Some properties of natural language

• Contextualized
• Goal-driven
• Adaptive
• Underspecified: "I saw her duck"

Goals

The Turing Test for artificial intelligence: a machine is intelligent if a human, asking questions (by typing), can't distinguish it from another human (1950).

Eliza (1966): a program that used simple pattern-matching and string substitution to simulate a Rogerian psychotherapist. Many people were fooled.

People treat computers as normal social entities:

• when a computer asked a human to evaluate how well the computer had been doing, the human gives more positive responses than when a different computer asks the same questions.
• people give computers higher performance ratings if the computer has recently said something flattering to the human.

Hal, Star Trek computer: NL as the standard mode of human-computer interaction

Some History

Symbolic paradigms developed in departments of linguistics and computer science: generative linguists such as Harris, Chomsky; AI researchers such as Minsky working on reasoning and logic

Stochastic paradigms developed in departments of statistics and electrical engineering: optical character recognition, letter probabilities. Later, speech recognition.

Logic-based paradigms lead to work on grammar formalisms which had consequences both in programming languages (Prolog) and formal theories of syntax (LFG, Unification Grammar)

Natural language understanding: predicate logic, human conceptual knowledge such as scripts, plans and goals

Discourse modelling: substructure in discourse, discourse focus, reference resolution, speech act interpretation