Producers for Cardinal and Ordinal?

[Alain Désilets]

Looking at this page:

https://dkpro.github.io/dkpro-core/releases/1.8.0/docs/typesystem-reference.html#type-de.tudarmstadt.ukp.dkpro.core.api.ner.type.Cardinal

I see that dkpro ner api defines types for Cardinal and Ordinal, but those don't have producers listed.

Is there no component in dkpro that can tag those?

If not, I am willing to add one. Any suggestion as to which library I should use to add this producer?

[Richard Eckart de Castilho]

Hi,

On 19. Nov 2019, at 13:20, Alain Désilets <alainde...@gmail.com> wrote:
>
> Looking at this page:
>
> https://dkpro.github.io/dkpro-core/releases/1.8.0/docs/typesystem-reference.html#type-de.tudarmstadt.ukp.dkpro.core.api.ner.type.Cardinal
>
> I see that dkpro ner api defines types for Cardinal and Ordinal, but those don't have producers listed.
>
> Is there no component in dkpro that can tag those?

At some point - probably even before DKPro Core ever was open sourced - we had some custom-trained models which could produce these. But this was a long long time ago.

There is even an issue about dropping these unused types altogether:

https://github.com/dkpro/dkpro-core/issues/78

That doesn't mean that e.g. cardinal numbers or ordinal numbers couldn't be annotated anymore - just that there wouldn't be any types for them anymore. I.e. one would use a regular "Named entity" annotation and just set the "value" feature to e.g. "ordinal number".

> If not, I am willing to add one. Any suggestion as to which library I should use to add this producer?

E.g. CoreNLP can do such types:

https://stanfordnlp.github.io/CoreNLP/ner.html

So probably the CoreNlpNamedEntityRecognizer could be extended/configured to generate them. Actually - check if it may already produce entity annotations for numeric types and just doesn't map them to the Cardinal / Ordinal annotation types (cf. CoreNlpNamedEntityRecognizer.PARAM_APPLY_NUMERIC_CLASSIFIERS and the mapping files in dkpro-core-corenlp-gpl/src/main/resources/org/dkpro/core/corenlp/lib).

Cheers,

-- Richard