Performance on existing tagging datasets

[dennlinger]

Hi Paul,
I'm a researcher working with temporal tagging, and was wondering if you do have any information on the tagging performance of Timexy on some of the existing datasets, such as TempEval (for English) or KRAUTS (for German).

I'm generally curious if you could comment on what particular tags/temporal expressions your package works well with, and what it might struggle on (normalization, language-specific expressions, etc.) or how it stacks up against existing packages like Heideltime. Let me know if you wanted some of the evaluation results in your future work, I'm happy to help out with that (currently busy this and next week, but free to contribute after).

Best,
Dennis

[paulrinckens]

Hi Dennis,

there is no evaluation of Timexy on public datasets yet.

Currently, the temporal expression extraction and normalization is based on rules only, which makes it easy to assess and track on which expressions it works well. For expressions of type date, you can refer to the existing Timexy language modules to see transparently which patterns are covered.

However, if you plan to evaluate Timexy on any public datasets, your contribution is very much welcomed.

Best,

Paul

[chlor]

Hi Paul,

What is the right configuration to use the Timexy Language modules?

If I try to use timexy to parse German language data strings (e.g., 10.10.2023), I have an error like this:

Span could not be retrieved for annotation of type timexy for datestring 10.10.2023 with character offsets (14, 24). Skipping the match.

Do you have a code example to configure the detection or annotation of German data strings?

Best, Christina