
==== Front
0020713
6083
Neuropsychologia
Neuropsychologia
Neuropsychologia
0028-3932
1873-3514

38971370
10.1016/j.neuropsychologia.2024.108948
nihpa2014020
Article
Revisiting which language declines more in Spanish-English bilinguals with Alzheimer’s disease: Longitudinal decline patterns on the multilingual naming test
Neveu Anne a*
Goldrick Matthew b
Kleinman Daniel c
Salmon David P. d
Gollan Tamar H. a
a Department of Psychiatry, University of California, San Diego, La Jolla, CA, 92093, USA
b Department of Linguistics, Northwestern University, Evanston, IL, 60208, USA
c Child Study Center, Yale University, New Haven, CT, 06519, USA
d Department of Neurosciences, University of California, San Diego, La Jolla, CA, 92093, USA
* Corresponding author. aneveu@health.ucsd.edu (A. Neveu).
11 8 2024
09 9 2024
04 7 2024
09 9 2024
202 108948108948
https://creativecommons.org/licenses/by-nc/4.0/ This is an open access article under the CC BY-NC license (https://creativecommons.org/licenses/by-nc/4.0/).
Theories of bilingual language production predict that bilinguals with Alzheimer’s disease (AD) should exhibit one of two decline patterns. Either parallel decline of both languages (if decline reflects damage to semantic representations that are accessed by both languages), or asymmetrical decline, with greater decline of the nondominant language (if decline reflects reduced ability to resolve competition from the dominant language with disease progression). Only two previous studies examined decline longitudinally with one showing parallel, and the other asymmetrical, decline. We examined decline over 2–7 years (3.9 on average) in Spanish-English bilinguals (N = 23). Logistic regression revealed a parallel decline pattern at one year from baseline, but an asymmetrical decline pattern over the longer decline period, with greater decline of the nondominant language (when calculating predicted probabilities of a correct response). The asymmetrical decline pattern was significantly greater for the nondominant language only when including item-difficulty in the model. Exploratory analyses across dominance groups looking at proportional decline relative to initial naming accuracy further suggested that decline of the nondominant language may be more precipitous if that language was acquired later in life, but the critical interaction needed to support this possibility was not statistically significant in a logistic regression analysis. These results suggest that accessibility of the nondominant language may initially be more resilient in early versus more advanced AD, and that AD affects shared semantic representations before executive control declines to a point where the ability to name pictures in single-language testing block is disrupted. Additional work is needed to determine if asymmetrical decline patterns are magnified by late age of acquisition of the nondominant language, and if more subtle impairments to executive control underlie impairments to language switching that occur in the earliest stages of AD (even preclinically).

Bilingualism
Alzheimer’s disease
Language decline
Language dominance
Picture naming
==== Body
pmc1. Introduction

The population of bilinguals with Alzheimer’s Disease (AD) is growing, yet much remains to be understood about how languages decline in older bilinguals with AD. Existing models of bilingual language production (e.g., Abutalebi and Green, 2007; Kroll et al., 2014; Pickering and Garrod, 2013) predict either parallel decline of the two languages or that the nondominant language should decline faster than the dominant language. Parallel decline is predicted because of damage to semantic representations, which are shared across languages (e.g., Francis, 2005; Kroll and Stewart, 1994), and support production in both languages. Greater decline of the nondominant language is predicted because more executive control may be needed to suppress competition from the dominant language, allowing the nondominant language to be produced (Calabria et al., 2018; Green, 1998; Green and Abutalebi, 2013). This may be especially true in bilinguals with one clearly dominant language, henceforth, unbalanced bilinguals.

Cross-sectional comparisons of bilinguals with AD relative to matched controls have shown mixed results including parallel effects (Costa et al., 2012; Gómez-Ruiz et al., 2012; Manchon et al., 2015; Salvatierra et al., 2007), and the expected asymmetrical pattern with AD affecting the nondominant more than the dominant language (Friedland and Miller, 1999; Meguro et al., 2003; de Picciotto and Friedland, 2001). A different asymmetrical pattern that no model predicts has also been observed, in which the dominant language was more affected by AD than the nondominant language (Gollan et al., 2010; Ivanova et al., 2013). One of these studies (Gollan et al., 2010), compared cognitively healthy Spanish-English bilinguals to bilinguals with AD on ability to name pictures in the Boston Naming Test (BNT: Kaplan et al., 1983). Language dominance was determined by asking participants which language they preferred for neuropsychological testing. Bilinguals with AD who preferred English had lower naming scores than controls, especially in the dominant language (English). However, in the same study, bilinguals with AD who preferred Spanish exhibited parallel decline of both languages. This finding should be interpreted with caution though, as the BNT was designed for use with English speakers, and use of this test to assess naming ability in Spanish (whether as the dominant or as the nondominant language) may have affected the results in systematic ways that obscured true decline patterns.

Two other studies found parallel decline patterns. One examined decline in Spanish versus in English on category and phonemic fluency in unbalanced Spanish-dominant bilinguals, including 11 cognitively healthy bilinguals and 11 bilinguals with AD, and found parallel decline of the two languages relative to controls (Salvatierra et al., 2007). Another compared 12 cognitively healthy bilinguals to 13 bilinguals with AD, with Italian, Spanish or German as the native language and French as the second language (Manchon et al., 2015). Multiple different language tests showed equally sized deficits for bilinguals with AD in the two languages (including picture naming, verbal fluency, and semantic and syntactic comprehension; Manchon et al., 2015). However, participants in this study had about equal and high proficiency level in both languages (i.e., they were balanced bilinguals even though they had learned French on average relatively late in life as adolescents).

A more recent study identified a more complex pattern of results. Gollan et al. (2023) compared 26 bilinguals with AD to 39 cognitively healthy controls on the Multilingual Naming Test (MINT; Gollan et al., 2012), a picture naming test that was designed specifically for use with Spanish-English bilinguals. In this study, the pattern of decline depended on item difficulty. In the dominant language, AD affected naming on difficult items while easier items were at ceiling levels in both patients and controls. In the nondominant language however, bilinguals with AD exhibited lower naming scores than controls at all difficulty levels. Therefore, while AD affected naming in both languages, the pattern of decline could not be simply labeled as either parallel or asymmetric as it followed more complex patterns depending on item difficulty.

Thus, cross-sectional studies to date identified multiple decline patterns including parallel decline, asymmetrical decline with greater decline of the dominant language, asymmetrical decline with greater decline of the nondominant language, and a mixed pattern of decline that was modulated by item difficulty. Differences in findings between different cross-sectional studies could reflect factors related to stage of disease progression, variability in age of acquisition of the nondominant language, proficiency level in the nondominant language, use of tests developed for one but not both of the bilinguals’ languages, and differences in tasks used to examine the effects of AD. Another important consideration is that between-subjects comparisons will always be less powerful than within-subject comparisons, perhaps particularly when measuring abilities in nondominant languages, which vary widely between individuals (Garcia and Gollan, 2022; Gollan et al., 2017). Longitudinal studies should provide a more definitive answer on the question of dual-language decline in AD as multiple data points are collected within-individuals, thereby controlling for variation in ability at the first testing point. Longitudinal studies also remove the need to match patients and controls on all factors which might have affected proficiency level in the nondominant language before disease onset which is nearly impossible given the many factors that affect proficiency in a nondominant language.

Only two longitudinal studies have been reported to date. One tested 50 relatively balanced bilinguals, who had learned both Catalan and Spanish on average before the age of five. A small number had Mild Cognitive Impairment (MCI; n = 15) and the majority were diagnosed with AD (n = 35; Calabria et al., 2017). Participants were tested at three time points separated by only six months between testing sessions on picture naming, word-picture matching, and word translation tests. Bilinguals with MCI exhibited no decline on any test in either language and bilinguals with AD exhibited parallel decline of the two languages on all tests. The second longitudinal study tested unbalanced Spanish-English bilinguals including 14 cognitively healthy and 12 with AD on the BNT with three years between the first and last testing points (Ivanova et al., 2014). At baseline, cross-sectional comparisons of patients to controls showed a greater difference between patients and controls in the dominant than in the nondominant language (replicating Gollan et al., 2010), whereas longitudinally, the nondominant language declined more than the dominant language. However, there were only 12 bilinguals with AD in Ivanova et al. (2014) and the longitudinal pattern seemed to be driven by just four Spanish-dominant bilinguals with AD, who had learned English relatively late in life (but there were insufficient subjects to run separate analyses by language dominance group). Moreover, testing was done with the BNT, which as noted above, was not developed for Spanish speakers or bilinguals. Finally, due to time constraints, only half of the BNT items (about 30 items) were administered, and slightly different versions of the BNT (with nonoverlapping items) were used across different subjects – introducing yet another possible source of uncontrolled variance.

One additional anecdotal study speaks to the possibility of longitudinal decline. This study interviewed caregivers of Spanish-dominant bilinguals, and another group of bilinguals who were dominant in a variety of other languages (Mendez et al., 1999). All participants in this study had acquired English after age 13, and all were living in the USA. Caregivers reported that over time, since onset of symptoms, bilinguals with AD became more reluctant to speak in their nondominant language (English) while the dominant language remained relatively more intact. Thus, anecdotally, decline appeared to be asymmetrical for these participants with the nondominant language more affected than the dominant language, but objective measures of speaking ability were not obtained in this study.

The present study examined longitudinal decline in Spanish-English bilinguals with AD using the MINT, which as noted above was designed to assess naming abilities in both dominant and nondominant languages (Garcia and Gollan, 2022; Gollan et al., 2012). Additionally, most of the bilinguals were unbalanced in their relative proficiency in the two languages, providing a strong manipulation of language dominance. Finally, time between the first and last testing points was also longer on average relative to previous studies, which may match anecdotal reports more closely given that cognitive changes over short periods should be more difficult to detect (Weintraub et al., 2012). In exploratory analyses, we also examined the role of age of acquisition of the languages as our sample contained about equal numbers of English-dominant bilinguals who had learned both languages earlier in life and Spanish-dominant bilinguals, most of whom had learned English later in life.

2. Methods

2.1. Participants

We included 23 Spanish-English bilinguals from a longitudinal study at the Alzheimer’s Disease Research Center (ADRC) at the University of California, San Diego. At the first session, five participants were classified as cognitively healthy, five as mild cognitive impairment (MCI), one as Dementia with Lewy bodies (DLB), and 12 as probable Alzheimer’s Disease (AD). By the last session, all had a diagnosis of probable AD. The gap between first and last session ranged from two to seven years (M = 3.86; SD = 1.74). Diagnoses were based on criteria developed by the National Institute of Neurological Disorders and Stroke (NINDS) and the Alzheimer’s Disease and Related Disorders Association (ADRDA; McKhann et al., 1984).

Participants were tested on the full 68 item Multilingual Naming Test (MINT; Gollan et al., 2012) in both languages as part of their annual ADRC neuropsychological evaluation. Language dominance was determined based on average MINT scores available across all testing sessions: about half of the participants were Spanish-dominant, the other half were English-dominant.1 Seven bilinguals (from an original sample of 30) were excluded for the following reasons: one had identical naming scores in the two languages, two did not have a diagnosis of probable AD at the last testing session, and four had diagnoses of probable AD but did not decline (scores were greater than or equal to their scores at the first testing session) in naming accuracy in either language by the last testing session. Participant characteristics are summarized in Table 1. Spanish-dominant bilinguals rated their proficiency level in the nondominant language as significantly lower relative to English-dominant bilinguals, although their nondominant language naming scores were not significantly lower at the first testing session (in Table 1). Thus, despite being immersed in their nondominant language, Spanish-dominant bilinguals were not more proficient or balanced bilinguals relative to English-dominant bilinguals at the initial testing point.

The study procedures were approved by the UCSD Institutional Review Board and informed consent was obtained from patients and caregivers prior to neuropsychological testing and after discussing all study procedures.

3. Materials

The MINT includes 68 black-and-white pictures presented in increasing order of difficulty. Participants’ spontaneous responses were recorded. If a participant had difficulty recognizing a picture, a semantic cue was given. An answer was marked as correct if the expected answer was provided before or after this cue. If the picture name was not produced after this point, a phonetic cue was provided, and the item was marked as incorrect.

3.1. Procedure

Testing took place at the ADRC and was conducted by a proficient Spanish-English bilingual examiner. The MINT was administered first in the dominant language and then in the nondominant language in the latter half of a testing session that was part of the annual ADRC neuropsychological evaluation.

3.2. Analyses

We report a mixed-effects logistic regression analysis with by-item level accuracy (0 or 1) as the dependent variable. Fixed factors were language (dominant, coded as 0.5, or nondominant, coded as − 0.5), time point (first or last session, coded respectively as 0.5 and − 0.5), and following Gollan et al. (2023), we included item difficulty as a continuous centered fixed effect. Item difficulty was derived from an independent study with naming accuracy averaged across languages in 39 cognitively healthy older Spanish-English bilinguals (from Gollan et al., 2023; 28 were English-dominant and 11, Spanish-dominant); these mean accuracy rates were subtracted from 1, so that higher values of this variable represent pictures that were more difficult to name. We included all 2-way and the 3-way interactions (see model formula below Table 3). The model failed to converge with the maximal random-effects structure; we resolved convergence and singularity issues by removing by-subject and by-item random slopes so that only by-subject and by-item random intercepts were retained (Brauer and Curtin, 2018). Significance of fixed effects was measured using likelihood ratio tests.

4. Results

The model included 6255 observations. There was a main effect of language, such that bilinguals named significantly more pictures in the dominant compared to the nondominant language (b = 2.54, SE = 0.11, χ2 (1) = 559.12, p <0.001), a main effect of time point, such that bilinguals named more pictures correctly during the first session, compared to the last session (b = 1.09, SE = 0.10, χ2 (1) = 116.34, p <0.001), and a main effect of item difficulty, such that bilinguals named fewer pictures as item difficulty increased (b = − 10.41, SE = 0.43, χ2 (1) = 591.93, p <0.001). The interaction of language and time point was significant (b = 0.42,2 SE = 0.20, χ2 (1) = 4.45, p = 0.03). To evaluate this interaction, we calculated predicted probabilities for each cell at the mean item difficulty. The results showed a greater decline across time in probability of correct responses for the nondominant (first session: 69% vs. last session: 48%, difference of 21%) as compared to the dominant language (first session: 97% vs. last session: 90%, difference of 7%). The interaction of item difficulty by time point was marginally significant (b = − 0.86, SE = 0.51, χ2 (1) = 2.84, p = 0.09, such that the effect of item difficulty was stronger at the last than at the first session. The interaction of language by item difficulty and the three-way interaction were not significant (see Fig. 1 and Table 3). Proportion correct on MINT items is presented by difficulty level in Table 2. In Fig. 1 and Table 2, we dichotomized it into easy and hard items for visualization purposes, based on pictures named correctly by cognitively healthy older bilinguals (taken from Gollan et al., 2023), but note that item difficulty was entered as a continuous predictor in statistical analyses.

4.1. Exploratory analysis: age of acquisition

In our previous study (Ivanova et al., 2014), the asymmetrical decline pattern was driven primarily by a small number of Spanish-dominant bilinguals (n = 4). Because our sample included a balanced number of English-dominant bilinguals (n = 12), who learned both of their languages early in life, and Spanish-dominant bilinguals (n = 11), who learned a second language significantly later in life, we ran an exploratory analysis to look at age of acquisition effects. To do so, we added group (Spanish- versus English-dominant) to the model above; the critical interaction of Group x Language x Time was not significant (OR = 0.56). While acknowledging the limitations of post-hoc power analyses, we note that analyses with the “simr” (Green and MacLeod, 2016) and “mixedpower” (Kumle et al., 2024) packages in R with 100 simulations suggested that to detect a significant 3-way interaction, we would have needed 500 participants (22 times larger than our sample size) to achieve 85% power. These analyses show that both language dominance groups showed a tendency towards the same decline pattern, differing only in degree of this tendency – making it exceedingly difficult to detect group differences.

Since we were underpowered and to avoid interpreting a model with a four-way interaction (which is exceedingly complex), we further explored these data by separating our sample into dominance groups.

This sub-group analysis replicated the significant interaction between language and time in Spanish-dominant bilinguals, but the interaction was not significant in English-dominant bilinguals (see Supplemental Materials, Tables B and C and Figure B). However, because the sample sizes were still small for models of this complexity, we ran non-parametric sign test analyses. To this end, we calculated percent decline scores by subtracting the MINT score in the last testing session from the MINT score in the first available testing session (the starting point) divided by the MINT score in the first testing session. These were submitted to a sign test comparing the direction of decline in each language relative to the other for each participant. The sign test in the Spanish-dominant group was significant: the nondominant language declined more than the dominant language between the first and last testing sessions (S (11) = 1, p = .01, n = 11). Fig. 2 shows individual subject data; 10/11 Spanish-dominant bilinguals showed greater proportional decline of the nondominant than the dominant language. By contrast, the sign test within the English-dominant group overall showed no significant difference in median percentage of decline across the dominant versus the nondominant languages (S (12) = 4, p = .39, n = 12). However, Fig. 2 shows that more than half (7/12) of English-dominant bilinguals showed tendency for greater decline in the nondominant than the dominant language, while a minority (4/12) showed parallel decline (and 1/12 showed more decline in the dominant language). Thus, following Ivanova et al. (2014) the Spanish-dominant bilinguals seemed to exhibit a more robust asymmetrical decline pattern, but in the present study the English-dominant bilinguals exhibited nonsignificant tendencies in the same direction. Note that 10/12 English-dominant bilinguals acquired English at or before age 6. Only two bilinguals had a later age of acquisition, with one acquiring English at age 10 and the other at age 15. Only one of these two was one of the four bilinguals in the top panel of Fig. 2 who showed much greater decline of the nondominant language relative to the other English-dominant participants.

5. Discussion

The results of the present study revealed parallel decline of the two languages after one year, but an asymmetrical decline pattern after almost four years (on average), with greater decline of the nondominant language over time. Importantly, the asymmetrical pattern was only present when considering item difficulty – without it, decline was parallel. Exploratory analyses suggested that the asymmetrical decline pattern might be stronger (was significant on its own) in Spanish-dominant bilinguals, while English-dominant bilinguals exhibited more varying decline patterns (and parallel decline overall).

Why might the nondominant language decline more precipitously than the dominant language? It is possible that bilinguals may need more executive control to regulate activation of a dominant language when attempting to name pictures in a less proficient language (Abutalebi and Green, 2007; for further discussion, see Gollan et al., 2023; Ivanova et al., 2014; Stasenko et al., 2021). If executive control declines more rapidly than semantic representations in AD (see Guarino et al., 2019 for a review) this would affect retrieval of the nondominant language more than the dominant language. A less proficient language might also be used less often, and since low frequency of use disproportionately affects naming accuracy in AD compared to controls (Thompson-Schill et al., 1999), this could have affected that language relatively more.

The finding that decline was parallel when examining naming ability only over one year suggests that accessibility of the dominant and the nondominant language was similar early on after disease onset. Differential decline patterns only emerge in more advanced stages of the disease. This suggests incremental steps in cognitive decline, where shared semantic representations are affected first, causing parallel decline of the languages, and executive functioning declines next. This then leads to difficulties in inhibiting the dominant language to use the nondominant language, in turn impacting naming ability in the nondominant language. Note that more subtle impairments to executive control likely emerge in very early stages of disease progression (even pre-clinically; Garcia and Gollan, 2024), but these can only be observed in tasks that place greater demands on executive control (e.g., language switching tasks). Controlling activation of the nontarget language will be much easier when naming pictures in just one language during a testing block (as was done in the present study - see Smirnov et al., 2019, where the extent of damage to the anterior cingulate cortex - hypothesized to be a common locus for domain-general cognitive control - in bilinguals with AD predicts the ability to name pictures in the nondominant language.)

A powerful factor in the present study was item difficulty, which seems to be generally important for understanding the progression of language decline in bilinguals with AD (see also Gollan et al., 2023). Repeating the analysis without item difficulty as a factor returned parallel instead of asymmetrical decline (i.e., the interaction between language and time point was no longer significant, p = 0.41). Ivanova et al. (2014) suggested a complex U-shaped back and forth between languages with respect to which items decline first. On this view, only the most difficult (lowest frequency) items that bilinguals tend to know in the dominant language may decline first while relatively easier items remained unaffected. This might be followed by a period in which the nondominant language declines at a faster rate until it reaches floor levels, after which point only the dominant language will continue to decline. The dependence of the decline pattern on item difficulty level in the present study seems generally consistent with this view. Spanish-dominant older bilinguals may have been closer to this middle stage of faster decline of easier items in the nondominant language (see Table 2). By contrast, in English-dominant bilinguals, easy items in the nondominant language remained relatively intact between the first and last testing sessions (see Table 2). Although the two groups did not differ in DRS scores at the first or last testing sessions (see Table 1), it is possible that difficulty level influenced each group differently at the same stage of cognitive decline given that Spanish-dominant bilinguals tended to have lower education level than English-dominant bilinguals (see Table 1).

The results of the present study reconcile seemingly disparate findings relative to those reported by Calabria et al. (2017), in which bilinguals exhibited parallel decline of the two languages at one year past baseline. Critically, in the present study we too observed parallel decline when we repeated our analyses but examining decline only between the very first and second testing session one year later (see Supplemental Materials, Table A and Figure A), as was reported in Calabria et al. (2017). Thus, asymmetrical decline patterns may be visible only over several years. Additional evidence is needed to confirm whether the parallel decline pattern holds over many years of testing for bilinguals like those tested by Calabria et al. (2017).

Another possibly critical difference between previous work was that in Calabria et al. (2017), all participants were Catalan-dominant with Spanish (Castilian) as the nondominant language and started using both languages regularly before age five. In our study, English-dominant bilinguals had a similar profile (they began using English and Spanish regularly before age five), but Spanish-dominant bilinguals had on average a 15.4-year gap between acquisition of Spanish and English. This pattern was similar to Ivanova et al. (2014) in which English-dominant bilinguals acquired both languages within the first year of life, whereas Spanish-dominant bilinguals acquired English on average approximately 13.75 years after Spanish. Exploratory analyses also matched findings reported by Ivanova et al. (2014), in which the asymmetrical pattern was driven by four Spanish-dominant bilinguals who had learned English relatively later in life and exhibited greater decline of English than of Spanish, even though they were immersed in English (their nondominant language) at the time of testing. By contrast, English-dominant bilinguals who had learned both languages early in life and were immersed in the dominant language at the time of testing, exhibited parallel decline of the two languages with progression of AD over time.3

The differential pattern of decline across English- and Spanish-dominant bilinguals is in line with what we found in exploratory sign tests. The nondominant language declined more steeply than the dominant language only for the Spanish-dominant group, who had learned their second language later in life compared to the English-dominant group, who had learned both languages early in life, and exhibited parallel decline of their languages. However, as noted above, a slim majority of English-dominant bilinguals also exhibited a tendency towards asymmetrical decline. Later-learned lexical representations may be harder to access in picture naming even when equating early versus late learned items for frequency of use (Barry et al., 2001). This raises the possibility that age of acquisition may have influenced the effects reported here. The current data cannot resolve this issue, as we did not have sufficient power to systematically examine the effects of bilingual subgroup, and because we did not systematically control for age of acquisition (some of the Spanish-dominant bilinguals had learned English before adolescence; one even reported regularly using English since birth, and the standard deviations for age of regular use were large, see Table 1).

The issue of age of acquisition merits further investigation also because of well-documented anecdotal evidence that later-learned languages are more vulnerable to decline in AD (Mendez et al., 1999), and for theoretical reasons. In the Declarative/Procedural model (Ullman, 2001, 2020), which relates memory systems to language processing, grammatical processing depends on procedural memory whereas lexical processing depends on declarative memory. Moreover, later-learned languages are proposed to be more reliant on declarative memory for both lexical and grammatical processing. We did not test participants’ ability to speak in full sentences, but based on this model, it seems possible that producing full sentences might become progressively more difficult with cognitive decline especially in a later-learned language. If so, this would likely reduce inclination to converse in that language; and in turn, reduce the ability to access individual names in that language as well. To directly test this assumption, it would be necessary to assess bilinguals’ ability to produce connected speech longitudinally, perhaps with Oral Proficiency Interviews (OPIs) (Gollan et al., 2012; Garcia and Gollan, 2022, 2024). This could help understand the trajectory of possibly interconnected patterns of decline in retrieval of picture names versus ability to speak in syntactically well-formed multiword utterances over time.

A puzzling finding in our study was that decline patterns did not seem to be linked to current immersion. The group that showed greater decline of the nondominant language over time was also immersed in the nondominant language at the time of testing (and had been immersed in it for years; see Table 1). Normally, bilinguals who are immersed in the nondominant language should have an easier time maintaining proficiency in the nondominant language. In the present study, if anything the opposite was true. By contrast, here English-dominant bilinguals tended to have higher MINT scores in their nondominant language than did Spanish-dominant bilinguals (see Tables 1 and 2), and Spanish-dominant bilinguals self-rated their proficiency of the nondominant language as significantly lower than did English-dominant bilinguals. This could be because of a tendency for Spanish-dominant bilinguals to have lower education levels, or Spanish-dominant bilinguals may have declined in the nondominant language more than did English dominant bilinguals even before the first testing point in the present study. Importantly, Fig. 2 suggests that Spanish-dominant bilinguals (bottom panel) were more likely to exhibit asymmetrical decline than English-dominant bilinguals (top panel) at all proficiency levels (including those with very high proficiency based on the first nondominant language MINT score). Thus, it is unlikely that a lower initial proficiency level in the nondominant language for Spanish-dominant bilinguals could fully explain the between group difference in decline pattern (i.e., parallel versus asymmetrical).

5.1. Limitations

The results of the present study must be interpreted with caution given the small number of participants tested, and the presence of some complex and not necessarily expected interactions. Additionally, more control over the length of time between first and final testing sessions would be needed to strengthen findings. Note however that we doubled the number of participants and items relative to our previous study on this topic with a partially overlapping sample (Ivanova et al., 2014) and came to the same conclusion: decline is asymmetrical overall, not parallel. Additionally, our results replicate and reconcile seemingly distinct patterns across different labs (Ivanova et al., 2014 vs. Calabria et al., 2017) when matching for the length of longitudinal follow-up. Though we had a small number of participants it is important to consider that there are practical limitations on the sample sizes that can be obtained for hard-to-reach populations followed longitudinally for several years. Rigorous assessment of such populations, which reflect the true diversity of the human experience better than samples of monolingual speakers only, is essential for making accurate generalizations about language processing and language impairments – even when our samples are underpowered. The only other study on this topic in the literature (Calabria et al., 2017) may have reached an incomplete conclusion because participants were followed only for a year. Given that the parallel decline pattern at the one-year follow-up point is replicated here, the present study makes a powerful point that decline patterns need to be examined for more than one year, as picture naming abilities seem to decline progressively in AD, over the course of several years.

6. Conclusion

In sum, our longitudinal analyses converge to suggest that the nondominant language declines more than the dominant language in bilinguals with AD, when examining decline for longer than just a single year of time between testing points. More research is needed to confirm this pattern (with larger numbers of participants) and to investigate what cognitive mechanism underlies the precipitous drop in picture naming ability in the nondominant language when it does occur. It is also notable that there were substantial individual differences in which individual bilinguals did versus did not exhibit the precipitous drop (see Fig. 2). The dependence of asymmetrical decline on item difficulty also merits further investigation. These results suggest that the ability to access the nondominant language remains relatively intact in initial stages of AD until executive control declines to the point where it affects the ability to name pictures in single-language testing blocks. Additional research will be needed to understand the cognitive mechanisms underlying age of acquisition effects and language proficiency, and how they are related to decline in AD.

Supplementary Material

Supplementary Materials

Acknowledgements

This research was supported by grants from the National Institute on Aging (R01 AG076415; P30 AG062429).

Data availability

The data and script that support the findings of this study are openly available on OSF at: https://osf.io/dsnu6/?view_only=c766b21b79d24bab956ae26f02df022c.

Fig. 1. Probability of correct response on MINT scores as a function of language (dominant, nondominant), time point (first, last testing session), and item difficulty across easy (top) and hard items (bottom). Dots represent actual data and lines represent predicted data.

Fig. 2. Proportion decline by dominance group at the last testing session relative to the first testing session.

Note. The x-axis shows English age of regular use in ascending order (i.e. earlier ages of acquisition are on the left and later ages of acquisition are on the right). Additionally, corresponding MINT scores of each individual participant in the nondominant language at the first session is added below. Negative values in percent decline indicate an increase in scores between first and last session. In Spanish-dominant bilinguals, all but one participant showed greater decline of the nondominant than the dominant language. In English-dominant bilinguals decline patterns varied more between participants (e.g., four showed parallel decline, one showed greater decline of the dominant than the nondominant language).

Note. MDP = missing data point. Arbitrarily positioned at the far-right of the x-axis (age of regular use for this person was not available).

Table 1 Participant characteristics, sample of older bilinguals divided by language dominance.

	Language dominance	
		
Characteristic	English (n = 12)	Spanish (n = 11)	t-test	p-value	
	
Gender (female/male)	6/6		8/3		<1a	0.49	
	M	(SD)	M	(SD)			
Age at first session	78.9	(9.4)	72.0	(13.8)	1.42	0.17	
Gap in years between first and last session	4.0	(2.8)	3.8	(2.1)	−0.18	0.86	
Education in years	13.2	(4.4)	10.6	(3.0)	1.60	0.13	
Dementia Rating Scale score						
 First session	121.2	(13.2)	120.1	(12.8)	<1	0.84	
 Last session	107.3	(12.4)	101.5	(19.3)	<1	0.39	
Age of regular useb							
 English	4.3	(4.6)	16.8	(14.8)	−2.58d	0.03	
 Spanish	1.0	(1.8)	1.4	(1.3)	<1	0.61	
Years lived in Spanish-speaking countryb	13.0	(16.8)	33.9	(20.4)	−2.63	0.02	
Average self-rated proficiencyb,c						
 Dominant language	6.4	(0.8)	6.6	(0.6)	<1	0.48	
 Nondominant language	5.5	(1.1)	4.2	(1.6)	2.28	0.03	
Dominant language MINTe score						
 First session	59.0	(4.1)	58.2	(3.1)	<1	0.60	
 Last session	53.9	(7.3)	52.0	(9.2)	<1	0.58	
Nondominant language MINTe score						
 First session	46.5	(10.9)	37.8	(15.6)	1.56	0.13	
 Last session	39.5	(12.9)	29.2	(16.4)	1.68	0.11	
a For the variable of gender, Pearson’s Chi-squared test was run instead as the data are categorical.

b Data missing from one participant. We report age of regular use because some bilinguals in both groups reported acquiring both languages from birth, but this may not have reflected the age of immersion which may be most relevant for present purposes.

c Average of scores ranging from 1 to 7 of speaking, understanding, reading, and writing skills.

d Welch’s t-test was used as variances across groups were unequal.

e Multilingual Naming Test.

Table 2 MINT item proportion correct by difficulty level, dominance group and time, across languages.

	Dominant			Nondominant		
			
	First session	Last session	Diff.	First session	Last session	Diff.	
	
All bilinguals (N = 23)						
Easier items	0.99	0.96	0.03	0.86	0.74	0.12	
Harder items	0.73	0.59	0.14	0.37	0.26	0.11	
English-dominant (n = 12)						
Easier items	0.98	0.96	0.02	0.91	0.82	0.09	
Harder items	0.75	0.62	0.13	0.44	0.33	0.11	
Spanish-dominant (n = 11)						
Easier items	0.99	0.96	0.03	0.81	0.66	0.15	
Harder items	0.71	0.56	0.15	0.29	0.19	0.10	
Note. The easy versus hard items division was derived from a median split of average accuracy (proportion correct) in an independent study (Gollan et al., 2023). There were 35 easy and 33 hard items.

Table 3 Model output for mixed-effects logistic regression analysis of accuracy on the MINT as a function of language (dominant or nondominant), time point (first and last session), item difficulty, and their interaction, in N = 23 Spanish-English bilinguals with AD.

Accuracy					
	
	b	SE	χ2	p-value	
	
Intercept	1.62	0.27	36.97	<0.001	
Language	2.54	0.11	559.12	<0.001	
Time point	1.09	0.10	116.34	<0.001	
Item difficulty	−10.41	0.43	591.93	<0.001	
Language × Time point	0.42	0.20	4.45	0.03	
Language × Item difficulty	−0.62	0.57	1.18	0.28	
Time point × Item difficulty	−0.86	0.51	2.84	0.09	
Language × Time point × Item difficulty	−1.16	1.00	1.34	0.25	
Note. Model: Accuracy ~ Language × Time point × Item difficulty + (1|Participant) + (1|Item).

Random effects for Item: 0.24 and for Participant: 1.45.

Declaration of competing interest

None.

CRediT authorship contribution statement

Anne Neveu: Writing – review & editing, Writing – original draft, Visualization, Project administration, Investigation, Formal analysis, Data curation. Matthew Goldrick: Writing – review & editing, Visualization, Formal analysis. Daniel Kleinman: Writing – review & editing, Visualization, Formal analysis. David P. Salmon: Writing – review & editing, Funding acquisition. Tamar H. Gollan: Writing – review & editing, Funding acquisition, Formal analysis, Conceptualization.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.neuropsychologia.2024.108948.

1 In most cases, objective (i.e., picture naming test scores) and subjective measures of language dominance matched (i.e., averaging self-rated proficiency for ability to speak, read, write, and comprehend spoken language). Specifically, subjective and objective language dominance matched for 7/12 English-dominant bilinguals and for 9/11 Spanish-dominant bilinguals (n.b.: self-rated proficiency level was missing for one Spanish-dominant participant). An additional 4/12 English-dominant and 1/11 Spanish-dominant bilinguals rated themselves as balanced bilinguals but named between 2.25 and 34 more pictures in one language on average (across each bilinguals’ testing sessions, aggregated over all testing sessions). There was only one bilingual we classified as English-dominant even though by self-ratings this person was Spanish-dominant (and objectively this person was relatively balanced; s/he named two more pictures in English than in Spanish over three testing sessions, which corresponded to a difference in averaged scores across sessions of <1 point).

2 While the beta coefficient for this interaction is positive, suggesting larger decline in the dominant language, values get distorted at the extremes of the probability space. Therefore, we use the predicted probability method to interpret this effect. Predicted probabilities are calculated using the regression equation and the appropriate coding for categorical variables (e.g. for calculating the predicted probability of accuracy at the first session, coded as 0.5, and in the dominant language, coded as 0.5, beta values are input as follows: intercept beta + (time beta*0.5)+(language beta*0.5)+(interaction of time by language beta*0.5*0.5). The value obtained is a natural log (logit); we take its inverse to get the odds by using the exponential function. Probabilities are next obtained by taking odds/(1+odds) (see R Studio script, section titled “Mixed effects logistic regression n = 23” on OSF repository linked in this paper for an example).

3 Note that 7/12 of the English-dominant bilinguals and 2/11 of the Spanish-dominant bilinguals in the present study were also participants in Ivanova et al. (2014).
==== Refs
References

Abutalebi J , Green D , 2007. Bilingual language production: The neurocognition of language representation and control. J. Neurolinguistics 20 (3 ), 242–275.
Barry C , Hirsh KW , Johnston RA , Williams CL , 2001. Age of acquisition, word frequency, and the locus of repetition priming of picture naming. J. Mem. Lang. 44 (3 ), 350–375.
Brauer M , Curtin JJ , 2018. Linear mixed-effects models and the analysis of nonindependent data: A unified framework to analyze categorical and continuous independent variables that vary within-subjects and/or within-items. Psychol. Methods 23 (3 ), 389.29172609
Calabria M , Cattaneo G , Marne P , Hernández M , Juncadella M , Gascón-Bayarri J , , 2017. Language deterioration in bilingual Alzheimer’s disease patients: A longitudinal study. J. Neurolinguistics 43 , 59–74.
Calabria M , Costa A , Green DW , Abutalebi J , 2018. Neural basis of bilingual language control. Ann. N. Y. Acad. Sci. 1426 (1 ), 221–235.
Costa A , Calabria M , Marne P , Hernández M , Juncadella M , Gascón-Bayarri J , , 2012. On the parallel deterioration of lexico-semantic processes in the bilinguals’ two languages: Evidence from Alzheimer’s disease. Neuropsychologia 50 (5 ), 740–753.22261401
de Picciotto J , Friedland D , 2001. Verbal fluency in elderly bilingual speakers: Normative data and preliminary application to Alzheimer’s disease. Folia Phoniatrica Logop. 53 (3 ), 145–152.
Francis WS , 2005. Bilingual semantic and conceptual representation. In: Kroll JF , De Groot AMB (Eds.), Handbook of Bilingualism: Psycholinguistic Approaches. Oxford University Press, pp. 251–267.
Friedland D , Miller N , 1999. Language mixing in bilingual speakers with Alzheimer’s dementia: A conversation analysis approach. Aphasiology 13 (4–5 ), 427–444.
Garcia DL , Gollan TH , 2022. The MINT Sprint: Exploring a fast administration procedure with an expanded multilingual naming test. J. Int. Neuropsychol. Soc. 28 (8 ), 845–861.34463235
Garcia DL , Gollan TH , 2024. Language switching and speaking a nondominant language challenge executive control: Preliminary data for novel behavioral markers of Alzheimer’s risk in Spanish–English bilinguals. Neuropsychology 38 (4 ), 322–336. 10.1037/neu0000943.38330361
Gollan TH , Salmon DP , Montoya RI , Da Pena E , 2010. Accessibility of the nondominant language in picture naming: A counterintuitive effect of dementia on bilingual language production. Neuropsychologia 48 (5 ), 1356–1366.20036679
Gollan TH , Stasenko A , Li C , Salmon DP , 2017. Bilingual language intrusions and other speech errors in Alzheimer’s disease. Brain Cognit. 118 , 27–44.28753438
Gollan TH , Stasenko A , Salmon DP , 2023. Which language is more affected in bilinguals with Alzheimer’s disease? Diagnostic sensitivity of the Multilingual Naming Test. Neuropsychology 37 (5 ), 595.36931816
Gollan TH , Weissberger GH , Runnqvist E , Montoya RI , Cera CM , 2012. Self-ratings of spoken language dominance: A Multilingual Naming Test (MINT) and preliminary norms for young and aging Spanish–English bilinguals. Biling. Lang. Cognit. 15 (3 ), 594–615.
Gómez-Ruiz I , Aguilar-Alonso Á , Espasa MA , 2012. Language impairment in Catalan-Spanish bilinguals with Alzheimer’s disease. J. Neurolinguistics 25 (6 ), 552–566.
Green DW , 1998. Mental control of the bilingual lexico-semantic system. Biling. Lang. Cognit. 1 (2 ), 67–81.
Green DW , Abutalebi J , 2013. Language control in bilinguals: The adaptive control hypothesis. J. Cognit. Psychol. 25 (5 ), 515–530.
Green P , MacLeod CJ , 2016. simr: An R package for power analysis of generalised linear mixed models by simulation. Methods Ecol. Evol. 7 (4 ), 493–498. 10.1111/2041-210X.12504. https://CRAN.R-project.org/package=simr.
Guarino A , Favieri F , Boncompagni I , Agostini F , Cantone M , Casagrande M , 2019. Executive functions in Alzheimer disease: A systematic review. Front. Aging Neurosci. 10 , 437.30697157
Ivanova I , Salmon DP , Gollan TH , 2013. The multilingual naming test in Alzheimer’s disease: Clues to the origin of naming impairments. J. Int. Neuropsychol. Soc. 19 (3 ), 272–283.23298442
Ivanova I , Salmon DP , Gollan TH , 2014. Which language declines more? Longitudinal versus cross-sectional decline of picture naming in bilinguals with Alzheimer’s disease. J. Int. Neuropsychol. Soc. 20 (5 ), 534–546.24725624
Kaplan E , Goodglass H , Weintraub S , 1983. The Boston Naming Test. Lea & Febiger, Philadelphia.
Kroll JF , Gollan TH , 2014. Speech planning in two languages: What bilinguals tell us about language production. In: Goldrick M , Ferreira V , Miozzo M (Eds.), The Oxford Handbook of Language Production. Oxford University Press, pp. 165–181.
Kroll JF , Stewart E , 1994. Category interference in translation and picture naming: Evidence for asymmetric connections between bilingual memory representations. J. Mem. Lang. 33 (2 ), 149–174.
Kumle L , Võ ML-H , Draschkow D , 2024. Mixedpower: Pilotdata Based Simulations for Estimating Power in Linear Mixed Models. R package version 0.1.0, commit 69680b3668d5d5ffc201fadb705dcfbd91129567. https://github.com/DejanDraschkow/mixedpower.
McKhann G , Drachman D , Folstein M , Katzman R , Price D , Stadlan EM , 1984. Clinical diagnosis of Alzheimer’s disease: Report of the NINCDS-ADRDA work group. Neurology 34 , 939–944.6610841
Manchon M , Buetler K , Colombo F , Spierer L , Assal F , Annoni JM , 2015. Impairment of both languages in late bilinguals with dementia of the Alzheimer type. Biling. Lang. Cognit. 18 (1 ), 90–100.
Meguro K , Senaha ML , Caramelli P , Ishizaki J , Chubacci RY , Meguro M , , 2003. Language deterioration in four Japanese–Portuguese bilingual patients with Alzheimer’s disease: A trans-cultural study of Japanese elderly immigrants in Brazil. Psychogeriatrics 3 (2 ), 63–68.
Mendez MF , Perryman KM , Pontón MO , Cummings JL , 1999. Bilingualism and dementia. The Journal of Neuropsychiatry and Clinical Neurosciences 11 (3 ), 411–412.10440021
Pickering MJ , Garrod S , 2013. An integrated theory of language production and comprehension. Behav. Brain Sci. 36 (4 ), 329–347.23789620
Salvatierra J , Rosselli M , Acevedo A , Duara R , 2007. Verbal fluency in bilingual Spanish/English Alzheimer’s disease patients. Am. J. Alzheimer’s Dis. Other Dementias 22 (3 ), 190–201.
Smirnov DS , Stasenko A , Salmon DP , Galasko D , Brewer JB , Gollan TH , 2019. Distinct structural correlates of the dominant and nondominant languages in bilinguals with Alzheimer’s disease (AD). Neuropsychologia 132 , 107131.31271821
Stasenko A , Kleinman D , Gollan TH , 2021. Older bilinguals reverse language dominance less than younger bilinguals: Evidence for the inhibitory deficit hypothesis. Psychol. Aging 36 (7 ), 806.34166027
Thompson-Schill SL , Gabrieli JD , Fleischman DA , 1999. Effects of structural similarity and name frequency on picture naming in Alzheimer’s disease. J. Int. Neuropsychol. Soc. 5 (7 ), 659–667.10645707
Ullman MT , 2001. A neurocognitive perspective on language: The declarative/procedural model. Nat. Rev. Neurosci. 2 (10 ), 717–726.11584309
Ullman MT , 2020. The Declarative/Procedural Model: A neurobiologically motivated theory of first and second Language. In: VanPatten B , Keating GD , Wulff S (Eds.), Theories in Second Language Acquisition. Routledge, pp. 128–161.
Weintraub S , Wicklund AH , Salmon DP , 2012. The neuropsychological profile of Alzheimer disease. Cold Spring Harbor Perspect. Med. 2 (4 ), a006171.
