DMRS transformations to RDF is broken

[arademaker]

SENT: A person is riding the bicycle on one wheel

We have 3 analyses but each analysis should be a separate graph. But notes are being reused.. BTW, <http://ibm.com/sick-b/22/nodes#node10002#predicate> is not a valid URI (no two #)

[yfaria]

I think the code behavior when we have more than one is not well described yet (our tests were made with the gold profile of ERG, so it wasn't in our tests). The profile is the same from #17, isn't it? What was the command that you used in the CLI?

[yfaria]

Checking the sick-b, the problem that appears when there is more than 1 analysis for each sentence is due to both having the same id in the profile; so the nodes end up with the same names. I can think of two major ways to solve this problem:

• create the notion of profile and make the transform depend on those profile specific things in a schema;
• just abandon the id in the profile and rename the URI with a unique number for each analysis, it can be a unique number associated with the id as well (v.g., if we have two analysis with the id 5104, we could put 5104-01 and 5104-02).

I believe that the first is more scalable, but will increase the RDF output in size (not much tough).

About the URIs with two #, the problem was solved in the url-check. I'll PR when I solve the ids problem

[arademaker]

Can you please enumerate here all the URIs that we need to generate in the MRS-RDF and DRMS-RDF? So we can decide about the patterns we should use.

[arademaker]

For instance, if we take {prefix} to be an URI for the profile:

1. The DMRS: <{prefix}/item/{num}-{analysis}>
2. A node <{prefix}/item/{num}-{analysis}/node/{number}>
3. An edge <{prefix}/item/{num}-{analysis}/edge/{number}>

etc

[arademaker]

I believe we can close it now. This is about a bug now solved. Additional improvements being discussed in #21