Use c-amr with gold relation and concept labels

[silvianacmp]

Hello,

I have a use-case in which I would like to evaluate the performance of c-amr using smatch, but when predicting the AMR using gold standard labels for the relations and concepts. I have seen that you have reported such results in the paper explaining your solution so I was wondering whether it is possible to do this using the code publicly available in this repository. Thank you!

[Juicechuan]

Hi,

I don't have a function to do that in the current version. But you can try tweaking it bit by modifying the evaluate function in graphstate.py:

    def evaluate(self):
        num_correct_arcs = .0
        num_correct_labeled_arcs = .0

        parsed_tuples = self.A.tuples()
        ....
        num_parsed_arcs = len(parsed_tuples)
        gold_tuples = self.gold_graph.tuples()
        num_gold_arcs = len(gold_tuples)

        num_correct_tags = .0
        num_parsed_tags = .0
        num_gold_tags = .0
        visited_nodes = set()
        for t_tuple in parsed_tuples:
            p,c = t_tuple
            p_p,c_p = p,c
            # comparing for evaluating
            ....

where this function is basically doing evaluation by comparing each concept and relation, you can just assign the gold concept and relation to the parsed graph.

[zy-zhangying]

Have you solved this problem?