Skip to content

Understanding Humans

The world needs a proper model of human comprehension of natural language. This one is based partly on the OSI Model, partly on standard compiler design. Conceptually, it’s an interface stack: a set of layers of functionality. Each layer can talk freely within itself, and has a well-defined interface to the layers above and below, but never calls otherwise.

  1. Lexer / Phoneme Analyzer: This just tokenizes the input stream, whether it be audio or text. These are actually separate branches for the different types of input, but functionally do the same thing. Operates at the character/raw sound level.
  2. Parser: Checks grammar and syntax of input tokens. Generates all possible interpretations of homonym/homophone possibilities. Operates at the word level. Generates sentences.
  3. Local Contextual Integrator: Considers local details: formatting of text, source and quality of audio, etc. Gives a small, local “big picture” to frame the input in question. Operates at the word level, but by nature considers a variety of external cues.
  4. Literal Semantic Analyzer: Given the tokens and their context, decides what the literal meaning of a given sentence is. Operates at the sentence level.
  5. Source Knowledge Integrator: The source of a given communication is important. A message from a family member might be more trusted than a random internet article. A sentence from a very literal, precise person is more likely to mean exactly what it says than one from an excitable teenager. Operates at the sentence level.
  6. Conceptual Accumulator: Collects a bunch of related sentences into a paragraph-level concept. Decides what sentences are related and how they fit together.
  7. General Semantic Analyzer: Decides what the author probably meant in a particular paragraph. Resolves logical contradictions and paradox. Operates at the paragraph/concept level.
  8. General Contextual Integrator: Integrates a concept with a worldview. This is the level which decides if someone is lying, wrong, or otherwise speaking falsehoods. Operates at the concept level, though by nature includes a wide variety and broad scope of external information.
  9. Cognition: normal thought about ideas. Operates at the conceptual level. Can modify the rules of lower levels, for example when learning a new language or updating the current model of the current language.

Humans accomplish layers 1-5 automatically and unconsciously. Layers 6-8 are like breathing: generally subconscious, but can be consciously overridden. Layer 9 is when we start getting into id, ego, superego stuff: possibly subconscious but generally sentient behavior. Computers right now are quite good at steps 1 and 2 using artificial languages, and rubbish at them for natural languages. Layers 3 and above may exist as research projects, but are above the current state of the art.

When I go back to school, there is a high probability that this is the stuff I will focus on, trying to push the state of the art in computer thought up, level by level. It is fascinating!

RSS feed

Comments

No comments yet.

Sorry, the comment form is closed at this time.