liveBook · Manning

Natural language

As explained in Machines that Think

A human language used for communication, as distinct from a formal language designed for computation or logic.

Natural language is a human language used to communicate, unlike a formal language designed for computation or logic. Through speech and writing, people coordinate actions, share knowledge, and express ideas across time and distance. Meaning depends not only on words, but also on context, intent, and the surrounding conversation. section 2.0

Language helps people name and compare experiences, connect ideas, and reason about things beyond direct perception. It can also guide attention: Guugu Yimithirr speakers use cardinal directions to describe locations, while Russian distinguishes light and dark blue as separate basic colors. section 2.1.1

For computers, language must be converted into machine-readable representations. Early approaches moved from explicit rules to statistical methods that counted words and predicted sequences; later representations mapped units of text into numerical spaces that capture patterns of meaning. section 2.0 section 2.3

Several features make language difficult to model reliably:

  • Ambiguity and implication: words and sentences can have multiple interpretations, while intentions often remain unstated and must be inferred from context. section 2.6.3
  • Long-range relationships and sparsity: meaning can depend on distant words, and rare terms may be important despite appearing infrequently in data. section 2.6
  • Linguistic diversity: languages encode information through different combinations of word order, morphology, tone, and script, so a technique effective in one may perform worse in another. section 2.6.5

Language is a useful interface for observing and using AI: people can describe a request, inspect the response, and refine it without learning specialized commands. But fluent language alone does not establish understanding. Human thought also includes spatial reasoning, visual imagination, and motor planning, which are not exhausted by linguistic expression. section 8.1.1 section 2.1.3 section 5.1.1

How Machines that Think covers it

How deep it goes
Traces language representation from rules and word counts to numerical spaces, while outlining ambiguity, long-range dependencies, and multilingual diversity.
Examples
Illustrates linguistic differences with Guugu Yimithirr cardinal directions and Russian light-blue and dark-blue categories.
What you can do after
Evaluate AI language behavior without equating fluent responses with human-like understanding or broader intelligence.
How it connects
Connects language to intelligence, pragmatic inference, statistical natural language processing, and large language models.

Also explained in

Compare how 5 books explain natural language →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Machines that Think. The section links point to the parts of the book it draws on.