19:00 FRDCSA Project, Perl5/6 as AI Knowledge Representation Language, etc.etc.

This talk will serve to discuss the FRDCSA mission, status etc.

Then I will present on the status of new research into goal of arbitrary document parsing based on the usage of Perl5/6 as a knowledge representation glue language. The undecidability of parsing Perl, and the PPI parser, are stepping stones into this task. One of the core systems for this approach is called "Sayer". Sayer is based on search over possible meanings (among many techniques). It uses "deductions", that is, memoized deterministic function/args/results triples, and machine learning over this data to predict and suggest deductions, in order to "prove" things about data structures.

For instance, rather than a classical knowledge representation approach to representing, for instance, the fact that the string 'Jill' represents a female name, i.e.

(#$isa "Jill" #$FemalePersonName)

the system records the success of the following function call

Function: Aranea::Modules::LookupPerson::do_Module Args: ["Jill"] Results: [1]

A large library of real-life function calls is compressed and stored by the "Learner" system, using "FreeKBS" to store triples.

Simultaneously, it knows the results of the string "Jill" being passed to many other functions.

Various similar "deductions" are used as features to a machine learning system which generates further tests to run on the data, factoring in expected run time,

with heuristics learned by machine learning over memoized function arguments/results on large software systems and "inferences" represented as function/args/results triples and chains of these triples