Bug in Span.sents

[nrodnova]

When a Doc's entity is in the second to the last sentence, and the last sentence consists only of one token, entity.sents includes that last 1-token sentence (even though the entity is fully contained by the previous sentence.

How to reproduce the behaviour

text = "This is a sentence. This is another sentence. Third"
doc = nlp.tokenizer(text)
doc[0].is_sent_start = True
doc[5].is_sent_start = True
doc[10].is_sent_start = True
doc.ents = [('ENTITY', 7, 9)] # "another sentence" phrase in the second sentence
entity = doc.ents[0]
print(f"Entity: {entity}. Sentence: {entity.sent} Sentences: {list(entity.sents)}")

Output:

Entity: another sentence. Sentence: This is another sentence. Sentences: [This is another sentence., Third]

Your Environment

• Operating System:
• Python Version Used:
• spaCy Version Used:
• Environment Information: