. . . . . . . . . . . . . . RAMSCAR . . . . . . . . . LAB . . . . . . . . . PAPERS . . . . . . . . . LAB PEOPLE . . . . . . . . . RESEARCH . . . . . . . . . LIBRARIES
cognition language & learning lab
melody explains...
language as a predictive process
department of psychology, stanford university

Language is Uncertain

Let us adopt the view that we learn words through an on-going process of refining our understanding of the sets of features (in the world and in language) that cue them. If words are "categories," we learn to discriminate between them and use them correctly through trial and error, arriving over time at an ever finer approximation of what aptly constitutes a given word's semantic space - that is, its most predictive set of cues. Cues here can be understood to be both cues in the physical world (e.g., long necked, yellow and spotted might cue giraffe) and cues in language (words are more or less likely to occur given the words that precede them, e.g., hit the nail on the ____). In learning the set of cues that predict a given word, we get better and better at understanding how that word is used. As Wittgenstein or Quine would say, we don't learn meanings for words; rather we learn the criteria for their use.

Under this notion, the cognitive representation of a word is a predictive representation. (For example, we might come to know that this likely counts as a 'this' because it has x-y-z features that have reliably cued 'this' in the past.) Cognitive representations are not, by this definition, discrete; rather, they are statistical representations of likelihood based on prior experience and cue competition. This is why children often confuse objects with overlapping cues (cats and dogs) and adults may confuse an unfamiliar exemplar of a familiar category (e.g., a postmodern 'chair').

(Importantly, this is a strictly non-referential account of language; that is, words do not refer to their meanings or instatiations of their meanings, nor do they attach to them in even a figuative sense; rather, they are abstractions of what they mean.)