Log10 transformed frequency of each word was used to scale the error derivatives. This level of frequency compression was employed to reduce the training time in this large model (Plaut et al., 1996). The error (difference between the target and the output patterns) was estimated with the cross-entropy method (Hinton, 1989). No error was backpropagated from a unit if the difference between

the output and the target was <0.1 (i.e., zero-error radius parameter was set to 0.1). Momentum was not used in this study. All the units in the hidden and output layers had a trainable bias link, except for the copy and Elman layers. Weights were initialized to random values between −1 and 1 (0.5 for recurrent connections). Weights Ruxolitinib nmr from the bias unit to hidden units were initialized at −1, so as to avoid strong hidden unit activation early in training (Botvinick and Plaut, 2006). A sigmoid activation function was used for each unit with activation ranging from 0 to 1. At the beginning of each trial, activations for all units in the hidden layer (including vATL-output layer) were set to 0.5, and for all units in the insular-motor output layer to zero. This work was supported selleckchem by an

MRC programme grant to M.A.L.R. (G0501632), a Royal Society travel grant to M.A.L.R. and S.S., and a Study Visit grant from the Experimental Psychology Society to T.U. T.U. was supported by an Overseas Research Scholarship (UK) and an Overseas Study Fellowship from the Nakajima Foundation (Japan). “
“Language processing depends not only on cortical regions, but also on the

white matter fiber bundles that connect them (Geschwind, PD184352 (CI-1040) 1965, Wernicke, 1874 and Friederici, 2009). Traditionally the arcuate fasciculus was considered to be the main pathway connecting frontal and temporal language areas (Geschwind, 1965). However, recent studies using diffusion tensor imaging (DTI) have revealed that frontal and temporal language regions are connected by multiple dorsal and ventral tracts. Dorsal tracts include not just the arcuate fasciculus, but also other branches of the superior longitudinal fasciculus (SLF) (Catani et al., 2005, Frey et al., 2008, Glasser and Rilling, 2008, Makris et al., 2005 and Makris and Pandya, 2009). Ventral tracts include the extreme capsule fiber system (ECFS), which connects the frontal operculum to mid-posterior temporal cortex, and the uncinate fasciculus (UF), which connects the orbitofrontal region to anterior temporal cortex (Anwander et al., 2007, Croxson et al., 2005, Frey et al., 2008, Friederici et al., 2006, Makris and Pandya, 2009, Parker et al., 2005 and Saur et al., 2008). Syntax is one important component of language and has been shown in functional imaging studies to depend on both frontal and temporal language regions (Bornkessel et al., 2005, Wilson et al., 2010a and Pallier et al., 2011).