in this project, and linked here when available. Related study should look closely at what Conceptual Graphs bring to the table.