Support for user-defined column labels in CoNLL formats?

[Christian Chiarcos]

Dear all,

right now, Inception support for CoNLL formats seems to be limited to one exactly fixed set of columns per format. Would it be possible to extend this to support custom CoNLL formats? The reason I'm asking is that we are routinely merging different CoNLL formats (https://github.com/acoli-repo/conll-merge), and after every merge, we have a different set of columns, say, three concatenated CoNLL-U files after each other to represent the output of different parsers, with a total of 30 columns. (But we can append other stuff, too.)

Thanks,

Christian

[Richard Eckart de Castilho]

Hi Christian,

> On 27. Jul 2026, at 16:13, Christian Chiarcos <christian...@gmail.com> wrote:
>
> right now, Inception support for CoNLL formats seems to be limited to one exactly fixed set of columns per format. Would it be possible to extend this to support custom CoNLL formats?

WebAnno TSV is/was a tab-separated format loosely inspired by CoNLL which supports custom annotation layers to quite a large degree.

I wouldn't call it a CoNLL format though. The traditional CoNLL formats were those used by the respective CoNLL shared tasks around
the first decade of the 2000s. And those have a fixed set of columns. Then there is CoNLL-U which also has a fixed set of columns
and a rather lax set of rules how custom information can be encoded.

What is the goal that you seek to achieve?

Cheers,

-- Richard

[Christian Chiarcos]

Hi Richard,

I think WebAnno TSV should solve most of our problems. Thank you. My immediate concern are two CoNLL formats designed for (a) the Cuneiform Digital Library Initiative, and (b) Role and Reference Grammar. Both could be extended to conform to CoNLL-U Plus, but they don't have the obligatory DEPS, MISC and FEAT columns, and they use non-UD dependency labels.

As a side-note on CoNLL formats, you are right. More precisely this is to be referred to as the CoNLL format family, not as a single format. It is true that most CoNLL formats have fixed columns (CoNLL-2004, CoNLL-2005, later PropBank skel formats don't, not sure about the CoNLL-2015/16 TSV formats), but there are many of them, and when aggregating annotations in multi-layer corpora, there is a need for some generic technology for the format family in general (which we have, but for backend technologies, mostly, e.g., with https://github.com/acoli-repo/conll-rdf and https://github.com/acoli-repo/conll-merge).

Best,

Christian

--
You received this message because you are subscribed to a topic in the Google Groups "inception-users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/inception-users/1IFcqD0DK3M/unsubscribe.
To unsubscribe from this group and all its topics, send an email to inception-use...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/inception-users/6777F319-9BBB-4C82-9CED-20EB1EEEFA2E%40gmail.com.

[Richard Eckart de Castilho]

Hi Christian,

> On 28. Jul 2026, at 10:04, Christian Chiarcos <christian...@gmail.com> wrote:
>
> I think WebAnno TSV should solve most of our problems. Thank you. My immediate concern are two CoNLL formats designed for (a) the Cuneiform Digital Library Initiative, and (b) Role and Reference Grammar. Both could be extended to conform to CoNLL-U Plus, but they don't have the obligatory DEPS, MISC and FEAT columns, and they use non-UD dependency labels.

In case you have not found it yet, here is a specification of the format:

https://inception-project.github.io/releases/41.2/docs/user-guide.html#sect_webannotsv

And here is a comprehensive test suite for implementations of the format:

https://github.com/inception-project/inception/tree/main/inception/inception-io-webanno-tsv/src/test/resources/tsv3-suite

Cheers,

-- Richard