morph2vec requires a vocabulary file containing lines of the form:
<word> <semantic-class> <syntactic-class> <count>

For the CBOW and CBSOW models, word, semantic-class and syntactic-class should all be the same - just the word itself.
For the CBOM and CBSOWM models, semantic-class is a semantic stem of word, and syntactic-class is a syntactic suffix of word.
You could use other classes: POS tags or wordnet synsets.

The option -seq determines whether a sequential model is used for the syntactic component. 
For the CBSOW and CBSOWM models this should be set to 1.

The option -lambdas sets the parameter lambda. 
For the CBSOWM model this should 0.1.

example:
./morph2vec -train text8 -read-vocab s8 -output cbsowm_200_100.bin -sem_size 200 -syn_size 100 -lambdas 0.1 -seq 1 -window 5 -sample 1e-3 -threads 12 -binary 1






compute-accuracy has been modified to use the output from morph2vec.

The option -add determines whether the original additive approach is used, or whether the new approach should be used.
For the CBSOW and CBSOWM models this should be set to 0.

example:
./compute-accuracy cbsowm_200_100.bin 30000 0 < questions-words.txt > cbsowm_200_100_accuracy





The original word2vec package can be found here: https://code.google.com/p/word2vec/
The CBSOW, CBOM and CBSOWM models are described in Orthogonality of syntax and semantics within distributional spaces, Mitchell & Steedman, 2015


