Automatic Identification of Academic Phrases for Czech

Abstract

The aim of this study is to automatically extract academic phrases in Czech using data-mining techniques as a first step towards creating a dictionary of academic words and phrases targeting university-level students (L1 and L2). The decision to use data mining was based on excellent results of data mining in automatic recognition of single-word and multi-word terms [10]. This method has identified various types of academic phrases: structurally incomplete lexical bundles with specific functions in texts (e.g. na druhou stranu – on the other hand), collocations (e.g. podrobna analýza – detailed analysis) or combinations of a content word and a typical function word (e.g. zaměřený na - focused on; podobný jako - similar to). The final list of automatically identified academic phrases is quite extensive and consists of 7,300 bigrams. Manual evaluation of the output data sample showed that precision of the automatic identification method is more than 72% and recall is 81%. The list of identified academic phrases is a very good starting point for the planned dictionary because the majority of the extracted bigrams constitute collocations typically used for academic texts. Such collocations are useful for the target audience, that is, university students interested in academic writing.