Development of acoustic and linguistic markers for the automatic detection of cognitive states in older Quechua-speaking adults

The use of natural language processing (NLP) tools has been having a significant impact on healthcare. Much of this development is focused on the automatic detection of emotional and cognitive states, but primarily in “major” languages such as English, with very little attention paid to languages considered to be under-resourced, such as Quechua. The progressive aging of the population brings with it a series of social and health challenges, including diseases such as dementia. These difficulties are more pronounced in rural populations, since diagnostic procedures are highly costly (e.g., MRI), the number of specialists is limited, many of them have not been validated in ethnic minorities, and they have low ecological validity (stimuli that are far removed from reality). In contrast, natural language tools are relatively low-cost, can be applied by non-specialized personnel, and have high ecological validity (e.g., recounting a happy memory). Therefore, we believe that advances in the development of these algorithms could have a significant impact on supporting the assessment and differentiation of Quechua-speaking patients with cognitive impairment. For this reason, we have set out to create a dataset in Quechua Collao and to validate the extracted acoustic and linguistic markers that are sensitive to detecting older adults with signs of cognitive decline and dementia. We hope to develop robust and highly scalable measures that will enable us to support dementia diagnostic processes, as well as to develop tools based on a Quechua Collao speech corpus that will allow us not only to advance research but also to contribute to its preservation.