The field of Intelligent Virtual Agents (IVAs) has evolved immensely with respect to design and development of agents over the years. This has brought the following questions: What processes and methods are used to design virtual agents, and in particular, to what extent and how are users involved in the design process of virtual agents? In this paper, we review papers from the conference “Intelligent Virtual Agents” for the last five years (2015 - 2019) to shed light on these questions. The review included 308 short and long papers, with a focus on the interactive aspects of design. The review showed that only 14% of 308 papers explicitly mentioned/referred to user participation during the design of an IVA. User involvement is classified into two categories: one-time and iterative. The few studies that mention design only report on the use of standard Human-Computer Interaction (HCI) design methods to a rather limited degree.

Since Bacil Kirtley in 1958 proposed that Bram Stoker’s Count Dracula, the best known literary character ever, shared his historical past with the Wallachian Voivode Vlad III Dracula, an intense debate about this connection has developed and other candidates have been suggested, like the Hungarian General János Hunyadi – a proposal resurfacing in the most recent annotated Dracula edition by Leslie Klinger (2008). By close-reading Stoker’s sources, his research notes and the novel, I will demonstrate that Stoker’s narrative initially links his Count to the person of Vlad III indeed, not Hunyadi, although the novelist neither knew the ruler’s first name, nor his father’s name, nor his epithet “the Impaler”, nor the cruelties attributed to him.

Still – or maybe for this very reason – Stoker did not wish to uphold this traceable identity: In Chapter 25, shortly before the decisive chase, he removes this link again, by way of silent substitution, cloaked by Professor van Helsing’s clownish distractions. Like the Vampire Lord Ruthven, disappearing through the “vampire trap” constructed by James R. Planché for his play The Brides of the Isles in the English Opera House, later renamed to Lyceum Theatre and run by Stoker, the historical Voivode Vlad III Dracula is suddenly removed from the stage: In the final chapters, the Vampire Hunters pursue a nameless double.

Smoothly performed, this piece of stage magic has gone unnoticed for more than one hundred years now. As a consequence, most of the arguments related to the Count’s lifetime identity turn out to suffer from ignoratio delenchi (the fallacy of irrelevant conclusion). The “marriage” of Count Dracula and Vlad the Impaler needs no divorce, as filed for by Toronto Prof. Em. Elizabeth Miller in 1998: As Stoker revoked this bond before his book went to print, it was never consummated and can be annulled without much ado.

Bram Stoker's novel Dracula was published in 1897. In ten weeks, the centenary of Stoker's death will be commemorated. Today, Dracula is one of the best-known narratives worldwide. Since the 1970's, a wealth of secondary literature has developed. Several annotated editions of the novel have been published, as well as Stoker's personal research notes. The location of the vampire's headquarters, Castle Dracula, however, could not be reconstructed until now. Rather, it is generally assumed Stoker had no exact position in mind while writing the story.

Based on the novel's travel descriptions, this paper will narrow down the Castle's territory with the help of modern satellite-based maps, interactive route navigators and high-resolution scans of contemporary military maps of Transylvania and Moldavia, until only two mountain peaks remain. An analysis of Stokers notes will reveal that Stoker's actually had one of these two peaks in mind while designing the final chapters of his book, but choose not to mention its name:

Mount Izvorul Călimanului, 2,033 m high, located in the Transylvanian Kelemen Alps near the border with Romania, at 47°08'03" North, 25°17'19" East.

Good tools are critical to successful examination of digital evidence, yet the quality of such tools is very often an unknown. Independent testing is currently the only way to assess the performance and quality of the tools available to digital forensic examiners, and independent testing is critical in order to know whether tools deliver what they promise, and to assess the level of trust one can place in them.   This project has aimed to thoroughly assess the quality of the disk imaging functionality in EnCase 6.8 and LinEn 6.1 under a variety of conditions and with respect to the requirements of Swedish law enforcement. The project has been conducted in four phases: survey, requirements, test planning and testing and analysis.   Details of the project results are published in separate reports related to each phase. The overall conclusions are as follows:

• No independent, public evaluations of EnCase version 6 other than this one exist.
• EnCase 6.8 performs as expected on the Windows platform when using a hardware write blocker. Operation without a write blocker was not assessed, as this was not a requirement. Hidden sectors were, as expected, not acquired.
• LinEn 6.1 performs as expected only when operating in “BIOS mode”, but fails many tests when operating under “direct ATA mode”. Hidden sectors were, as expected, not acquired in “BIOS mode”.

Based on the test results, we conclude that provided the limitation of not acquiring hidden sectors is acceptable, EnCase 6.8 is an adequate tool for imaging ATA drives when using a hardware write blocker, whereas LinEn 6.1 should be avoided if complete and accurate acquisition is a desired.

All the work was conducted by experienced staff in test labs belonging to the Security and Networks group within the Division for Database and Information Techniques (ADIT) at the Department of Computer and Information Science (IDA) at Linköping University under the supervision of Professor Nahid Shahmehri.

Linköping, March 5, 2008

Denna rapport ger en bild av potential och förutsättningar för att utveckla former för energirådgivning som tyglar och kanaliserar den långa svansen, som tar ett steg mot att komplettera dagens expertbaserade energirådgivning med olika former av användargenererad information och rådgivning. Den har som ambition att utifrån forskningsrön och praktiska exempel väcka tankar om hur den statsfinansierade energirådgivningen skulle kunna röra sig i riktning mot web 2.0-former. Exemplet energirådgivning kan förhoppningsvis också väcka tankar avseende andra ämnesområden.

Denna rapport beskriver slutsatserna från en studie av Stockholms Läns Landstings satsning på elektronisk kontakt mellan invånare och vården. Rapporten bygger på kontakter med projektgruppen och projektägaren, vårdpersonal på vårdcentraler (intervjuer och enkäter), användningsstatistik ur systemet, seminarier med företrä­dare för andra svenska landsting, studium av websajter i utlandet, intervju med projektledaren för en intressant amerikansk satsning och med en svensk utvärderare av svenska företags kundtjänstef­fektivitet. Studien var en del i ett flerårigt forskningsprojekt med samarbete mellan landstinget och forskarna Mats Edenius, Han­delshögskolan i Stockholm och Alf Westelius och finansiering från VINNOVA och landstinget. Denna rapport är författad av Alf Westelius i maj 2004. Andra publikationer från projektet är Alf Westelius och Mats Edenius, Gaps between intended and actual use – turning problems into opportunities in health informatics. Inter­national Journal of Public Information Systems, (2006), Vol 2 (2). Tillgänglig på http://www.ijpis.net; Mats Edenius och Alf Westelius, E-service and Knowledge Formation: the Use of a Web-based E-Messaging System in HealthCare. Kapitel 1 i Managing Customer Relationships on the Internet, Angelika Lindstrand, Jan Johanson, Dharma Sharma (Red.), Elsevier, 2006, ISBN: 0-08-044124-6; samt Mats Edenius och Alf Westelius, Patients’ knowl­edge formations through a healthcare e-messaging system, Journal of Human Resource Costing & Accounting, 2004, Vol 8 (1) 21-34.

This article is based on a longitudinal case study of a major, complex, multinational Enterprise Systems venture in a growing, international group. The process covers a decade, and is interesting both because of the differences between different groupings and companies in the Group, and because the Enterprise System implementation is an important part of a larger, strategic reorientation in the Group. Standardisation, a benefit normally associated with Enterprise Systems ventures, is problematised and explored. It can serve efficiency and learning, but also has implications for the power distribution and can give rise to conflicts. The shifting rationale for the project that seems to lead its own life is analysed and the roles of important actors in conducting, guiding and supporting the project are discussed. A finding is that the project participants, rather than being representatives of their “home organisation”, acting on its behalf and having its approval, became aliens, representing the project, its solutions and values. A proposition derived from this case is that benefits from Enterprise System implementation can stem from the learning the Enterprise System project has generated in the organisation and from specific applications made possible by the common platform the Enterprise System constitutes, rather than from the use of the Enterprise System in itself.

Level set methods [OS88] have proved very successful for interface tracking in many different areas of computational science. However, current level set methods are limited by a poor balance between computational efficiency and storagerequirements. Tree-based methods have relatively slow access times, whereas narrow band schemes lead to very large memory footprints for high resolution interfaces.   In this paper we present a level set scheme for which both computational complexity and storage requirements scale with the size of the interface. Our novel level set data structure and algorithms are fast, cache efficient and allow for avery low memory footprint when representing high resolution level sets. We use a time-dependent and interface adapting grid dubbed the .Dynamic Tubular Grid. or DT-Grid. Additionally, it has been optimized for advanced nite difference schemes currently employed in accurate level set computations. As a key feature of the DT-Grid the associated interface propagations are not limited to any computational box and can expand freely. We present several numerical evaluations, including a level set simulation on a grid with an effective resolution of 10243.

Over the last decade, highly accurate mobile 3D measurement technologies have become available and are widely used now in industry and entertainment. In the cultural heritage field, various 3D pilot projects for conservation and restauration purposes have been initiated or completed already. My Digital Sculpture Project, started in November 2001, focuses on establishing an efficient workflow for creating high-resolution geometric models of both small and large plaster, terra-cotta and bronze sculptures, with a limited budget and a small support team, in improvised, non-laboratory situations and within narrow time windows, as encountered in the course of a significant series of museum visits all over Europe. Specific requirements and scanning strategies for scanning complex sculptures are discussed, along with a series of scanner tests. The article explains possible applications of 3D documentation methods, especially their relevance for comparative morphological analysis and issues of originality and authenticity. To demonstrate the use of 3D difference maps, this paper presents an exact and objective comparison of two monumental plasters of Auguste Rodin's 'Thinker', located in France and Poland respectively, conducted in December 2003.

This paper proposes a computational model for children's word acquisition based on inductive logic programming. There are three fundamental features in our approach. Firstly, we incorporate cognitive biases developed recently to explain the efficiency of children's language acquisition. Secondly, we design a co-evolution mechanism of acquiring concept definitions for words and developing concept hierarchy. Concept hierarchy plays an important role of defining contexts for later word learning processes. A context switching mechanism is used to select a relevant set of attributes for learning a word depending on the category which it belongs to. On the other hand, during acquiring definitions for words, concept hierarchy is developed. Thirdly, we pursue resemblance to human brain in functional level.We developed an experimental language acquisition system called WISDOM (Word Induction System for Deriving Object Model) and conducted virtual experiments or simulations on acquisition of words in two different categories. The experiments shows feasibility of our approach.

There are a variety of ways to represent information and each representation scheme typically has associated tools to manipulate it. In this paper, we present a single, generic representation that can accommodate a broad range of information representation schemes (i.e., structural models), such as XML, RDF, Topic Maps, and various database models.We focus here on model-based information where the information representation scheme prescribes structural modeling constructs (analogous to a data model in a database). For example, the XML model includes elements, attributes, and permits elements to be nested. Similarly, RDF models information through resources and properties.

Having a generic representation for a broad range of structural models provides an opportunity to build generic technology to manage and store information. Additionally, we can use the generic representation to exploit a formally defined mapping language to transform information, e.g., from one scheme to another. In this paper, we present the generic representation and the associated mapping formalism to transform information and discuss some of the opportunities and challenges presented by this work.

The key to applying machine inference to scientific problems is to have an appropriate representation of the problem. The most successful applications of inductive logic programming (ILP) have been in the field of learning molecular quantitative structure activity relationships (QSARs). The basic representation used in these applications has been based on atoms linked together by bonds. This representation, although successful and intuitive, ignores the most basic discovery of 20th century chemistry, that molecules are quantum mechanical objects. Here we present a new representation using quantum topology (StruQT) for machine inference in chemistry which is firmly based on quantum mechanics. The new representation is based on Richard F.W. Bader's quantum topological atoms in molecules (AIM) theory. Central to this representation is the use of critical points in the electron density distribution of the molecule. Other gradient fields such as the Laplacian of the electron density can also be used. We have demonstrated the utility of the use of AIM theory (in a propositional form) in the regression problem of predicting the lowest UV transition for a system of 18 anthocyanidans. The critical point representation of fields can be easily mapped to a logic programming representation, and is therefore well suited for ILP. We have developed an ILP-StruQT method and are currently evaluating it for QSAR problems.

The main point of proposed approach to study ants cognitive abilities is that our experiments provide a situation in which insects have to transmit information quantitatively known to the experimentalist in order to obtain food. One may estimate some properties of ant intelligence by measuring complexity of tasks they solve in order to pass definite pieces of information from scouts to foragers.Our previous experiments, basing on ideas of Information Theory, have shown that ants are able to memorize and transmit messages concerning sequence of turns toward a trough of syrup and use the simplest regularities to compress the information. To reveal counting and number related skills, we suggested red wood ants Formica polyctena to transmit information on the number and coordinates of objects. One of the experimental set-ups consisted of a tree trunk with branches that ended in empty troughs, except for one which was filled with syrup. Another set-up consisted of a lattice which simulated Cartesian coordinates. The foragers of F. polyctena were separated into teams of 5-8 individuals , each with one scout. All laboratory ants were marked with coloured labels. To start the experiment, an ant scout was placed at the randomly numbered trough containing food and then returned to the nest on its own. The duration of the contact between foragers and the scout was measured. Then we removed the scout and the foragers had to search for the food by themselves. The experiments were so devised as to eliminate all possible ways that may help to find food, except for distant homing. It turns out that the ants are able to count within several tens, and transmit this information to their nestmates. The analysis of time duration of ants contacts enable us to create a hypothesis of how they use numbers and coordinates in their communication. We suppose that only a few highly social ant species use such a complex communication system based on cognitive processes. At the same time, we believe that the experimental schemes described can be used to study the communication systems and numerical competence of other animals.

We investigate a few approaches that have been considered in the simulation and modeling of networks describing cell behavior. By simulation it is meant the direct problem of determining cell behavior when given a graph (network) specifying the interaction among genes. By cell behavior we mean determining the amount of byproducts (mRNA or protein) that each gene generates with time as it interacts with other genes. We refer to modeling as the inverse problem namely, inferring the network graph when given the data describing the cell's behavior. The modeling problem has acquired significant importance in view of the present high volume of cell data available from micro-array experiments. The emphasis of the paper is in using the constraint logic programming paradigm to describe the simulation of cell behavior. In that paradigm the same program describes both a problem and its inverse. Basically one defines multi-dimensional regions, transitions (specifying how control is transferred from one region to the other), and trajectories (sequences of transitions describing cell behavior). The paradigm is applied to several approaches that have been proposed to study simulation and modeling. Several logic programs have been developed to prototype those approaches under the same proposed paradigm. They include considering Boolean and discrete domains. In each case the potential of obtaining practical solutions to the inverse problem are discussed. The proposed paradigm is related to machine learning and to the synthesis of finite-state automata.

This paper investigates whether Chomsky-like grammar representations are useful for learning cost-effective, comprehensible predictors of members of biological sequence families. The Inductive Logic Programming (ILP) Bayesian approach to learning from positive examples is used to generate a grammar for recognising a class of proteins known as human neuropeptide precursors (NPPs). Collectively, five of the co-authors of this paper, have extensive expertise on NPPs and general bioinformatics methods. Their motivation for generating a NPP grammar was that none of the existing bioinformatics methods could provide sufficient cost-savings during the search for new NPPs. Prior to this project experienced specialists at SmithKline Beecham had tried for many months to hand-code such a grammar but without success. Our best predictor makes the search for novel NPPs more than 100 times more efficient than randomly selecting proteins for synthesis and testing them for biological activity. As far as these authors are aware, this is both the first biological grammar learnt using ILP and the first real-world scientific application of the ILP Bayesian approach to learning from positive examples.A group of features is derived from this grammar. Other groups of features of NPPs are derived using other learning strategies. Amalgams of these groups are formed. A recognition model is generated for each amalgam using C4.5 and C4.5rules and its performance is measured using both predictive accuracy and a new cost function, Relative Advantage (  RA  ). The highest  RA  was achieved by a model which includes grammar-derived features. This  RA  is significantly higher than the best  RA  achieved without the use of the grammar-derived features. Predictive accuracy is not a good measure of performance for this domain because it does not discriminate well between NPP recognition models: despite covering varying numbers of (the rare) positives, all the models are awarded a similar (high) score by predictive accuracy because they all exclude most of the abundant negatives.

It remains unclear what principles underlie a protein sequence/structure adopting a given fold. Local properties such as the arrangement of secondary structure elements adjacent in sequence or global properties such as the total number of secondary structure elements may act as a constraint on the type of fold that a protein can adopt. Such constraints might be considered "signatures" of a given fold and their identification would be useful for the classification of protein structure. Inductive Logic Programming (ILP) has been applied to the problem of automatic identification of structural signatures. The signatures generated by ILP can then be both readily interpreted by a protein structure expert and tested for their accuracy. A previous application of ILP to this problem indicated that large insertions/deletions in proteins are an obstacle to learning rules that effectively discriminate between positive and negative examples of a given fold. Here, we apply an ILP learning scheme that reduces this problem by employing the structural superposition of protein domains with similar folds. This was done in three basic steps. Firstly, a multiple alignment of domains was generated for each type of fold studied. Secondly, the alignment was used to determine the secondary structure elements in each of those domains that can be considered equivalent to one another (the "core" elements of that fold). Thirdly, an ILP learning experiment was conducted to learn rules defining a fold in terms of those core elements.

We aim to partially automate some aspects of scientific work, namely the processes of forming hypotheses, devising trials to discriminate between these competing hypotheses, physically performing these trials and then using the results of these trials to converge upon an accurate hypothesis. We have developed ASE-Progol, an Active Learning system which uses Inductive Logic Programming to construct hypothesised first-order theories and uses a CART-like algorithm to select trials for eliminating ILP derived hypotheses. We have developed a novel form of learning curve, which in contrast to the form of learning curve normally used in Active Learning, allows one to compare the costs incurred by different leaning strategies.We plan to combine ASE-Progol with a standard laboratory robot to create a general automated approach to Functional Genomics. As a first step towards this goal, we are using ASE-Progol to rediscover how genes participate in the aromatic amino acid pathway of Saccharomyces cerevisiae. Our approach involves auxotrophic mutant trials. To date, ASE-Progol has conducted such trials in silico. However we describe how they will be performed automatically in vitro by a standard laboratory robot designed for these sorts of liquid handling tasks, namely the Beckman/Coulter Biomek 2000.

Although our work to date has been limited to trials conducted in silico, the results have been encouraging. Parts of the model were removed and the ability of ASE-Progol to efficiently recover the performance of the model was measured. The cost of the chemicals consumed in converging upon a hypothesis with an accuracy in the range 46-88% was reduced if trials were selected by ASE-Progol rather than if they were sampled at random (without replacement). To reach an accuracy in the range 46-80%, ASE-Progol incurs five orders of magnitude less experimental costs than random sampling. ASE-Progol requires less time to converge upon a hypothesis with an accuracy in the range 74-87% than if trials are sampled at random (without replacement) or selected using the naive strategy of always choosing the cheapest trial from the set of candidate trials. For example to reach an accuracy of 80%, ASE-Progol requires 4 days while random sampling requires 6 days and the naive strategy requires 10 days.

With the completion of the sequencing of genomes of an increasing number of organisms, the focus of biology is moving to determining the role of these genes (functional genomics). To this end it is useful to view the cell as a biochemical machine: it consumes simple molecules to manufacture more complex ones by chaining together biochemical reactions into long sequences referred to as metabolic pathways. Such metabolic pathways are not linear but often intersect to form a complex network. Genes play a fundamental role in this network by synthesising the enzymes that catalyse biochemical reactions. Although developing a complete model of metabolism is of fundamental importance to biology and medicine, the size and complexity of the network has proven beyond the capacity of human reasoning. This paper presents intermediate results in the Robot Scientist research programme that aims to discover the function of genes in the metabolism of the yeast Saccharomyces cerevisiae. Results include: (1) the first logical model of metabolism; (2) a method to predict phenotype by deductive inference; and (3) a method to infer reactions and gene function by abductive inference. We describe the in vivo experimental set-up which will allow these in silico inferences to be automatically tested by a laboratory robot.

Recently, there has been a wide interest in using ontologies on the Web. As a basis for this, RDF Schema (RDFS) provides means to dene vocabulary, structure and constraints for expressing metadata about Web resources. However, formal semantics are not provided, and the expressivity of it is not enough for full-fledged ontological modeling and reasoning.In this paper, we will show how RDFS can be extended in such a way that a full knowledge representation (KR) language can be expressed in it, thus enriching it with the required additional expressivity and the semantics of this language. We do this by describing the ontology language OIL as an extension of RDFS. An important advantage of our approach is a maximal backward compatability with RDFS: any meta-data in OIL format can still be partially interpreted by any RDFS-only-processor.

The OIL extension of RDFS has been carefully engineered so that such a partial interpretation of OIL meta-data is still correct under the intended semantics of RDFS: simply ignoring the OIL specic portions of an OIL document yields a correct RDF(S) document whose intended RDFS semantics is precisely a subset of the semantics of the full OIL statements. In this way, our approach ensures maximal sharing of meta-data on the Web: even partial interpretation of meta-data by less semantically aware processors will yield a correct partial interpretation of the meta- data. We conclude that our method of extending is equally applicable to other KR formalisms.

According to recent several reports, the number of human genes is estimated fewer than 40,000, which, surprisingly, is only double the number (19,000) present in Caenorhabditis elegans. This fact indicates that the correlation between complexity of a species and its gene number appears to be loose, motivating studies on other sources of complexity such as alternatively splicing and cis-regulatory machinery. The first step towards this research direction would be to list candidates of alternatively spliced transcripts and promoters. Because of the release and updates of the human genome, this enumeration could be achieved by aligning millions of expressed sequence tags (ESTs, for short) to a draft human genome and then organizing alignments into related groups, which is however computationally costly. We have developed an efficient algorithm for this purpose, taking only one day to align three millions of ESTs to a newly revised draft genome from scratch. Analysis of 2.2 millions of alignments identifies about 8,000 groups of alternatively spliced transcripts. A typical group contains tens of alignments, demanding a method of selecting representatives. All the results are accessible through the WWW at our Gene Resource Locator web site http://grl.gi.k.u-tokyo.ac.jp/

It is necessary to integrate various types of information in order to recognize the genes in DNA sequences. The information might be the four letters of DNA sequences, stochastic frequency of the corresponding codons, homology search scores, splice cite signal strengths. We have developed a software system of multi-stream Hidden Markov Models with which those information can be easily integrated with a consistent measure of probabilities. The output symbols of HMMs are the signals that we can observe. In the field of bioinformatics, the output symbols of the HMMs are mostly the four letters of nucleic acids or the twenty letters of amino acids. However, the output symbols can be anything, as far as we can attach their probability distributions. They can be discrete symbols, real values, real valued vectors, and multiple streams of those values. We propose gene annotation using HMMs with multiple streams, which combine the sequence and other pre-processed information. The important feature of multi-stream HMMs is that the weights of the streams can be optimized for each model. The multi-stream HMM with adjustable weight permits very flexible design of the gene finding systems.

Semantic Annotation is a basic technology for intelligent content and is bene cial in a wide range of content-oriented intelligent applications, esp. in the area of the Semantic Web. In this paper we present our work in ontology-based semantic annotation, which is embedded in a scenario of a knowledge portal application. Starting with seemingly good and bad manual semantic annotation, we describe our experiences made within the KA2 initiative. The experiences gave us the starting point for developing an ergonomic and knowledge base-supported annotation tool. Furthermore, the annotation tool described are currently extended with mechanisms for semi-automatic information-extraction based annotation. Supporting the evolving nature of semantic content we additionally describe our idea of evolving ontologies supporting semantic annotation.This paper has been presented at the COLING-2000 Workshop on Semantic Annotation and Intelligent Content, Centre Universitaire, Luxembourg, 5.-6. August, 2000.

We summarize and discuss the work accomplished through the HypothesisCreator project, an ongoing effort whose aim is to develop systematic methods to accelerate the computational knowledge discovery process.A novel approach to describe the knowledge discovery process, focusing on a generalized form of attribute called view, is presented. It is observed that the process of knowledge discovery can, in fact, be modeled as the design, generation, use, and evaluation of views, asserting that views are the fundamental building blocks in the discovery process. Also, we note that the trial-and-error cycle inherent in any knowledge discovery process, can be formulated as the composing and decomposing of views. To assist this cycle, a programming language called VML (View Modeling Language) was designed, providing facilities for this purpose. Following this view oriented perspective, we describe our approach to the problem of characterizing N-terminal protein sorting signals in which we have obtained significant results.

This paper introduces a framework where the semantic access to Web resources is realised by a mediator module. Users can express their knowledge needs using the terms of a shared ontology without taking into account context-specific locations, encodings, and vocabulary. Integration among shared ontologies and Web meta data descriptions is achieved by inserting an intelligent meta-level layer using lifting rules. Lifting rules can be used to automatically decompose semantic information needs into concrete information requests. Wrapper and facilitator agents are used to find suitable information on the Web.Labelled directed acyclic graphs (DAGs) allow a natural modelling of syntactical, structural and semantic properties of standardised XML Web data meta data, and ontologies, and their operations. However, graph-based data models are only a first step towards intelligent tools for automated knowledge extraction from the WWW. Planning and design synthesis of technical systems involve dealing with interacting constraints, imposed by selected tasks, such as cost or compatibility constraints, or general constraints.

As a step to cope with these requirements, a novel approach combining constraint processing and reasoning with Web (meta-)data is presented. Canonical graphterms are used for the encoding of labelled DAGs, which are embedded into constraint logic formulas, thus representing Web (meta-)data and ontologies. Graph rewriting and application of the built-in reasoning facilities such as matching, unification, subsumption and constraint solving of a constraint-logic language can then be used to efficiently automate mediation tasks.

This paper describes a conversational agent, called "ProBot", that uses a novel structure for handling context. The ProBot is implemented as a rule-based system embedded in a Prolog interpreter. The rules consist of patterns and responses, where each pattern matches a user's input sentence and the response is an output sentence. Both patterns and responses may have attached Prolog expressions that act as constraints in the patterns and can invoke some action when used in the response.

In this paper we discuss several issues related to automated text classification of web sites. We analyze the nature of web content and metadata in relation to requirements for text features. We find that HTML metatags are a good source of text features, but are not in wide use despite their role in search engine rankings. We present an approach for targeted spidering including metadata extraction and opportunistic crawling of specific semantic hyperlinks. We describe a system for automatically classifying web sites into industry categories and present performance results based on different combinations of text features and training data. This system can serve as the basis for a generalized framework for automated metadata creation.

In this article we address the question of design principles for software individuals, and approach it as a software design issue. We use the term 'software individuals' to designate aggregates of programs and data that have the following properties.They exist in a population of similar but not identical in dividuals.Individuals are able to interact with their surrounding environment, with each other, and/or with people. While doingso they may modify their internal state.Each individual contains the safeguards that may be re quired in order to select which in uences to accomodateand which ones to ignore.The aggregate of programs and data that de ne an individual, and that in particular define its behavior, is a part ofits internal state and can therefore be modified as the result of the interactions where the individual is involved.Individuals or small groups of individuals are able to create new individuals that inherit the features of their parents.The program/data aggregate that dfi nes an individual is symbolic in character. The ability for knowledge represen tation is designed into individuals from the start.

RDF(S) constitutes a newly emerging standard for metadata that is about to turn the World Wide Web into a machine-understandable knowledge base. It is an XML application that allows for the denotation of facts and schemata in a web-compatible format, building on an elaborate object-model for describing concepts and relations. Thus, it might turn up as a natural choice for a widely-useable ontology description language. However, its lack of capabilities for describing the semantics of concepts and relations beyond those provided by inheritance mechanisms makes it a rather weak language for even the most austere knowledge-based system. This paper presents an approach for modeling ontologies in RDF(S) that also considers axioms as objects that are describable in RDF(S). Thus, we provide exible, extensible, and adequate means for accessing and exchanging axioms in RDF(S). Our approach follows the spirit of the World Wide Web, as we do not assume a global axiom speci cation language that is too intractable for one purpose and too weak for the next, but rather a methodology that allows (communities of) users to specify what axioms are interesting in their domain.

Recently there has been an unexpected rebirth of Turing's imitation game in the context of commercial demand. To meet the new requirements the following is a minimal list of what must be simulated.Real chat utterances are concerned with associative exchange of mental images. They are constrained by contextual relevance rather than by logical or linguistic laws. Time-bounds do not allow real-time construction of reasoned arguments, but only the retrieval of stock lines and rebuttals, assembled Lego-like on the fly.

A human agent has a place of birth, age, sex, nationality, job, family, friends, partners, hobbies etc., in short a profile. Included in the profile is a consistent personality, which emerges from expression of likes, dislikes, pet theories, humour, stock arguments, superstions, hopes, fears, aspirations etc. On meeting again with the same conversational partner, a human agent is expected to recall not only the profile, but also the gist of previous chats, as well as what has passed in the present conversation so far.

A human agent typically has at each stage a main goal, of fact-provision, fact elicitation, wooing, selling, "conning" etc. A human agent also remains ever-ready to maintain or re-establish rapport by switching from goal mode to chat mode. Implementation of this last feature in a scriptable conversational agent will be illustrated.

Representing semantic information embedded within documents is important to the Semantic Web vision of having machines automatically process data found on the Web, but not nearly sufficient. Many Web-based information management service tools, such as GeoWorlds, deal with collections of documents and the services that operate upon them. Semantic modeling of document collections and Web services is essential. We describe techniques for representing semantics of both collections and services, using a lightweight multi-form ontology. These techniques improve the efficiency and reusability of users' work with Web-based information management systems. They help users to set up complex analyses and structurings of information collections to adapt their work for other analyses or for different collections, and to obtain automatic refreshing and updating for collections with content that changes over time. Our semantic representation facilitates identifying and sequencing appropriate analysis and visualization services for a given task. Templates capture these sequences in terms of active semantic relations between document collections created and manipulated during those tasks. Templates can be dynamically modified and instantiated to generate document collections for similar tasks, or to refresh an information space with time-varying document membership. They can also be exchanged, allowing others to reapply them.

The term Semantic Web was coined by Tim Berners-Lee to describe his proposal for "a web of meaning," as opposed to the "web of links" that currently exists on the Internet. To achieve this vision, we need to develop languages and tools that enable machine understandable web pages. The SHOE project, begun in 1995, was one of the rst to begin exploring these issues. In this paper, we describe our experiences developing and using the SHOE language. We beginby describing the unique features of the World Wide Web and how they must in uence potential Semantic Web languages. We then discuss why web standards such as XML and RDF are currently insufficient for the Semantic Web. We present SHOE, a language which allows web pages to be annotated with semantics, describe its syntax and semantics, and discuss our approaches to handling characteristics of the Web such as distributed authority and rapid evolution. We discuss the implementation issues of such a language, and describe some generic tools that we have built to aid in its use. Finally, we demonstrate the language and tools by applying them to two di erent domains. The language, tools, and details of the applications are all available on the World Wide Web at http://www.cs.umd.edu/projects/plus/SHOE/.

I describe an architecture for linking perception and action in a robot. It consists of three “towers” of layered components. The “perception tower” contains rules that create increasingly abstract descriptions of the current environmental situation starting with the primitive predicates produced by the robot’s sensory apparatus. These descriptions are deposited in a “model tower” thich is continuously kept faithful to the current environmental situation by a “truth-maintenance” system. The predicates in the model tower, in turn, evoke appropriate action-producing programs in the “action tower.” It is proposed that the actions be written as “teleo-reactive” programs – ones that react dynamically to changing situations in ways that lead inexorably toward their goals. Programs in the the action tower are organized more-or-less hierarchially – bottoming out ni programs that cause the robot to take primitive actions in its environment. The effects of the actions are sensed by the robot’s sensory mechanism, completing a sense-model-act cycle that is quiescent only at those times when the robot’s goal is perceived to be satisfied. I illustrate the operation of the architecture using a simple block-stacking task.

The problem of predicting a sequence  x1  ,  x2  , .... where each  xi  belongs to a finite alphabet  A  is considered. Each letter  xt+1  is predicted using information on the word  x1  ,  x2  ....,  xt  only. We use the game theoretical interpretation which can be traced to Laplace where there exists a gambler who tries to estimate probabilities for the letter  xt+1  in order to maximize his capital . The optimal method of prediction is described for the case when the sequence  x1  ,  x2  ....is generated by a stationary and ergodic source. It turns out that the optimal method is based only on estimations of conditional probabilities. In particular, it means that if we work in the framework of the ergodic and stationary source model, we cannot consider pattern recognition and other complex and interesting tools, because even the optimal method does not need them. That is why we suggest a so-called nonprobabilistic approach which is not based on the stationary and ergodic source model and show that complex algorithms of prediction can be considered in the framework of this approach.The new approach is to consider a set of all infinite sequences (over a given finite alphabet) and estimate the size of sets of predictable sequences with the help of the Hausdorff dimension. This approach enables us first, to show that there exist large sets of well predictable sequences which have zero measure for each stationary and ergodic measure. (In fact, it means that such sets are invisible in the framework of the ergodic and stationary source model and shows the necessity of the new approach.) Second, it is shown that there exist quite large sets of such sequences that can be predicted well by complex algorithms which use not only estimations of conditional probabilities.

Support in the exchange of data, information, and knowledge is becoming a key issue in current computer technology. Ontologies may play a major role in supporting the information exchange processes, as they provide a shared and common understanding of a domain. However, it is still an important question how ontologies can be applied fruitfully to online resources. Therefore, we will investigate the relation between ontology representation languages and document structure techniques (schemas) on the web. We will do this by giving a detailed comparison of OIL, a proposal for expressing ontologies in the Web, with XML Schema, a proposed standard for describing the structure and semantics of XML based documents. We will argue that these two refer to different levels of abstraction, but that, in several cases, it can be advantageous to base a document schema on an ontology. Lastly, we will show how this can be done by providing an translation procedure from an OIL ontology to a specific XML Schema. This will result in a schema that can be used to capture instances of the ontology.

A new architecture for the World Wide Web is emerging, known as the Semantic Web. In broad terms, it encompasses efforts to populate the Web with content which has formal semantics; this will enable automated agents to reason about Web content, and produce an intelligent response to unforeseen situations. We believe that to build the Semantic Web, the sharing of ontological information is required. This allows agents to reach partial shared understanding and thus interoperate. We are proposing frame-based representation as a suitable paradigm for building ontologies as well as the World Wide Web Consortium's RDF-formalism (and its extensions, such as the DARPA Agent Markup Language) as a manifestation of frame-based representation for the Web. The paper will discuss required and desirable features of ontological languages, giving examples of the possible usage of frame-based representation and ontologies on the Semantic Web.

The Modelica standard library and available Modelica tools contain some facilities for specification of 3D geometry and 3D graphics. Geometry and graphics is associated with physical objects included in simulated Modelica models. However, important graphics properties are missing from this model. In particular, physical objects cannot change their shape (geometry) and rendering features (graphics) dynamically. The physics of simulation, is often not affected by geometry of physical objects. For instance, a body is often approximated by its center of mass under certain conditions. Either simple predefined shapes or specifications of geometry via external files are used. The last facility leads to separation between the model and the corresponding graphics and geometry. Our proposal is to integrate 3D geometric and graphical features with Modelica models of physical objects. The 3D graphics information is specified explicitly via annotations containing certain graphics primitives or using instances from a specially designed geometry class library. The motivation, syntax and implementation outline for this approach are discussed in this report.

This paper introduces the notion of fitness-for-purpose, presents a tractable, approximate approach to the recognition of fitness-for-purpose, and describes a working implementation using constraint programming.The property of fitness-for-purpose states whether running a software component with a supplied set of inputs can satisfy a given goal. Our interest is to assess whether a chosen problem solver, together with one or more knowledge bases, can satisfy a given problem-solving goal. In general, this is an intractable problem. We therefore introduce an effective, practical, approximation to fitness-for-purpose based on the plausibility of the goal. We believe that constraint (logic) programming provides a natural approach to the implementation of such approximations. We took the Common LISP constraints library SCREAMER and extended its symbolic capabilities to suit our purposes.

In this paper, a framework of a belief revision agent in a multi-agent environment is presented. In the agent knowledge kernel, by separating domain knowledge from social knowledge, various level of belief revision capabilities has been modularized into multiple interactive layers. A vertical 4-layer 2-pass architecture is adopted for the BR engine. Therefore, the framework can model various levels of belief revision, including Single Belief Revision, BR using information from Multiple Sources and Multi-Agent Belief Revision. In order to achieve multi-agent belief revision in a heterogenous society, a Shared Knowledge Structure(SKS) is proposed, which allows the sharing of knowledge as well as protecting private knowledge. Using graded knowledge in the SKS, the process of multi-agent belief revision is implemented using knowledge migration, which is the procedure that reshapes the knowledge structure triggered by the new information.

This paper is a description of a technique to allow the manipulation of conceptual scales by domain experts through interaction and manipulation of a folding Hasse diagram. By providing a set of manipulations, the domain expert, with only an intuitive understanding of the underlying representational theory - formal concept analysis - is able to construct scales to investigate text data. The method shows how a knowledge representation language, and its diagrammatic rendering, can be used to navigate a document collection.

The Resource Description Framework (RDF) is intended to be used to capture and express the conceptual structure of information offered in the Web. Interoperability is considered to be an important enabler of future web applications. While XML supports syntactic interoperability, RDF is aimed at semantic interoperability. Interoperability is only given if different users/agents interpret an RDF data model in the same way. Important aspects of the RDF model are, however, expressed in prose which may lead to misunderstandings. To avoid this, capturing the intended semantics of RDF in first-order logic might be a valuable contribution and may provide RDF with a formalization allowing its full exploitation as a key ingredient of the evolving Semantic Web. This paper seeks to express the concepts and constraints of the RDF model in first-order logic.

We explore which kinds of nonmonotonic inference relations naturally arise when using similarity-based implication and consistency measures to rank propositions à la Gäardenfors and Makinson. There is no surprising result the main interest being to provide a new perspective to nonmonotonic reasoning from a field which has not been traditionally considered within the uncertainty formalisms, but which is indeed very close, at least to Possibility theory.

In this paper we present a new approach to a symbolic treatment of quantified statements having the following form "Q A's are B's", knowing that A and B are labels denoting sets, and Q is a linguistic quantifier interpreted as a proportion evaluated in a qualitative way. Our model can be viewed as a symbolic generalization of statistical conditional probability notions as well as a symbolic generalization of the classical probabilistic operators. Our approach is founded on a symbolic finite M-valued logic in which the graduation scale of M symbolic quantifiers is translated in terms of truth degrees. Moreover, we propose symbolic inference rules allowing us to manage quantified statements.

Automated theory formation involves, amongst other things, the production of examples, concepts and statements relating the concepts. The HR program has been developed to form theories in mathematical domains, by calculating examples, inventing concepts, making conjectures, and settling conjectures using the Otter theorem prover and MACE model generator. In addition to providing a plausible model for automated theory formation in pure mathematics, HR has been applied to other problems in Artificial Intelligence. We discuss HR's application to inducing definitions from examples, scientific discovery, problem solving and puzzle generation. For each problem, we look at how a theory formation approach can be applied and mention some initial results from the application of HR. Our aim is not to describe the applications in great detail, but rather to provide an overview of how HR is used for these problems. This will facilitate a comparison of the problems and discussion of the effectiveness of theory formation for these tasks. Our second aim is to compare HR with the Progol machine learning program. We do this first by looking at the concept formation these programs perform. Also, by suggesting how Progol could be used for the applications mentioned above, we compare the programs in terms of how they can be applied.

Bayesian networks provide an elegant formalism for representing and reasoning about uncertainty. They are a probabilistic extension of propositional logic and, hence, inherit some of the limitations of propositional logic, such as the difficulties to represent objects and relations. The main contribution of this extended abstract is to introduce a new approach, called Bayesian logic programs, to overcome the limitations. It combines Bayesian networks with definite clause logic, i.e. "pure" Prolog, by establishing a one-to-one mapping between ground atoms and random variables. Thus, Bayesian logic programs combine the advantages of definite clause logic and Bayesian networks. This includes the separation of quantitative and qualitative aspects of the world. Furthermore, Bayesian logic programs generalize both Bayesian networks as well as logic programs, many ideas developed in both areas can be adapted.

This paper describes a conceptual scheme for classifying conceptual situations from the viewpoint of similarities under a goal concept. As the similarity notion depends on a context, we use a goal concept to specify it. Each situation and the goal is represented as a concept graph from which an explanation of the goal is extracted using a typed domain theory. A notion of abstractions is then used to define what is a valid similarity with respect to the explanation of goal. In a word, given an initial situation and a goal subsuming it, a situation is similar to the initial situation if it is a specialization of some analogue of the initial one. We observe that generalization is also needed to obtain a non-trivial or useful similarity. Based on this observation, we present an experimental result on a system for finding similarities between situations under a goal concept.

The problem of merging multiple sources information is central in many information processing areas such as databases integrating problems, multiple criteria decision making, expert opinion pooling, etc. Recently, several approaches have been proposed to merge classical propositional bases, or sets of (non-prioritized) goals. These approaches are in general semantically defined. Like in belief revision, they use priorities, generally based on Dalal's distance, for merging the classical bases and return a new classical base as a result.An immediate consequence of the generation of a classical base is the impossibility of iterating the fusion process in a coherent way w.r.t. priorities since the underlying ordering is lost.

This paper presents a general approach for fusing prioritized bases, both semantically and syntactically, when priorities are represented in the possibilistic logic framework. Different classes of merging operators are considered including conjunctive, disjunctive, reinforcement and adaptive operators.We show that the approaches which have been recently proposed for merging classical propositional bases can be embedded in this setting. The result is then a prioritized base, and hence the process can be coherently iterated. Moreover, we also provide a syntactic counterpart for the fusion of classical bases.

Faceted Knowledge Representation provides a formalism for implementing knowledge systems. The basic notions of faceted knowledge representation are "unit", "relation", "facet", and "interpretation". Units are atomic elements and can be abstract elements or refer to external objects in an application. Relations are sequences of matrices of 0's and 1's (binary matrices). Facets are relational structures that combine units and relations. Each facet represents an aspect or viewpoint of a knowledge system. Interpretations are mappings that can be used to translate between different representations. This paper introduces the basic notions of faceted knowledge representation. The formalism is applied here to an abstract modeling of a faceted thesaurus as used in information retrieval.

Incremental updates to multiple inheritance hierachies are becoming more prevalent with the increasing number of persistent applications supporting complex objects. Efficient computation of lattice operations such as greatest lower bound (GLB), least upper bound (LUB), and subsumption subsequently is becoming more and more important. General techniques for compact encoding of a hierarchy are presented that support the operations, and are flexible enough to allow incremental updates to the hierarchy. One such method is to plunge the given ordering into a boolean lattice of binary words, leading to an almost constant time complexity of the lattice operations. The method is based on an inverted version of the encoding of Aït-Kaci et al. to allow incremental update. Simple grouping is used to reduce the code space while keeping the lattice operations efficient. Comparisons are made to an incremental version of the range compression scheme of Agrawal et al., where each class is assigned an interval, and relationships are based on containment in the interval. The result is two incoding methods which have their relative merits.

Collision detection and response is one of the most difficult areas in simulation of multibody systems. Two known approaches, the impulse-based method and the force-based (penalty) method, can be applied for multibody simulation in Modelica. The impulse-based method requires instantaneous modification of some variables, but such modification is not always possible in Modelica. The force-based method leads to stiff ODE, which can be handled by solvers used with Modelica. We suggest a new way to express the penalty coefficients. The force-based method, however, requires computation of penetration depth which is time-consuming.We also suggest a method that combines the distance between bodies and the penetration depth into a single quantity used for force computation.

Calling external functions is a preferable method integrate collision detection algorithms with practical physical models, since body geometry is stored externally. We describe an interface with collision detection tool SOLID.

Agents in a multiagent system may need to share information and services. For this, they need to be able to interleave deliberative planning with execution of actions. The deliberative planning is needed to decide which actions to perform to achieve an objective, whereas execution of some of the actions is needed to make a more informed decision on the other actions and to access services provided by other agents.HITaP is a planner that interleaves planning and execution: using HITaP an agent can, during planning, gather information by either direct inspection of the domain or by firing queries to other agents and recording their answers. Interleaving planning and execution, as provided by HITaP, plays a crucial role in an agent's ability to construct shared plans with other agents and to manage the negotiation process that leads to agreement with the agent's teammates on these plans.

HITaP is implemented and currently used as planning module for agents in the RETSINA multiagent system. These agents cooperate to solve problems in different domains that range from portfolio management to command and control decision support systems