IWCS-4 Conference Report
From SIGSEM
A conference report by Kristina Striegnitz
Contents |
Do you know what this is?
Is it:
(a) the map of an 18th century park?
or
(b) the logo of a conference on computational semantics?
You answered a)? - You are probably from Tilburg or fanatic about French style parks. Read on, if you want to discover new uses for old park maps. If your answer was b), you really know what's going on in computational semantics. Read on to find out more about this year's IWCS, the fourth International Workshop on Computational Semantics, which took place from January 9 to 12 at Tilburg University. If you answered a) and b), you probably attended IWCS-4. But maybe you would like to refresh your memories?
IWCS is, as Harry Bunt, the IWCS-4 chairman, writes in the preface to the proceedings, "the conference where researchers involved in any aspect of the computation of meaning in natural language meet and discuss the development in this exciting area of research". And that computational semantics indeed is an exciting area of research was evident in Tilburg. The field was presented by a wide variety of high quality talks and poster presentations. Three invited speakers gave riveting talks. And many stimulating discussions developed after the talks as well as during the coffee and lunch breaks.
All this was possible in a welcoming environment created by Carol McGregor, Elias Thijsse, Ielka van der Sluis (who is also the one responsible for the logo), Huub Prust, and Reinhard Muskens. Many thanks to them and everybody else involved in the perfect organization of IWCS-4.
If you want to know more about IWCS-4, if you are curious to find out more about the surprise that was scheduled for Thursday afternoon, read on.
Wednesday
Each day started with an invited lecture. On Wednesday, James Allen talked about grammars for dialog systems. For efficiency and accuracy reasons, such grammars have to contain domain specific semantic restrictions which usually makes it impossible to reuse them for other applications. As a solution to this problem, James Allen proposed to generate application specific grammars from a generic base grammar and a domain specific ontology.
The following talk by Shalom Lappin was also concerned with the special needs a dialog setting imposes on the grammar. He presented an HPSG based treatment of fragmentary utterances.
Then, Ivana Kruijff-Korbayovà presented an analysis of discourse connectives that accounted for the influence of information structure on their interpretation. She illustrated this approach by giving a detailed analysis of the discourse connective "although".
Rodger Kibble proposed a mechanism for reflecting discourse structure in the way updating of the context is done in dynamic frameworks such as update semantics. He argued that a variety of phenomena that usually are problematic for dynamic semantics can be treated with this approach.
The morning session ended with short introductions to the posters that could be looked at and discussed during lunch. Elena Karagjosova investigated the effect of modal particles in German as devices for expressing beliefs and intentions. Isabelle Debourges proposed a new method for exploring large corpora. Based on a user query, information about the relevant concepts is extracted from the corpus and presented to the user as a semantic network, which can then be refined further. Jan van Kuppevelt presented a new approach for determining discourse structure by building on the topic-comment structure. And Birte Loenneker addressed the problem of extracting the background knowledge relevant in a given situation from a frame based representation of world knowledge.
The afternoon session started with a talk by Emiel Krahmer, in which he reported on perception and production experiments providing evidence for the existence of contrastive accent and metalinguistic negation. He argued that such experiments are of general use for examining the relation between meaning and intonation.
In the next talk, Alastair Butler discussed the context dependence of exhaustification and proposed a treatment that correctly derives maximality, minimality or uniqueness.
Then, Robert van Rooy addressed the problem of determining whether an answer to a certain question should be exhaustive or not. In his account questions are related to decision problems, such that the quality of an answer depends on its utility to the questioner.
After a short coffee break, Nicolas Denand presented the results of an experiment testing the answers to 'Where' questions, based on which he identified a number of criteria influencing the choice of reference object.
Patrick St. Dizier then presented a compositional framework for representing the semantics of prepositions which captures the interaction of prepositions with their arguments and the modified constituent.
The final talk on this first workshop day was given by Andrea Zielinski, who argued that quantifier scoping should not be left underspecified in machine translation. She described a system which determines the preferred scoping on the basis of heuristics and then does transfer on these structures.
The official program of this first day ended with a reception featuring not only drinks but also lekkere hapjes and providing an opportunity for informal meeting and talking.
Thursday
The program on Thursday morning started again with an invited talk. Alex Lascarides argued that discourse structure plays an important role in the way natural language utterances are interpreted. She showed how this influence of discourse structure on the interpretation process can be account for in a dynamic semantic formalism.
Then, Michel Gagnon presented a DRT treatment of temporal anaphora, which he applied to the intricate Portuguese progressive.
The next talk, by Helen Seville, was concerned with the distinction between sense and reference of referring expressions. She described an approach in which the dependency of sense on context is captured by the different discourse models that a referring expression may give rise to in different contexts.
After the coffee break Massimo Poesio presented the results of a corpus study and a psycholinguistic experiment investigating the conditions under which pronouns may be used in contexts in which they cannot be unambiguously resolved.
In the following talk, by Diego Mollá, a more application oriented view on computational semantics was presented. Diego used a flat logical representation for answer extraction.
Then, there were again four short presentations before lunch. Gregers Koch described an approach to inducing a syntax-semantics interface for a given grammar from example pairs consisting of an input string and a target logical form. Allan Ramsay investigated the types of discourse relations introduced by discourse connectives and proposed a treatment to account for the different consequences that follow from different uses of a discourse connective. Based on a form of psychotherapy that focusses on the structure of the language used by patients, Jon Patrick applied methods of computational semantics to analyze changes in the language use of patients during a therapy. Lena Sokolova discussed factors responsible for different realizations of one semantic concept in different languages.
After the lunch break, Dick Crouch presented an approach to transfer in machine translation on the basis of glue language meaning constructors. He showed how to deal with cases of structural misalignment, such as head switching, which are problematic for previous approaches to transfer in this formalism.
He was followed by Theo Janssen, who proposed a new interpretation of Hintikka's I(denpendence)F(riendly) logic, which gives consideration to computational aspects and does away with some other problems inherent to previous designs of If logic semantics.
The next item on the program: SURPRISE. The only hint that had leaked out until then was that it was better to bring an umbrella. However, at least this time, the Dutch unjustly hadn't trust their weather, which was cold but beautifully sunny. We had a refreshing walk through the <a href="iwcs4map.html">"Oude Warande"</a>, an 18th century park situated right next to the conference venue. Park designer Jan van Summeren guided the tour telling us about the long history of the <a href="iwcs4logo.html">IWCS-4 logo</a>.
After having warmed up with some coffee and tea, everybody gathered again for the last two talks in this afternoon. They were both on applications of type theory. René Ahn used type theory to model an agent's knowledge about the world by formalizing the way the agent observes and classifies it.
In the second talk, Pascal Boldini analyzed mass terms in a type theoretic framework and showed how various cases of reference of mass terms can be dealt with.
Following the SIGSEM business meeting, a bus took us to the restaurant, where dinner in form of an enormous buffet was being prepared.
Friday
The last invited talk on Friday morning was given by Jan van Eijck, who proposed a type theoretic formulation of context semantics suited for representing natural language semantics. He showed how a notion of salience can be encoded in this framework supporting a treatment of pronoun resolution.
Then there were two talks concerned with the generation of referring expressions. In the first one, Kees van Deemter explored the limits of Dale and Reiter's incremental algorithm and proposed techniques for exceeding them.
Then Kristina Striegnitz presented a treatment of bridging anaphora in generation. She concentrated on the contextual inferences necessary for appropriately generating expressions referring to enitities which have not been explicitly mentioned in the discourse before.
The next two talks were concerned with the the treatment of tense. Frank van Eynde compared a rule-based and a constraint-based approach to modelling the syntax-semantics interface by comparing how they deal with a treatment of Dutch tense. He argued that the constraint-based approach is better suited.
Sven Verdoolaege then presented an implementation of this treatment of Dutch tense and showed how abductive reasoning can deal with the linguistic knowledge, world knowledge and default assumptions influencing the interpretation of tense.
In the very last talk, George Demetriou inquired into the question what a tag set for semantic tagging should look like and argued that semantic tags should not be atomic but a set of features. He then showed how such a tagger can be derived from a lexicon.
>> back to Conference Reports
