What conceptual graph workbenches need for natural language processing

An important capability of the conceptual graph knowledge engineering tools now under development will be the transformation of natural language texts into graphs (conceptual parsing) and its reverse, the production of text from graphs (conceptual generation). Are the existing basic designs adequate for these tasks? Experience developing the BEELINE system's natural language capabilities suggests that good entry/editing tools, a generous but not unlimited storage capacity and efficient, bidirectional lexical access techniques are needed to support the supply of data structures at both the linguistic and conceptual knowledge levels. An active formalism capable of supporting declarative and procedural programs containing both linguistic and knowledge level terms is also important. If these requirements are satisfied, future text-readers can be included as part of a conceptual knowledge workbench without unexpected problems