What does it take (for a robot) to be a ‘linguistic agent’?

  • A linguistic agent can be visualized as a robot navigating, interacting with, and ultimately contributing to the building (“terraforming”) of a Universal Knowledge Graph.
  • The primary criteria of success is not simply guessing the correct answer, but demonstrating a full understanding of what is being said – by generating the full logical form for any sentence or phrase that is being processed in a dialog. The Logical Form must state BOTH the denotation of a sentence, AND also the set of relevant presuppositions.
  • Conversational agents always have to deal with other such agents. A multi-agent environment is the natural habitat of a linguistic agent. In a multi-agent setting, each agent exchanges messages with other agents. This goes to the very essence of natural language.
  • Importantly, it must have access to a scientific linguistic parser, able to handle ‘ellipsis’ and other kinds of gaps and recover the “understood” material.
  • As is well known, quantifiers play a very central role in the semantics of natural language. In order to support quantifiers, it is necessary to handle bound variables.
  • Finally, no conversational agent can get away with failing to understand elementary logic – it is a basic component of the normal use of language. To qualify as a linguistic agent, a learning agent must be conversant in Second Order Predicate Calculus (and some elementary set theory) – even BEFORE starting to learn its own field of specialization. This requires machine learning to be in some way combined with symbolic reasoning.