International Conference on e-Learning and the Knowledge Society - e-Learning’10 - 179 - Natural Language Based Concept Map Building Marks Vilkelis, Janis Grundspenkis, Normunds Gruzitis Abstract: The paper represents the research on transformation of natural (Latvian) language texts into concept maps. The main point of this work is to overview and to discuss some ideas in order to describe a number of already solved tasks, linked with the topic. There are many examples of how and what to do in automated “translation” of common linguistic structures and patterns into the formal data structure. The paper gives also an overview of authors’ previous research, which encouraged starting the current work. Key words: Natural Language Processing, Concept Map, Concept Structure, Ontology. INTRODUCTION One of the ultimate goals of early research directions in artificial intelligence is to reach a perfect man-machine interaction using natural language. Nowadays computers mainly can understand only specific machine language, programmed by specific people – programmers. The fact, that computer industry and applications are growing fast, makes humans to be involved in it more and more. Thus, the need of communication in natural language is extremely high. The lowest level computers can operate on is mathematics. The problem is that natural language has words and sentences with no explicit mathematics in it. To operate with words (concepts) and their mutual relations, we have to use more advanced data types, which could be understandable both for a human and for a computer. There are many formal data structures, which support appropriate relationships between aspects and concepts: mind maps, frames, graphs, UML diagrams, concept maps [2] etc. The basic aim of the research is to recognize key concepts and their relations. The next step is to transform the recognized words into concept map [2] and store it in order to create ontologies [4]. Then we could operate with this data further by answering questions, seeking relations among known concepts and even by inferring new relations. We want to make computer to understand what human says. The work described in this paper has just started, so there are more questions than answers. However, the authors would like to share some results they already have and to introduce some ideas and ways how to solve the problems faced with during the research. The paper is organized as follows. Section 2 gives a short overview of the designed and implemented intelligent knowledge assessment system (IKAS), which usage inspired the authors of this paper to start the current research. This section also describes the concept map term. At the end of the section the main task to solve the stated problem is formulated. Already solved subtasks and successful solutions for natural language transformation into the formal data structure are described in section 3. The end of the section 3 discusses problems authors faced with during the research. This section has also some theoretical proposals. The paper ends with conclusions and future work. PROBLEM DESCRIPTION IKAS overview Since year 2005 the Department of Systems Theory and Design of the Faculty of Computer Science and Information Technology of Riga Technical University (RTU) has been developing the concept map based knowledge assessment system IKAS (Intelligent Knowledge Assessment System), which usage inspired the authors of this paper to start the current research [5, 7, 8]. The system works in the following way. The teacher defines knowledge assessment area and creates concept map for it. The process of the creation of a concept map consists of the specification of relevant concepts and relationships. Thus the concept map includes all concepts and relationships among them, which are taught