Sentence multi-root problem

[andreizammit]

From the readme, it is stated that you managed to solve the multi-root problem:
Having a Unique ROOT: It has just one ROOT for addressing each sentence

If it is possible, can you please indicate me where the problem is located?
The work of Eliyahu Kiperwasser has the multi-root problem which was discussed but it seems that it was never tackled/solved.

[jujbob]

The problem was already solved by email but it is for people who will face the problem (more than two people have asked)

I have solved the problem with two ways.

Solution1, Based on the output values of arc MLP, you can take tokens have the relation as a ROOT (tokens have the highest probability of ROOT). If the number of ROOT is more than 2 you can compare all the tokens pointing ROOT and pick up the highest one as a unique ROOT. And then get rid of the probability of ROOT for the other tokens pointing out ROOT.

Solution2, (that I adapted), it is a heuristic method but better than Solution1 in terms of performance. I have conducted some experiments what makes multi-ROOT effects for overall performance. And finally found out that it is highly related to the word order and most of the case the first or last ROOT showed in the ordinary sequence is correct one.

You can find the solution in:
https://github.com/jujbob/multilingual-bist-parser/blob/master/bist-parser/bmstparser/src/mstlstm.py
From the source code, you may see 311~317.