Figure 1. Information flow in local and unified architectures for speech processing.

(A) Schematic characterization of the linguistic units used to characterize speech. The same phoneme can be invoked as part of a sublexical phoneme sequence, ph_k, or as part of word_j, ph_j,i. (B) Each box stands for a level of representation, characterized by its output and a probability distribution describing the level’s use of context. For example, the sublexical level’s output is an estimate of the current phoneme, ph_k, and the distribution for ph_k is estimated as probability for different phonemes based on the sound input and a sublexical phoneme history. At the sentence level, sentence_j,i stands for a temporary representation of the sentence at time j,i. Boxes represent functional organization rather than specific brain regions. Arrows reflect the flow of information: each level of representation is updated incrementally, combining information from the same level at the previous time step (horizontal arrows) and the level below (bottom-up arrows). (C) The unified architecture implements a unified, global context model through information flowing down the hierarchy, such that expectations at lower levels incorporate information accumulated at the sentence level. Relevant differences from the local context model are in red. Note that while the arrows only cross one level at a time, the information is propagated in steps and eventually crosses all levels.