13:30-14:00 Semantic Alignment for Agent Interactions: making communicationmeaningful in open environments Paula Chocro, Artificial Intelligence Research Institute (IIIA-CSIC)
The fact that the meaning of words depends on thecontext in which they are used is evident for any speaker: ifsomeone asks for chips in a cafeteria, she will unlikely beexpecting to get electronic circuits. In human dialogues this kindof semantic alignment happens permanently and has been extensivelystudied. In this talk I will discuss how these ideas can also beapplied to help achieve meaningful communication in artificialmulti-agent systems, in which heterogeneous interlocutors willlikely use different vocabularies. I will start by presenting anotion of context that is based on the formal specifications ofthe tasks performed by agents. I will then show how this contextcan be used by the agents to align their vocabularies dynamically,by learning mappings from the experience of previousinteractions. In doing so, we will also rethink the traditionalapproach to semantic matching and its evaluation, tackling thefollowing questions: What does it mean for agents to “understandeach other”? When is an alignment good for a particularapplication? How can the interaction context helpinteroperability?
14:00-14:30 From Human Instructions to Executable Linked Data Paolo Pareti, University of Edinburgh, UK; Taiger, Spain
Linking the unstructured to the structured has alwaysbeen one of the fundamental problems of ArtificialIntelligence. While most humans can communicate in naturallanguage, only a few can communicate with machines using formallanguages, for example by programming. This talk describes anapproach to link informal instructions written by humans forhumans (such as those that can be found on websites like wikiHow)to machine understandable concepts and machine executablefunctionalities. The machine understanding of instructions canincrease gradually and opportunistically, generating robustsystems where human and machine reasoning can be interleaved. Iwill show potential uses of such knowledge that range fromimproved visualisation and exploration of instructions tomulti-agent collaboration between humans and machines.
14:30-15:00 Generating Linked Data from Unstructured Text Natthawut Kertkeidkachorn, SOKENDAI (The Graduate University for Advanced Studies)
Linked Data resources play an important role in variousapplications. Nevertheless, most of the publishing data isunstructured data, which is not feasible for constructing LinkedData. Recently, there are many studies proposed approaches totransform unstructured data to structured data in order toconstruct Linked Data resources. Still, two major issues, entitymapping and predicate mapping, are not solved yet. We thereforeproposed the framework for automatic generating Linked Data fromUnstructured Text. In the framework, two sub-frameworks, namelyHMILDs and T2KG, are proposed for dealing with entity mapping andpredicate mapping respectively.
15:30-16:00 Ontology for Knowledge Representation Lihua Zhao, Artificial Intelligence Research Center, National Institute ofAdvanced Industrial Science and Technology (AIST)
An ontology is a description of the concepts andrelationships that enables machines to understand informationabout the world. In this presentation, we will introduce adecision making system for intelligent vehicles that usedontologies to represent knowledge of driving environment andtraffic situations. We will also introduce our current research,which is to extend ontology-based knowledge graph using wordembedding approach.
16:00-16:30 Expressive Schema Languages for Probabilistic Knowledge Graphs Melisachew Wudage Chekol Data and Web Science Group, University of Mannheim, Germany
Recent advances in open information extraction havepaved the way for automatic construction of knowledge graphs(KGs). Often the extraction tools used to construct KGs producefacts and rules along with their confidence scores, leading to thenotion of uncertain KGs. The facts and rules contained in thesegraphs tend to be noisy and erroneous due to either the accuracyof the extraction tools or uncertainty in the source data. In thiswork, we use Markov logic networks to provide formal syntax andsemantics for uncertain KGs. Moreover, we propose a schemalanguage based on description logics, that extend the underlyingschema of the KGs and help in resolving conflictingfacts. Finally, we discuss experimental results.