The primary objective of this chapter will be to examine the few models of multilingual speech production that are currently available. Since most of these models are based on proposals originally made in relation to monolingual and bilingual speakers, I shall first summarize two key models of monolingual speech production (Dell, 1986 and Levelt, 1989) before proceeding to examine those models of bilingual speech production which are argued to account for multilingual production as well.

Two Influential Monolingual Speech Production Models: Dell (1986) and Levelt (1989)

Among the most influential models of monolingual speech production are those proposed by Dell (1986) and Levelt (1989). These two models substantially differ in their assumptions of how feedback between levels of encoding operates, and in their views of how the speech production process takes place.

From an interactionist perspective, Dell (1986) believes that encoding occurs by way of activation spreading from node to node between levels of encoding. Activation is the driving force behind the production process, and the various levels of encoding interact with one another passing on feedback in two directions: from higher to lower levels of encoding (feedforward), and from lower to higher levels of encoding (feedback). In contrast, Levelt (1989) maintains that processing is strictly modular and no feedback between levels occurs. Due to this inflexibility with respect to the flow of information, models of this kind are sometimes called feedforward models, in that they only allow input to be passed on to the level of encoding that is next in line.

Dell's (1986) model is based on the principle of spreading activation, which is a retrieval mechanism by which speakers are able to select the relevant information during the entire speech production process, from message conceptualization to the articulation of speech. If we take picture naming as an example, spreading activation would work as follows.