Currently, we return an extraction_keyword with all extraction bricks, which is passed to the function of the brick and then returned again, which is not efficient. Instead it might make more sense to return <value>, <start_char_idx>, <end_char_idx>. This need to be then applied for all bricks.