Poster B86, Tuesday, August 20, 2019, 3:15 – 5:00 pm, Restaurant Hall
An analysis of the dual route theory of reading using neural decoding
Aidan Curtis1,2, Oscar Woolnough2, Cristian Donos2, Patrick Rollo2, Nitin Tandon2,3;1Rice University, 2University of Texas Health Science Center at Houston, 3Memorial Hermann Hospital, Texas Medical Center
The dual route theory of reading continues to be the most credible description of the neural organization to enable the mapping of word orthography to its pronunciation. The dual route model originates from data on patients who, due to brain lesions, rely exclusively on one of two routes - A phonological route for grapheme-to-phoneme conversions or a lexico-semantic route for direct lexical access. However, in healthy subjects both routes are engaged simultaneously – this implies that network level representation rather than focal neural activation in any individual region would be more informative of overall processing. To evaluate the validity of the dual route model, we combined the high spatiotemporal resolution of intracranial recordings with a logistic regression model, to examine the dynamic nature of the phonological and lexical streams of reading. 35 intractable epilepsy patients implanted with depth or subdural electrodes to localize the focus of epilepsy, were asked to read word stimuli comprising of (i) regular words, (ii) exception words (orthographically irregular), (iii) pseudo-homophones (orthographically novel but phonologically familiar) and (iv) pseudowords (orthographically and phonologically novel). A logistic regression temporal neural decoding model was used to track the spatial distribution of task-relevant information over time. Features including filtered time courses, analytic envelopes of activation, and pairwise connectivity were used as input into a linear neural decoding model, that was designed to maximize classification accuracy of stimuli from the time-localized brain state. Model hyperparameters and input feature mappings were selected via a Bayesian hyperparameter optimization algorithm to maximize stimuli distinguishability. We found high word type classification accuracy between words and pseudowords in the first 250 ms following stimulus onset. Analysis of the decoding model’s coefficients at this time showed a high contribution to the decoding accuracy from electrodes in the anterior superior temporal gyrus and mid-fusiform cortex. Around the time of articulation, we saw higher word vs. pseudoword accuracy (both pre- and post-articulation) than when aligned to stimulus onset. Decoding performance around the time of articulation was mainly influenced by the inferior frontal gyrus, and a high gamma differential was found between words and pseudowords in this region. This word-pseudoword distinguishability lends credence to the dual route theory, which proposes that these two classes of words utilize separate mechanisms for word identification and verbalization. Temporal neural decoding enables the tracking of task-relevant information without imposing prior assumptions about anatomical or oscillatory features in the data and can give insight into which cortical regions and nonlinear feature mappings are useful for identifying differences between trial classes.
Themes: Reading, Methods
Method: Electrophysiology (MEG/EEG/ECOG)