The Relation Between the Composition of Corpora (Genre Balance and Representativeness) and Their Reliability in Compiling General Explanatory Dictionary

This paper aims to examine the genre composition of certain Slovenian corpora as sources for lexicographic analysis (especially when compiling dictionaries such as eSSKJ, the general explanatory dictionary), particularly of the largest corpus, Gigafida 2.0 (divided into two sub-corpora: a sub-corpus of non-fiction and literary texts and a sub-corpus of journalistic texts), the Corpus of Slovenian School Texts, the Corpus of Scientific Texts of Contemporary Slovenian, as well as the KRES corpus. We argue that corpora with major discrepancy in the proportions between different text genres used as lexicographic resources do not reflect the proportions between meanings which originate in semantic extension processes. Thus, one of the largest corpora available for Slovene, Gigafida (in both versions, 1.0 and 2.0, updated in 2019), could hardly be regarded as a reference source of data for a general explanatory dictionary. This is because various journalistic texts and web texts are predominant in Gigafida, while the share of non-fiction and literary texts does not exceed 10% in total. We suggest that a corpus should be at least approximately balanced, which could in turn provide its representativeness.