Roberto Micalizio

AI*IA 2013: Advances in Artificial Intelligence

ion in Markov Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145 Lorenza Saitta Improving the Structuring Capabilities of Statistics–Based Local Learners . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 Slobodan Vukanović, Robert Haschke, and Helge Ritter Kernel-Based Discriminative Re-ranking for Spoken Command Understanding in HRI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 Roberto Basili, Emanuele Bastianelli, Giuseppe Castellucci, Daniele Nardi, and Vittorio Perera Natural Language Processing A Natural Language Account for Argumentation Schemes . . . . . . . . . . . . . 181 Elena Cabrio, Sara Tonelli, and Serena Villata Deep Natural Language Processing for Italian Sign Language Translation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193 Alessandro Mazzei, Leonardo Lesmo, Cristina Battaglino, Mara Vendrame, and Monica Bucciarelli A Virtual Player for “Who Wants to Be a Millionaire?” based on Question Answering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 Piero Molino, Pierpaolo Basile, Ciro Santoro, Pasquale Lops, Marco de Gemmis, and Giovanni Semeraro The Construction of the Relative Distance Fuzzy Values Based on the Questionnaire Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217 Jedrzej Osiński Process Fragment Recognition in Clinical Documents . . . . . . . . . . . . . . . . . 227 Camilo Thorne, Elena Cardillo, Claudio Eccher, Marco Montali, and Diego Calvanese

Monitoring the Execution of a Multi-Agent Plan: Dealing with Partial Observability

The paper addresses the task of monitoring and diagnosing the execution of a multi-agent plan (MAP) which involves actions concurrently executed by a team of cooperating agents. The paper describes a weak commitment strategy to deal with cases where observability is only partial and it is not sufficient for inferring the outcome of all the actions executed so far. The paper discusses the role of target actions in providing sufficient conditions for inferring the pending outcomes in a finite time window. The action outcome provides the basis for computing plan diagnosis and for singling out the goals which will not be achieved because of an action failure.

On-line monitoring of plan execution: A distributed approach

The paper introduces and formalizes a distributed approach for the model-based monitoring of the execution of a plan, where concurrent actions are carried on by a team of mobile robots in a partially observable environment. Each robot is monitored on-line by an agent that has the task of tracking all the possible evolutions both under nominal and faulty behavior of the robot and to estimate the belief state at each time instant. The strategy for deriving local solutions which are globally consistent is formalized. The distributed monitoring provides on-line feedback to a system supervisor which has to decide whether building a new plan as a consequence of actions failure. The feasibility of the approach and the gain in the performance are shown by comparing experimental results of the proposed approach with a centralized one.

Cooperative monitoring to diagnose multiagent plans

Diagnosing the execution of a Multiagent Plan (MAP) means identifying and explaining action failures (i.e., actions that did not reach their expected effects). Current approaches to MAP diagnosis are substantially centralized, and assume that action failures are independent of each other. In this paper, the diagnosis of MAPs, executed in a dynamic and partially observable environment, is addressed in a fully distributed and asynchronous way; in addition, action failures are no longer assumed as independent of each other. The paper presents a novel methodology, named Cooperative Weak-Committed Monitoring (CWCM), enabling agents to cooperate while monitoring their own actions. Cooperation helps the agents to cope with very scarcely observable environments: what an agent cannot observe directly can be acquired from other agents. CWCM exploits nondeterministic action models to carry out two main tasks: detecting action failures and building trajectory-sets (i.e., structures representing the knowledge an agent has about the environment in the recent past). Relying on trajectory-sets, each agent is able to explain its own action failures in terms of exogenous events that have occurred during the execution of the actions themselves. To cope with dependent failures, CWCM is coupled with a diagnostic engine that distinguishes between primary and secondary action failures. An experimental analysis demonstrates that the CWCM methodology, together with the proposed diagnostic inferences, are effective in identifying and explaining action failures even in scenarios where the system observability is significantly reduced.

ACTION FAILURE RECOVERY VIA MODEL-BASED DIAGNOSIS AND CONFORMANT PLANNING

A plan carried on in the real world may be affected by a number of unexpected events, plan threats, which cause significant deviations between the intended behavior of the plan executor (i.e., the agent) and the observed one. These deviations are typically considered as action failures. This paper addresses the problem of recovering from action failures caused by a specific class of plan threats: faults in the functionalities of the agent. The problem is approached by exploiting techniques of the Model‐Based Diagnosis (MBD) for detecting failures (plan execution monitoring) and for explaining these failures in terms of faulty functionalities (agent diagnosis). The recovery process is modeled as a replanning problem aimed at fixing the faulty components identified by the agent diagnosis. However, since the diagnosis is in general ambiguous (a failure may be explained by alternative faults), the recovery has to deal with such an uncertainty. The paper advocates the adoption of a conformant planner, which guarantees that the recovery plan, if it exists, is executable no matter what the actual cause of the failure. The paper focuses on a single agent performing its own plan, however the proposed methodology takes also into account that agents are typically situated into a multiagent scenario and that commitments between agents may exist. The repair strategy is therefore conceived to overcome the causes of a failure while assuring the commitments an agent has agreed with other team members.

Exploiting Social Commitments in Programming Agent Interaction

Modeling and regulating interactions among agents is a critical step in the development of Multiagent Systems (MASs). Some recent works assume a normative view, and suggest to model interaction protocols in terms of obligations. In this paper we propose to model interaction protocols in terms of goals and commitments, and show how such a formalization promotes a deliberative process inside the agents. The proposal is implemented via JaCaMo+, an extension of JaCaMo, in which Jason agents can interact, while preserving their deliberative capabilities, by exploiting commitment-based protocols, reified by special CArtAgO artifacts.

Empowering Agent Coordination with Social Engagement

Agent coordination based on Activity Theory postulates that agents control their own behavior from the outside by using and creating artifacts through which they interact. Based on this conception, we envisage social engagements as first-class resources that agents exploit in their deliberative cycle (as well as beliefs, goals, intentions), and propose to realize them as artifacts that agents create and manipulate along the interaction, and that drive the interaction itself. Consequently, agents will base their reasoning on their social engagement, instead of relying on event occurrence alone. Placing social engagement at the center of coordination promotes agent decoupling and also the decoupling of the agent specifications from the specification of their coordination. The paper also discusses JaCaMo+, a framework that implements this proposal.

Agent cooperation for monitoring and diagnosing a MAP

The paper addresses the tasks of monitoring and diagnosing the execution of a Multi-Agent Plan, taking into account a very challenging scenario where the degree of system observability may be so low that an agent may not have enough information for univocally determining the outcome of the actions it executes (i.e., pending outcomes). The paper discusses how the ambiguous results of the monitoring step (i.e., trajectory-set) are refined by exploiting the exchange of local interpretations between agents, whose actions are bounded by causal dependencies. The refinement of the trajectory-set becomes an essential step to disambiguate pending outcomes and to explain action failures.

Plan Diagnosis and Agent Diagnosis in Multi-agent Systems

The paper discusses a distributed approach for monitoring and diagnosing the execution of a plan where concurrent actions are performed by a team of cooperating agents. The paper extends the notion of plan diagnosis(introduced by Roos et al. for the execution of a multi-agent plan) with the notion of agent diagnosis. While plan diagnosis is able to capture the distinction between primary and secondary failures, the agent diagnosis makes apparent the actual health status of the agents. The paper presents a mechanism of failure propagation which captures the interplay between agent diagnosis and plan diagnosis; this mechanism plays a critical role in the understanding at what extent a fault affecting the functionalities of an agent affects the global plan too. A relational formalism is adopted for modeling both the nominal and the abnormal execution of the actions.

Intelligent Supervision for Robust Plan Execution

The paper addresses the problem of supervising the execution of a plan with durative actions in a just partially known world, where discrepancies between the expected conditions and the ones actually found may arise. The paper advocates a control architecture which exploits additional knowledge to prevent (when possible) action failures by changing the execution modality of actions while these are still in progress. Preliminary experimental results, obtained in a simulated space exploration scenario, are reported.