Yes, the posterior predictive distribution is the relevant distribution in both of these cases. See the bottom of page two of this paper:
https://arxiv.org/abs/1507.04544
Each term in the sum for lpd is just the the log of the posterior predictive density evaluated at one of the data points.