Automatic Collocation Extraction and Classification of Automatically Obtained Bigrams

This paper focuses on automatic determination of the distributional preferences of words in Russian. We present the comparison of six different measures for collocation extraction, part of which are widely known, while others are less prominent or new. For these metrics we evaluate the semantic stability of automatically obtained bigrams beginning with singletoken prepositions. Manual annotation of the first 100 bigrams and comparison with the dictionary of multi-word expressions are used as evaluation measures. Finally, in order to present error analysis, two prepositions are investigated in some details.