Towards "Natural" Natural Language in Machine Cognition

Machine understanding of language would obviously offer numerous business opportunities such as truly conversational human-robot interfaces, conversational information systems, automated text understanding and summarization, information search, junk mail filtering and eventually true thinking machines. Traditionally the problem of automated language understanding has been approached from the linguistic point of view; the utilization of formal syntax and lexicons. So far, the results have left much to be desired. Obviously the brain does not process language in this way and there must be another, natural way to do it. It is proposed that machine understanding of a natural language should be based on the emulation of the cognitive processes of the brain instead of preprogrammed formal syntax and vocabulary. Meaning should be grounded to real world entities. Same processes for perceiving and understanding the world and linguistic understanding should be used, no specific linguistic machinery is assumed to exist. In the brain sensory processes provide information about the world and this information is integrated into a continuously updated multimodal “situation model”. A natural language is seen as means for the description and evocation of these inner “situation models”. The author’s proposition for the artificial implementation of these principles is called “The multimodal model of language” and its main application area would be the use and understanding of natural language in cognitive machines and robots.