Extraction of Training Sets for Experimentation with Cross Language Information Retrieval Systems

In this paper we focus on methods, models and tools for the extraction of bilingual training / test sets useful for the (semi) automatic classification of textual documents. Such documents could be tutorials, technical specifications, articles, personal notes, etc. Another motivation for our research is the need for managing corpus of classified texts and especially parallel corpora (texts). We discuss the usage of pre-selected key-phrases as attributes for classification, and methods for classifying new documents. These methods could be applied to training data and produce (infer) the corresponding models. We also describe and discuss the classification of various document (textual) types, which is supported by our prototype tool.