On True Language Understanding

Abstract

Despite the relative successes of natural language processing in providing some useful interfaces for users, natural language understanding is a much more difficult issue. Natural language processing was one of the main topics of AI for as long as computers were put to the task of generating intelligent behavior, and a number of systems that were created since the inception of AI have also been characterized as being capable of natural language understanding. However, in the existing domain of natural language processing and understanding, a definition and consensus of what it means for a system to “truly” understand language do not exist. For a system to understand an idea, firstly it has to ground the meaning of the concepts in the idea that it manipulates - the concepts that are associated with the words it inputs and outputs. However, there has not been any standardized consensus on what constitutes adequate semantic grounding. This paper presents a spatio-temporal representational method as a basis for a specification of what constitutes adequate semantic grounding, particularly in connection with certain words and concepts related to grounding of physical concepts and mental constructs. This research has critically important implication for learning – true language understanding will usher in an era of learning through language instruction, which is how humans learn, to rapidly accumulate a vast amount of knowledge critical to the propagation of the species and the advancement of its civilization.