Questions concerning Semantic Roles Annotation

[Olanrewaju Samuel]

Good day, everyone

I am a linguist with interests overlapping with computational linguistics. I am working on semantic role annotation for my native language, and I have been trying to work with Inception, but I have some challenges. 

1. The format of the annotation is not in the format that I want, as shown here:

Arg0: I

REL : man

Arg1: the gate

What is the best format to export my annotation to?

2. Do you know how to link my semantic roles and their verbs for each sentence?

3. Is there anyone who has done semantic role labeling in inception that will be happy to share their experience with me?

I will be happy to hear from you.

Thanks,

Samuel.

[Richard Eckart de Castilho]

Hi,

> On 31. Mar 2025, at 18:03, Olanrewaju Samuel <olasamue...@gmail.com> wrote:
>
> 1. The format of the annotation is not in the format that I want, as shown here:
>
> Arg0: I
> REL : man
> Arg1: the gate
> What is the best format to export my annotation to?

IMHO the best export format is CAS JSON - because that fully supports all functionalities of INCEpTION and is easy to handle e.g. with DKPro Cassis in Python or UIMA in Java. But your mileage my vary.

> 2. Do you know how to link my semantic roles and their verbs for each sentence?

Create a project using the "Classic linguistics" project template or ensure that you have added the "Predicate argument structure" layers to your project (SemPred and SemArg).

Then annotate the verb using "SemPred", add slots to your annotation for the arguments, and fill them. I hope the attached video makes it through to the list.

[Richard Eckart de Castilho]

I don't know if the last part of my mail made it to the list, but it doesn't show up in the Google Group web UI - so here it is again.

> 3. Is there anyone who has done semantic role labeling in inception that will be happy to share their experience with me?

INCEpTION inherits the functionality for predicate-argument structure annotation from WebAnno. Although the following paper is about WebAnno, the principle, it still works in the same way in INCEpTION.

https://www.inf.uni-hamburg.de/en/inst/ab/lt/publications/webanno3-tl4dh2016.pdf

Cheers,

-- Richard

[Olanrewaju Samuel]

Hi Richard!

Thank you so much for your reply. This is incredibly helpful. The attachment clarified a whole lot of things. I have not read the pdf carefully, but on the high level, I have got the gist.

I have another follow-up question. When exported in the format you suggested, it comes with other parts of the linguistic annotation that are blank. Is it possible to get a format that exports only the specific semantic field? Additionally, does this platform help to build propbank? 

I will be happy to hear from you. 

Thank you so much.

Olanrewaju.

[Richard Eckart de Castilho]

Hi,

> On 1. Apr 2025, at 19:28, Olanrewaju Samuel <olasamue...@gmail.com> wrote:
>
> I have another follow-up question. When exported in the format you suggested, it comes with other parts of the linguistic annotation that are blank. Is it possible to get a format that exports only the specific semantic field?

I did an UIMA CAS JSON export of the example file I created as part of the video.
I'm afraid, I don't know what blank parts you refer to.
In any case, can't you easily just ignore the annotations that are not relevant to you?

> Additionally, does this platform help to build propbank?

You can build annotate texts with semantic-argument structures similar as to what has been done in propbank.

Cheers,

-- Richard