Basic theoretical and practical information about PGEN++ system is given. The system allows: a) generating solving programs both for natural language and visual problem definition, b) reconstructing generating and/or other semantic models of programs/texts, c) performing intellectual transformation of programs. Generating object and event models (OSM) are described. Recognizing OSMs, which allow extracting knowledge from text statement, and transform them, generating output models, on which the program is generated, are offered. The problem of applying the layer of grammatical parsing of input texts is analyzed, a Turing-complete algorithmic microlanguage for interaction with such layer is constructed (on the basis of usual/constructing XPath). Inverse/direct logic output scripts are used to process the extracted knowledge. A new parallel direct sense output machine based on a system of weak XPath-like constraints is proposed, controlled by the Markov model, whose transitions matrix is synthesized by a neural network based on case studies. The approaches are tested in tasks of generating programs, verification of generating scripts and paralleling program transformations.