Matcher pattern with NOT_IN and extension attributes

[vito-detullio-finconsgroup]

I'm trying to write a Matcher pattern that use a pair of extension attributes that I set in my pipeline, but I have errors:

what I'm trying to achieve is a pattern that matches every sequence of tokens that is not an end_phrase and a section_marker (that I have implemented as extension attributes). So if a token is an end_phrase, but not a section_marker (or viceversa, or neither) should be accepted, but if is both of them, should stop the match.

I tried to write something like the following snippet:

nlp = ... 
pattern = [{
	"OP": "+",
	"_": {
		"NOT_IN": [{
				"end_phrase": True
			},
			{
				"section_marker": True
			}
		]

	}
}]
matcher = Matcher(nlp.vocab, validate=True)
matcher.add('mymatch', [pattern], greedy='LONGEST')

As I have already used NOT_IN in other scenarios (with POS or ENT_TYPE values, mostly), but I have this error:

  File "spacy\matcher\matcher.pyx", line 128, in spacy.matcher.matcher.Matcher.add
  File "spacy\matcher\matcher.pyx", line 720, in spacy.matcher.matcher.init_pattern
TypeError: an integer is required

What is the "right" way to write the pattern?

[adrianeboyd]

The _ values can only be specified individually, so you'd want something like:

{"_":` {"end_phrase": True, "section_marker": True}}

[vito-detullio-finconsgroup]

Maybe I'm missing something, but this is exactly the opposite of the pattern I want - as in: this pattern capture only tokens that are both with end_phrase and section_marker as true.
How can I use this?

[adrianeboyd]

You'd have to add multiple patterns for each combination that you want to match.

Edited to add: Or to keep a pattern with only one attribute, you could add a third custom extension with a getter that combines the values however you need for your pattern.