People can “get by” in a new language with only a couple of thousand words. So why don’t today’s systems scale to understand language?
An obvious way to scale a system to use human language is to start with a machine that has some of the capabilities of a human brain and then learn as human children do, asking questions when needed.
But what does a brain store to use language? Last time we looked at the science of semiotics that splits language into an arbitrary sign, and one or more interpretants/definitions (to interpret the sign) that represents an object.
Definitions can be one of two categories of meaning: referents or predicates. Referents are the things we experience — cars, people, streets, countries and products, for example. Predicates relate referents as states or activities — happy, running, tall, or rich, for example.
Let’s introduce one more concept from Patom (brain) theory: specific versus general[i]. You can only ever experience specific things, but you can then describe them with general terms. Coke is a drink, a Big Mac is food and a 911 is a car. Drink, food and car are general terms (you can only see examples of these), while Coke, a Big Mac and a 911 are specific (you can see one).
