Is it possible to pass nlp object to registry functions via config?

[kinghuang]

I'm using a SpanRuler with a custom spans_filter. A registry function is used in the config to create the filter function.

[components.custom_span_ruler]
factory = "span_ruler"
spans_filter = {"@misc":"some_name.some_spans_filter.v1"}

@registry.misc("some_name.some_spans_filter.v1")
def make_some_spans_filter():
    def some_spans_filter(
        existing_spans: Iterable[Span],
        new_spans: Iterable[Span],
        sequence_patterns: dict=sequence_patterns,
    ) -> List[Span]:
        return new_spans

    return some_spans_filter

Is there a way in the config to refer to the nlp (Language) object, so that I can pass it to the registry function as an argument on make_some_spans_filter? I tried looking at the confection and spacy code, but couldn't tell if there was some magical variable or keyword that would provide such a reference (the ??? below).

[components.custom_span_ruler]
factory = "span_ruler"
spans_filter = {"@misc":"some_name.some_spans_filter.v1", "nlp": "???"}

@registry.misc("some_name.some_spans_filter.v1")
def make_some_spans_filter(nlp: Language):
   …

[adrianeboyd]

Off the top of my head, I don't think this is possible. What information do you need from nlp for the spans filter?

[kinghuang]

I'd like to get the StringStore (nlp.vocab.strings) and also access another pipeline component for some reference data.

[adrianeboyd]

You can access the same vocab from within a span under span.doc.vocab.

Maybe you can have the other component add a custom attribute to the doc with the reference data?

It's fine for components to depend on annotation from an earlier component, but having one component try to dig into the state or settings of another component is not going to be recommended.