Data analysis

Reaction times (RTs) and hit rates (HRs) were computed separately for each type of stimulus (An, Vn and AV) and side (left and right), and analyzed using a repeated-measures analysis of variance (ANOVA) with type (An, Vn, and AV) and side (left and right) as subject factors. All statistical analysis was performed using SPSS software (version 22, IBM Inc., UAS). The α level was set to 5%.

The processing of each type of stimulus (An, Vn, AV and Fn) is described in Table 1. The effect of early sensory integration can be removed by comparing the audiovisual interaction in AV+Fn condition relative to those in the Vn+An condition, allowing one to extract the effect involved in the semantic integration. For all left sessions, the comparison of (AV+Fn) and (Vn+An) presented on left side reflected the effect of semantic integration with attention, while the comparison of that presented on right side reflected the effect of semantic integration without attention. The situations were similar for all right sessions. The regions associated with the modulation effect of attention were identified by comparing the semantic integration with and without attention.

The fMRI imaging data were analyzed using the SPM12 software package (Wellcome Department of Cognitive Neurology, London, UK) running under Matlab2012a (MathWorks Inc., Natick, Massachusetts, USA). Six scans at the beginning of the measurement were removed automatically from the data set. Functional data were slice time-corrected, motion-corrected, normalized into standard stereotactic space using the Montreal Neurological Institute (MNI) template, and smoothed using a 6.0-mm full-width half-maximum Gaussian kernel. To reduce motion-related artifacts, session-specific realignment parameters from preprocessing were used as first-level covariates. Statistical analysis was performed at the individual participant level by using the general linear model framework, and the blood oxygen level-dependent response was modeled as the neural activity convolved with a canonical hemodynamic response function. The contrast of (AV+Fn > An+Vn) with attention and (AV+Fn > An+Vn) without attention in all left-sessions and right-sessions were implemented. All individual functional localization data were then used for the group-level statistics. One-sample t-tests were used to construct statistical parametric maps at the group level for (AV+Fn > An+Vn) contrasts, determining the voxels in which activity differed significantly from zero, i.e., the voxels that showed significant activity in the processes of audiovisual semantic integration with and without attention.