The source of the research is the database of the Dictionary of the Lithuanian Nominal Phrases, which contains ca 69,000 phrases of varying length from 2 to 16 words. Phrases were extracted by the statistical method of Gravity Counts from the Corpus of the Contemporary Lithuanian Language (100 million running words). The analyzed data contain 3,704 two-word, 3,702 three-word and 870 four-word multi-word units (MWUs), which were detected automatically as being of one particular grammatical form; therefore, in this study, we refer to such MWUs as stable uniform MWUs. The results revealed that among stable uniform MWUs, there are fully fixed expressions (idioms, sayings, compound nominals, etc.). However, the majority of stable MWUs are collocations, which were extracted as one particular grammatical form due to the fact that this particular form had the highest frequency in the corpus. The stability of the form of such collocations is due to the governing constituents, e.g., a) cases of nominals (XX a. pirmaisiais dešimtmečiais) or b) prepositions, pronouns, numerals, adverbs or other inflected verb forms (sulig kiekviena diena, kartą per savaitę). Regarding the meaning of collocations, they can be classified into the semantic groups of place, time, manner, quality, object, and subject. From the point of view of usage, groups of specific discourse-bound collocations were identified, i.e., phrases from prayers, legal documents, horoscopes, etc. It can be concluded that the prevalence of a specific form is typical not only of idiomatic and semantically non-compositional MWUs but also of collocations usually defined as groups of semantically compositional and less stable MWUs.

In this paper, we analyze two-word phrases which consist of at least one noun. The phrases for the analysis were selected from the Dictionary of Lithuanian Nominal Phrases. The morphological analysis has revealed that almost a half of the phrases are built from two nouns. Adjective + noun is the second most frequent pattern (32,2%), and the verb + noun pattern (9,99%) is the third one concerning frequency. The classification according to the head of a phrase has shown the domination of noun phrases (96,4% of all two-word phrases). After the analysis of two-word phrases by their syntactic role, the results are as follows: 95% of all phrases are attributive phrases linked by the subordination relation, object phrases with subordination relation are the second most frequent phrases (3,42%), whereas phrases expressing other syntactic relations make up less than one percent each. The results have shown that the head of a phrase is usually used as the left constituent of a phrase, except for attributive phrases linked by subordination and interdependent object phrases. Variability of the phrases was investigated by the analysis of the inflectional and derivational features of the left and / or right constituent of a phrase. The results suggest that the most phrases (46%) are with two variable constituents; however, phrases with the left fixed constituent are also frequent (44%). The presented statistical data based on authentic usage complement the descriptions of the morphological and syntactic features of two-word phrases provided in Lithuanian grammars and theoretical studies and are relevant to quantitative evaluation of linguistics aspects and to automatic syntactic analysis.