This supplementary material contains two networks from our dataset. The first one, "insurance", is about the assessment of car insurance seekers. The second one, "hailfinder", is about severe weather forecasting in a mountainous area. 

We provide both JSON and ProbLog files.

The JSON files contain the following entries:
"nodes": All nodes (= random variables) that make up the Bayesian network (BN) are listed here. Also, all possible states for each node are provided.
"connectivity": This lists all incoming edges for every node in the BN.
"numeric_premises": Here, all numeric background premises are listed that are expected to be parsed by LLM-based models to build up the probabilistic model.
"wep_based": This entry contains the same amount of background premises, but all percentages were replaced with "words of estimative probabilities" (WEP)
"evidence_query_pairs": These are the instances that allow for Bayesian reasoning. Evidences refer to the observations, i.e., "Given that [...]". Query is then the question for the probability of a particular random variable taking a single state.

The ProbLog files contain the symbolic representations of the BNs. They show how categorical distributions can be modeled in ProbLog.