Predictive coding (PC) is a well-known model of human cortical processing which states that prediction error measurements are the key driver of learning and inference. PC’s neural operation has been posited in a number of perceptual domains (including spoken word recognition, SWR). Despite PC’s popularity and plausibility, confirming its operation remains difficult. Here, a diagnostic signal reported to support the operation of PC in the brain during a SWR task (Blank & Davis, 2016) is investigated. We find that the computational models used in the report are highly non-standard in their architectures, undermining the reliability of the diagnostic signal as reported. We take efforts to restore these models to standardized (fully bi-directional) architectures, and find that even with restored models, this diagnostic signal is not present appreciably. We hold that PC can still serve as a valuable contributor to SWR theory, but that its operation in the brain during this SWR task cannot be supported using the previously reported evidence by Blank & Davis (2016). The importance of systematic, pre-publication content review is discussed here, as well as the value of considering the scientific publication as part of an open dialogue.
Bryce is a fourth year bioengineering Ph.D student in Kevin Brown’s Lab. He came to OSU from a benchtop biology and chemistry background, with the hopes of learning more Python to simulate biological systems. He works with computational models (both deep learning and deterministic) which are designed to emulate aspects of neuropsychology, in order to study the mathematics of the brain, particularly during speech recognition. He enjoys going to church to worship Jesus, spending time with his 1-year-old old son, and playing blues on the guitar.