No abstract

Here, we describe our work in developing Indonesian Mind Map Generator that employs several Indonesian natural language understanding tools as its main engine. The Indonesian Mind Map Generator 1 aims to help the user in easily making a Mind Map object. The system consists of several Indonesian natural language understanding tools such as Indonesian POS Tagger, Indonesian Syntactic Parser, and Indonesian Semantic Analyzer. The methods used for developing each of Indonesian natural language understanding tools are devised to such an extend that they are enable to alleviate the low availability of Indonesian language resources. For Indonesian POS Tagger, we employed HMM and subsequently enhanced the result by using affix tree. As for the Indonesian Syntactic Parser, we compared the performance of CYK and Earley parser, which are known as common dynamic algorithms in PCFG. The Indonesian Semantic Analyzer consists of several components such as lexical semantic attachment, reference resolution, and Semantic Analyzer itself that transforms the parse tree result into first order logic representation. In our work, instead of using a rich resource on semantic information for each vocabulary, we defined several rules for the lexical semantic attachment based on POS Tags and certain words. Finally, to develop the Mind Map generator, we used the radial drawing method to visualize the first order logic representation and we also built a Mind Map editor to allow a user in modifying the Mind Map result. To evaluate the result, we conducted the experiments for each component mentioned previously. The POS Tagger accuracy achieved 96.5%, the Syntactic Parser achieved accuracy of 47.22%, and the Semantic Analyzer achieved accuracy of 62.5%. The final result of Mind Map object was evaluated by 5 respondents. The results of evaluationshowed that, for the simple sentence, the Mind Map object can be easily understood.

No abstract

In this research, we build a question answering system that can answer comparative questions. Our system contains two components, question analysis and answer processing. In question analysis, we use information extraction method to extract entities, aspects, relations, and constants from the question. We also classify the question into five question types such as entity-mentioned, entity-other, entity-all, aspect, and yes/no. These processes in question analysis component are solved by machine learning technique. In answer processing component, we process the comparison by using word lists and rules. The result of comparison processing is used to find the answer by generating query to get relevant data from the database. The query generation process is performed by using rules based on the question type. After that, the data from database is also processed based on question type to generate the answer. Based on the experiment results, our proposed method for comparison question answering system is promising.