Over the past 20 years, one of the most contentious issues in language production has concerned the degree to which lexical access occurs in discrete stages. Theorists agree that words are retrieved first as semantic-syntactic entities, and then later spelled out in terms of their phonological forms (e.g., Garrett Reference Garrett and Bower1975; Kempen & Huijbers Reference Kempen and Huijbers1983). But is the first of these steps – lemma or word-access – entirely separate from the second one – phonological access? Three possibilities are debated. The discrete-stage or modular view (Levelt et al. Reference Levelt, Roelofs and Meyer1999) holds that the first step must be completed before any activation of phonological forms takes place. During the first step of retrieval of the word “cat,” the lemmas for CAT and DOG may both be active, but the phonological forms of these will not be. Not until the first step has completed its selection of CAT can /k/, /æ/, and /t/ gain activation. The cascade hypothesis blurs the distinction between the steps by allowing for activation of phonological forms of potential lemmas (e.g., the forms of both “cat” and “dog”) before the first step has been completed. The interactive hypothesis permits cascading, but also bottom-up feedback (e.g., Dell Reference Dell1986). Activated phonological units send activation upwards to lexical units. This loop of cascading and feedback between units at adjacent levels of the system is assumed to operate regardless of whether the lexical access process is engaged in word or phonological access. Currently, there is a considerable amount of evidence for cascading (e.g., Cutting & Ferreira Reference Cutting and Ferreira1999), but little consensus on the degree to which the system is interactive (see Harley Reference Harley2008, for review).

I suggest that Pickering & Garrod's (P&G's) proposed use of forward modeling and the predicted perceptions that result from it map onto the notions of cascading and feedback in interactive models of production. Cascading consists of a prediction by processing level i of what needs to be active on the next lower level i+1, and feedback from that level delivers the anticipated “sensory” consequences of that prediction back to level i.

In P&G's view, the advance prediction of representational components of an utterance and their sensory consequences allow for each production decision to be coordinated with other decisions. They illustrate by showing how heavy noun phrase (NP) shift and phonological error monitoring could result from this system. Generally speaking, forward modeling during production helps make the many parts of an utterance mesh for accurate fluent speech. That is also the function of cascading and feedback in hierarchical production models, except that representational levels rather than utterance parts are what are being meshed. Cascading of activation to lower levels prepares the way for the construction of representations at those levels. The resulting feedback allows for decisions at the higher representational level to be sensitive to information at the lower level. For example, feedback from phonological forms to the word/lemma level allows for word selection at the higher level to reflect the retrievability of the form (Dell et al. Reference Dell, Schwartz, Martin, Saffran and Gagnon1997). A phonological form that is more easily available will feed back more activation to its lemma than a form that is difficult to retrieve will. Hence, the system will be biased to select lemmas whose forms will be available. Feedback also enables representations at a lower level to mesh with higher-level information, as seen when feedback from phonological to lexical levels biases the phonological level activations toward lexical outcomes, functioning as a lexical editor (e.g., Nozari & Dell Reference Nozari and Dell2009).

P&G emphasize that forward predictions are not actual production representations. They are “easier-to-compute ‘impoverished’ representations” (target article, sect. 3.1, para. 6). This is also true of cascading in interactive models. Units that are active through cascading are less active than they would be if they were committed parts of a representation. Furthermore, unlike committed representational elements, they have yet to be bound to structural frames, at least in activation-based models that use such frames (e.g., Dell Reference Dell1986). So, although units activated through cascading may soon be fully part of an utterance's representation at a particular level, they are not there yet.

P&G have outlined a compelling integrated theory of production and comprehension, showing how each contributes to the other. With this commentary, I invite them to consider whether the notions of cascading and feedback in production are part of the picture, and particularly whether they can be considered to be reflections of the forward modeling system operating between processing levels.