Predictions

In addition to producing descriptive statistics on the observed corpus data, as presented in the previous section, the statistical analysis also involved estimating and applying a model to predict the mean response of each dependent variable to the independent variables, including values not observed in the corpus. If our corpus is considered representative, the predictions produced in this way can be generalized to other similar texts, talks and speeches.

The descriptive analysis presented in the previous section included two structural sub-variables: difference in the branching direction of relative clauses and difference in the branching direction of complement clauses. For the predictive analysis, those two sub-variables were combined to produce a single independent variable. That combined variable, structural difference of the language pair, represents the difference between English and each of the five target languages in the typical branching direction of subordinate clauses in general. The other two independent variables in the predictive analysis were the same as in the descriptive analysis: mode and sentence complexity. The three dependent variables were also the same: reordering, nesting changes and changes in semantic relations. For nesting changes, predictions for changes in double nestings are presented, but not included in the statistical comparisons because of the low values involved. The observed data was insufficient to yield any predictions for changes in triple nestings.

The first step in the predictive analysis was to check for significant interactions between our three independent variables. To do that, the p‑value, or chance of randomness, was computed for the three pairwise interactions between those variables. A p-value below 0.05 is generally regarded as indicating that an interaction is significant. The figures involved in calculating p-values for interactions between our three independent variables – mode, structural difference and sentence complexity – are shown in table 13.

Table 13
p-values for interactions between mode, structural difference and sentence complexity
(R = Reordering, N1 = Changes in single nestings, S = Changes in semantic relations)

Interaction	Statistic			Degrees of freedom			p-value
	R	N1	S	R	N1	S	R	N1	S
mode and structural difference	91.437	60.756	88.210	6	6	6	0	0	0.00
mode and sentence complexity	25.457	45.916	66.663	2	2	2	0	0	0.00
structural difference and sentence complexity	37.047	48.312	8.331	3	3	3	0	0	0.04

Table 13 shows that each of the pairwise interactions between mode, structural difference and sentence complexity yielded p-values of less than 0.05 for all three indicators of difficulty and can therefore be regarded as significant for all three indicators. In other words, each independent variable interacts significantly with the other two, with less than a 5% chance that those interactions are due to chance.

When such interactions are found to exist, statistical tests avoid investigating the effect of any single independent variable acting on its own, to find trends like which one is the strongest predictor. That’s because the results of such a test would be uninformative. To understand why, let’s take the example of the significant interaction found here between structural difference and mode in their effect on reordering. That interaction means that the effect of structural difference on reordering changes depending on mode. It could even be that structural difference has a positive effect on reordering in one mode and a negative effect in another mode. In that case, if we tested the effect of structural difference on reordering without considering its interaction with mode, we might obtain an average value close to 0, since the positive effect in one mode would be offset by the negative effect in another mode. By testing the simple effect of one independent variable without taking into account its interaction with the other one, we would wrongly conclude that structural difference had no impact on reordering.

As we’ve seen, each of our three independent variables – mode, structural difference and sentence complexity – interacts significantly with the other two. So, to produce informative results, our statistical tests investigated the interaction between each pair of those independent variables in turn, and finally all three of them acting together. The results of those four sets of tests are presented below.

Let’s start with the predicted mean response of the three indicators of difficulty to interactions between mode and structural difference of the language pair. The figures are shown in table 14.

Table 14
Predicted mean response of indicators of difficulty
to interactions between mode and structural difference of language pair
(R = Reordering, N1/N2 = Changes in single/double nestings, S = Changes in semantic relations)

Mode	Structural difference of language pair	Response			Degrees of freedom: reordering 5660 / single nestings 5661 / double nestings 5671 / semantic relations 5660
					Standard error			Lower confidence level			Upper confidence level
		R	N1 / N2	S	R	N1 / N2	S	R	N1 / N2	S	R	N1 / N2	S
legal translation	same	0.121	0.122 / 0.006	0.107	0.017	0.015 / 0.004	0.016	0.092	0.096 / 0.002	0.080	0.160	0.154 / 0.020	0.142
	somewhat different	0.969	0.731 / 0.078	0.222	0.060	0.047 / 0.016	0.023	0.092	0.644 / 0.051	0.180	1.094	0.829 / 0.118	0.272
	moderately different	1.131	0.968 / 0.307	0.386	0.066	0.057 / 0.046	0.034	1.008	0.862 / 0.229	0.325	1.269	1.087 / 0.412	0.459
	opposite	|3.095|	|1.395| / 0.270	0.529	0.130	0.069 / 0.036	0.034	2.850	1.267 / 0.207	0.467	3.361	1.537 / 0.352	0.599
subtitle translation	same	0.100	0.098 / 0.001	0.386	0.018	0.015 / 0.001	0.038	0.070	1.267 / 0.000	0.318	0.142	0.131 / 0.004	0.470
	somewhat different	0.167	0.168 / 0.016	0.290	0.024	0.020 / 0.005	0.032	0.127	0.134 / 0.009	0.233	0.220	0.212 / 0.028	0.361
	moderately different	0.344	0.346 / 0.062	0.557	0.036	0.031 / 0.016	0.048	0.281	0.290 / 0.038	0.470	0.422	0.412 / 0.103	0.660
	opposite	1.651	0.661 / 0.055	0.709	0.083	0.042 / 0.014	0.045	1.495	0.584 / 0.034	0.626	1.823	0.748 / 0.089	0.803
simultaneous interpretation	same	0.024	0.105 / 0.001	0.788	0.012	0.022 / 0.001	0.086	0.009	0.069 / 0.000	0.636	0.064	0.158 / 0.004	0.976
	somewhat different	0.235	0.198 / 0.014	1.637	0.041	0.032 / 0.005	0.141	0.167	0.144 / 0.007	1.383	0.330	0.272 / 0.029	1.938
	moderately different	0.292	0.440 / 0.055	1.168	0.047	0.053 / 0.01	0.112	0.214	0.348 / 0.028	0.968	0.400	0.557 / 0.107	1.409
	opposite	1.183	0.618 / 0.048	|1.946|	0.092	0.056 / 0.016	0.135	1.015	0.517 / 0.025	1.698	1.378	0.738 / 0.093	2.231

In table 14, the highest value in the column showing the mean response of each indicator of difficulty (R, N1 and S) to each mode is underlined. A comparison of each underlined value to the other values in the same column corresponding to the same mode shows that, in all three modes, the mean rates of all three indicators of difficulty is predicted to be lowest when the language pair has similar complex sentence structure, and to increase as the structural difference of the language pair increases.

Of the underlined values in the column showing the mean rates for each indicator of difficulty, the highest value is also in a box. Those boxed values show that reordering and nesting changes are predicted to be most frequent in legal translation, where the mean rate for reordering is predicted to reach more than 3 place shifts per sentence when the language pair has opposite structure, and the mean rate for changes in single nestings is predicted to reach nearly 1½ changes per sentence when the language pair has opposite structure. Changes in semantic relations are predicted to be especially characteristic of simultaneous interpretation, where the mean rate is predicted to reach nearly 2 changes per sentence when the language pair has opposite structure.

The predicted mean response of each indicator of difficulty to interactions between mode and structural difference of the language pair are visualized in chart 7.

Chart 7. Predicted mean response of indicators of difficulty
to interactions between mode and structural difference

Let’s look next at the predicted mean response of the three indicators of difficulty to interactions between mode and sentence complexity, as measured by the number of functionally subordinate or reported propositions in an original English sentence. The sentence complexity variable can take any integer value. The predictive analysis considered five representative values, ranging from simple (3 subordinate or reported propositions per sentence) to extremely complex (15 subordinate or reported propositions per sentence). The figures are shown in table 15.

Table 15
Predicted mean response of indicators of difficulty
to interactions between mode and sentence complexity
(R = Reordering, N1/N2 = Changes in single/double nestings, S = Changes in semantic relations)

Mode	Sentence complexity	Response			Degrees of freedom: reordering 5660 / single nestings 5661 / double nestings 5671 / semantic relations 5660
					Standard error			Lower confidence level			Upper confidence level
		R	N1 / N2	S	R	N1 / N2	S	R	N1 / N2	S	R	N1 / N2	S
legal translation	simple	0.739	0.552 / 0.020	0.247	0.041	0.031 / 0.004	0.017	0.663	0.495 / 0.013	0.216	0.824	0.616 / 0.031	0.281
	somewhat complex	1.457	0.948 / 0.039	0.437	0.066	0.045 / 0.008	0.024	1.333	0.863 / 0.027	0.392	1.593	1.041 / 0.059	0.487
	moderately complex	|2.872|	1.626 / 0.079	0.775	0.154	0.098 / 0.016	0.047	2.586	1.446 / 0.054	0.688	3.191	1.829 / 0.117	0.872
	very complex	5.662	2.790 / 0.160	1.373	0.419	0.234 / 0.034	0.108	4.897	2.367 / 0.105	1.177	6.546	3.290 / 0.242	1.603
	extremely complex	11.161	4.788 / 0.322	2.434	1.109	0.538 / 0.079	0.254	9.185	3.841 / 0.198	1.985	13.562	5.969 / 0.522	2.986
subtitle translation	simple	0.280	0.223 / 0.004	0.413	0.020	0.015 / 0.001	0.024	0.243	0.195 / 0.002	0.369	0.323	0.255 / 0.007	0.463
	somewhat complex	0.703	0.541 / 0.008	1.004	0.069	0.057 / 0.002	0.089	0.580	0.440 / 0.005	0.844	0.851	0.666 / 0.014	1.194
	moderately complex	1.763	1.313 / 0.016	2.440	0.267	0.229 / 0.005	0.362	1.310	0.933 / 0.009	1.825	2.371	1.848 / 0.028	3.263
	very complex	4.421	3.186 / 0.033	5.930	0.939	0.796 / 0.010	1.270	2.915	1.953 / 0.018	3.897	6.705	5.199 / 0.059	9.024
simultaneous interpretation	simple	0.184	0.237 / 0.004	1.127	0.027	0.023 / 0.001	0.071	0.138	0.196 / 0.002	0.996	0.246	0.285 / 0.007	1.277
	somewhat complex	0.551	0.817 / 0.007	3.982	0.101	0.129 / 0.003	0.464	0.384	0.600 / 0.004	3.169	0.790	1.114 / 0.014	5.005
	moderately complex	1.646	|2.822| / 0.014	|14.067|	0.427	0.741 / 0.005	2.776	0.990	1.687 / 0.007	9.554	2.738	4.723 / 0.029	20.711

In table 15, we don’t have figures for all combinations of mode and sentence complexity. There were too few non‑zero values in the data to make predictions for extremely complex sentences (15 subordinate or reported propositions per sentence) in subtitle translation, or for very or extremely complex sentences (12 or 15 subordinate or reported propositions per sentence) in simultaneous interpretation. So those rows are missing from the table.

To compare like with like, we can compare the values in the column showing the mean response of each indicator of difficulty (R, N1 and S) to simple, somewhat complex and moderately complex sentences – the three degrees of complexity for which we have figures for all three modes. Of those values, the highest value in the column showing the mean response of each indicator to each mode is underlined. A comparison of each underlined value to the other values in the same column corresponding to the same mode shows that, in all three modes, the mean rates for all three indicators of difficulty are predicted to increase as sentence complexity increases.

Of the underlined values in the column showing the mean rates for each indicator of difficulty, the highest value is also in a box. Those boxed values show that reordering is predicted to be most frequent in legal translation, where the mean rate for reordering is predicted to reach nearly 3 place shifts per sentence in moderately complex sentences (sentences with 9 subordinate or reported propositions) and more than 11 place shifts per sentence in extremely complex sentences (sentences with 15 subordinate or reported propositions). Nesting changes and changes in semantic relations are predicted to be especially characteristic of simultaneous interpretation, where the mean rate for changes in single nestings is predicted to reach nearly 3 changes per sentence in moderately complex sentences, and the mean rate for changes in semantic relations is predicted to reach more than 14 changes per sentence in moderately complex sentences.

The predicted mean response of each indicator of difficulty to interactions between mode and sentence complexity are visualized in chart 8. No bars are shown for complex or extremely complex sentences (12 or 15 subordinate or reported propositions per sentence), since there were too few non-zero values in the data to predict comparable rates for them all.

Chart 8. Predicted mean response of indicators of difficulty
to interactions between mode and sentence complexity

Now let’s see the predicted mean response of the three indicators of difficulty to interactions between structural difference and sentence complexity. The figures are shown in table 16.

Table 16
Predicted mean response of indicators of difficulty
to interactions between structural difference and sentence complexity
(R = Reordering, N1 / N2 = Changes in single/double nestings, S = Changes in semantic relations)

Structural difference of language pair	Sentence complexity	Response			Degrees of freedom: reordering 5660 / single nestings 5661 / double nestings 5671 /semantic relations 5660
					Standard error			Lower confidence level			Upper confidence level
		R	N1 / N2	S	R	N1 / N2	S	R	N1 / N2	S	R	N1 / N2	S
same structure	simple	0.060	0.096 / 0.001	0.287	0.011	0.009 / 0.000	0.021	0.042	0.080 / 0.000	0.248	0.086	0.117 / 0.002	0.331
	somewhat complex	0.137	0.244 / 0.001	0.717	0.026	0.026 / 0.001	0.054	0.095	0.199 / 0.000	0.619	0.199	0.300 / 0.003	0.830
	moderately complex	0.315	0.617 / 0.002	1.792	0.067	0.089 / 0.001	0.195	0.207	0.465 / 0.001	1.448	0.479	0.820 / 0.007	2.219
	very complex	0.722	1.563 / 0.004	4.483	0.186	0.310 / 0.003	0.697	0.435	1.060 / 0.001	3.305	1.197	2.306 / 0.014	6.081
	extremely complex	1.654	3.957 / 0.008	11.212	0.516	1.018 / 0.005	2.315	0.897	2.390 / 0.002	7.480	3.047	6.553 / 0.028	16.805
somewhat different structure	simple	0.303	0.265 / 0.006	0.422	0.024	0.019 / 0.002	0.026	0.260	0.230 / 0.004	0.373	0.354	0.304 / 0.010	0.476
	somewhat complex	0.726	0.575 / 0.013	1.101	0.064	0.050 / 0.003	0.074	0.611	0.486 / 0.008	0.965	0.862	0.681 / 0.020	1.255
	moderately complex	1.736	1.250 / 0.026	2.872	0.201	0.157 / 0.006	0.281	1.384	0.977 / 0.016	2.370	2.179	1.599 / 0.041	3.480
	very complex	4.154	2.717 / 0.052	7.495	0.636	0.471 / 0.014	1.037	3.076	1.934 / 0.031	5.715	5.608	3.817 / 0.087	9.829
	extremely complex	9.936	5.906 / 0.104	19.559	1.931	1.326 / 0.031	3.559	6.788	3.803 / 0.058	13.691	14.545	9.171 / 0.188	27.942
moderately different structure	simple	0.435	0.475 / 0.025	0.571	0.030	0.026 / 0.005	0.031	0.381	0.427 / 0.017	0.513	0.497	0.530 / 0.037	0.636
	somewhat complex	1.086	1.162 / 0.051	1.336	0.084	0.084 / 0.009	0.085	0.933	1.009 / 0.035	1.180	1.265	1.339 / 0.072	1.514
	moderately complex	2.713	2.841 / 0.102	3.129	0.292	0.320 / 0.019	0.303	2.197	2.277 / 0.070	2.587	3.351	3.543 / 0.147	3.783
	very complex	6.777	6.944 / 0.205	7.325	0.988	1.109 / 0.044	1.011	5.092	5.077 / 0.134	5.588	9.019	9.497 / 0.313	9.601
	extremely complex	16.926	16.973 / 0.412	17.149	3.167	3.546 / 0.107	3.118	11.729	11.270 / 0.248	12.007	24.426	25.563 / 0.685	24.492
opposite structure	simple	1.624	0.741 / 0.022	0.809	0.057	0.031 / 0.004	0.033	1.515	0.682 / 0.015	0.747	1.740	0.804 / 0.032	0.877
	somewhat complex	4.301	1.925 / 0.044	1.997	0.229	0.121 / 0.007	0.106	3.874	1.702 / 0.032	1.800	4.774	2.177 / 0.062	2.216
	moderately complex	11.390	5.003 / 0.089	4.928	1.021	0.526 / 0.016	0.424	9.554	4.072 / 0.063	4.163	13.577	6.148 / 0.127	5.834
	very complex	30.164	13.005 / 0.180	12.160	3.912	1.975 / 0.037	1.510	23.392	9.657 / 0.120	9.532	38.895	17.514 / 0.271	15.511
	extremely complex	|79.884|	|33.803| / 0.363	|30.003|	13.651	6.760 / 0.092	4.916	57.144	22.840 / 0.221	21.760	111.673	50.027 / 0.595	41.370

In table 16, the highest value in the column showing the mean response of each indicator of difficulty (R, N1 and S) to each degree of structural difference is underlined. A comparison of each underlined value to the other values in the same column corresponding to the same degree of structural difference shows that, for all degrees of structural difference, the mean rates for all three indicators of difficulty are predicted to be lowest in simple sentences, and to increase as sentence complexity increases.

Of the underlined values in the column showing the mean rates for each indicator of difficulty, the highest value is also in a box. Those boxed values show that the rates for all three indicators are predicted to increase greatly as structural difference increases. When the language pair has opposite structure, the mean rate for reordering is predicted to reach nearly 80 place shifts per sentence in extremely complex sentences, the mean rate for changes in single nestings is predicted to reach more than 33 changes per sentence in extremely complex sentences, and the mean rate for changes in semantic relations is predicted to reach 30 changes per sentence in extremely complex sentences (sentences with 15 subordinate or reported propositions). All three indicators are predicted to respond especially strongly to the combined effect of high sentence complexity and great structural difference of the language pair.

The predicted mean response of each indicator of difficulty to interactions between structural difference and sentence complexity are visualized in chart 9.

Chart 9. Predicted mean response of indicators of difficulty
to interactions between structural difference and sentence complexity

Putting all the above data together, we can see the combined effect of all three independent variables. Table 17 shows the predicted mean response of the three indicators of difficulty to interactions between structural difference of the language pair, sentence complexity and mode. There were too few non-zero values in the data to make predictions for extremely complex sentences in subtitle translation, or for very or extremely complex sentences in simultaneous interpretation.

Table 17
Predicted mean response of indicators of difficulty
to interactions between structural difference of language pair, sentence complexity and mode
(R = Reordering, N1/N2 = Changes in single/double nestings, S = Changes in semantic relations)

Structural difference of language pair	Sentence complexity	Mode	Response			Degrees of freedom: reordering 5660 / single nestings 5661 / double nestings 5671 / semantic relations 5660
						Standard error			Lower confidence level			Upper confidence level
			R	N1 / N2	S	R	N1 / N2	S	R	N1 / N2	S	R	N1 / N2	S
same structure	simple	legal tr.	0.113	0.114 / 0.002	0.100	0.017	0.014 / 0.001	0.015	0.085	0.089 / 0.000	0.074	0.150	0.145 / 0.005	0.134
		subtitle tr.	0.091	0.087 / 0.000	0.348	0.016	0.013 / 0.000	0.035	0.064	0.065 / 0.000	0.286	0.129	0.117 / 0.001	0.423
		interpretation	0.021	0.090 / 0.000	0.678	0.011	0.019 / 0.000	0.073	0.008	0.059 / 0.000	0.548	0.057	0.136 / 0.001	0.838
	somewhat complex	legal tr.	0.208	0.203 / 0.003	0.179	0.024	0.020 / 0.002	0.023	0.165	0.167 / 0.001	0.139	0.261	0.247 / 0.010	0.229
		subtitle tr.	0.212	0.221 / 0.001	0.852	0.041	0.038 / 0.000	0.104	0.144	0.158 / 0.000	0.671	0.311	0.309 / 0.002	1.083
		interpretation	0.059	0.323 / 0.001	2.417	0.031	0.080 / 0.000	0.360	0.021	0.199 / 0.000	1.805	0.163	0.525 / 0.002	3.238
	moderately complex	legal tr.	0.382	0.363 / 0.006	0.320	0.045	0.038 / 0.004	0.039	0.304	0.296 / 0.002	0.251	0.481	0.444 / 0.020	0.407
		subtitle tr.	0.497	0.558 / 0.001	2.091	0.041	0.128 / 0.001	0.374	0.308	0.356 / 0.000	1.472	0.800	0.876 / 0.004	2.970
		interpretation	0.165	1.162 / 0.001	8.617	0.092	0.385 / 0.001	1.951	0.055	0.608 / 0.000	5.528	0.490	2.224 / 0.004	13.432
	very complex	legal tr.	0.703	0.648 / 0.012	0.572	0.104	0.085 / 0.007	0.079	0.527	0.501 / 0.004	0.436	0.939	0.836 / 0.040	0.750
		subtitle tr.	1.163	1.411 / 0.003	5.129	0.358	0.428 / 0.002	1.271	0.635	0.778 / 0.001	3.155	2.127	2.558 / 0.009	8.337
	extremely complex	legal tr.	1.295	1.157 / 0.025	1.023	0.251	0.197 / 0.015	0.172	0.885	0.829 / 0.008	0.736	1.894	1.615 / 0.083	1.421
somewhat different structure	simple	legal tr.	0.897	0.695 / 0.019	0.206	0.057	0.046 / 0.005	0.022	0.792	0.609 / 0.012	0.166	1.017	0.792 / 0.031	0.254
		subtitle tr.	0.150	0.154 / 0.004	0.260	0.021	0.018 / 0.001	0.029	0.114	0.122 / 0.002	0.209	0.198	0.194 / 0.007	0.324
		interpretation	0.207	0.173 / 0.003	1.402	0.036	0.028 / 0.001	0.118	0.147	0.127 / 0.002	1.189	0.290	0.237 / 0.007	1.654
	somewhat complex	legal tr.	1.724	1.065 / 0.039	0.384	0.090	0.060 / 0.008	0.035	1.557	0.955 / 0.025	0.322	1.910	1.189 / 0.059	0.459
		subtitle tr.	0.368	0.334 / 0.008	0.665	0.057	0.048 / 0.002	0.087	0.271	0.251 / 0.004	0.515	0.499	0.443 / 0.014	0.860
		interpretation	0.602	0.535 / 0.007	5.216	0.124	0.110 / 0.003	0.677	0.403	0.357 / 0.003	4.044	0.900	0.801 / 0.014	6.727
	moderately complex	legal tr.	3.314	1.634 / 0.078	0.717	0.193	0.114 / 0.016	0.063	2.956	1.426 / 0.051	0.603	3.715	1.873 / 0.118	0.852
		subtitle tr.	0.900	0.723 / 0.016	1.702	0.176	0.148 / 0.005	0.309	0.614	0.485 / 0.009	1.193	1.319	1.079 / 0.028	2.429
		interpretation	1.755	1.651 / 0.014	19.401	0.484	0.490 / 0.005	4.039	1.022	0.923 / 0.007	12.900	3.013	2.955 / 0.029	29.178
	very complex	legal tr.	6.368	2.507 / 0.156	1.339	0.499	0.245 / 0.036	0.137	5.462	2.071 / 0.100	1.096	7.426	3.035 / 0.244	1.636
		subtitle tr.	2.200	1.569 / 0.032	4.357	0.544	0.433 / 0.010	1.064	1.355	0.914 / 0.017	2.699	3.573	2.695 / 0.059	7.032
	extremely complex	legal tr.	12.238	3.846 / 0.315	2.499	1.279	0.504 / 0.082	0.319	9.971	2.974 / 0.189	1.946	15.021	4.973 / 0.524	3.210
moderately different structure	simple	legal tr.	1.042	0.908 / 0.076	0.364	0.063	0.055 / 0.014	0.033	0.925	0.805 / 0.052	0.305	1.174	1.023 / 0.110	0.434
		subtitle tr.	0.308	0.311 / 0.015	0.505	0.032	0.028 / 0.004	0.044	0.252	0.261 / 0.009	0.426	0.378	0.371 / 0.026	0.598
		interpretation	0.256	0.380 / 0.014	1.013	0.041	0.045 / 0.005	0.095	0.534	0.302 / 0.007	0.842	1.137	0.479 / 0.026	1.218
	somewhat complex	legal tr.	2.090	1.566 / 0.152	0.609	0.104	0.081 / 0.024	0.046	1.896	1.415 / 0.112	0.525	2.304	1.732 / 0.207	0.706
		subtitle tr.	0.787	0.759 / 0.031	1.159	0.097	0.091 / 0.008	0.128	0.619	0.600 / 0.019	0.933	1.001	0.961 / 0.050	1.440
		interpretation	0.779	1.320 / 0.027	3.381	0.150	0.228 / 0.009	0.466	0.188	0.941 / 0.014	2.581	0.350	1.852 / 0.052	4.429
	moderately complex	legal tr.	4.194	2.701 / 0.307	1.020	0.235	0.170 / 0.046	0.080	3.758	2.387 / 0.229	0.875	4.681	3.056 / 0.412	1.189
		subtitle tr.	2.009	1.852 / 0.062	2.661	0.339	0.341 / 0.016	0.444	1.443	1.290 / 0.038	1.919	2.797	2.659 / 0.103	3.690
		interpretation	2.371	4.582 / 0.055	11.282	0.630	1.248 / 0.019	2.418	1.408	2.686 / 0.028	7.411	3.992	7.816 / 0.107	17.174
	very complex	legal tr.	8.415	4.659 / 0.618	1.708	0.635	0.407 / 0.106	0.167	7.258	3.925 / 0.442	1.411	9.756	5.529 / 0.864	2.068
		subtitle tr.	5.128	4.518 / 0.126	6.111	1.159	1.166 / 0.035	1.421	3.293	2.725 / 0.073	3.873	7.986	7.492 / 0.218	9.641
	extremely complex	legal tr.	16.883	8.036 / 1.243	2.860	1.700	0.940 / 0.264	0.360	13.859	6.389 / 0.820	2.235	20.568	10.106 / 1.884	3.662
opposite structure	simple	legal tr.	2.831	1.299 / 0.067	0.495	0.124	0.066 / 0.012	0.033	2.599	1.175 / 0.047	0.435	3.085	1.436 / 0.094	0.563
		subtitle tr.	1.468	0.591 / 0.014	0.639	0.073	0.037 / 0.003	0.040	1.331	0.523 / 0.008	0.564	1.619	0.667 / 0.022	0.723
		interpretation	1.030	0.529 / 0.012	1.678	0.078	0.047 / 0.004	0.113	0.887	0.445 / 0.006	1.471	1.196	0.630 / 0.023	1.915
	somewhat complex	legal tr.	6.024	2.383 / 0.134	0.873	0.220	0.105 / 0.019	0.047	5.608	2.186 / 0.102	0.785	6.471	2.598 / 0.177	0.970
		subtitle tr.	3.974	1.532 / 0.027	1.546	0.325	0.154 / 0.006	0.143	3.385	1.258 / 0.017	1.290	4.665	1.866 / 0.044	1.853
		interpretation	3.323	1.954 / 0.024	5.902	0.441	0.299 / 0.008	0.701	2.562	1.448 / 0.013	4.676	4.309	2.637 / 0.045	7.450
	moderately complex	legal tr.	|12.815|	4.370 / 0.270	1.541	0.595	0.248 / 0.036	0.091	11.701	3.910 / 0.207	1.372	14.035	4.885 / 0.352	1.731
		subtitle tr.	10.756	3.973 / 0.055	3.741	1.500	0.676 / 0.014	0.567	8.183	2.846 / 0.034	2.779	14.139	5.546 / 0.089	5.036
		interpretation	10.718	|7.214| / 0.048	|20.758|	2.407	1.867 / 0.016	4.129	6.902	4.343 / 0.025	14.055	16.646	11.983 / 0.093	30.659
	very complex	legal tr.	27.264	8.016 / 0.543	2.720	1.803	0.645 / 0.087	0.213	23.949	6.846 / 0.397	2.333	31.037	9.386 / 0.743	3.171
		subtitle tr.	29.116	10.305 / 0.111	9.054	5.881	2.533 / 0.030	1.971	19.596	6.365 / 0.065	5.909	43.261	16.683 / 0.189	13.872
	extremely complex	legal tr.	58.002	14.702 / 1.093	4.801	5.186	1.590 / 0.223	0.498	48.677	11.894 / 0.733	3.917	69.114	18.173 / 1.631	5.884

In table 17, we don’t have figures for all combinations of sentence complexity and mode in the second and third columns. For very complex sentences, we only have figures for legal translation and subtitle translation; for extremely complex sentences, we only have figures for legal translation. To compare like with like, we can compare the values in the column showing the mean rates for each indicator of difficulty (R, N1 and S) in simple, somewhat complex and moderately complex sentences – the three degrees of complexity for which we have figures for all three modes. Of those values, the highest value in the column showing the mean response of each indicator to each combination of structural difference and sentence complexity is underlined.

A comparison of the underlined values in each column corresponding to each degree of structural difference shows that, for each degree of structural difference, the mean rates for all three indicators of difficulty are predicted to increase as sentence complexity increases. This trend is confirmed by the still higher mean rates for indicators in very complex and extremely complex sentences, for those modes for which we have figures. Moreover, a comparison of the underlined values in each column corresponding to the same degree of sentence complexity shows that, for each degree of sentence complexity, the mean rates for all three indicators of difficulty are predicted to increase as structural difference increases.

Of the underlined values in the column showing the mean rates for each indicator of difficulty, the highest value is also in a box. Those boxed values show that the mean rates for all three indicators of difficulty are predicted to increase greatly as structural difference and sentence complexity increase, especially when the two aggravating factors are combined.

The underlined values in the column for reordering (R) are all for legal translation. This shows that, for each combination of structural difference and sentence complexity, reordering is predicted to be most frequent in legal translation, where the mean rate for reordering is predicted to reach nearly 13 place shifts per sentence in moderately complex sentences and 58 place shifts per sentence in extremely complex sentences when the language pair has opposite structure.

Some of the underlined values in the column for changes in single nestings (N1) are for legal translation and some – including the highest one – are for simultaneous interpretation. This shows that nesting changes are predicted to be more frequent in those two modes – especially in simultaneous interpretation, where the mean rate for changes in single nestings is predicted to reach more than 7 changes per sentence in moderately complex sentences when the language pair has opposite structure.

The underlined values in the column for changes in semantic relations (S) are all for simultaneous interpretation. This shows that, for each combination of structural difference and sentence complexity, changes in semantic relations are predicted to be especially characteristic of simultaneous interpretation, where the mean rate is predicted to reach nearly 21 changes per sentence in moderately complex sentences when the language pair has opposite structure.

The predicted mean response of each indicator of difficulty to interactions between structural difference, sentence complexity and mode are visualized in chart 10. No bars are shown for complex or extremely complex sentences, since there were too few non-zero values in the data to predict comparable rates for them all.

Chart 10. Predicted mean response of indicators of difficulty
to interactions between structural difference, sentence complexity and mode

Reordering

Reordering: predicted to increase greatly with structural difference and sentence complexity –
a bit more in legal translation than in the other modes

Changes in single nestings

Changes in single nestings: predicted to increase greatly with structural difference and sentence complexity
(no clear pattern by mode)

Changes in double nestings

Changes in double nestings: predicted to increase greatly with structural difference and sentence complexity,
especially in legal translation

Changes in semantic relations

Changes in semantic relations: predicted to increase with structural difference and sentence complexity,
especially in simultaneous interpretation

The last two sections have presented the main findings from our statistical analysis. In addition, a reliability check was carried out on 10% of the sentences in the corpus, to confirm the validity of the data recorded. The results of that check are presented in the next section.