Applying Matcher to Spans with as_spans set to True

[thiippal]

Hey,

I'm a bit puzzled by the behaviour of the Matcher class. See the example below:

import spacy
from spacy.matcher import Matcher

nlp = spacy.load("en_core_web_sm")

matcher = Matcher(vocab=nlp.vocab)

matcher.add("verbs", patterns=[[{"POS": "VERB"}]])

doc = nlp("This sentence seeks to provide an example, which shows the weird behaviour.")

matches = matcher(doc, as_spans=True)

print(matches)

# Define some arbitrary Spans and add to list 'spans'
spans = [doc[0:3], doc[3:6], doc[6:13]]

for span in spans:

	results = matcher(span)

	for result in results:

		print(span[result[1]:result[2]])

for span in spans:

	results_2 = matcher(span, as_spans=True)

	print(results_2)

This gives the following output:

[seeks, provide, shows]
seeks
provide
shows
[seeks]
[sentence]
[to]

What happens with the second loop over the list spans when matching Span objects while setting the attribute as_spans to True?

[adrianeboyd]

Hi, that looks like a bug in how the offsets are handled when the spans are created. We will look into it!

[adrianeboyd]

Hopefully addressed by #7992.