Let$$X=(X_1,\cdots,X_k)\sim Dir(\alpha_1,\cdots,\alpha_k)$$
According to this reference, the independence of the two vectors, $$\bigg(\frac{X_1}{X_1+\cdots+X_j},\cdots,\frac{X_j}{X_1+\cdots+X_j}\bigg)$$and$$\bigg(\frac{X_{j+1}}{X_{j+1}+\cdots+X_k},\cdots,\frac{X_k}{X_{j+1}+\cdots+X_k}\bigg),$$ follows from the neutrality and aggregation property of the Dirichlet distribution.
I'm wondering to what extent can this result be generalized?
More specifically, for three-subvector cases, let $\{\pi_1,\pi_2,\pi_3\}$ be a partition over $\{1,\cdots,k\}$. e.g., $k=3, \pi_1=\{1\},\pi_2=\{2\}$, and $\pi_3=\{3\}$. Can we say the following normalized subvectors are mutually independent$$\bigg(\frac{X_k}{\sum_{j\in\pi_1}X_j}\bigg)_{k\in\pi_1},\bigg(\frac{X_k}{\sum_{j\in\pi_2}X_j}\bigg)_{k\in\pi_2},\bigg(\frac{X_k}{\sum_{j\in\pi_3}X_j}\bigg)_{k\in\pi_3}$$