\section{Description Logics} \label{app:dl} In this section, we recall the expressive description logic $\mathcal{ALC}$ \cite{DBLP:journals/ai/Schmidt-SchaussS91}. We refer to \cite{DBLP:journals/ws/LukasiewiczS08} for a detailed description of $\mathcal{SHOIN}(\mathbf{D})$ DL, that is at the basis of OWL DL. %%ALC Let $\mathbf{A}$, $\mathbf{R}$ and $\mathbf{I}$ be sets of \emph{atomic concepts}, \emph{roles} and \emph{individuals}. A \emph{role} is an atomic role $R \in \mathbf{R}$. \emph{Concepts} are defined by induction as follows. Each $C \in \mathbf{A}$, $\bot$ and $\top$ are concepts. If $C$, $C_1$ and $C_2$ are concepts and $R \in \mathbf{R}$, then $(C_1\sqcap C_2)$, $(C_1\sqcup C_2 )$, $\neg C$, $\exists R.C$, and $\forall R.C$ are concepts. Let $C$, $D$ be concepts, $R \in \mathbf{R}$ and $a, b \in \mathbf{I}$. An \emph{ABox} $\cA$ is a finite set of \textit{concept membership axioms} $a : C$ and \textit{role membership axioms} $(a, b) : R$, while a \emph{TBox} $\cT$ is a finite set of \textit{concept inclusion axioms} $C\sqsubseteq D$. $C \equiv D$ abbreviates $C \sqsubseteq D$ and $D\sqsubseteq C$. A \emph{knowledge base} $\cK = (\cT, \cA)$ consists of a TBox $\cT$ and an ABox $\cA$. A KB $\cK$ is assigned a semantics in terms of set-theoretic interpretations $\cI = (\Delta^\cI , \cdot^\cI )$, where $\Delta^\cI$ is a non-empty \textit{domain} and $\cdot^\cI$ is the \textit{interpretation function} that assigns an element in $\Delta ^\cI$ to each $a \in \mathbf{I}$, a subset of $\Delta^\cI$ to each $C \in \mathbf{A}$ and a subset of $\Delta^\cI \times \Delta^\cI$ to each $R \in \mathbf{R}$. \section{DISPONTE} \label{app:disponte} In the field of Probabilistic Logic Programming (PLP for short) many proposals have been presented. An effective and popular approach is the Distribution Semantics \cite{DBLP:conf/iclp/Sato95}, which underlies many PLP languages such as PRISM~\cite{DBLP:conf/iclp/Sato95,DBLP:journals/jair/SatoK01}, Independent Choice Logic \cite{Poo97-ArtInt-IJ}, Logic Programs with Annotated Disjunctions \cite{VenVer04-ICLP04-IC} and ProbLog \cite{DBLP:conf/ijcai/RaedtKT07}. Along this line, many reserchers proposed to combine probability theory with Description Logics (DLs for short) \cite{DBLP:journals/ws/LukasiewiczS08,DBLP:conf/rweb/Straccia08}. DLs are at the basis of the Web Ontology Language (OWL for short), a family of knowledge representation formalisms used for modeling information of the Semantic Web TRILL follows the DISPONTE \cite{RigBelLamZese12-URSW12,Zese17-SSW-BK} semantics to compute the probability of queries. DISPONTE applies the distribution semantics \cite{DBLP:conf/iclp/Sato95} of probabilistic logic programming to DLs. A program following this semantics defines a probability distribution over normal logic programs called \emph{worlds}. Then the distribution is extended to queries and the probability of a query is obtained by marginalizing the joint distribution of the query and the programs. In DISPONTE, a \emph{probabilistic knowledge base} $\cK$ is a set of \emph{certain axioms} or \emph{probabilistic axioms} in which each axiom is independent evidence. Certain axioms take the form of regular DL axioms while probabilistic axioms are $p::E$ where $p$ is a real number in $[0,1]$ and $E$ is a DL axiom. The idea of DISPONTE is to associate independent Boolean random variables to the probabilistic axioms. To obtain a \emph{world}, we include every formula obtained from a certain axiom. For each probabilistic axiom, we decide whether to include it or not in $w$. A world therefore is a non probabilistic KB that can be assigned a semantics in the usual way. A query is entailed by a world if it is true in every model of the world. The probability $p$ can be interpreted as an \emph{epistemic probability}, i.e., as the degree of our belief in axiom $E$. For example, a probabilistic concept membership axiom $ p::a:C $ means that we have degree of belief $p$ in $C(a)$. A probabilistic concept inclusion axiom of the form $ p::C\sqsubseteq D $ represents our belief in the truth of $C \sqsubseteq D$ with probability $p$. Formally, an \emph{atomic choice} is a couple $(E_i,k)$ where $E_i$ is the $i$th probabilistic axiom and $k\in \{0,1\}$. $k$ indicates whether $E_i$ is chosen to be included in a world ($k$ = 1) or not ($k$ = 0). A \emph{composite choice} $\kappa$ is a consistent set of atomic choices, i.e., $(E_i,k)\in\kappa, (E_i,m)\in \kappa$ implies $k=m$ (only one decision is taken for each formula). The probability of a composite choice $\kappa$ is $P(\kappa)=\prod_{(E_i,1)\in \kappa}p_i\prod_{(E_i, 0)\in \kappa} (1-p_i)$, where $p_i$ is the probability associated with axiom $E_i$. A \emph{selection} $\sigma$ is a total composite choice, i.e., it contains an atomic choice $(E_i,k)$ for every probabilistic axiom of the probabilistic KB. A selection $\sigma$ identifies a theory $w_\sigma$ called a \emph{world} in this way: $w_\sigma=\cC\cup\{E_i|(E_i,1)\in \sigma\}$ where $\cC$ is the set of certain axioms. Let us indicate with $\mathcal{S}_\cK$ the set of all selections and with $\mathcal{W}_\cK$ the set of all worlds. The probability of a world $w_\sigma$ is $P(w_\sigma)=P(\sigma)=\prod_{(E_i,1)\in \sigma}p_i\prod_{(E_i, 0)\in \sigma} (1-p_i)$. $P(w_\sigma)$ is a probability distribution over worlds, i.e., $\sum_{w\in \mathcal{W}_\cK}P(w)=1$. We can now assign probabilities to queries. Given a world $w$, the probability of a query $Q$ is defined as $P(Q|w)=1$ if $w\models Q$ and 0 otherwise. The probability of a query can be defined by marginalizing the joint probability of the query and the worlds, i.e. $P(Q)=\sum_{w\in \mathcal{W}_\cK}P(Q,w)=\sum_{w\in \mathcal{W}_\cK} P(Q|w)p(w)=\sum_{w\in \mathcal{W}_\cK: w\models Q}P(w)$. \begin{example} \label{people+petsxy} \begin{small} Consider the following KB, inspired by the \texttt{people+pets} ontology \cite{ISWC03-tut}: {\center $0.5\ \ ::\ \ \exists hasAnimal.Pet \sqsubseteq NatureLover\ \ \ \ \ 0.6\ \ ::\ \ Cat\sqsubseteq Pet$\\ $(kevin,tom):hasAnimal\ \ \ \ \ (kevin,\fluffy):hasAnimal\ \ \ \ \ tom: Cat\ \ \ \ \ \fluffy: Cat$\\} \noindent The KB indicates that the individuals that own an animal which is a pet are nature lovers with a 50\% probability and that $kevin$ has the animals $\fluffy$ and $tom$. Fluffy and $tom$ are cats and cats are pets with probability 60\%. We associate a Boolean variable to each axiom as follow %\begin{small} $F_1 = \exists hasAnimal.Pet \sqsubseteq NatureLover$, $F_2=(kevin,\fluffy):hasAnimal$, $F_3=(kevin,tom):hasAnimal$, $F_4=\fluffy: Cat$, $F_5=tom: Cat$ and $F_6= Cat\sqsubseteq Pet$. %\end{small}. The KB has four worlds and the query axiom $Q=kevin:NatureLover$ is true in one of them, the one corresponding to the selection $ \{(F_1,1),(F_2,1)\} $. The probability of the query is $P(Q)=0.5\cdot 0.6=0.3$. \end{small} \end{example} \begin{example} \label{people+pets_comb} \begin{small} Sometimes we have to combine knowledge from multiple, untrusted sources, each one with a different reliability. Consider a KB similar to the one of Example \ref{people+petsxy} but where we have a single cat, $\fluffy$. {\center $\exists hasAnimal.Pet \sqsubseteq NatureLover\ \ \ \ \ (kevin,\fluffy):hasAnimal\ \ \ \ \ Cat\sqsubseteq Pet$\\} \noindent and there are two sources of information with different reliability that provide the information that $\fluffy$ is a cat. On one source the user has a degree of belief of 0.4, i.e., he thinks it is correct with a 40\% probability, while on the other source he has a degree of belief 0.3. %, i.e. he thinks it is correct with a 30\% probability. The user can reason on this knowledge by adding the following statements to his KB: {\center$0.4\ \ ::\ \ \fluffy: Cat\ \ \ \ \ 0.3\ \ ::\ \ \fluffy: Cat$\\} The two statements represent independent evidence on $\fluffy$ being a cat. We associate $F_1$ ($F_2$) to the first (second) probabilistic axiom. The query axiom $Q=kevin:NatureLover$ is true in 3 out of the 4 worlds, those corresponding to the selections $ \{ \{(F_1,1),(F_2,1)\}, \{(F_1,1),(F_2,0)\}, \{(F_1,0),(F_2,1)\}\} $. So $P(Q)=0.4\cdot 0.3+0.4\cdot 0.7+ 0.6\cdot 0.3=0.58.$ This is reasonable if the two sources can be considered as independent. In fact, the probability comes from the disjunction of two independent Boolean random variables with probabilities respectively 0.4 and 0.3: $ P(Q) = P(X_1\vee X_2) = P(X_1)+P(X_2)-P(X_1\wedge X_2) = P(X_1)+P(X_2)-P(X_1)P(X_2) = 0.4+0.3-0.4\cdot 0.3=0.58 $ \end{small} \end{example} \section{Inference} \label{app:inf} Traditionally, a reasoning algorithm decides whether an axiom is entailed or not by a KB by refutation: the axiom $E$ is entailed if $\neg E$ has no model in the KB. Besides deciding whether an axiom is entailed by a KB, we want to find also explanations for the axiom, in order to compute the probability of the axiom. \subsection{Computing Queries Probability} The problem of finding explanations for a query has been investigated by various authors \cite{DBLP:conf/ijcai/SchlobachC03,DBLP:journals/ws/KalyanpurPSH05,DBLP:conf/semweb/KalyanpurPHS07,Kalyanpurphd,extended_tracing,Zese17-SSW-BK}. It was called \emph{axiom pinpointing} in \cite{DBLP:conf/ijcai/SchlobachC03} and considered as a non-standard reasoning service useful for tracing derivations and debugging ontologies. In particular, in \cite{DBLP:conf/ijcai/SchlobachC03} the authors define \emph{minimal axiom sets} (\emph{MinAs} for short). \begin{definition}[MinA] Let $\cK$ be a knowledge base and $Q$ an axiom that follows from it, i.e., $\cK \models Q$. We call a set $M\subseteq \cK$ a \emph{minimal axiom set} or \emph{MinA} for $Q$ in $\cK$ if $M \models Q$ and it is minimal w.r.t. set inclusion. \end{definition} \noindent The problem of enumerating all MinAs is called \textsc{min-a-enum}. \textsc{All-MinAs($Q,\cK$)} is the set of all MinAs for query $Q$ in knowledge base $\cK$. A \emph{tableau} is a graph where each node represents an individual $a$ and is labeled with the set of concepts $\cL(a)$ it belongs to. Each edge $\langle a, b\rangle$ in the graph is labeled with the set of roles to which the couple $(a, b)$ belongs. Then, a set of consistency preserving tableau expansion rules are repeatedly applied until a clash (i.e., a contradiction) is detected or a clash-free graph is found to which no more rules are applicable. A clash is for example a couple $(C, a)$ where $C$ and $\neg C$ are present in the label of a node, i.e. ${C, \neg C} \subseteq \cL(a)$. Some expansion rules are non-deterministic, i.e., they generate a finite set of tableaux. Thus the algorithm keeps a set of tableaux that is consistent if there is any tableau in it that is consistent, i.e., that is clash-free. Each time a clash is detected in a tableau $G$, the algorithm stops applying rules to $G$. Once every tableau in $T$ contains a clash or no more expansion rules can be applied to it, the algorithm terminates. If all the tableaux in the final set $T$ contain a clash, the algorithm returns unsatisfiable as no model can be found. Otherwise, any one clash-free completion graph in $T$ represents a possible model for the concept and the algorithm returns satisfiable. To compute the probability of a query, the explanations must be made mutually exclusive, so that the probability of each individual explanation is computed and summed with the others. To do that we assign independent Boolean random variables to the axioms contained in the explanations and defining the Disjunctive Normal Form (DNF) Boolean formula $f_K$ which models the set of explanations. Thus $ f_K(\mathbf{X})=\bigvee_{\kappa\in K}\bigwedge_{(E_i,1)}X_{i}\bigwedge_{(E_i,0)}\overline{X_{i}} $ where $\mathbf{X}=\{X_{i}|(E_i,k)\in\kappa,\kappa\in K\}$ is the set of Boolean random variables. We can now translate $f_K$ to a Binary Decision Diagram (BDD), from which we can compute the probability of the query with a dynamic programming algorithm that is linear in the size of the BDD. In \cite{DBLP:journals/jar/BaaderP10,DBLP:journals/logcom/BaaderP10} the authors consider the problem of finding a \emph{pinpointing formula} instead of \textsc{All-MinAs($Q,\cK$)}. The pinpointing formula is a monotone Boolean formula in which each Boolean variable corresponds to an axiom of the KB. This formula is built using the variables and the conjunction and disjunction connectives. It compactly encodes the set of all MinAs. Let's assume that each axiom $E$ of a KB $\cK$ is associated with a propositional variable, indicated with $var(E)$. The set of all propositional variables is indicated with $var(\cK )$. A valuation $\nu$ of a monotone Boolean formula is the set of propositional variables that are true. For a valuation $\nu \subseteq var(\cK)$, let $\cK_{\nu} := \{t \in \cK |var(t)\in\nu\}$. \begin{definition}[Pinpointing formula] Given a query $Q$ and a KB $\cK$, a monotone Boolean formula $\phi$ over $var(\cK)$ is called a \emph{pinpointing formula} for $Q$ if for every valuation $\nu \subseteq var(\cK)$ it holds that $\cK_{\nu} \models Q$ iff $\nu$ satisfies $\phi$. \end{definition} In Lemma 2.4 of \cite{DBLP:journals/logcom/BaaderP10} the authors proved that the set of all MinAs can be obtained by transforming the pinpointing formula into a Disjunctive Normal Form Boolean formula~(DNF) and removing disjuncts implying other disjuncts. Irrespective of which representation of the explanations we choose, a DNF or a general pinpointing formula, we can apply knowledge compilation and \textit{transform it into a Binary Decision Diagram (BDD)}, from which we can compute the probability of the query with a dynamic programming algorithm that is linear in the size of the BDD. We refer to \cite{Zese17-SSW-BK,ZesBelRig16-AMAI-IJ} for a detailed description of the two methods.