Sociodemographic and linguistic variables.
There were no significant differences in sociodemographic measures or premorbid linguistic factors (age of acquisition, proficiency, or language usage) between the two groups. For (post-morbid) language switching behavior, BWAs reported higher degrees of switching into their dominant (BWAs: M
= 9.86, SD
= 1.86; controls: M
= 7.80, SD
= 1.14; p
= 0.012) and non-dominant languages (BWAs: M
= 9.57, SD
= 0.79; controls: M
= 6.30, SD
= 1.57; p
< 0.001) compared to controls (see Table 2
). For each participant, we compared the scores of the BAT-C of the two languages using a Chi-squared test with Yates’ correction; five out of seven patients showed parallel language deficits (Pt 2 showed a significantly more impaired score in their non-dominant compared to their dominant language, and Pt 3 the opposite).
Switching frequencies. Both experimental groups switched between their languages in approximately half the trials (BWA: M = 49.65%, SD = 13.15; Controls: M = 47.97%, SD = 7.32), with no significant differences between groups, t(15) = 0.34, p = 0.739; U = 40.00, p = 0.669. Additionally, participants did not show any differences in switching associated with cognate status [cognates: M = 50.06%, SD = 3.43; non-cognates: M = 49.94%, SD = 3.43; t(16) = 0.07, p = 0.943] or language dominance [DL: M = 50.01%, SD = 1.07; NDL: M = 49.99%, SD = 1.07; t(16) = 0.01, p = 0.99].
Naming latencies. Analyses revealed a main effect of Trial Type, F(2, 30) = 3.26, p = 0.05, ηp2 = 0.18. Post hoc analyses showed that participants demonstrated significant switch costs, as they were slower in naming switch trials (M = 1176 ms, SD = 55) than naming repeat trials (M = 1072 ms, SD = 35, p < 0.05). However, they did not show significant mixing costs, as naming latencies between repeat trials (M = 1113 ms, SD = 38) and single trials (M = 1072 ms, SD = 35) were not significantly different (p = 0.31).
The main effect of Cognate Status was also significant, F
(1,15) = 30.09, p
< 0.001, ηp2
= 0.67, where cognate trials (M
= 1089 ms, SD
= 34) were named faster than non-cognate trials (M
= 1151 ms, SD
= 40). Conversely, the main effect of Language was not significant, F
(1,15) = 3.01, p
= 0.10. Finally, the main effect of Group was significant, F
(1,15) = 44.97, p
< 0.001, ηp2
= 0.75, with greater naming latencies for BWA (M
= 1363 ms, SD
= 55) than for controls (M
= 877 ms, SD
= 46). See Figure 2
for comparison between groups.
The interaction between Cognate Status and Group was also significant, F(1,15) = 15.21, p = 0.001, ηp2 = 0.50. An independent samples t-test was performed on the difference in magnitude of the cognate effects between the two groups and we found that this effect was larger in patients (M = 107 ms, SD = 69) than in controls (M = 18 ms, SD = 18; t(16) = 3.01, p = 0.001).
Furthermore, both the Language × Cognate Status × Trial Type interaction, F(2,30) = 6.66, p < 0.01, ηp2 = 0.31, as well as the Language × Cognate Status × Trial Type × Group interaction, F(2,30) = 3.54, p < 0.05, ηp2 = 0.19, were significant. To address these complex interactions, follow-up analyses were conducted by performing repeated-measures ANOVAs for cognates and non-cognates separately, both including Language and Trial Type as within subject factors. Said analyses were also separated between control and patient groups.
In controls, the main effect of Trial Type was significant for both cognates, F(2,18) = 3.46, p = 0.05, ηp2 = 0.44, and non-cognates, F(2,18) = 5.52, p = 0.01, ηp2 = 0.38, in the dominant language; and for cognates, F(2,18) = 7.89, p = 0.003, ηp2 = 0.47, but not for non-cognates, F(2,18) = 3.25, p = 0.06, in the non-dominant language. Significant switch costs were found for cognates (repeat trials: M = 850 ms, SD = 72; switch trials: M = 900 ms, SD = 91; p = 0.03) and non-cognates (repeat trials: M = 870 ms, SD = 54; switch trials: M = 917ms, SD = 82; p = 0.02) in the dominant language and only for cognates in the non-dominant language (repeat trials: M = 877 ms, SD = 56; switch trials: M = 917 ms, SD = 96; p = 0.04). Mixing costs were only significant for cognates in the non-dominant language (single trials: M = 811 ms, SD = 75; repeat trials: M = 877 ms, SD = 56; p = 0.03).
In patients, the main effect of Trial Type was neither significant in the dominant [cognates words: F(2,12) = 0.11, p = 0.89; non-cognates words: F(2,12) = 1.11, p = 0.36] nor in the non-dominant language [cognates words: F(2,12) = 3.41, p = 0.07; non-cognates words: F(2,12) = 1.17, p = 0.34].
Individual level analyses for naming latencies.
As we observed a great variability in the patient group for switching and mixing effects, we ran individual level analyses for BWAs. We first calculated individual proportional switching and mixing effects and we used a modified t-test described by Crawford and Howell (1998) for independent samples to compare each individual’s performance to the controls’ mean (switch cost: M
= 3.2, SD
= 3.4; mixing cost: M
= 2.2, SD
= 3.3). The t values were calculated as follows:
is the individual’s performance, X2
is the mean of the control sample, s2
is the standard deviation of the control group, and N
is the sample size.
The results of the analyses showed that 2 patients (Pt 2: 24.5%, p
< 0.001; Pt 5: 13.3%, p
< 0.01) had larger switch costs as compared to controls. Moreover, 3 patients (Pt 2: 25.6%; Pt 6: 15.2%, Pt 7: 15.7%; all p
-values < 0.001) had larger mixing costs than controls and one (Pt 4: −31.2%, p
< 0.001) had a significant mixing benefit as compared to controls (see Figure 3
Accuracy. The main effect of Trial Type was significant, F(2,30) = 19.01, p < 0.001, ηp2 = 0.56, and post hoc analyses revealed a significant decrease in accuracy for single trials (M = 86.65%, SD = 10.67) compared to both repeat (M = 94.27%, SD = 4.68; p < 0.001) and switch trials (M = 94.38%, SD = 6.37; p < 0.001) trials. The main effects of Language, F (1,15) = 0.37, p = 0.56, and Cognate Status, F(1,15) = 2.27, p = 0.18, were not significant. Finally, the main effect for Group was significant, F(1,15) = 25.91, p < 0.001, ηp2 = 0.62, where control responses (M = 97.21%, SD = 2.58) were more accurate than BWA responses (M = 86.47%, SD = 6.42; p < 0.001).
Furthermore, the Cognate Status × Group interaction was significant, F(1,15) = 5.86, p = 0.03, ηp2 = 0.28. Post hoc analyses showed that controls performed with the same accuracy for cognates (96.20%) and non-cognates (97.40%, p = 0.37), whereas patients were more accurate in naming cognates (88.25%) than non-cognates (84.70%, p = 0.04).
The Trial Type × Group interaction was also significant, F(2,30) = 3.74, p = 0.04, ηp2 = 0.20. Post hoc analyses showed that in the BWA group, single trial accuracy (M = 78.93%, SD = 11.22) was significantly lower than both repeat (M = 90.58%, SD = 3.50; p < 0.05) and switch (M = 90.14%, SD = 7.52, p < 0.05) trials. Similarly, single trial accuracy in controls (M = 94.04%, SD = 4.67) was significantly lower than both repeat (M = 98.42%, SD = 2.50; p < 0.01) and switch (M = 98.43%, SD = 2.55, p < 0.01) trials. However, the difference in accuracy between single and repeat trials was significantly larger in patients (11.51%) than in controls (4.38%, p = 0.05).
To parse the distribution of error types for experimental groups within single-language blocks, the number of incorrect trials for each error type was expressed as a percentage of total single-language trials for each participant and then averaged within groups (Table 4
). Although there is a clear increase in errors on single-language trials overall, BWAs show a specific rise in the proportion of cross-language intrusions, accounting for an average of ~5% of all trials compared to 0.67% for controls. Of note, 11 out of the 12 auto-correction errors committed by BWAs (accounting for an average of 1.43% of single trials) were errors where they first produced the picture name in the incorrect language and then corrected with the target language word, or a “cross-language auto-correction” hybrid. Consequently, cross-language intrusions and these auto-corrections were grouped together as cross-language errors for BWAs, totaling to an average of 6.79% of single-language trials. Of note, these cross-language errors were distributed equally across languages: 52% were NDL intrusions in DL trials and 48% were DL intrusions in NDL trials.
Broken down to individual patient performance on single-language trials (Table 5
), cross-language intrusions and auto-corrections combined were the most common errors for 3 BWAs (Pt 1, Pt 5, Pt 6), the second most common errors behind omissions for 2 BWAs (Pt 3 and Pt 7), and were not as common among the remaining 2 BWAs (Pt 2 and Pt 4).
Word duration. The main effect of Group was found to be significant, F(1,15) = 15.44, p = 0.001, ηp2 = 0.51, where the BWA group produced larger word durations (M = 730 ms, SD = 174) than controls (M = 479 ms, SD = 86). No other main effects were significant. The 3-way interaction between Language × Cognate Status × Trial Type, F(2,30) = 4.18, p = 0.03, ηp2 = 0.22, and the 4-way interaction of Language × Cognate Status × Trial Type × Group, F(2,30) = 3.96, p = 0.03, ηp2 = 0.21, were both significant, but no comparisons remained significant in post hoc analyses.