Spontaneous speech undergoes subtle but significant changes years before the onset of Alzheimer's dementia (AD). In monolinguals, these changes, or linguistic markers of AD, include the use of syntactically simpler structures, reduced lexical diversity, reduced semantic detail/specificity, and increased disfluencies (Ostrand & Gunstad, 2020; Slegers et al., 2018; Venneri et al., 2018). No studies have examined if bilinguals exhibit similar changes in their language output prior to diagnosis of AD though this question has important clinical relevance and can also shed light on which cognitive abilities decline first with AD pathology. Of particular interest, changes in semantic representations might affect both languages (because semantics are shared between the two), but changes in executive control might be more prominent in the nondominant language (because of interference from the dominant language).

Seventeen older Spanish-English bilinguals completed an interview in which they described a picture in each language and answered a series of questions beginning with "warm-up" questions and progressing to questions that elicited higher level language (e.g., defending an opinion). All participants were considered cognitively healthy at the time of testing, but 8 participants later developed Alzheimer's Disease (i.e., converters) on average after 4.1 (SD=2.5) years, while 9 matched controls remained cognitively healthy on average for 5.7 (SD=3.6) years (for as long as they were followed). Converters and controls were matched for age, education, language proficiency, and cognitive status at the time of testing. Language samples were transcribed word for word and analyzed using the Systematic Analysis of Language (Miller & Iglesias, 2012).

Converters and controls were compared on measures of syntactic complexity, lexical diversity, abandoned utterances, errors, and disfluencies. In the dominant language, the number of different words (using a moving window average; a measure of lexical diversity), showed promise for classifying who would eventually convert (Area Under the Curve = 77), though the difference between converters and controls was significant only in a 1-tailed test (t(15)=-1.96, p=.034). In the nondominant language, converters showed a higher percent of Maze words compared to controls (2-tailed t (15) = 2.27, p = 0.039). Mazes combine repetitions, filled pauses, and revisions. Further exploration of Maze subcomponents revealed that filled pauses and revisions produced no differences between groups in either language (all ps3.18), but converters produced more repetitions (e.g., "the the boy" or "the cou-counter") than controls, (2-tailed t-tests in both languages were significant; ps <.03). However, variability in repetitions was high, making it less sensitive in the ROC analysis.

Changes in bilinguals' spoken language output occur years before diagnosis, in agreement with literature on monolinguals. However, in bilinguals, the two languages may be differentially affected by cognitive changes. The dominant language may be more sensitive for discriminating groups possibly reflecting semantic decline and decreased ability to quickly access a variety of words. But changes in the nondominant language reveal a broader nature of cognitive deficits in prodromal AD, including decreased circumlocution abilities to avoid disfluencies when faced with word-finding difficulties.