PROCEEDINGS of the 14Thstudent SESSION of THE
Total Page:16
File Type:pdf, Size:1020Kb
PROCEEDINGS OF THE 14th STUDENT SESSION OF THE EUROPEAN SUMMER SCHOOL FOR LOGIC,LANGUAGE, AND INFORMATION JULY 20-31, BORDEAUX,FRANCE THOMAS ICARD (EDITOR) Preface The Student Session has been a part of the ESSLLI tradition every year since its inaugu- ral session in 1996 in Prague, making this the fourteenth. The quality of papers in these proceedings is a testament to the fact that the Student Session continues to be an excellent venue for students to present original work in quite diverse areas of logic, language, and computation. This year’s authors also comprise a very international group, with students from universities in the Netherlands, Germany, France, Spain, United Kingdom, United States, Japan, and India. I would first like to thanks each of the co-chairs, as well as our distinguished area ex- perts, for all of their help in the reviewing process and organizational matters in general. Thanks are also due to the long list of generous anonymous reviewers. Other individuals who have helped on numerous counts from FoLLI and ESSLLI Organizing Committee, and from previous Student Sessions, include Ville Nurmi, Kata Balogh, Christian Retor´e, Richard Moot, Uwe M¨onnich, and especially Paul Dekker and Sophia Katrenko. As in previous years, Springer-Verlag has graciously offered prizes for ‘Best Paper’ awards, and for this we are very grateful. Finally, thanks to all those who submitted papers, as they are what really make the Student Session such an exciting event. Thomas Icard Chair of the 2009 Student Session Stanford, June 2009 ii Program Committee Chair Thomas Icard (Stanford University) Area Co-Chairs Logic and Language Salvador Mascarenhas (New York University) Language and Computation Bruno Mery (University of Bordeaux) Logic and Computation Marija Slavkovik (University of Luxembourg) Area Experts Logic and Language Nathan Klinedinst (University College London) Reinhard Muskens (Tilburg University) Language and Computation Makoto Kanazawa (National Institute of Informatics, Japan) Jens Michaelis (University of Bielefeld) Logic and Computation Arnon Avron (Tel-Aviv University) Alexandru Baltag (Oxford University) iii Table of Contents Szymon Klarman DESCRIPTION LOGICS FOR RELATIVE TERMINOLOGIES Or: Why the Biggest City is not a Big Thing 1 Yavor Nenov ANOTE ON THE THEORY OF COMPLETE MEREOTOPOLOGIES 12 Amaldev Manuel LTL WITH A SUB-ORDER 21 Stefan Wintein ON LANGUAGES THAT CONTAIN THEIR OWN UNGROUNDEDNESS PREDICATE 29 Franc¸ois Bouchet CHARACERIZATION OF CONVERSATIONAL ACTIVITIES IN A CORPUS OF ASSISTANCE REQUESTS 40 Daniel Lassiter THE ALGEBRAIC STRUCTURE OF AMOUNTS:EVIDENCE FROM COMPARATIVES 51 Ekaterina Ovchinnikova FRAME-ANNOTATED CORPUS FOR EXTRACTIONS OF THE Argument-Predicate RELATIONS 62 iv Thomas Graf TOWARDS A FACTORIZATION OF STRING-BASED PHONOLOGY 72 Daphne Theijssen VARIABLE SELECTION IN LOGISTIC REGRESSION: THE BRITISH ENGLISH DATIVE ALTERNATION 85 Pere Pardo BASE BELIEF REVISION FOR FINITARY MONOTONIC LOGICS 96 Arno Bastenhof EXTRACTION IN THE LAMBEK-GRISHIN CALCULUS 106 Kostadin Cholakov TOWARDS MORPHOLOGICALLY ENHANCED AUTOMATED LEXICAL ACQUISITION 117 Xuchen Yao, Jianqiang Ma, Sergio Duarte, and C¸a˘gri C¸¨oltekin UNSUPERVISED SYNTAX LEARNING WITH CATEGORIAL GRAMMARS USING INFERENCE RULES 127 Sumiyo Nishiguchi EXTENDED QUALIA-BASED LEXICAL KNOWLEDGE FOR DISAMBIGUATION OF JAPANESE POSTPOSITION No 137 Simon Charlow CAN DP BE A SCOPE ISLAND? 148 Armin Buch MILDLY NON-PLANAR PROOF NETS FOR CCG 160 v Copyright !c , the authors. vi DESCRIPTION LOGICS FOR RELATIVE TERMINOLOGIES OR WHY THE BIGGEST CITY IS NOT A BIG THING Szymon Klarman Department of Computer Science, Vrije Universiteit Amsterdam Abstract. Context-sensitivity has been for long a subject of study in linguistics, logic and computer science. Recently the problem of representing and reasoning with con- textual knowledge has been brought up in the research on the Semantic Web. In this paper we introduce a conservative extension to Description Logic, the formalism underlying Web Ontology Languages, supporting representation of ontologies con- taining relative terms, such as ‘big’ or ‘tall’, whose meaning depends on the selection of comparison class (context). The solution rests on introduction of modal-like op- erators in the language and an additional modal dimension in the semantics, which is built upon the standard object dimension of the Description Logic languages and whose states correspond to selected subsets of the object domain. We present the syntax and semantics of the extended language and elaborate on its representational and computational features. 1. Introduction It is a commonplace observation that the same expressions might have different mean- ings when used in different contexts. A trivial example might be that of the concept The Biggest. Figure 1 presents three snapshots of the same knowledge base that focus on different parts of the domain. The extension of the concept visibly varies across the three takes. Intuitively, there seem to be no contradiction in the fact that individual Lon- don is an instance of The Biggest, when considered in the context of European cities, CITY EUROPEAN_CITY AUSTRALIAN_CITY !THE_BIGGEST !THE_BIGGEST !THE_BIGGEST ! Amsterdam ! London ! Amsterdam THE_BIGGEST THE_BIGGEST THE_BIGGEST ! Sydney ! New_York ! London ! Sydney Figure 1: Example of a context-sensitive concept The Biggest. an instance of The Biggest, when contrasted with all cities, and finally, not belonging ¬ to any of these when the focus is only on Australian cities. Natural language users re- solve such superficial incoherencies simply by recognizing that certain terms, call them relative, such as The Biggest, acquire definite meanings only when put in the context of specified comparison classes (Shapiro 2006, van Rooij to appear, Gaio 2008). The problem of context-sensitivity has been for a long time a subject of studies in lin- guistics, logic and even computer science. Recently, it has been also encountered in the research on the Semantic Web (Bouquet, et al. 2003, Caliusco, et al. 2005, Benslimane, et al. 2006) where the need for representing and reasoning with imperfect information becomes ever more pressing (Lukasiewicz & Straccia 2008, Laskey, et al. 2008). Relativ- ity of meaning appears as one of common types of such imperfection. Alas, Description 1 Logics (DLs), which form the foundation of the Web Ontology Language (Horrocks, et al. 2003), the basic knowledge representation formalism on the Semantic Web, were origi- nally developed for modeling crisp, static and unambiguous knowledge, and as such, are incapable of handling the task seamlessly. Consequently, it has become clear that it is necessary to look for more expressive, ideally backward compatible languages to meet the new application requirements on the Semantic Web. Current proposals focus mostly on the problems of uncertainty and vagueness (Lukasiewicz & Straccia 2008, Straccia 2005), with several preliminary attempts of dealing with different aspects of contextu- alization of DL knowledge bases (Grossi 2007, Goczyla, et al. 2007, Benslimane et al. 2006). In this paper we propose a simple, conservative extension to the classical DLs, which is intended for representation of relative, context-sensitive terminologies, where by contexts we understand specifically the comparison classes with respect to which the terms acquire precise meanings. To take a closer look at the problem consider again the scenario from Figure 1. On a quick analysis it should become apparent there is no straightforward way of modeling the scenario within the standard DL paradigm. Asserting both London : The Biggest and London : The Biggest in the same knowledge base results in an immediate contradic- ¬ tion, which is obviously an unintended outcome. To avoid this consequence one can resort to the luring prospect of indexing, and instead assert London : The BiggestEuropean City and London : The Biggest , with an implicit message that the two indexed concepts ¬ City are meant to be two different ‘variants’ of The Biggest, corresponding to two possible contexts of its use. The contradiction is indeed avoided, but unfortunately the baby has been thrown out with the bath water, for the two ‘variants’ become in fact two unrelated concept names, with no common syntactic or semantic core. More precisely, using this strategy one cannot impose global constraints on the contextualized concepts, for instance, to declare that regardless of the context, The Biggest is always a subclass of Big. Even if this goal was achieved by rewriting constraints over all individual contexts, another source of problems is reasoning about the contexts themselves, for example, deciding whether an individual occurs in a given comparison class or not. The extension proposed in this paper is, to our knowledge, unique in addressing this particular type of context-sensitivity in DL, and arguably, it cannot be simulated within any of the approaches present in the literature. Technically, the solution rests on the pres- ence of special modal-like operators in the language and a second modal dimension in the semantics of the language, which is defined over the standard object dimension of DL and whose states correspond to selected subsets of the object domain. In the following section we formally define the language, next we elaborate on some of its basic representational and computational features, and finally, in the last two sections, we shortly position our work in a broader perspective and conclude the presentation. 2. Representation