2nd CoLogNET-ElsNET Symposium - Questions and Answers: Theoretical and Applied Perspectives

Logic and Natural Language Processing
Valentin Jijkoun
LIT, University of Amsterdam

Introduction
The symposium, organized by the Logic and Natural Language Processing Area of the European Network of Excellence in Computational Logic and Network of Excellence in Human Language Technology, was a very successful meeting bringing together researchers from very different, at the first glance, areas but all interested in the same ubiquitous and challenging phenomenon: certain strings of tokens being answers to other strings of tokens called “questions”.

The program of the one-day symposium was as dense (3 invited talks, 7 paper presentations and 4 presentations at the poster session) as it was diverse. The topics discussed ranged from architectures of Question Answering (QA) systems, the role of syntactic resources and semantic ontologies in QA, to possible definitions of the logical concept of answerhood. It was interesting and informative to see how similar problems and tasks are addressed in different contexts and similar research questions are answered from very different and often complementary perspectives.

In the invited talk, Karen Spark Jones gave a broad historic overview of the Question Answering task, which since its emergence more than 40 years ago has been considered a good application for investigating computational theories of language processing and meaning representation.
While being tractable (as the impressive performance of state-of-the-art QA systems shows) without being trivial, the QA task raises many important issues: what an appropriate response of the system should look like or whether it is always rational and natural from the user’s point of view to look for single, correct and precise answer.

The problem of defining the notion of a “good” answer re-appeared in the invited talk by Jeroen Groenendijk. He described how interpretation of questions as partitions of a logical space of possibilities gives rise to different notions of entailment and answerhood, which have syntactic characterizations and are computationally attractive.

Apart from relations between questions and answers, the nature of relations inside the set of possible answers to the same question is another interesting and important research topic.

As Tiphaine Dalmas argued in her talk, careful handling of different types of these relations (two answers can be different correct answers, or re-formulations of each other, or one refinement of the other) can be very useful in the open-domain QA.

Domain-specific (e.g. medical) Question Answering has always attracted many researchers, due to availability of high-quality specialized resources and significant interest in these systems.

In his invited talk, Werner Ceusters presented strategies as well as open issues in using semantic ontologies in restricted domains. He showed that although the very definition of what “ontology” is, might need be revised to include individuals in addition to classes, in order for it to be useful in real information access systems, sophisticated manually created and semi-automatically updated ontologies are indispensable in domain-specific applications.

I’d like to thank the organizers, the program committee and, of course, the speakers for this inspiring and fruitful meeting.