Referring Expressions in Artificial Intelligence and Knowledge Representation Systems
The tutorial introduces the audience to the concept of referring expressions, formulae that can be used to communicate identities of otherwise abstract objects. The formalism provides foundations for a successful and unambiguous exchange of information about individuals between agents sharing common knowledge about such individuals, a task that is indispensable in most modern applications of knowledge representation and semantic technologies.
Location and Time: | Monday August 1, 9:00 at KR/FLoC 2022, Haifa. (FLoC program) |
Tutorial Slides: | refexp-tutorial.pdf | (to be updated)
Tutorial Synopsis
A referring expression in linguistics is any noun phrase identifying an object in a way that will be useful to interlocutors. In the context of knowledge representation and information systems, constant symbols occurring in an underlying knowledge base are the artifacts usually used to identify a subset of the objects for which the knowledge base captures knowledge.
This tutorial explores how objects that can be usefully identified can be extended by allowing more general expressions in the underlying language of the knowledge base, called singular referring expressions, to replace constants as syntactic identifiers of such objects.
Expanding the possibilities of identifying (possibly implicitly defined) objects serves numerous purposes, ranging from allowing query answers to contain additional tuples (which are typically eliminated due to lack of constant symbols denoting components of such tuples), to answers that are more informative, to decisions on how to communicate references to objects among various cooperating agents, and to identification issues related to physical data representation in computer storage (such as relying on addresses in main-memory databases).
The idea of referring expressions itself circumvents the need for artificially-invented identifiers that are commonly opaque to the user that interacts with the knowledge base.
Goals of the Tutorial
- to review the history of approaches to identifying individuals/objects, in particular in the area at the intersection between logic-based approaches to knowledge representation and the areas of natural languages and philosophy, approaches that can be traced back to Russell, Frege, and Strawson, and
- to present a modern unifying approach to referring expressions based on logical methods in AI and Knowledge Representation with clear and direct applications to many practical areas of AI and Computer Science in general.
 
Detailed Outline
- What is a referring expression? We start with an introduction
and overview of how well formed formulae that satisfy certain
properties can serve as referring expressions in
information systems whose underlying ontologies correspond to
first order knowledge bases.
- Background. We introduce formal properties of referring expressions and
show how they can be determined. We discuss how referring expressions can
be computed, in particular when the knowledge base conforms to a decidable
fragment of first order logic.
We also review past work on determining referring expressions in the
context of knowledge bases and position these approaches among other
approaches designed to indirectly and/or symbolically capture
identities of (sets of) objects.
- Referring expressions, types, and query answering. We show how
referring expressions can be used to enrich query answers over
knowledge bases by allowing to refer to answer objects that may not
have an explicit name within the knowledge base, or for which a more
preferred way of communicating its identity is available. To control the
form of the answers, we define a type language that describes
varieties of referring expressions desired in query answers.
- Referring expressions and types in conceptual modelling.
We then we explore the benefits of adopting referring expression
types for use in information systems derived from conceptual
modelling. In particular, we show how this approach can separate
purely conceptual ontology design from issues connected with how
objects are identified within an eventual information system based
on the design.
- Open problems. We conclude the tutorial with an outline of
directions for further research, and with a list of open issues
related to the use of referring expressions in ontology-based
information systems.
Audience and Background
The topics covered in the tutorial are of interest to wide range of AI researchers and to members of the general public with an interest in knowledge representation. In particular, the tutorial targets the following groups:
- Undergraduate and graduate students and junior researchers: the tutorial introduces this group to state-of-the-art approaches to addressing identification issues in knowledge bases and to modern techniques that address these issues;
- Researchers in the area of knowledge representation and other areas of AI: the tutorial provides bridges to many areas of AI, ranging from natural language issues, where the idea of referring expressions has originated, to philosophical underpinnings of object identification to unambiguous communication of information between agents, and to identity management in information and semantic WEB systems;
- Industry practitioners and developers: the tutorial provides ideas how development of software systems, in particular in the critical phase of conceptual modelling, can be improved and what tools are available to aid this goal;
- Members of the general public, with an interest in logical underpinnings of logic-based information management and in technologies based on these ideas.
Relevance to KR/FLoc 2022
The tutorial focuses on foundational issues relating to object identification in knowledge bases, ontologies, and information systems based on ontologies, and on how such issues can be comprehensively addressed. Since every design of an information system faces decisions relating to how external entities will be identified within such a system (in addition to representing various properties of such entities), a general approach to this problem is of interest to ontology developers/engineers and data scientists. Interestingly, the approach to object identification discussed in the tutorial naturally and seamlessly complements standard approaches in conceptual and ontology design methodologies. The tutorial is thus of interest both to researchers in knowledge representation and to practitioners in the wide area of information management.
 
About the Authors
Dr. David Toman and Dr. Grant Weddell are professors of Computer Science at the University of Waterloo, Canada. They have published and presented results in the area of knowledge representation over the last 20 years at premier AI conferences (including a Reiter Prize at KR 2010 and Best Paper Prize at ISWC 2013); Dr. Toman has also given tutorials in the area of temporal representation and reasoning and temporal databases and information systems that has led to an invited chapter in the Handbook of Temporal Reasoning in Artificial Intelligence.
Presenters' Background in the Area of the Tutorial
The authors, together with Alexander Borgida (Rutgers), have introduced referring expressions in the area of Knowledge Representation and Ontology-based Data Access and were awarded the Ray Reiter Best Paper prize at KR 2016 for this work. They have extended this work to the area of conceptual modelling and other areas connected with ontological reasoning and knowledge representation. Subsequently, with their coauthors, they were awarded the 2018 Bob Wielinga Best Paper Award for the paper furthering the use of referring expressions in conceptual modelling.
The authors have recently presented tutorials on the topic of referring expressions in knowledge representation and information systems:
- Referring Expressions in Ontologies and Query Answering at the 10th International Conference on Formal Ontology in Information Systems, FOIS 2018 (in Cape Town, South Africa, September 2018), and
- Managing and Communicating Object Identities in Knowledge Representation and Information Systems at the 31st Australasian Joint Conference on Artificial Intelligence AI 2018 (in Wellington, New Zealand, December 2018).
Contact
Name: | David Toman and Grant Weddell |
Affiliation: | Cheriton School of Computer Science, University of Waterloo |
Address: | 200 University Ave W., Waterloo, ON N2L3G1, Canada |
E-mail: | {david,gweddell}@uwaterloo.ca |
WWW: | cs.uwaterloo.ca/~{david,gweddell} |