ISSN Volume 17, number 6 December, EDITORIAL ADDRESS BNVKI newsletter

ISSN 1566 - 8266 BOARD BNVKI: Joost Kok (chair) Rineke Verbrugge (member) Wiebe van der Hoek (member) Yao-Hua Tan (member) Floris Wiesman (member) Luc DeHaspe (member) Walter Daelemans (member) Bas Zinsmeister (member)

BNVKI-SECRETARY Rineke Verbrugge Rijksuniversiteit Groningen Cognitive Science and Engineering Grote Kruisstraat 2/1 9712 TS Groningen [email protected]

Volume 17, number 6 December, 2000 EDITORIAL BOARD: Floris Wiesman (Editor in Chief) Jaap van den Herik Bart de Boer Shan-Hwei Nienhuys-Cheng Cees Witteveen Antal van den Bosch (section editor) Joris Van Looveren (editor Belgium) Richard Starmans (section editor) Radboud Winkels (section editor) EDITORIAL ADDRESS BNVKI newsletter Joke Hellemons Universiteit Maastricht FdAW, Department of Computer Science P.O.Box 616, 6200 MD Maastricht Telephone: 043 388 34 77 / 35 04 Fax: 043 388 48 97 E-mail: [email protected] http://www.cs. unimaas.nl/~bnvki

Editorial

BNVKI Newsletter

128

December 2000

TABLE OF CONTENTS (Editor-in-Chief) ....................................................................................................................................................... Table of Contents ....................................................................................................................................................... BNVKI-Board News (Joost Kok) ............................................................................................................................. ................................................................................................................................................................................... ................................................................................................................................................................................... ................................................................................................................................................................................... ................................................................................................................................................................................... ................................................................................................................................................................................... ................................................................................................................................................................................... Section Computational Linguistics (Antal van den Bosch) ....................................................................................... ......................................................................................................................................................................... SIKS (Richard Starmans)........................................................................................................................................... Section Knowledge Systems in Law and Computer Science (Radboud Winkels)..................................................... ......................................................................................................................................................................... ......................................................................................................................................................................... Call for Papers............................................................................................................................................................ ......................................................................................................................................................................... Conferences, symposia, Workshops........................................................................................................................... Email addresses, Board Members/ Editors BNVKI newsletter / How to become a member?/ Submissions............. Advertisement ............................................................................................................................................................

The BNVKI is sponsored by AHOLD and by BOLESIAN In 2000, the publication of the BNVKI newsletter is also supported by the Division of Computer Science Research in the Netherlands (previously called SION, now ACI)

BNVKI

129

December 2000

BNVKI- BOARD NEWS Joost Kok

NEURAL NETWORKS 1 Report by Bert Kappen Katholieke Universiteit Nijmegen SpikeProp: Backpropagation for Networks of spiking neurons S.M. Bohte, H. La Poutré and J.N. Kok This session contains three contributions on neural networks. The presentations are as diverse as the field of neural networks itself the first contribution has a neurobiological motivation, the second paper discusses cognitive aspects and the third paper treats the neural network as a statistical tool. Backpropagation is a well-known learning rule for feed-forward neural networks. A shortcoming of these networks is that neural activity is modelled as rate coding. Instead, more and more neurophysiological evidence indicates that the exact timing of spikes is also of importance or information processing in the brain. This paper presents an extension of the Backpropagation learning rule for spiky (integrate-and-fire) neurons. Trading off perception with internal state reinforcement learning and analysis of Q-Elman Networks in a markovian task Bram Bakker and G. van der Voort van der Kley There is a long experimental tradition in experimental psychology on measuring reaction times of subjects performing various tasks. From these studies it has become clear that the time to execute a task increases with task complexity. This suggests that complex tasks are implemented as the sequential execution of elementary steps. In a simplified way, these elementary steps may be thought of as hardwired feed-forward neural networks. This paper demonstrates that the Elman network behaves in the above way. An Elman network is a feed-forward network augmented with delayed feedback connections from the hidden layer to the input layer. Therefore, the Elman network can implement tasks either as an elementary feed-forward step or as a sequence of such steps. The paper demonstrates that the Elman network learns the parallel (simply feedforward)

BNVKI newsletter

130

December 2000

commitment to more law-specific concepts would make modelling with the proposed ontology easier?

solution for simple tasks and chooses the sequential solution when the task complexity increases. Thus, the behaviour of the Elman network is similar to the neural network human. In addition, it shows that although some tasks can in principle be solved by a pure feed-forward implementation, the sequential solution provides a simpler alternative.

Towards a generic model of trust for electronic commerce Y.H. Tan and W. Thoen This first of two contributions in this session from the Erasmus Universiteit consisted of two halves: in the first halve, the authors presented a general analysis of the notion of trust. They discussed different definitions of trust, culminating in their own definition of level of trust. They distinguished different types of trust and different mechanisms for trust. In the second half of the presentation, they provided a formal analysis of a particular trust scenario, namely the mechanism for creating trust in the shipping industry via bills of lading. Unfortunately, the formal analysis was limited to this particular case. An obvious agenda for further work is to provide a formal account of trust that is as general as the informal analysis in the first part of the presentation.

Ensembles of nonconformist neural networks M.C. van Wezel, M.D. Ont, and W.A. Kosters This paper shows a scheme for training an ensemble of feed-forward neural networks. It is well-known, that the expected prediction error of the mean output of an ensemble is smaller than the mean prediction error of the outputs of each of the individual networks in the ensemble. The difference is proportional to the diversity among the networks in the ensemble. The diversity can be measured by the covariance in the network outputs over the training set. The authors propose to stimulate diversity during training by adding the covariance to the training error, and performing gradient descent on this cost criterion. Numerical results indicate improvement over training the networks independently.

DocLog: an electronic contract representation language Y.H. Tan and W. Thoen The same two authors presented a rather different type of work, namely one aiming at a machine readable representations of legal documents. In itself this is nothing new, but the approach in this paper was innovative because it distinguished different levels at which to represent such a legal document: • a data-layer, which aims at the kind of representation typically used in ERP systems. • a natural language layer, that contains the text of the document structured by a predefined set of XML tags. • a semi-formal layer, using concepts like agents, actions, and norms to describe the semantics of the legal document in a machine processable form. Clearly, such multi-layered, partly machine accessible representations of legal documents have a great potential in many application areas.

KNOWLEDGE REPRESENTATION AND SYSTEMS Report by Frank van Harmelen Vrije Universiteit Amsterdam The Law as a dynamic interconnected system of states of affairs J. Hage and B. Verhey Jaap Hage and Bart Verheij from the University of Maastricht presented a (very) abstract model of the law. The purpose of this model is to serve as a toplevel ontology of the law. Much existing work on legal ontologies (e.g. by McArty, by Valente, etc) can then be presented as specialisations of this very general top-level legal ontology. The basic modelling primitives of their ontology are states of affairs, with events connecting these states, possibly governed by rules. The authors showed how a number of existing legal concepts and scenario’s can be modelled in such a very general ontology.Your reporter liked the general aim of the paper (i.e. providing a top-level ontology to unify various existing more detailed ontologies), but wondered whether the proposed primitives (states, events, and rules) were not so general that this would cover almost any domain. Perhaps more

BNVKI newsletter

NATURAL LANGUAGE PROCESSING 2 Report by Walter Daelemans Universiteit Antwerpen The organizers of BNAIC 2000 have done a great job in making the conference attractive for computational linguistics researchers, for example by inviting Yorick Wilks as keynote speaker (unfortunately he didn’t make it because of illness),

131

December 2000

and by devoting two sessions to NLP work. In the second of these, Bart de Boer, a previous BNVKI award winner, could not present his paper, unfortunately.

P. van der Putten The speaker showed how to combine a customer database (many customers, not so many variables) and a market survey (not so many people, but many variables) in order to extend the customer database to predict the behaviour of the customers on variables not in the customer database, but important for a datamining task. His first results are promising.

An analysis of multiple-word naming games J. Van Looveren This works builds incrementally on the groundbreaking work on lexicon evolution in populations of agents playing naming games, undertaken at the AI-LAB in Brussels. In this particular study, it was shown that the general approach still holds when allowing the communicating agents to use expressions consisting of more than one word. This results in smaller lexicons and does not decrease communicative success in the population.

Fuzzy Clustering Based Target Selection U. Kaymak and M. Setnes The method of extended fuzzy c-means clustering was described and applied on two cases with real-world data. Gain charts showed the effect of the clustering on the target selection. Also this talk and paper concluded with remarking the transparency of the obtained profiles. The paper makes a solid impression.

Efficient parsing of domain language K. Sima’an This paper describes part of the dissertation research of Khalil Sima’an, a dissertation for which he received the FoLLI Outstanding Dissertation Award, recently. He received the BNVKI award for his paper at the BNAIC as well. Given all these honours, I can only advise you to read at least the paper, preferably also the dissertation, for yourself. In his research, Sima’an shows that broad-coverage grammars and statistical parsers can effectively be specialised to specific domain language, making the statistical model considerably smaller and more efficient, without substantial loss in accuracy.

LOGIC AND REASONING 1 Report by Bart Verhey Universiteit Maastricht Semantics of Input-Consuming Logic Programs A. Bossi, S. Etalle and S. Rossi

MACHINE LEARNING 4 Report by Ida Sprinkhuizen-Kuyper Universiteit

The abstract was presented by Sandro Etalle, who works at the IKAT institute of the Universiteit Maastricht. The full paper was presented at CL2000, the First International Conference on Computational Logic, and deals with input-consuming logic programs. Such logic programs are a special kind of moded programs. A logic program is moded when the arguments of the atoms are labelled as input or output ones. Input-consuming programs are then those moded programs in which the derivation steps are exactly those for which the input arguments are not instantiated by unification with the clause's head. The authors proved that the operational semantics of the inputconsuming resolution rule is correct and

In this session three papers on datamining techniques were presented. Classification and Target Group Selection based upon Frequent Patterns W. Pijls and R. Potharst

This talk reported on research inspired by association rules. The conclusion of the paper is that the methods introduced especially result in transparent results: it is easy to state why an object is classified as it is, or why an individual is selected in a mailing group. This is an important aspect for practical applications of datamining. Data Fusion: A Way to Provide More Data to Mine in?

BNVKI newsletter

132

December 2000

classical logic as the intended semantics. The abstract was presented by Cees Witteveen, who informed the audience that the full paper contains an error with respect to its relation with the AGM framework of theory revision.

complete with a particular denotational semantics for logic programs, and investigated properties of that semantics. The context of the work is the use of socalled delay declarations, that are used to avoid nontermination and inefficient computation that can result from dynamic rule selection.

MACHINE LEARNING 2 Report by Hendrik Blockeel Katholieke Universiteit Leuven

Relating Protocols for Dynamic Dispute with Logics for Defeasible Argumentation H. Prakken

An Initial Approach to Wrapped Input Selection using Least Squares Support Vector Machine Classifiers: Some Empirical Results B. Baesens, S. Viaene, T. van Gestel, J.A.K. Suykens, G. Dedene, B. de Moor and J. Vanthienen

The paper, to appear in Synthese, deals with disputes in which arguments and counterarguments are exchanged. A way of evaluating the arguments used in such disputes is in terms of games: the proponent of an argument for a claim must have a winning strategy. The author discusses how this evaluation criterion is affected when the body of information is not fixed at the start of the debate. The author defines notions of soundness and fairness for the protocols underlying such dynamic disputes. In the talk, the focus was on some of the author's other work, focusing on the speech acts involved in disputes.

In the first talk of this session, Bart Baesens presented a machine learning approach in which two learning techniques are combined: the use of least squares support vector machines (LS-SVMs) and of wrapped feature selection. LS-SVM's are a modification of the standard support vector machines. The distinguishing feature of LS-SVMs is that the error function that is minimised is the sum of squared errors instead of the sum of absolute errors; this gives rise to a more efficiently solvable system of equations and hence ensures a better scalability. Wrapped feature selection refers to a feature selection approach that involves a greedy search for a minimal subset of features that allows to train LS-SVMs with an estimated predictive performance not significantly worse than the one achieved by an LS-SVM using all features; this estimation is done using crossvalidation. Experimental results on several benchmark datasets were presented that show the value of this approach.

Classical and General Frameworks for Recovery W. van der Hoek and C. Witteveen

The paper, that appears in the proceedings of ECAI2000, the 14th European Conference on Artificial Intelligence, deals with theory recovery. The idea is that the consistency of a theory is restored with respect to an intended semantics. In the authors' approach, a second semantics, the backup semantics, is used to guide the process of recovery. In the process, the theory that is not interpretable in the intended semantics, is replaced by an interpretable theory that has the same meaning from the point of view of the backup semantics. In previous work, the focus was on theories that were at least classically consistent. In contrast, the present paper allows classically inconsistent theories and deals also with BNVKI newsletter

A Unified Approach for Practical Applications of Fuzzy Clustering U. Kaymak Uzay Kaymak talked about fuzzy clustering, which differs from normal clustering in that elements not necessarily belong to a single cluster, but instead are assigned some degree of membership of several clusters. One fuzzy clustering algorithm is the socalled "fuzzy-c-means" algorithm. Dr. Kaymak discussed several problems with this algorithm, such as its tendency to behave suboptimally when used with highly skewed data distributions, and the fact that a good choice for the number of clusters c must be made in advance. He then presented an extended fuzzy c-means (EFCM) algorithm, which has the property that during the clustering process clusters that are highly similar may be merged, and 133

December 2000

networks. She explained that this type of networks is used to represent knowledge in terms of observables and their probabilistic relations. The existing algorithms to determine the effects of observables on each other can only handle a single observation appropriately. A single observation is entered and propagated through a qualitative probabilistic graph and the effect of the observation is determined. The standard way to deal with multiple observations is by entering them in sequential order. However, the order of entering observables yields different and weaker results. The authors proposed an algorithm to propagate multiple observations in qualitative probabilistic networks such that the order of observations does not generate weaker results. This is done by disregarding the so-called “intercausal influences”. The speaker demonstrated their algorithm by working out an example.

also uses an extended notion of prototypes of clusters. The EFCM algorithm was demonstrated to work better than FCM on several example data sets. Rough Sets and Ordinal Classification J.C. Bioch and V. Popova Viara Popova presented an approach to ordinal classification (the learning of monotone classification functions) that is based on rough sets. After introducing the audience to some concepts of rough set theory, she showed how this theory can be extended to cater for ordinal classification by defining monotone versions of existing concepts such as discernibility matrix, and reducts. She proceeded by presenting some existing results on the use of monotone discrete functions for ordinal classification, and then presented several theorems that clarify the relationship between monotone discrete functions and monotone rough set concepts. The talk ended with a brief discussion of experiments that support the theoretical results.

Variational Approximations Between Mean Field Theory and the Junction Tree Algorithm W. Wiegerinck The last speaker of this session was Wim Wiegerinck who presented a method for approximate reasoning in graphical models such as Bayesian networks. He argued that the exact reasoning in large and complex graphical models is computationally intractable and that approximate reasoning is therefore essential. Large, complex and intractable graphical models are approximated by models with tractable structures. To this end, he proposed an extension of mean field method by using approximating model that factorises according to given structure such as potential clusters. In this way, he claims to bridge between mean-field theory and exact computations. This method is presented by some examples.

GRAPHICAL MODELS Report by Mehdi Dastani

Vrije Universiteit Amsterdam Bayesian Model-Based Diagnosis P. Lucas This session began with Peter Lucas who presented an extension of the model-based diagnosis framework. According to him, traditional consistency-based diagnosis systems do not provide appropriate representation to deal with the involved uncertainty when they generate diagnosis. On the other hand, Bayesian networks, designed to represent and reason about uncertainties, are not appropriate to represent and reason about the structure and the behaviour of systems. The presented framework, called Bayesian model-based diagnosis, is a way to integrate both logical and probabilistic reasoning to determine and generate diagnosis. In such a framework, it can be reasoned about the structure and the behaviour of systems while the involved uncertainties are taken into account.

INFORMATION BROKERING Report by Yao-Hua Tan Erasmus Universiteit Small-World Semantic Networks E. Postma and F. Wiesman

In their research Eric Postma, Floris Wiesman and Jaap van den Herik studied semantic networks for knowledge representation. Semantic networks consist of conceptual nodes and association links between these nodes. Knowledge is retrieved from these networks by using these association links between the nodes that connect concepts to related concepts. For example, by having a direct link

Propagation of Multiple Observations in Qualitative Probabilistic Networks S. Renooij, L.C. van der Gaag, and

S. Parsons The second speaker was Silja Renooij who presented a paper on qualitative probabilistic BNVKI newsletter

134

December 2000

dimensional database that is modelled in the DESIRE framework. The Iceberg system uses a taxonomy of types to assist the user in finding the right search terms. Based on the reasoning process of the user the Iceberg system can narrow down questions, or broaden the scope of a question to generate alternative solutions that could help the user to formulate his questions more precisely. For example, when the user asks for a course how to learn how to give presentations, the system could reply by asking whether it is for preparing slides or how to cope with stage fright. The DESIRE framework is very suitable to model this reasoning process of the user, because it is based on a collection of autonomous reasoning modules that can simulate the human reasoning. After the presentation Jonker also gave a demonstration of the system.

between Newton and apple, the person Newton is directly associated with the apple that supposedly led to the invention of the gravitation law. In their research they studied the problem what the characteristics are of an efficient semantic network. Based on some fundamental results from graph theory they postulated that the so-called small-world networks are the most efficient ones. Small-world networks are networks that are subdivided in smaller densely connected sub-networks with relatively few links between these sub-networks. They validated these results with a comparison between a natural human network and an artificial man-made network. The human network was a database of word association as produced by free association of human test persons. The man-made network was the ARIA database of the Amsterdam Rijksmuseum that was developed to classify the art objects in the collection of this museum. Interestingly, the human network had a larger small-worldliness than the manmade network, which seems a corroboration of their conclusion about the efficiency of small-world networks.

KNOWLEDGE REPRESENTATION AND SYSTEM 2 Report by Jaap van den Herik Universiteit Maastricht

The session consisted of three presentations treating different aspects of the theme. The speakers presented their knowledge well and in an animated way. They neatly kept the time schedule and raised an interesting discussion. The first speaker was Frank van Harmelen (Vrije Universiteit Amsterdam). He is co-author of the publication Oil in a Nutshell, written by D. Fensel, I. Horrocks, F. van Harmelen, S. Decker, M. Erdmann and M. Klein. One month earlier the SIKScommunity had already been extensively informed on the propositions for standardization of this illustrious team (beside the VU Amsterdam, the Universities of Manchester, Karlsruhe and Stanford are represented in the consortium). Van Harmelen first explained why we need a uniform standard for the description of ontologies. Thereafter, issues were addressed as to how should we describe an ontology in a standard way?

ICEBERG: Exploiting Context in Information Brokering Agents C. Jonker and A. Vollebregt

Catholijn Jonker reported research that she had done jointly with Arjen Vollebregt on the development of an information broker that assists the user in finding personal development courses. The problem in this domain is that people can have a wide variety of perspectives on these courses. For example, a question to learn how to give a presentation might be either a request for a course on how to use PowerPoint software to prepare a slide presentation, or a request how to cope with the stage fright. The information broker should be able to handle both questions, hence it has to store the data about courses from many different perspectives. The Iceberg architecture is a kind of multiBNVKI newsletter

135

December 2000

but not by the logical hierarchy with which the components are connected to each other. Bos addressed this problem in his talk and also gave an answer to a question on this subject. He did not have a solution, but the problem did have the attention of the research group.

What are the common aspects and what can be exchanged? The team proposes to come to an OIL, i.e., an Ontology Interchange Language. OIL works with the well-known modelling primitives from the frame-based world and is based on the ontologies that are Description Logic oriented. Furthermore, it is an extension of the Web standards XML and RDF. In his talk Van Harmelen addressed both the syntactical and semantic aspects. Neither did he forget practice. If we may believe his enthusiastic presentation OIL will be the standard of the future (for ontologies). From a political point of view Van Harmelen and his group have ensured themselves of renowned support. Moreover, the concepts shown are convincing. The floor is now to the users.

Ben Kröse (Universiteit van Amsterdam) gave the last presentation of this session. His co-authors, Rob van den Boogaard and Niek Hietbrink, are from the same University. Their publication is titled: Programming robots is fun: Robocup Jr. 2000. During the RoboCup Euro 2000 the group of Kröse organized a tournament for schools, in which they used robots made of LEGO. They called this tournament: Robocup Jr. The idea was borrowed from Henrik Lund and Luigi Pagliarini, who organized the first tournament in 1999 during the RoboCup ’99 (Stockholm). The idea is to have school children (13-16 years) gain experience in programming robots. In this talk (and in the publication) various educational aspects were discussed, and the reasons why something was done and why something else not, were weighed against each other (e.g., the use of LEGO light sensors in stead of the use of a camera). We will limit the description here to some realized facts, because these will show best the success of the whole project and also the enthusiasm with which the group has organized the Robocup Jr. Twelve teams of 8 schools participated (11 Dutch teams and 1 German team), consisting of approximately 50 children. In the morning a general introduction was given and in the afternoon the children could work on the robot. There was one teacher for three groups. Especially the children with some fear for programs and programming quickly (in 15 minutes) lost their diffidence. The presence of the “assignments”: Find the ball, Make a sound, Look for the ball, made it easier for the pupils (certainly the beginners group)

Model-based diagnosis support for satellitebased instruments André Bos

The co-authors were Arjen van Gemund en Cees Witteveen (all three of the Delft University of Technology). The Delft group has ample experience with modelbased monitoring and the diagnosing of a satellite-based instrument. In the course of their research they have come to the conclusion that there are two explicit problems which recur repeatedly, i.e., the complexity and the testframe. The group therefore decided to improve the support of the diagnosis by a special support tool named UpTime. This tool contains a large number of detailed component descriptions that link up to the model-based diagnoses. Obviously, a hierarchy has been made in the components. This also reduces the complexity that occurs when, in a certain situation, a component has to be investigated more closely before use. In short, the tool clearly simplifies the process of argumentation, but unfortunately at the expense of the model management. Several descriptions of behaviour of the model have to be maintained and also linked to each other, BNVKI newsletter

136

December 2000

to learn how to program. Unfortunately, I nowhere read in the description: Kick the ball, or Shoot the ball in the goal. Neither did I hear this in the presentation. But the fact that Ben Kröse, who cheered up the talk with a film of very active children playing ‘robot football’, told that there was a winner (Montessori Lyceum Amsterdam), several goals must have been scored. The winners received a LEGO Mindstorms set which was again an encouragement to continue activities and publications in this direction. The BNAICaudience heard/read with pleasure that the second place was won by a very simple program, that was not hindered by knowledge but benefited from the speed (of the loops) and was therefore often first at the ball. In summary, here we see a nice educational opposite between knowledge (and the obstacles to deal with it properly) and rapid search (with little knowledge). The advice of Kröse cum suis to call this field edutainment is gladly taken over and we hope that the industry and academia will continue to support it.

BNVKI newsletter

137

December 2000

originating from the Knowledge and Agent Technology group of Maastricht with the title Approximation of the optimal solution for Dynamic CSPs. Many AI problems can be modelled as constraint satisfaction problems (CSPs). However, the CSP problem definition may change over time because of the environment, the user preference or agent’s behaviour in a distributed system. After a change the infringed solution must be repaired. A Repair-Based algorithm with arcconsistency (RB-AC) is available. However, the time complexity RB-AC is higher than that of an algorithm that constructs a solution from scratch. In the presentation (and the paper) two alternative algorithms that modify the original RB-AC algorithm are proposed. Experimental results show the two proposed algorithms perform better with respect to the approximation of the optimal solution in a limited period of time. A very clear presentation, but maybe an improvement is possible by adding one or more practical relevant examples to demonstrate the basic ideas and the usefulness of the two proposed repair algorithms.

LOGIC AND REASONING 2 Session Chair H.C.M. de Swart Reasoning with Modularly Pointwise Preferential Relations O. Arieli In this longer contribution the author considers a family of preferential consequence relations, defined by a general and natural semantics. The common property shared by all these relations is that their underlying preference criteria are based on modular partial orders. These relations share many desirable properties for common-sense reasoning, such as paraconsistency, plausibility, adaptivity and rationality. Input-Output Logics L.W.N. van der Torre Roughly speaking, there are two kinds of black boxes. The box may stop some inputs, while letting others through, perhaps in modified form. And inputs may be conditions, with outputs expressing what is deemed desirable in those conditions. The purpose of this short contribution is to develop a general theory of propositional input/output operations, covering both kinds of examples. Particular attention is given to the case where outputs may be recycled as inputs. Learning in Description Logics by Refining Concepts Shan-Hwei Nienhuys-Cheng Since Description Logics (DL) are a fragment of first-order logic, one could in principle translate DL expressions into prenex conjunctive normal form (PCNF) and use the PCNF-refinement operator to refine the PCNF encoding of DL expressions. The author explains that this straightforward approach has two disadventages. In order to circumvent these problems, the author develops a refinement operator working directly on DL formulas. Since there exists no complete and minimal refinement operator, the operator proposed by the author is one of the best one can hope for.

The second presentation ‘An Approximation Algorithm for a Logistic Planning Problem’ from the Information Technology and Systems faculty of the University of Delft deals with a logistic planning problem. However, this logistic problem is (nearly) decomposable into a number of subproblems. The research question is a more or less traditional: how well can this (planning) problem be solved by iteratively scheduling the local problems according to a certain scheduling policy. An algorithm designed as a multi-agent system that uses a specific policy to solve the logistic problem was explained an approximation results were presented. The time improvements were very impressive. The plan quality comparison is less clear because the Black-box approach optimise for a minimum amount of time steps, the presented approach optimise for the plan length. However, the quality of the plans seems good compared with the Black-box approach. Because the local approximation algorithms are still NP-hard, an open question seems the performance of the new approach on more complex real-world problems.

PLANNING AND SCHEDULING Session Chair Ton Weijters

After disentangle some communication problems between notebooks and beamers the planning and scheduling session started with the two planned presentations. YongPing Ran presented the first paper BNVKI newsletter

138

December 2000

In conclusion, two fine presentations combined with some clarifying discussions.

BNVKI newsletter

139

December 2000

and to deal with technologies such as multicast, fluents, and for rapid agent system prototyping. Frits Vaandrager’s lectures on Verification of Timed and Hybrid Systems concentrated on the use of timed automata technology for model checking and on hybrid systems. Again, a tool was used to demonstrate the practicality of some issues. Peter Lee gave a lecture on Proof-Carrying Code: a mechanism for certifying the safety of a program supplied by an untrusted source. Such mechanism uses a language of predicates to express the verification conditions and the specification of inference rules: in this way an interesting link was set between the high level specifications and the low level code (assembler for x86). More formal courses were held by Dale Miller (A Logic for Reasoning about Logic Specifications) on linear logic and sequent calculus, and by Moreno Falaschi (Optimization of Declarative Languages), who gave an overview on Functional Logic Programming (FLP), partial evaluation and static analysis of functional logic programs based on abstract interpretation techniques. Finally, Thom Fruehwirth, (Constraint Programming and Reasoning) discussed with wealth of practical examples the fundamentals of Constraint Logic Programming (CLP), Concurrent-CLP, and Constraint Handling Rules (CHR), which allow one to define declaratively new constraints propagation method. The deep level of detail used to explore certain issues did not let loose sight of the general framework that was being depicted: the lecturers themselves had care to stress the connections holding among the different topics. In this way, both the novice and the more experienced students benefited from participating. The organization has been faultless. The choice of putting attendees and lecturers together in a hotel, as in other past GULP schools, fostered frequent and fruitful interactions, minimizing any logistic problem and leaving little room for distractions. The scholastic obligations, though, have been relieved by some social events, and especially by the delightfulness of the location. More information about the 2000 ISCL courses is available at the following URL: http://www.cs.unimaas.nl/ ~etalle /school 2000/index.html

ISCL 2000 International Summer School in Computational Logic Acquafredda di Maratea, Basilicata September 3-8 2000 Marco Gavanelli DIF - University of Ferrara [email protected] Paolo Torroni LIA - DEIS - University of Bologna [email protected]

ISCL 2000, the International Summer School in Computational Logic held at Acquafredda di Maratea, Basilicata (Italy) on September, 3rd-8th, 2000, represents the prosecution of the biennial summer schools organized by GULP, the Italian association for Logic Programming established in 1987 in Pisa, Italy (two former schools had been held in ‘96 and ‘98). This year’s edition opened to an international perspective thanks to the auspices of ALP, the Association for Logic Programming, and of COMPULOG, the ESPRIT Network of Excellence in Computational Logic. Computational Logic constitutes nowadays a foremost research stream. Glaring achievements of the most recent years range from Constraint Logic Programming to Internet applications and verification techniques, where solid mathematical foundations allowed dealing with more and more complex problems. The school organizers, Sandro Etalle and Maurizio Gabbrielli, decided to give it a strong characterization with respect to new technologies and applications of Logic Programming, still leaving an adequate margin to more fundamental and theoretical issues. The school counted 25 participants, mostly from academic research areas: half of them were Italian, the others came from France, Spain, Estonia, Japan, Israeli, UK, Russia, Czech Republic, Sudan. In general, the courses were very high level, and covered a wide range of topics. ISCL 2000 was composed of six lectures of seven hours each, with the possibility of proficiency grades at the end for those interested. The first day was devoted to the course presentations. The lectures had a large participation, despite the tight scheduling (eight hours per day). An application oriented course was held by Paul Tarau (Logic Programming and Internet Technologies), where through the description of several development tools it has been shown how Logic Programming can be used to create Internet applications (HTML, VRML, CGI)

BNVKI newsletter

140

December 2000

BNVKI newsletter

141

December 2000

combine 3APL with low-level controllers to study the possibility to move cognitive robotics out of artificial worlds and towards complex real worlds. Since 1998 there is a large involvement of students in the RoboLab through both curricular and extra-curricular activity. Within the second year project-based course, the so-called software project, there have been assignments to write robotic control software. Besides this course, the largest part of the activity in the lab has been extra-curricular, due to the efforts of a group of enthusiastic students interested in robotics. These students have different backgrounds: cognitive artificial intelligence(CKI), technical artificial intelligence(TKI), and computer science. This group has developed a robotic toolkit, RU-SMART, which was used to participate in RoboCup 2000 in Amsterdam.

ROBOTIC RESEARCH IN UTRECHT

Walter de Back INTRODUCTION

Agent technology and autonomous robotics are active research topics. In both areas there is a quest for control methods providing fast, accurate, robust behaviour. At the Institute for Information and Computing Science (ICS) at the University of Utrecht we have a modest robot laboratory (RoboLab) for this kind of research. Currently, there are two parallel lines of ongoing research. (1) The group Intelligent Systems (IS), interested in cognitive robotics, has developed an agent programming language, 3APL, for the design and construction of intelligent agents like Personal Assistants. (2) A group of students, working in the RoboLab, has developed a toolkit for robotic control, RU-SMART, for the design and implementation of intelligent behaviour for real-world robots. The latter system was used in building a soccer-robot for participating in the European Championships RoboCup 2000 in Amsterdam. This article provides an overview of the activities of the RoboLab at the University of Utrecht.

The RoboLab uses four robots: three Pioneers 1 (P1) and one Pioneer 2 (P2) from ActivMedia[1]. All our robots have two driven wheels and a castor wheel. The P1s have eight sonar sensors in front and a simple camera. They do not have on-board computers, but are controlled by an offboard computer via a modem connection. The P2 has sixteen sonar sensors all around, a pan-tilt-zoom camera, a compass, and a laser range finder. It has a Pentium2 on-board computer for all processing. For communication with an off-board computer or other robots, it is provided with wireless ethernet.

ROBOLAB

In 1997, the discipline group IS of the Institute for ICS founded a robot laboratory. The main goals of this RoboLab are: (1) to subject the theoretical efforts of the group (especially with respect to cognitive robotics) to experimental testing and (2) to get students interested and involved in the research on intelligent systems in general and robotics in specific. Cognitive Robotics views programs as intelligent agents acting on our behalf. To support this kind of programming, the group IS has proposed the agent-based programming language 3APL. In the RoboLab, there are ideas to BNVKI newsletter

3APL

In cognitive robotics, one is concerned with designing robots encompassing highlevel cognitive functions: robots that sense, reason about and act in a constantly changing world. 3APL is a high-level agent programming language, that takes the idea that agents have different mental states (such as beliefs, goals) seriously; the constructs of the language immediately 142

December 2000

address these states. 3APL include features for: ‘representing and querying the agent’s beliefs; belief updating for incorporating new and removing existing information in the agent’s belief base; and goal updating to facilitate practical reasoning, that is, for planning and the reconsideration of adopted plans’[2].

of soccer as a central topic of research, aiming at innovations to be applied for socially significant problems and industries. The Dutch RoboCup team is a joint project of the University of Amsterdam (UvA), the Technical University of Delft (TUD), and the University of Utrecht. The resulting robot team has participated in the European Championship RoboCup 2000 held in Amsterdam[8]. The UvA and the TUD were responsible for developing a goalkeeper and defenders, for which they used Nomadic Scout robots. The effort of our RoboLab was concentrated on the research and development of the team’s attacker. For this task, we used our P2 robot, Dexter, because of its fast acceleration, the on-board computer and the laser range finder. Dexter was equipped with a pneumatic device for ball handling and kicking, which was developed at the TUD. For controlling the robot, we developed a soccer application in RU-SMART. This application is an implementation of a subsumption architecture in the task execution module (using the worldmap, motor control and communication modules provided in RU-SMART). No more than three weeks before the start of the RoboCup event, we had the hardware of Dexter up and running for the first time. At that point, we only had tested RU-SMART (with a simplified soccer application) on the simple P1 robots. Therefore, all Dexter’s behaviour had to be made in these couple of weeks. RU-SMART supports the bottom-up implementing of behaviour. The different behaviours can be implemented, tested and debugged separately in an incremental fashion. This feature allowed us to develop all soccer behaviour in this little time. Moreover, our approach to a subsumption based control proved to result in coherent, smooth behaviour.

RU-SMART

RU-SMART is an acronym for RoboLab Utrecht’s Smart Multi-purpose Autonomous Robot Toolkit. The toolkit, which is still in development, supports the design and implementation of robotic control for intelligent behaviour of realworld robots. The requirements set for the system include robot- and applicationindependancy, fast, robust control and easy implementation of behaviour[3]. Its open architecture allows implementation of different control structures. RU-SMART consist of a parallel modules for perception and world modelling, task execution, communication and motor control[4]. The task execution module can be constructed in different ways (while maintaining the rest). Approaches we have explored so far include a simple planning system and a behaviour-based subsumption architecture[5], which we used in the RoboCup application. Currently, we are exploring the possibilities to use evolutionary techniques in our robotic systems, e.g. evolutionary robotics [6]. In the near future, we are going to use a variety of neural networks and genetic algorithms to solve optimisation and action selection problems. ROBOCUP-SOCCER

RoboCup [7] is an international project to promote artificial intelligence, robotics, and related fields. It is an attempt to foster AI and intelligent robotics research by providing a standard problem where wide range of technologies can be integrated and examined. RoboCup chose to use the game BNVKI newsletter

143

December 2000

promising exchange of knowledge and information that, we hope, will result in better robotic systems for all parties. We are investigating possibilities to develop applications in RU-SMART using evolutionary techniques. Evolutionary robotics offers a variety of methods for adaptive control and optimisation of behaviour. Besides providing optimal use of the robots capabilities and fast, robust control, this is interesting for the sciences of (neuro-)ethology, cognitive psychology, artificial intelligence, and philosophy.

During the tournament, the team showed an improving performance. Both Dexter and the goalkeeper and defenders from the UvA and the TUD were getting better due to the great efforts of the members of the Dutch team. Unfortunately this was not enough to get to the top of the tournament. This was due to problems with Dexter’s hardware, some software problems within the team and the absence of communication between the robots. Almost half of Dexter’s play time was lost because of a broken axis. Even with this setback, our last game, versus Shariff from Iran, who later went on to become the ‘European Champions’, was a very exciting one. The Dutch team, also called ‘RobOranje’, has reached a disappointing sixth place in this tournament. However, considering this was the first time we participated in such a project, this result is not to be ashamed about and is encouraging to do better next time.

Don’t hesitate to contact us via [email protected] or [email protected] RoboLab Team members: Steven Anker Walter de Back Martijn ten Brink Jannes Faber Arne Koopman Marco Wiering Tijn van der Zant

FUTURE

Lars Zwanepol

Currently, we are making new plans for continuing our RoboCup efforts. We are evaluating our system and planning changes in the software to improve Dexter’s performance. To ensure continuation of our knowledge and experience, we are motivating new students to get involved in the development process of RU-SMART and the RoboCup application. To facilitate the growth of the group, we will extend the RoboLab. We will have a separate lab for simulation, meetings and system administration and a lab for the real-world robots with enough space to lay out a RoboCup field. Besides the cooperation with the UvA and the TUD, we have contacts with the AiS group (Autonomous intelligent Systems) of the German research institute GMD and with a group of employees of CFT (Centre for Industrial Technology) within the Philips concern. With both groups, there is a BNVKI newsletter

References

[1] http://www.activmedia.com [2] Hindriks K., de Boer F., van der Hoek W., Meyer J.J.-Ch., “Agent Programming in 3APL” [3] de Back W., “SMART: A Toolkit for Robotic Control” [4] Faber J., “SMART: Manual” [5] Brooks R. A., “A Robust Layered Control System for a Mobile Robot” [6] Nolfi S., Floreano D., “Evolutionary Robotics” [7] http://www.robocup.org [8] http://www.robocup.nl

144

December 2000

Below we provide you with the total number of the Ph.D. theses announced in 2000. For comparison we have given the list containing the numbers of the previous years. This list contains the adjusted numbers.

Jaap van den Herik IKAT, Universiteit Maastricht F.M. Waas (November 3, 2000) Principles of Probabilistic Query Optimization. Universiteit van Amsterdam. Promotor: Prof.dr. M.L. Kersten.

1994: 22 1995: 23 1996: 21 1997: 30 (for reasons see above) 1998: 21 1999: 28 2000:

H.P. Ditmarsch (November 20, 2000) Knowledge Games. Rijksuniversiteit Groningen. Promotores: Prof.dr. G.R. Renardel de Lavalette, Prof.dr. J.F.A.K. van Benthem.

As a courtesy to the 2000 newborn doctores I am pleased to honour them (again) by mentioning them below together with their promotion date.

J. Romein (January 18, 2001)

Multigame – An Environment for Distributed Game-Tree Search. Vrije Universiteit van Amsterdam. Promotor: Prof.dr.ir. H.E. Bal.

J. Kamps (10-3), F. Niesink (28-3), J. Hulstein (7-4), K. Holtman (29-5), C.M.T. Metselaar (20-6), S. Verbaten (22-6), E. de Jong (23-6), R.W. van der Pol (14-9), Z. Zouridis (15-9), V. Coupé (27-9), G. de Haan (10-10), C.E. Areces (12-10), R.M. van Eijk (18-10), N. Peek (30-10), P. Jones (7-11), F. Waas (3-11).

R. van der Goot (January 19, 2001)

K. Hindriks (February 5, 2001) Agent Programming Languages: Programming with Mental Models. Universiteit Utrecht. Promotor: prof.dr. J.-J.Ch. Meyer, copromotoren: dr. W. v.d. Hoek and dr. F.S. de Boer. E. Smirnov (February 22, 2001)

Conjunctive and Disjunctive Version Spaces with Instance-Based Boundary Sets. Universiteit Maastricht. Promotor: Prof.dr. H.J. van den Herik M. van Someren (March 1, 2001) Learning as problem solving. Universiteit van Amsterdam. Promotor: prof.dr. B.J. Wielinga. S. Renooij (March 12, 2001) Qualitative Approaches to Quantifying Probabilistic Networks. Universiteit Utrecht. Promotoren: Prof.dr. J.-J.Ch. Meyer and Prof.dr.ir. L.C. van der Gaag, co-promotor: dr. C.L.M. Witteman.

BNVKI newsletter

145

December 2000

was designed to support the bottom-up recognition process. Unfortunately, following this approach too many ambiguities emerged. The system many times could not extract the correct information. Therefore, the bottom-up procedure was combined with a top-down procedure, in which the standard act textual procedure was followed. As we all know, notaries are very diligent persons and even if they deviate from the standard path (and set their own standards) they do so in an easily to recognize fashion. The combination turned out to be a success and the demonstration can leave its pilotproject status soon.

THE SKBS PRIZE Jaap van den Herik Universiteit Maastricht The Foundation of Knowledge-Based Systems (SKBS) has significantly supported AI research in The Netherlands in the late 1980s and the 1990s. The Foundation still exists, but employs currently a limited number of activities. A notable point to mention is that, the last Ph.D. student recently has passed the thesis defence with success. Similarly to last year, the SKBS has put up a prize for the best demonstration during the BNAIC session on applications and demonstrations. As announced in the October issue of the BNVKI newsletter, there were ten Start-up Companies which would like to demonstrate their products. For some companies Start-up should be in brackets, but what really was meant was ‘new research leading to new demonstrations’. As such, the session was well organized and a report by Luc Dehaspe, the session chair, will reveal some particularities of the demonstrating Companies.

Another contestant for the SKBS-prize was Lube Select: A fuzzy-logic decision support system for bearing lubrication selection via Internet.The program was demonstrated by Gerard Schram, from SKF Engineering & Research Centre B.V., Nieuwegein. The main idea is that appropriate (grease) lubrication is crucial in bearing applications. The knowledge incorporated was demonstrated in three groups: (1) understanding on grease lubrication, (2) showing empirical test results, and (3) showing applications experience. An example of the latter group is that the system showed to be aware of the possibility whether the lubrication was food approved. The system made a very sophisticated impression and its performance was alike. The system was to the point, as were its interfaces. The questions and answers were very well distributed over the screen and nowhere the user lost overview.

A jury consisting of Jaap van den Herik (chair), Cees Witteveen, Bert Kappen, Peter van Lith and Bas Zinsmeister (coordinator), inspected the demonstrations and made a report. During a cursory inspection of the demonstrations it became clear that Bas Zinsmeister (Bolesian) should withdraw from the Jury since one of his stable mates was in the running for the SKBS prize, i.e., Edwin Zopfi with the demo of HKT, which stands for HerKennings Technology. The demonstration had as its domain notarial acts. The idea is to extract from notarial acts all relevant information for taking a decision. Emphasis was on transfer acts and mortgage deeds. A formal grammar BNVKI newsletter

LubeSelect was designed and specified according to SDF (System Development Framework), during the period April 1999 to August 2000. Since its implementation 1400 users have received advice from this system that is internally available via Internet.

146

December 2000

As can be seen from the descriptions above, it was very hard to compare these two systems. The first system contained a large amount of AI techniques and was exactly on its place at the BNAIC where AI researchers should be made enthusiastic as they see that their ideas actually find some implementations. However, the crown on an implementation is its sophisticated use in practice, even if the AI techniques are simple (in the case of LubeSelect, it was mainly a rule-based technique, of course adequately supported by the SDF design). The Jury discussed many pros and cons, and came to the conclusion that from the SKBS point of view the cons of both demonstrations were no cons, so that only the pros remained. They were judged to be equal. Therefore the decision was made that HKT (Edwin Zopfi, Bolesian) and LubeSelect (Gerard Schram, SKF Engineering & Research Centre) should share the SKBS prize. Each of them was awarded Dfl. 500.= Reviewing all demonstrations, the jury was impressed by the quality and diversity of ideas shown by the other participants. Owing to this success, the BNAIC 2001 organizers, i.e., the University of Amsterdam (de Rijke, Kröse and Schreiber) are advised to continue the session Start-up Companies and to give them the opportunity to demonstrate their products.

BNVKI newsletter

147

December 2000

this meeting were adopted by organizers of the current workshop.

REPORT OF THE DUTCH INFORMATION RETRIEVAL WORKSHOP

That IR is a broad area was well reflected in the various presentations at the workshop. The presentations covered divers topics such as information retrieval, automatic text classification, information extraction, generating links from concept hierarchies, information filtering, cross language retrieval, and question answering. Although the focus of the workshop was clearly on ongoing research, some attention was also paid to the teaching of information retrieval. In the next paragraph a short description will be given of each presentation. Koster (University of Nijmegen) presented an evaluation of the Winnow learning technique for the classification of documents. Dufour, Ha, de Haan and van der Meer (Delft University) described a novel type of search engine. This tool, called Kascade, is an environment that enables users to structure and search (part of) the information on the internet. Verbeek (ILLC, University of Amsterdam) presented a bayesian text classification method. The method is based on word groups instead of individual words. Monz, Ragletli and De Rijke (ILLC, University of Amsterdam) presented an interesting application of information retrieval. This application generates cross-references in books automatically. Tauritz (University of Leiden) and Sprinkhuizen (University of Maastricht) presented ongoing work on information filtering. Instead of using term indexing they use n-grams as a more flexible strategy to match a query with the data. Hiemstra (University of Twente) showed that developments from other natural language processing (NLP) areas, in this case a statistical language model developed for speech recognition, can be successfully applied to various information retrieval tasks. Kraaij and Polman (TNOTPD) applied Hiemstra's model to cross language information retrieval. They

13 September 2000, Maastricht

Drs. M.J.H. Puts Nijmegen Institute for Cognition and Information University of Nijmegen Dr. I.M. de Mönnink Dept. of Language and Speech University of Nijmegen The first Dutch Information Retrieval Workshop was hosted by IKAT/the Department of Computer Science of Maastricht University. It was organized by dr. F. Wiesman and dr. R. van der Pol and sponsored by the International Institute of Infonomics (IIoI), the Foundation for KnowledgeBased Systems, and the Belgian-Dutch Association for Artificial Intelligence (BNVKI). The main goal of the workshop was to provide a meeting place where researchers in the Netherlands and Belgium working in the field of Information Retrieval could meet and present and discuss their work in an informal atmosphere. The workshop included an invited talk by prof. dr. K. Järvelin from Tampere University (for a discussion of this presentation see elsewhere in this newsletter), ten presentations from various research teams, and was concluded by a general discussion on the future of Information Retrieval (IR) in the Netherlands and Belgium.

The organisation of the workshop was influenced by what was at first a separate initiative of prof. C. Koster of the Department of Computer Science at Nijmegen University. On June 20th, he brought together various Dutch and Flemish researchers involved in IR to present their work and to discuss possible future co-operations between teams. At this meeting an overview was given by the participants of ongoing research and future plans at the different sites. Attention was also paid to the initiative taken by the University of Twente and TNO to set up a TREC style information retrieval test collection of Dutch/Flemish data in the context of CLEF (Cross-Language Evaluation Forum). The topics presented at BNVKI newsletter

the

148

December 2000

share them with other parties interested. The current workshop can form such a platform.

discussed the results of a test of three different strategies to query translation. Moens gave an overview of ongoing and future research at the Interdisciplinary Centre for Law and Information Technology of the University of Leuven. She discussed existing challenges for text categorization, text clustering, text summarization and information extraction. In his presentation, Paijmans (Tilburg University) raised the question of how to discuss issues and developments in IR in a university course, more specifically a course developed for students in an arts faculty. Finally, De Mönnink discussed plans for future research at the Department of Language and Speech of Nijmegen University. She wants to find out in what way automatic syntactic annotation (parsing) can help to improve a Question Answering system.

(2) availability of software In the same way that there is a need for shared data collections, there is also a need for an overview of (publicly available) software for Dutch, such as IR-systems, part-of-speech taggers, lemmatizers, etc. Such an overview could also be provided by the current workshop, e.g. via the internet.

The workshop was concluded by a plenary discussion. The discussion focussed around four main topics:

(3) IR in education At various sites in the Netherlands and Flanders courses on IR are provided for students from different backgrounds. It would be a good idea to join forces and develop a general course that covers most important topics in IR using the existing courses as a starting point. The course could be offered to (PhD) students in the form of a master class (e.g. a SIKS master class) or via the internet.

(1) availability and collection of data The different presentations at this workshop evidence that many people use their own data collections for their research. These data collections are often not publicly available. For Dutch, a standard data collection is required which is publicly available and which can be used by researchers for the evaluation of different IR systems. A good starting point for such a collection is CLEF. CLEF is the continuation of the cross-language task in TREC (Text REtrieval Conference). The data required for CLEF (newspaper articles) are already available. What is needed now is the evaluation of the data for different IR tasks. The question is to establish this. Where can we find the people and the money? It is important that as many people as possible support the CLEF initiative in order to get funding. In general, it is a good idea to have a platform where people with data collections can

(4) the future of "IR in the Netherlands" From the previous points in the discussion it should be clear that there is ample need of a platform for IR in the Netherlands. This platform should bring together researchers in the Netherlands and Flanders working in the area of IR. It should support a website with information on ongoing research, available data, software, education, etc. And it should organize an (annual) workshop on the topic as a continuation of the current Dutch Information Retrieval Workshop (or perhaps we should reintroduce the acronym first opted by Prof. C. Koster: IRENE (Information REtrieval in the NEtherlands)). Although the platform was not given an official status, it is our firm conviction that such a platform is indeed in existence and is formed by the participants of this workshop and the previous meeting in Nijmegen. Practical initiatives to witness

BNVKI newsletter

149

December 2000

its existence are the construction of a website by Floris Wiesman (University of Maastricht) and the organisation of the next workshop by Arjen de Vries (CWI, Amsterdam) and Djoerd Hiemstra (University of Twente). Although on a very tight time schedule, the workshop was well-organized and it provided a good opportunity for researchers working in the broad area of Information Retrieval to get acquainted with (the work of) colleagues with similar interests working at other institutes and/or companies. The workshop deserves a good follow-up. We sincerely hope that the initiatives developed during this day will be realized.

BNVKI newsletter

150

December 2000

InQuery's special #syn-operator, which constructs on-the-fly synonyms from its operands. Illustrating this idea, Järvelin showed how the German word `Drogenpolitik' (definitely chosen in honour of our country's occurrence in the topic set) is transformed into the following expression:

BILINGUAL IR IN FINLAND: TWO TALKS BY JÄRVELIN by Arjen P. de Vries

More than a year after my short visit to Tampere, I suddenly got the opportunity to meet Kalervo Järvelin twice within a week: first at the Dutch workshop on IR (also described in this issue), followed by our joined attendance to the Cross-Language Evaluation Forum (CLEF) workshop in Portugal.

#syn( #4(drug policy) politics) #4(drug tactics) )

In this expression, the first #4-operator denotes an occurrence of the words `drug' and `policy' within a window of four words; the full expression therefore specifies informally that `drug policy' is synonymous of `drug politics' and `drug tactics'. Their experiments demonstrate a beneficial effect from using these operators to structure the query; especially for the Finnish-English bilingual runs this improves the quality of the results significantly (measured in average precision).

Järvelin chairs FIRE, the retrieval group at University of Tampere, Finland, specializing in experimental information retrieval research. For CLEF, they performed a series of bilingual information retrieval experiments; a task measuring the quality of English documents retrieved from topics (i.e. queries) stated in another language (non-English). The distinguishing aspect of their submitted bilingual runs is the usage of Finnish and Swedish topics, aside from the `normal' German topics.

Looking into the future of the next (2001) evaluation, Tampere intends to improve upon the default handling of proper names and other out-of-dictionary words. A problem to be tackled for most languages is the transliteration problem, like `Paris' becoming `Parijs' in Dutch. Handling proper names is even more complex in Finnish, due to inflectional forms of proper names; which cannot be normalized easily, because the morphological analysis programs cannot identify proper names. Also, Järvelin revealed plans to extend the CLEF collections with a Finnish (and possibly a Swedish) test collection. These plans were supported warmly by the Dutch participants, who intend to pursue a similar effort to create a Dutch collection as part of CLEF.

Like most participants, their system uses a dictionary-based approach: after looking up the translations of the topic words in a (preferably high-quality) bilingual dictionary, the translated query is submitted to a `standard' IR system. Given the characteristics of the three chosen topic languages, Järvelin and his researchers focused on addressing the compounding problem: reducing the number of out-ofdictionary words by splitting long words such as the German word `Informationsdienst' into its compounds `information' and `dienst'. After translating the topics into English queries, the InQuery retrieval engine (developed at University of Massachusetts at Amherst) retrieves the (hopefully) relevant documents.

Before I forget, here are the results of this year's evaluation (of course, CLEF is not a contest but a scientific evaluation forum... they say). In the official rankings, Tampere

The Tampere researchers concentrated on structuring the translated queries using BNVKI newsletter

#4(drug

151

December 2000

came out fourth with their German-English Bilingual runs. Unfortunately, this was one position above my own Dutch-English runs. TNO and University of Twente, who also attended both workshops, ranked second. Further information: http://galileo.iei.pi.cnr.it/DELOS/CLEF/

BNVKI newsletter

152

December 2000

SECTION COMPUTATIONAL LINGUISTICS Section-Editor Antal van den Bosch

Section Editor Richard Starmans

Coördinator Research School SIKS Dr. R.J.C.M. .Starmans P.O. Box 80089 3508 TB UTRECHT Tel: (030) 2534083/1454 Email: [email protected]

BNVKI newsletter

153

December 2000

iemand betrapt wordt tijdens het openbreken van een slot van bijvoorbeeld een brommer, dan kan hij, ook al is hij er uiteindelijk niet in geslaagd de brommer te stelen, strafbaar worden gesteld voor een poging tot het stelen van een brommer. De maximale hoogte van de straf wordt in dat geval wel met een derde verminderd (max. 2 jaar en acht maanden in plaats van 4 jaar).

SECTION KNOWLEDGE SYSTEMS IN LAW AND COMPUTER SCIENCE Section-Editor Radboud Winkels

Stel nu dat Hell A. een brommer uit de tuin van zijn buren meeneemt, met de bedoeling deze brommer te stelen. Echter, zonder dat hij daar weet van heeft, blijkt zijn vrouw die brommer eerder die dag al gekocht te hebben. Is deze dief strafbaar, is er hier sprake van een poging tot diefstal? Voor het beantwoorden van deze vraag maakt het uit welke pogingsleer men aanhangt. Als men de subjectivistische leer aanhangt, kijkt men enkel naar de intentie van de dader. Feitelijkheden van het geval spelen geen rol. In dit geval was de intentie misdadig en zou Hell A. dus strafbaar te zijn. Bij de objectivistische leer wordt juist naar de feitelijkheden gekeken. Hier had Hell A. de eigendom van de brommer al doordat zijn vrouw de brommer al gekocht had (er vanuit gaande dat ze in gemeenschap van goederen getrouwd waren). Uiteraard is het stelen van jezelf geen diefstal (fiscalisten zullen dit misschien tegenspreken), dus Hell A. is niet strafbaar.

HOW TO FOOL A LAWYER

JURIX-bijeenkomst 15 mei 2000, KUB Tilburg Jaap Dijkstra

Arno R. Lodder Computer/Law Institute, Vrije Universiteit Amsterdam [email protected] http://www.rechten.vu.nl/~lodder Sinds kort is Jaap Dijkstra als universitair docent Rechtsinformatica verbonden aan de RUG (groep van Kees de Vey Mestdagh). Daarvoor was hij twee jaar werkzaam bij het Telematicainstituut in Enschede. In de periode 1992-1997 heeft hij promotieonderzoek gedaan. Dit onderzoek vond plaats bij de werkgroep Sociaal-Wetenschappelijke Informatiekunde (SWI) en had als doel gebruikersaspecten van kennissystemen te bestuderen. Een van de experimenten handelde over juridische expertsystemen en droeg de provocatieve titel ‘How to fool a lawyer’. Het bleek namelijk dat juristen vrij klakkeloos de uitkomsten van twee niet helemaal goed werkende expertsystemen overnamen. Tijdens een bijeenkomst van JURIX op een zonnige maandag halverwege mei, deed hij verslag van dit onderzoek.

Dijkstra heeft twee expertsystemen gebouwd. In de ene representeerde hij de subjectivistische leer, in de andere de objectivistische leer. Het geldende Nederlandse recht neemt op dit punt een tussenpositie: een mengsel van de subjectivistische en objectivistische leer. Een casus oplossen vanuit enkel een van beide invalshoeken zal doorgaans niet tot de juiste oplossing leiden.

Binnen het recht zijn niet alleen voltooide delicten (een geslaagde moord, een gelukte diefstal) strafbaar, maar is het ook mogelijk dat daders in geval van een niet voltooid delict gestraft worden. Wanneer BNVKI newsletter

In het experiment werden 30 juristen (medewerkers RUG) opgesplitst in 2 154

December 2000

groepen. De ene groep beoordeelde 9 casus daarbij ondersteund door een objectivistisch expertsysteem (ES-obj), de andere groep kreeg hulp van een subjectivistisch expertsysteem (ES-subj). Voor elke casus moest aangegeven worden of de verdachte strafbaar was. De juristen die hulp hadden gekregen van ES-sub vonden de verdachte in 95 van de 135 gevallen strafbaar. De juristen die hulp hadden gekregen van ES-obj vonden de verdachte slechts in 20 van de 135 gevallen strafbaar. Zoals boven al bleek is iemand volgens de subjectivistische leer eerder strafbaar, vandaar de verschillen. De gebruikers namen vrij gemakkelijk, vaak volgens het geldende recht onjuiste, adviezen van het expertsysteem over.

Een op dit moment wellicht lastiger probleem dan de hierboven geschetste is hoe juristen zover te krijgen kennissystemen ook daadwerkelijk te gebruiken. Doorgaans blijken juridische kennissystemen niet of nauwelijks gebruikt te worden in de praktijk. Om dit enkel aan de onwil van juristen te wijten is te gemakkelijk, hoewel ook op dat punt winst te halen is. Misschien moeten we accepteren dat juristen niet de juiste doelgroep zijn voor kennissystemen. Als het domein eenvoudig is, kan een kennissysteem gemaakt worden maar voegt het voor de jurist niet veel toe. Als het domein lastig is, zou een kennissysteem voor de jurist veel toe kunnen voegen maar in dat geval is het maken van een kennissysteem niet (eenvoudig) mogelijk. Juridische kennissystemen (waarin eenvoudige domeinen zijn gerepresenteerd) zullen vooral voor niet juristen of juristen in wording (studenten) nuttig blijken. Het op een eenvoudig toegankelijke manier ontsluiten van grote hoeveelheden informatie lijkt voor de juristen althans op dit moment prioriteit te hebben. Maar, wat niet is kan nog komen…

Er was enige discussie over wat het experiment nu precies had aangetoond. Dat een expertsysteem als een betrouwbare bron wordt gezien? Dat mensen (in casu juristen) makkelijk te beïnvloeden zijn? Dat de bereidheid uitkomsten van expertsystemen inhoudelijk te controleren niet erg groot is? Een positieve uitleg is dat wanneer er goede expertsystemen gebouwd worden, de uitkomsten daarvan ook overgenomen zullen worden. In ieder val zijn er de volgende problemen met gebruikers van expertsystemen: • advies van expertsysteem wordt nauwelijks op inhoud gecontroleerd; •

groot (misplaatst) vertrouwen in expertsysteem;

•

gebruiker kijkt niet verder dan de modellering die in het expertsysteem is opgenomen (‘Tunneling effect’)

END SECTION KNOWLEDGE IN LAW AND COMPUTER SCIENCE

De oplossingen voor deze problemen moeten in de volgende richtingen gezocht worden (zie ook oratie van Anja Oskamp cf. Dijkstra): transparante systemen, simpele systemen, kundige gebruikers (lees: juristen), verantwoorde gebruikers. Kortom, er is meer dan hoop voor kennissystemen in het recht. BNVKI newsletter

155

December 2000

CALL FOR PAPERS

FIRST GENERAL CALL FOR PAPERS, OCT.-NOV. 2000 EUROPEAN SOCIETY FOR THE STUDY OF COGNITIVE

18th ANNUAL WORKSHOP, 18 (14.00 hrs) - 20 (17.00 hrs) May 2001, Maastricht, The Netherlands Multidisciplinary approach of cognition: Modelling, experiments, and applied research in the fields of: • psychology (cognitive, developmental), perception, • artificial intelligence (general aspects), • associative memory and neural networks, neuroscience, • linguistics (also computational), disorders of language, • education and instruction, • cognitive ergonomics, • philosophy, history of concepts. Fee NLG 70 (Dutch guilders, about 30 Euro for members of the ESSCS, NLG 90 (40 Euro) for others, students NLG 35 (16 Euro). For more information see webpage (to be updated 5 Nov. 2000):

http://www.ppsw.rug.nl/~esscs E-mail: A block booking of rooms has been made till 15 April 2001, later bookings may be possible. Address

and

telephone

numbers:

ESSCS Dept. of Psychology (E&A), University of Groningen Grote Kruisstraat 2/1, 9712 TS Groningen, The Netherlands) Telephone Netherlands +31-50-3636448 (G.J.Dalenoort) or +31-50-3636454, or 3636472, Fax +31-50-3636304

BNVKI newsletter

156

December 2000

June 18-23, 2001 The International Joint Conference on Automated Reasoning, Siena, Italy. Information: http://www.dii.uni-si.it/~ijcar/.

CONFERENCES, SYMPOSIA WORKSHOPS

June 26-29, 2001 Fourth International ICSC Symposia on Soft Computing and Intelligent Systems for Industry. Paisley, Scotland, U.K. Information: Email: [email protected]

Below, the reader finds a list of conferences and web sites or email addresses for further information. A more extensive list of conferences can be found in the Calendar 2000, as published in AI Communication and in the SIGART Newsletter.

July 28-29, 2001 Fourth Pacific Rim International Workshop on MultiAgents (PRIMA’2001). Grand Hotel Taipei, Taiwan. Information: http://www.lab7. kuis.kyoto-u.ac.jp/ prima2001/

December 12-15, 2000 Intelligent Systems and Applications (ISA’2000) University of Wollongong (near Sydney), Australia Information: http://www.icsc.ab.ca/isa2000.htm

August 20-24, 2001 Workshop ESSLLI XIII, Helsinki Information: http://www.lili.uni-bielefeld.de/~pkuehnle/ HELSINKI

December 14-15, 2000 JURIX 2000: Thirteenth International Conference on Legal Knowledge-Based Systems, Enschede, The Netherlands. Information: http://www.jurix.nl/jurix2000.html

September 5-7, 2001 IEEE Symposia on Human-Centric Computing Languages and Environments (HCC’01) Stresa, Italy. Information: http://cuising.unige.ch/HCC01.

March 22-24, 2001 European Conference on Computer-Supported Collaborative Learning Maastricht (The Netherlands). Information: http://www.mmi.unimaas.nl/euro-cscl. April 2-4, 2001 Workshop 2001 Agent Based Simulation II, Passau, Germany. Information: http://www.or.uni-passau.de/workshop2001.

September 6, 7 & 8, 2001 KES’2001. The Fifth International Conference on Knowledge-Based Intelligent Information Engineering Systems & Allied Technologies. Osaka-Kyoika University, Osaka, Japan (Sept.6 & 7) Nara-Ken New Public Hall, Nara, Japan (Sept.8) Information: http://www.brighton.ac.uk/kes/journal.

April 4-6, 2001 Fifth Dutch-German Workshop on Nonmonotonic Reasoning. DGNMR’01. Information: http://www.cs.uni-potsdam.de/DGNMRO1/

September 11-14, 2001 27th International Conference on Very Large Data Bases. Roma, Italy. Information: http://www.dia.uniroma3.it/vldb2001

April 25-26-27, 2001 ESANN’2001, 9th European Symposium on Artificial Neural Networks. Novotel Hotel, Katelijnstraat 65 B, 8000 Brugge, Belgium. Information: http://www.dice.ucl.ac.be/esann.

September 26-28, 2001 ASCM 2001, The Fifth Asian Symposium on Computer Mathematixs. Ehime University, Matsuyama, Japan. Information: http://www.hpc.cs.ehime-u.ac.jp/ãscm October 23-26, 2001 The Second Asia-Pacific Conference on Intelligent Agent Technology (IAT-2001). Maebashi Terrsa, Maebashi City, Japan. Information: http://kis.maebashiit.ac.jp/iat01

May 21-25, 2001 8th International Conference on Artificial Intelligence and Law. St. Louis, Missouri (USA). Submission deadline: January 12, 2001. Information: http://www.cs.wustl.edu/icail2001/ May 28-June 1, 2001 The fifth International Conference on Autonomous Agents. Montreal, Canada. Information: http://www.csc.liv.ac.uk/~agents2001/

November 29 – December 2, 2001 ICDM ’01: The 2001 IEEE International Conference on Data Mining. Silicon Valey, California, USA. Information: http://kais.mines.edu/~xwu/icdm/icdm01.html

June 10-13, 2001 7th International Meeting on DNA Based Computers. University of South Florida, Tampa FL. Information: http://www.cas.usf.edu/dna7/

January 15-18, 2002 First International ICSC Congress on Neuro-Fuzzy NF’2002. The Capitolio de la Habana, Cuba. Information: http://www.icsc.ab.ca/NF2002.htm.

June 14-16, 2001 Fifth Workshop on the Semantic and Pragmatics of Dialoque. Bielefeld University, Germany. Information: http://www.uni-bielefeld.de/BIDIALOG/

BNVKI newsletter

157

December 2000

Joris Van Looveren Vrije Universiteit Brussel, AI Lab Pleinlaan 2, 1050 Brussel, Belgium. Tel.: +32 (0) 6293702. E-mail: [email protected]

MAIL ADRESSES BOARD MEMBERS BNVKI Prof.dr. J. N. Kok Wiskunde en Natuurwetenschappen, Dept. of Computer Science Universiteit Leiden, Niels Bohrweg 1, 2333 CA Leiden Tel: (071) 5277057. E-mail: [email protected]

Dr. A. van den Bosch Katholieke Universiteit Brabant, Taal- en Literatuurwetenschap, Postbus 90153, 5000 LE Tilburg Tel.: (013) 4360911. E-mail: [email protected]

Dr. Y.H. Tan EURIDIS, Erasmus Universiteit Rotterdam Postbus 1738, 3000 DR Rotterdam Tel.: (010) 4082255. E-mail: [email protected]

Dr. R.J.C.M. Starmans Coordinator Research school SIKS, P.O. Box 80089, 3508 TB, Utrecht Tel.: (030) 2534083/1454. E-mail: [email protected]

Dr. E.O. Postma Universiteit Maastricht, IKAT Postbus 616, 6200 MD Maastricht Tel.: (043) 3883493. E-mail: [email protected]

Dr. B. de Boer Vrije Universiteit Brussel, AI Lab Pleinlaan 2, B-1050 Brussel, Belgium Tel.: +32 (0)2 6293703. E-mail: [email protected]

Dr. R. Verbrugge Rijksuniversiteit Groningen, Cognitive Science and Engineering Grote Kruisstraat 2/1, 9712 TS Groningen. Tel.: (050) 3636334. E-mail: [email protected]

HOW TO SUBSCRIBE The BNVKI/AIABN Newsletter is a direct benefit of membership in the BNVKI/AIABN. Membership dues are Fl. 75,- or BF 1.400 for regular members; NLG 50,- or BF 900 for doctoral students (AIO's); and NLG 40,- or BF 700 for students. In addition, members will receive two issues of the European journal AI Communications. The newsletter appears bimonthly and contains information about conferences, research projects, job opportunities, funding opportunities, etc., provided enough information is supplied. Therefore, all members are encouraged to send news and items they consider worthwhile to the editorial office of the BNVKI/AIABN newsletter. Subscription is done by payment of the membership due to RABO-Bank no. 11.66.34.200 or Postbank no. 3102697 for the Netherlands, or KBC Bank & Verzekering Veldwezelt No. 457-6423559-31, 20 Carabinierslaan 104, Belgie. In both cases, specify BNVKI/AIABN in Maastricht as the recipient, and please do not forget to mention your name and address. Sending of the BNVKI/AIABN newsletter will only commence after your payment has been received. If you wish to conclude your membership, please send a written notification to the editorial office before December 1 1999.

Dr. W. van der Hoek Universiteit Utrecht, Department of Computer Science, P.O. Box 80089, 3508 TB Utrecht Tel.: (030) 2533599. E-mail: [email protected] Dr. L. DeHaspe Katholieke Universiteit Leuven Dept. of Computer Science, Clestijnenlaan 200A,, B-3001 Heverlee, België Email:ldh@[email protected] G. Beijer BOLESIAN BV, Steenovenweg 19, 5708 HN Helmond Tel.: (0492) 502525. E-mail: [email protected] Dr. W. Daelemans Katholieke Universiteit Brabant, Vakgroep TaalLiteratuurwetenschap, Postbus 90153, 5000 LE Tilburg. Tel.: (013) 4663070. E-mail: [email protected]

en

EDITORS BNVKI newsletter

COPY

Dr. E.O. Postma (editor in chief) (See addresses Board Members)

The editorial board welcomes product announcements, book reviews, product reviews, overviews of AI research in business, and interviews. Contributions stating controversial opinions or otherwise stimulating discussions, are higly encouraged.

Prof. dr. H.J. van den Herik Universiteit Maastricht, IKAT Postbus 616, 6200 MD Maastricht Tel.: (043) 3883485. E-mail: [email protected] Dr. C. Witteveen Technische Universiteit Delft, Department Informatica, Julianalaan 132, 2628 BL Delft Tel.: (015) 2782521. E-mail: [email protected]

ADVERTISING It is possible to have your advertisement included in the BNVKI/AIABN Newsletter. For further information about pricing etc., please contact the editorial office.

Technische

CHANGE OF ADDRESS

Dr. R.G.F. Winkels Universiteit van Amsterdam, Rechtsinformatica Postbus 1030, 1000 BA Amsterdam Tel.: (020) 5253485. E-mail: [email protected]

The BNVKI/AIABN newsletter is sent from Maastricht. The BNVKI/AIABN board has decided that the BNVKI/AIABN membership administration takes place at the editorial office of the Newsletter. Therefore, please send address changes to: Editorial Office BNVKI/AIABN Newsletter Universiteit Maastricht, FdAW, Vakgroep Informatica, Postbus 616, 6200 MD Maastricht, Nederland, Tel: ++31-(0)43-3883477 E-mail: [email protected]

Dr. S.-H. Nienhuys-Cheng Erasmus Universiteit Rotterdam, Informatica Postbus 1738, 3000 DR Rotterdam Tel.: (010) 4081345. E-mail: [email protected]

BNVKI newsletter

158

December 2000