What can corpus tell us about noun phrases (on the example of the noun mreža)

Sažetak: Noun phrases in Croatian can differ in the degree of correlation between its constituents. Some constituents form a descriptive free word combinations (velik stol ʽlarge table’, sunčan dan ʽsunny day’, slatka kava ʽsweet coffee’, hladne ruke ʽcold hands’), while others form multiword units which concretize extra-linguistic content that can not be expressed in one word (crna kava ʽblack coffee’, krevet na kat ‘bunk bed’, kreditna kartica ‘credit card’, radno mjesto ‘workplace’). Dependent constituents can be adjectives, which are congruent with a noun (velika soba ‘big room’, radno mjesto ‘working place’), or they can be adverb phrase or prepositional phrase (korak naprijed ‘step ahead’, mnogo ljudi ‘many people’, malo prijatelja ‘a few friends’, četkica za zube ‘toothbrush’, roba s greškom ‘faulty good’). This paper will analyze noun mreža (with reach syntagmatic and semantic potential) and its co-occurrences – they can either form a collocation or a free combination of words. The lexicographic description will be compared with the corpus-data. The analyses will take into consideration a list of computationally obtained collocates (collocation candidates) of a node noun. The frequency and the strength between the words occurring within a particular span can differ. The list of collocates obtained from the corpus will be taken into account and we will examine how it coincides with the existing lexicographic description and with theoretical principles of word combination interpretations in Croatian. The aim of the study is to determine how the corpus analysis can improve the treatment of word-combination entries in lexicographic work.