viterbi_paths[pos, current_state] = np.argmax(
                    viterbi_scores[pos - 1, :]
                    + transition_scores[pos - 1, current_state, :]