The presence of semantic content in a visual recognition memory task reduces the severity of neglect