Pampac

[joancf]

One of the reasons to use GateNlp was Pampac. I find powerful the capacity to merge different kinds of annotations in the matching.

Now I'm facing a "simple" problem: Detec sentences that start with a Number and a : like
" 10.

I tried with:

r2 = Rule(
        Seq(AnnAt("Sentence"),Ann("Token", features=dict(is_digit=True)), Ann("Token", features=dict(lemma=".")) , name="seq2"),
        AddAnn(name="seq2", type="RefNum")
)
outset = doc.annset("Out2")
annSet=doc.annset("Original markups").with_type(["Paragraph","Sentence","Token"])
pars = annSet.with_type("Paragraph")
for par in pars:
        anns= annSet.within(par)
        Pampac(r2).run(doc, anns, outset=outset)
doc

But does not detect nothing (if I remove AnnAt("Sentence") then I get some matches (some of them not at the begining)

Thanks!!

[johann-petrak]

Could you please describe more exactly the pattern you intend to match?
I notice that the rule you try will look for, in order, a Sentence annotation, then Token annotation with the is_digit feature then a Token annotation withe lemma="." feature. But I think you want to match a Token that starts together with a sentence and has the is_digit feature true followed by another token etc.

In order to match an annotation at the beginning of a sentence you need a construct like

Ann("Token", features=dict(is_digit=True)).at("Sentence")

Note that the difference between Ann and AnnAt relates to something different: the annotations in the input set form a specific sequence, and even if e.g. Ann1 and Ann2 start at the same offset, the will occur in a certain order in the input annotation set. Ann will match the next annotation in that sequence, while AnnAt will match any annotation at the next offset where one or more annotations occur in the sequence.

BTW it is usually a good idea to restrict the input set to only annotations that are actually ever matched in a rule.

In your case something like this may work:

r2 = Rule(
        Seq(
            AnnAt("Token", features=dict(is_digit=True)).at("Sentence"), 
            Ann("Token", features=dict(lemma=".")) , name="seq2"),
        AddAnn(name="seq2", type="RefNum")
)

[joancf]

Thanks Johann,
It worked but even knowing the solution I don't know how to get there using the documentation ;-( . Ithink that each function needs an example.. (and maybe the presentation or the notebook need also more examples) at least each functionality should appear in one example. I would help writing them, but I don't have the knowledge to do so.

[johann-petrak]

Yes, you are right, currently a lot of what can be done is only documented in the PythonDoc of the code or through the code itself. We definitely need more examples and a more detailled documentation.
I am sorry, as we share this software as open source without explicit funding there is sometimes not enough time/resources to do this quickly.

[joancf]

I thank you for releasing it as open source, and documentation is always a problem!! You don't need to apologize for that!
But better documentation -> more users -> more feedback -> more bugs fixed ;-)