No description available.

No description available.

No description available.

国立情報学研究所で電子化したコンテンツを使用している。

Les ontologies sont diversement employées notamment dans les domaines du Web sémantique, de l’ingénierie des connaissances, … En effet, elles permettent de partager, de diffuser et d’actualiser les connaissances d’un domaine. Afin de construire ces ontologies, notre méthodologie utilise tout d’abord des méthodes de Traitement Automatique de la Langue Naturelle (TALN) et d'Extraction d'Information (IE) pour extraire des données préparées à partir de chaque ressource du domaine (corpus de textes, bases de données, thesaurus). Puis, ces données sont fouillées avec les méthodes de fouilles : l'Analyse Formelle de concepts (AFC) et l’Analyse Relationnelle de Concepts (ARC). L'AFC regroupe des objets partageant les mêmes attributs binaires dans des concepts d'un treillis. L'ARC est une extension de l'AFC qui permet de regrouper des objets partageant les mêmes attributs binaires, mais aussi les mêmes attributs relationnels. L’apposition de contextes (une propriété de l’AFC) permet d’associer ces attributs (binaires et relationnels) à un ensemble de classes prédéfinies et hiérarchisées par les experts du domaine. De cette façon, des définitions étendues sont proposées aux experts du domaine pour ces classes prédéfinies ainsi que de nouvelles classes inexistantes dans la hiérarchie initiale. Ces nouvelles classes peuvent être considérées pertinentes et ajoutées par les experts en tant que nouvelles « unités de connaissances ». Les treillis résultant des méthodes de fouille constituent ce que nous appelons schéma d’ontologie. Ce schéma d’ontologie est ensuite représenté par le langage FLE de la famille des logiques de descriptions afin d’avoir une ontologie. Cette ontologie, implémentée en OWL (Web Ontology Language), a permis à notre système de répondre automatiquement à différentes questions proposées par les experts du domaine (instanciation de concepts, comparaison de concepts,…). Des expériences pratiques ont été menées dans deux domaines d'application que sont l'astronomie et la microbiologie. / Ontologies are used in different fields like the semantic Web or the knowledge engineering. The ontologies allow to share, to diffuse and to update knowledge domain. This thesis propose a methodology to build ontologies using methods of Natural Language Processing (NLP) and Information Extraction (IE) for extracting prepared data from each kind of available resources in the domain (text corpora, databases, thesaurus). Then, these prepared data are mining with the mining methods : Formal Concepts Analysis (FCA) and Relational Concepts Analysis (RCA). The FCA regroups a set of objects sharing the same set of attributes in the same concept. The RCA, an extension of the FCA regroups a set of objects sharing the same attributes and the same relations (relational attributes) in the same concept. The apposition of contexts, a property of the FCA, affects a set of attributes and relational attributes to classes pre-defined and hierarchised by the domain experts. These affectations allow us to present classes and their definitions to the experts of domain as well as new nonexistent classes in the initial hierarchy. These new classes can be considered appropriate and added by experts as new « knowledge units ». The Lattices resulting from the data mining methods are considered as « ontology schema ». This ontology schema is represented in the FLE description logics language to obtain ontology. This ontology is implemented in the OWL language (Web Ontology Language) for allowing us to request it. This methodology was tested in different domains: Microbiology and Astronomy.

It is hoped that this exposition of the Platonic and Pragmatic systems of philosophy will aid in the understanding and interpretation of the highest Good, and what the implications are in modern education. The author has attempted to point out the identical phases and contrasting features, and to summarize the data in which research has been plentiful.

Digitized by Kansas Correctional Industries

Three issues from studies of primary children's addition are identif ied and investigated with children between 6 and 8 years old. The first is why descriptions of strategy use der i ved from chronometr ic studies vary so much from those based on observation and interview. The second and third concern the factors influencing strategy choice and are the relation between procedural and conceptual knowledge and characteristics of the sum. Whereas reaction time studies have been interpreted as showing that primary schoolchildren predominantly use one strategy, counting on from the larger addend (COL), interview and observation studies suggest each child uses a variety of strategies. In Experiment (1) children were given a large set of single digit additions and both their reaction times (RT) and their overt behaviour recorded. The best predictor of RT was a model based on COL. The tactic of recording both RTs and observations was continued in Experiments (2) and (3) and in Experiment (4) children were interviewed as well as timed. The major concern, however, was to investigate why children so rarely use decomposition (ie analysing a given sum into a known number fact and counting on the difference). To this end children were presented with number facts and asked to read them out prior to doing the sum. The relation between the given number fact and the sum was varied and the effect on RT determined. Presenting a number fact that was the same or a commuted version of the sum had major effects on observed and reported strategy and overall response latency. When the number facts differed by 1 or 2 from the sum children typically ignored them. Overall the utility of chronometric studies for identifying strategy use was questioned as there was much variation between studies in the best fitting RT model. Also, apart from COL, no RT model of a specific strategy made adequate predictions. Experiments (5) and (6) explored whether children do not use decomposition because they do not understand it. Awareness of the potential use of decomposition was assessed by presenting number facts and asking how these would help do particular sums. In Experiment (5), which was conducted with the participants in Experiment (4), the children were also asked to select which number facts would be useful. The same children who so rarely used decomposition demonstrated that they understood how to use it. How children's use of COL was related to understanding of commutativity was tested in Experiment (7). While children who used COL often typicallY passed commutativity tests there were some who did not.The influence of problem characteristics on strategy use was tested in Experiment (8) by using sums with very large second addends (eg 2 + 95). Every child who attempted these sums used the COL strategy on them whereas many never used COL on the more traditionally used single digit additions, but counted on from the first digit instead. This implies that the common practice of classifying children's strategy use on the basis of how they solve single digit sums may be misleading. In reviewing these and other findings it is concluded that what children know about number and addition strategies may bear little relation to how they solve simple addition sums. The explanation of why children choose a particular strategy may lie instead in the amount of 'cognitive effort' that is involved.

Theories of categorization typically assume that categories are represented by some set of features that describe the properties of category members. However this view of category representation is incomplete. This dissertation lays out a framework for category representation, following Markman and Stilwell (2001), that creates a taxonomy of categories based on different components of relational structures. Relational categories are categories of entire relational systems while, role-governed categories, are represented as the roles in these systems. Lastly, thematic-relation categories group entities together that play complementary roles within a system. Four experiments are presented in support of this framework. They contrast thematic-relation categorization with role-governed categorization. Thematic-relation categorization entails categorizing objects together that play different roles within a domain, while role-governed categorization entails categorizing two entities that play the same role across domains. When the two are put in direct conflict, people prefer to form a thematic-relation category because within-domain connections are easier to find than across-domain connections. The purpose of the four experiments is to examine ways to boost the preference for role-governed categorization, thus revealing underlying processes. Here, role-governed categorization is facilitated in two ways. Experiment 1 re-frames the question of category formation as novel word extension. Natural role-governed categories have labels while thematic-relation categories do not. This pattern is reflected in the measured behavior as novel labels are extended across members of role-governed categories more readily than across members of thematic-relation categories. By claiming relational structures are critical to category representation, the framework described in this dissertation predicts that role-governed categorization and analogical reasoning share underlying mechanisms. Experiments 2-4 examine how making an analogy between the members of role-governed categories facilitates forming such categories. When making an analogy, people align the relational representations of a pair of domains, putting entities into correspondence by role, ignoring featural dissimilarities. When analogical comparison is induced, the rate of role-governed categorization is shown to double as compared to a baseline with no such analogical processes. The thesis concludes by outlining several future lines of research generated by unifying the fields of analogy and concept learning. / text

No description available.