How spoken languages work in the absence of an inventory of discrete units

Historically, linguists and psychologists have generally assumed that language is a combinatoric process, thereby taking the idea that language users have access to inventories of discrete, combinable units (phonemes, morphemes, words, etc.) for granted, despite the fact that these units have tended to resist formal definitions. We propose a new approach to language understanding based on the psychological mechanisms that underpin context-sensitive processing. This new method is surprisingly simple, in large part because it embraces a view of learning that has been developed from studies of animal behavior and neuroscience. From this perspective, learning is seen as a systematic, discriminative process that seeks to reduce a learner's uncertainty in making moment-to-moment predictions. We suggest that language processing employs all the information available to the listener at any given moment to predict what will happen in the next moment, in the next couple of sentences, etc. This approach does not rely on any of the ambiguous traditional linguistic units because continuous-time processing simply acts to reduce a hearer's uncertainty about an actual message in relation to possible messages, rather than building up an interpretation out of elemental components. From this perspective, the conventional units of language - phonemes, morphemes, words - can be seen as idealizations of patterns that evolved for communicative efficiency that can serve the purposes of orthographic (and linguistic) description, rather than psychologically 'real' elements that are essential to language processing.