Understanding the semantic organization of animal fluency in mild Alzheimer's disease through time-course analysis and LDA topic modelling

Deterioration of semantic memory represents an important feature of Alzheimer's disease (AD), which has been widely addressed in neuropsychological research. A way to understand semantic integrity in AD patients is through a detailed analysis of verbal fluency (VF) performance. In the present study, we used an innovative methodology that combines measures of between-words latencies together with automatized identification of semantic clusters via Latent Dirichlet Analysis (LDA) to acquire a more precise understanding of the dynamics and semantic organization of VF in patients at early stages of AD. Importantly, and diverging from customary procedures, we included VF errors (i.e., repetitions and intrusions) across analyses. For comparison, a group of healthy older adults and young individuals were also examined. Standard parameters including total correct answers, number of clusters, mean cluster size (MCS), cluster duration, and within and out-of-cluster intervals were calculated. These parameters were expressed as mean values in 1-min VF trials and by calculating mean values in four 15-s time windows. Results for the 1-min trial demonstrated significantly larger mean cluster sizes (MCS) and fewer generated answers in AD patients compared to the healthy groups. No additional group differences were found neither on time intervals (both within and out-of-clusters), nor on the 15-s time windows analyses. These data suggest that the clustering ability of mild AD patients might be affected by executive impairments promoting larger MCS. At the same time, we found similar semantic processes and timings in patients and healthy participants. The main difference resides in the structure of the patients' clusters, which encompassed erroneous answers. We advance the idea that production of errors might not only be a consequence of executive dysfunction or working memory deterioration, but also a sign that associative semantic mechanisms are still active early in the disease, despite an evident loss of information.