my question is regarding collocation analysis (i'm using R). I have following data gathered from Polish corpus:
| collocation | chi2.score | word.type | count | |:-----------:|:-----------:|:---------:|:-----:| | szczególny | 31501785.92 | baczenie | 8 | | chory | 27.31 | rozum | 5 | | pojąć | 948513.60 | celowo | 13 | | odnosić | 338963.45 | wrazenie | 1909 |
Corpus automatically lists collocates with their Chi2 scores as an association measure. In total there are 66000 rows of data (66000 collocations for 58 unique word.types, each word.type has a different number of significant collocations).
I'm trying to statistically find out which of my word.types has similar collocations as one particular word.type (let's say that it is 'celowo'). Can I actually do anything with these Chi2 scores to compare a pair of word.types' collocations (i.e., using these scores to perform some clustering analysis), or should I focus solely on differences in nominal data column 'collocation'?
More general, though related, question is: having word frequency data and collocations from a linguistic corpus, how would you go about trying to find synonyms/equivalent phrases of a particular word?