In this study, 99 probable AD participants from the DementiaBank(Becker et al., 1994) were analyzed on the sentence construction task. They constructed a sentence with given words (pencil, tree) which is similar to Altmann’s(2004). Semantic distances between two different text corpora (DementiaBank vs. Wikipedia(or Blog)) were conducted by independent samples t-test and stepwise logistic regression to identify whether the verb clusters are associated with demographic factors.

In noun-verb semantic distance, the cosine similarity of ‘pencil’ (t₅₉₆ = -5.050, p = 5.881e-7) and ‘tree’ (t_24.98 = -7.888, p = 3.053e-8) were statistically higher in DementiaBank than the Wikipedia. Moreover, the results were similar between the DementiaBank and the Blog. The cosine similarity of ‘pencil’ (t₁₉₉ = -3.702, p = 2.764e-4) and ‘tree’ (t_24.98 = -7.888, p = 3.053e-8) were higher in the DementiaBank. 

In the verb clustering and regression analyses, education had a significantly positive effect on the choice of ‘write(baseline)’ or be-verb for the ‘pencil’ (B=.661, Wald=6.871, p=.009), but marginally positive on the choice of verb ‘write’(baseline) or ‘use’ for the noun ‘pencil’(B=.325, Wald=3.553, p =.059). For MMSE scores, it was significantly negative on the choice of verb(B =-.294, Wald=6.674, p =.01), either ‘write’(baseline) or ‘be’ for the ‘pencil’, and significantly positive on the choice of either ‘be’(baseline) or ‘grow’ for the noun ‘tree’(B=.274, Wald=3.921, p=.048).

To conclude, semantic distances between nouns and verbs are shorter in AD patients compared to existing big databases. Moreover, semantic weight of verbs that AD patients used was significantly related to the severity of dementia, indicating they tend to use more light verbs.