IWCS-9 Conference Report
A conference report by Elena Cabrio
What's new? IWCS in Oxford!
"Hibernation is certainly not one of the activities of computational semanticists. In the midst of winter they defy sub-zero temperatures, vicious snowstorms, and ice-covered roads to attend the mother of all computational semantics conferences: IWCS". This is the adventurous scenario as described by Johan Bos (University of Groningen) and Stephen Pulman (Oxford University), directors of the ninth episode of IWCS, that took place from January 12th to 14th 2011 in Oxford (UK). For the first time in its history, IWCS did not take place in the Netherlands, and was not organized by Harry Bunt, who initiated it in December 1994. Yet many characteristics and the spirit of the conference remained as they were: researchers from all over the world coming to enjoy an exciting selection of state-of-the-art work in computational semantics!
Wednesday 12th January
After a sandwich lunch to meet the participants, the organizers of the conference gave the go-ahead to the presentation of the research work selected by the program committee (30 long and 20 short papers). Laure Vieu, joint work with Nervo Verdezoto, described a semi-automatic method to improve WordNet, cleaning up the top level of its taxonomy. Then, Eduardo Blanco (joint work with Dan Moldovan) presented a model to compose semantic relations, based on an extended definition of semantic relations, including restrictions on domains, ranges and values for a set of semantic primitives.
Jerry Hobbs (joint work with James Blithe, Pedro Domingos, Rohit J. Kate and Raymond J. Mooney) proposed an approach that implements weighted abduction in Markov logic, which uses weighted first-order formulas to represent probabilistic knowledge. Martha Palmer presented then the last paper of this first session, written in collaboration with Susan W. Brown and Dmitriy Dligach. It describes a VerbNet classifier, that exploits rich syntactic and semantic features to label verb instances with their appropriate VerbNet class.
After the coffee break, Ekaterina Ovchinnikova (joint work with Niloofar Montazeri, Theodore Alexandrov, Jerry Hobbs, Michael McCord and Rutu Mulkar-Mehta), described a discourse processing framework based on weighted abduction. Andrey Gusev, presented an approach to learn the durations of events without annotated training data, employing web query patterns to infer duration distributions. In the final session of the day, Justin Washtell introduced a distributional model of compositional semantics, based on the simple addition of expectation vectors.
The day was concluded with a reception in the fascinating atmosphere of the Natural History Museum of Oxford. While drinks and food were served, it was possible to admire the beauty of collections of zoology, entomology, geology, and mineralogy, under the eyes of a 40ft Tyrannosaurus rex!
Thursday 13th January
In the morning session, Blake Stephen Howald (co-author: E. Graham Katz) presented the results of several machine learning tasks that exploit spatial information in narrative discourse. Then, Chee Wee Leong described an evaluation framework to quantify cross-modal semantic relationships existing between arbitrary pairs of words and images (joint work with Rada Mihalcea). Before the coffee break, Jerry Hobbs (joint work with Niloofar Montazeri) proposed a methodology to construct axioms defining event-related words, basing on theories of change of state and causality.
After the coffee break, Volha Petukhova (joint work with Harry Bunt), presented a machine learning-based approach to the incremental understanding of dialogue utterances, focusing on the recognition of their communicative functions. Then, Megumi Ohki described a work carried out with other colleagues at the Nara Institute of Science and Technology and at the Tohoku University, on recognizing confinement relations in Web Texts. Thursday's invited speaker was Eduard Hovy, whose brilliant talk argued for a new kind of semantics that combines traditional symbolic logic-based proposition-style semantics with Distributional Semantics (slides can be found here).
The afternoon session featured two parallel sessions. In Lecture Theatre A, a paper from Hiyan Alshawi, Pi-Chuan Chang and Michael Ringgaard was presented, where a method for training a statistical model for mapping sentences to underspecified semantic expressions was described. Then, Mathieu Morey (joint work with colleagues at INRIA and LORIA) proposed to use modular graph rewriting systems to compute semantics. Dan Garrette described an approach (co-authors: Katrin Erk and Raymond Mooney) to integrate logical representations with probabilistic information using Markov Logic.
In the meanwhile, in Lecture Theatre B, Rada Mihalcea presented a paper written in collaboration with Carmen Banea that explores the role of multilingual feature representation for word sense disambiguation. After Rada, Oliver Bott and Fabian Schlotterbeck presented some experiments (joint work with Jakub Szymanik) to investigate the computational complexity of German reciprocal sentences with different quantificational antecedents. Before the poster session, the posters' authors were asked to introduce their work in a 3-minute flash presentation, to give the audience a flavor of what their research was about. What an interesting idea!
After the coffee break, In Lecture Theatre A Edward Grefenstette and Mehrnoosh Sadrzadeh presented a joint work with colleagues at the Universities of Oxford and Cambridge. They provide a method to implement a compositional model of meaning for distributional semantics, by constructing a corpus-based vector space for the type of sentence. Then, Yoshihiko Hayashi described a framework for representing cross-lingual/interlingual lexical semantic correspondences, while Livio Robaldo compared two kinds of Maximalization proposed in the literature (Local and Global ones) for handling the proper truth values of independent set readings. On these bases presented his own solution to the problem. In the same time slot, in Lecture Theatre B Emiliano Raul Guevara introduced an approach to compute semantic compositionality in distributional semantics, while Hilke Reckman (joint work with Jeff Orkin and Deb Roy) described their use of data from a virtual world game to automatically learn words and grammatical constructions, and their meaning. Finally, Stephen Wu (joint work with William Schuler) proposed a structured vectorial semantic framework, in which semantic vectors are defined and composed in syntactic context.
One of the oldest colleges in Oxford opened its door to host the conference banquet. The Somerville College, founded in 1879 in response to a growing demand to provide an Oxford education for women, was the perfect place to gather IWCS participants, and to enjoy a traditional English dinner.
Friday 14th January
The morning session started with Sebastian Padò presenting a paper written in collaboration with Manaal Faraqui, where experiments to automatically acquire entailment pairs from newspaper corpora, pairing first sentences of articles with their titles are described. Then, Christian Meyer, on behalf of Elisabeth Niemann and Irina Gurevych, presented a method to automatically align WordNet synsets and Wikipedia articles to obtain a sense inventory of higher coverage and quality. Before the coffee break, Jason Utt (joint work with Sebastian Padò presented an approach to automatically classify polisemy, based on WordNet plus ontological classes provided by Corelex.
After the coffee break, the second invited speaker, i.e. Harry Bunt, presented an update semantics of dialogue acts, defined in terms of combinations of simple elementary update functions. Such approach is applied to dialogue acts representations (slides can be found here). Then, Chitta Baral described a system to translate sentences to formulas in a formal or a knowledge representation language, developed with other colleagues at the Arizona State University. The last presentation of the workshop was done by Aurelie Herbelot (joint work with Ann Capestake), who proposed a formalization that captures the underspecification of the quantifier in subject NPs.
Before the last lunch all together, the second Step meeting took place (organized by Johan Bos). Possible future shared tasks in the area of computational semantics were presented and discussed, respectively on the Deep Meaning Annotation Project (Johan Bos), on the annotation of the scope of negation cues (Roser Morante), and on QA4MRE, Question Answering for Machine Reading Evaluation (Anselmo Peñas).
IWCS: what's next?
So far, IWCS has been a biennial event. On the last day of the conference, participants were asked to fill in a form to find out how people feel about the frequency of the conference. It turned out that nearly 80% of the participants would like to see IWCS as an annual event! So... stay tuned!
>> back to Conference Reports