NLULP on Sunday

Detailed program
Sunday July 28th, 2002
See also the unified by-slot program

Sessions take place in auditorium 10 unless otherwise indicated.

09:00-09:30  Welcome

Chair: Shuly Wintner

Session 1: Invited talk

Chair: Shuly Wintner

09:30-10:30  Johan Bos, U Edinburgh, UK
Invited talk: Generating speech recognition grammars with compositional semantics from unification grammars
In this talk I will introduce a method to compile unification grammars into speech recognition packages. I will focus on how to transfer semantic operations (such as functional application), and in particular discuss the difficulties that arise for left-recursive productions. Used in practical applications, the resulting speech grammars will associate general domain independent meaning representations with recognised strings, making a subsequent traditional parsing processing step redundant. I will conclude with presenting some promising practical results.

Session 2: Formalisms

11:00-11:30  Mike Daniels, Ohio State U, USA
Detmar Meurers, Ohio State U, USA
Improving the efficiency of parsing with discontinuous constituents
We discuss a a generalization of Earley's algorithm to grammars licensing discontinuous constituents of the kind proposed by the so-called linearization approaches in Head-Driven Phrase Structure Grammar. We show that one can replace the standard indexing on the string position by bitmasks that act as constraints over possible coverage bitvectors. This improves efficiency of edge access and reduces the number of edges by constraining prediction to those grammar rules which are compatible with known linearization properties. The resulting parsing algorithm does not have to process the right-hand side categories in the order in which they cover the string, and so one can obtain a head-driven strategy simply by reordering the right-hand side categories of the rules. The resulting strategy generalizes head-driven parsing in that it also permits the ordering of non-head categories.

11:30-12:00  Katrin Erk, Saarland U, Germany
Geert-Jan M. Kruijff, Saarland U, Germany
A constraint-programming approach to parsing with resource-sensitive categorial grammar
Parsing with resource-sensitive categorial grammars is an NP-complete problem. The traditional approach to parsing with such grammars is based on generate and test and cannot avoid this high worst-case complexity. This paper proposes an alternative approach, based on constraint programming: Given a grammar, constraints formulated on an abstract interpretation of the grammar's logical structure are used to prune the search space during parsing. The approach is provably sound and complete. Calculations of its complexity show significant potential improvements on efficiency.

12:00-12:30  Chris Fox, U Essex, UK
Shalom Lappin, King's College London, UK
Carl Pollard, Ohio State U, USA
First-order, Curry-typed logic for natural language semantics
The paper presents Property Theory with Curry Typing (PTCT) where the language of terms and well-formed formulae are joined by a language of types. In addition to supporting fine-grained intensionality, the basic theory is essentially first-order, so that implementations using the theory can apply standard first-order theorem proving techniques. Some extensions to the type theory are discussed, including the possibility of adding type polymorphism

Session 3: Semantics

14:00-14:30  Barbara Gawronska, U Skövde, Sweden
Employing cognitive notions in multilingual summarization of news reports
The paper presents an approach to automatic text understanding inspired by speech act theory and cognitive semantics, especially by the notion of `mental spaces' (Fauconnier 1985), and by Pustejovsky's (1991a, 1991b, 1995) notion of `qualia' and his definition of formal and telic hyponymy. This approach is employed in an experimental system for understanding of news reports and multilingual generation of news summaries. The system, implemented in Prolog and Delphi, aims at analyses of English news reports in the domain of the world's news (military conflicts, terrorists attacks, natural disasters) and generation of summaries in Swedish, Danish, and Polish. The paper focuses on the understanding component and on the possibility of using WordNet as the main lexical knowledge resource for English. An appropriate semantic analysis and a successful summarization of English input texts require some modifications of the hyper-/hyponymy and holo-/meronymy relations that are encoded in WordNet. A combination of a cognitive analysis of certain lexical and phrasal categories (speech act phrases, epistemic phrase, prepositions) with qualia-based re-formulations of WordNet hierarchies is proposed and tested.

14:30-15:00  Marilisa Amoia, Saarland U, Germany
Claire Gardent, LORIA Nancy, France
Stephan Thater, Saarland U, Germany
Using set constraints to generate distinguishing descriptions
Algorithms such as (van Deemter 2000) which generate distinguishing descriptions for sets of individuals using positive, negative and disjunctive properties, do not always generate a minimal description. In this paper, we show that such an approach is cognitively inappropriate in that the descriptions produced might be unnecessary long and ambiguous and/or epistemically redundant. We then present an alternative, constraint-based algorithm which does produce minimal descriptions and compare its performance with the incremental algorithm.

15:00-15:30  Balder ten Cate, U Amsterdam, The Netherlands
Chung-chieh Shan, Harvard U, USA
Question answering: From partitions to Prolog
We implement Groenendijk and Stokhof's partition semantics of questions in a simple question answering algorithm. The algorithm is sound, complete, and based on tableau theorem proving. The algorithm relies on a syntactic characterization of answerhood: Any answer to a question is equivalent to some formula built up only from instances of the question. We prove this characterization by translating the logic of interrogation to classical predicate logic and applying Craig's interpolation theorem.

Session 4: Interpretation as deduction

16:00-16:30  Henning Christiansen, U Roskilde, Denmark
Abductive language interpretation as bottom-up deduction
A translation of abductive language interpretation problems into a deductive form is proposed and shown to be correct. No meta-level overhead is involved in the resulting formulas that can be evaluated by bottom-up deduction, e.g., by considering them as Constraint Handling Rules. The problem statement may involve background theories with integrity constraints, and minimal contexts are produced that can explain a discourse given.

16:30-17:30  Session 5: Panel: the future of NLP and LP

Chair: Shuly Wintner

20:00-22:00  NLULP Dinner

Room: Restaurant TBA