The general model of language underlying the course, and put over more
or less explicitly at different stages, was as follows. Learning a language is
essentially learning to make correct predictions. As soon as something is
said, one can make predictions about what is likely or unlikely to be said
next. Expectations may be broken, predictions may be wrong, and people
say unexpected things. However, unfulfilled predictions show that there
were predictions made, and whatever does occur is interpreted in the light of
what was expected. If something is not expected, this surprisal value is part
of its meaning. These points can be reformulated in terms of redundancy: if
an item is predictable, then this means that it is redundant. Linguists are
fully familiar with this information theory approach to language and meaning.
And they are fully familiar with the way in which it is equally applicable
to all levels of language: phonology, graphology, lexis (e.g. collocations),
syntax, semantics and discourse. It is also evident to linguists how this
concept may be reformulated into a concept of structure as constraints on
linear sequence. In one way or another, all modern linguistics is based on the
concept that language is polysystemic: that is, there are always constraintson linear sequence, and different paradigmatic choices are available at
different points in the sequence.