Cohort model

[12] As more speech segments enter the ear and stimulate more neurons, the competitors that no longer match the input to be kicked out or to decrease in activation.

These processes continue until an instant, called the recognition point,[9] at which only one word remains activated and all competitors have been kicked out.

[10] Finally, in the integration stage, the semantic and syntactic properties of activated words are incorporated into the high-level utterance representation.

When the hearer hears the first two phonemes /k/ and /æ/ ((1) and (2) in the image), he or she would activate the word "candle", along with competitors such as "candy", "canopy", "cattle", and numerous others.

[17] Since its original proposal, the model has been adjusted to allow for the role that context plays in helping the hearer rule out competitors,[9] and the fact that activation is "tolerant" to minor acoustic mismatches that arise because of coarticulation (a property by which language sounds are slightly changed by the sounds preceding and following them).

For example, in Pienie Zwitserlood's study of Dutch compared the words kapitein ("captain") and kapitaal ("capital" or "money"); in the study, the stem kapit- primed both boot ("boat", semantically related to kapitein) and geld ("money", semantically related to kapitaal), suggesting that both lexical entries were activated; the full word kapitein, on the other hand, primed only boot and not geld.

For example, some studies showed that "shadowers" (subjects who listen to auditory stimuli and repeat it as quickly as possible) could not shadow as quickly when words were jumbled up so they didn't mean anything; those results suggested that sentence structure and speech context also contribute to the process of activation and selection.

[6] Research in bilinguals has found that word recognition is influenced by the number of neighbors in both languages.