Abstract
The new method of automatic term recognition TERMIT is focused not only on the high number of correctly labeled terms, but also on the most important attributes of a term (in terms of their role in automatic term identification process). The method is based on data mining, i.e. finding meaningful information in very large corpus data.
It was able to both successfuly identify terms in academic texts and find constitutive features of a term as a terminological unit. The single-word term (SWT) can be characterized as a word with a low frequency in corpus (SYN2010) that occurs considerably more often in specialized texts of a given field than in non-academic texts, occurs in a small number of academic disciplines, its distribution in the corpus (SYN2010) is uneven as is the distance between its two instances.
The multi-word term (MWT) is a stable collocation consisting of words with low frequency and contains at least one (and often more) single-word term. Based on the characteristics of SWT and MWT, it is possible to classify individual tokens in texts as terms or non-terms with a success rate of more than 95 %.
Automatically identified terms can be used to identify percentage of SWT or MWT in different academic disciplines, as well as find terms shared by two or more domains in order to assess their relationship. In general, we can conclude that an automatic recognition of a languge phenomenon can contribute to its haracterization and that a purely quantitative approach (such as data mining) can be used to research linguistic (corpus) material.