Customized tokenizer reports 'AssertionError' with specific infixes

[Vimos]

When I trying to add '—-' to the infix_re, whole re is '[\[\]!&:,()\*—\-]', for a custom tokenizer, I got the following error.

Traceback (most recent call last):
  File "/home/vimos/Public/github/NLP/keras-nli/examples/squad/sequence_pointer/preprocessing.py", line 15, in <module>
    o.preprocessing()
  File "/home/vimos/Public/github/NLP/keras-nli/kerasnli/preprocessing/squad.py", line 223, in preprocessing
    self.to_sequence()
  File "/home/vimos/Public/github/NLP/keras-nli/kerasnli/preprocessing/squad.py", line 119, in to_sequence
    context = [word.text for word in nlp(p['context'])]
  File "/data/home/vimos/Public/git/github/NLP/spaCy/spacy/language.py", line 320, in __call__
    doc = self.make_doc(text)
  File "/home/vimos/Public/github/NLP/keras-nli/kerasnli/preprocessing/squad.py", line 36, in <lambda>
    return lambda text: tokenizer(text)
  File "tokenizer.pyx", line 152, in spacy.tokenizer.Tokenizer.__call__
  File "tokenizer.pyx", line 217, in spacy.tokenizer.Tokenizer._tokenize
  File "tokenizer.pyx", line 315, in spacy.tokenizer.Tokenizer._attach_tokens
  File "doc.pyx", line 469, in spacy.tokens.doc.Doc.push_back
AssertionError

I am using the head of spacy, but same error on an earlier version. Only '—-' is causing this problem.

After some hacking, I find a suspect for the error, trailing '-' in the string

# coding: utf-8
from __future__ import unicode_literals

from ... import load
from ...tokenizer import Tokenizer
from ... import util

import pytest


def test_customized_tokenizer_handles_infixes():
    def custom_tokenizer(nlp_model):
        prefix_re = util.compile_prefix_regex(nlp_model.Defaults.prefixes)
        suffix_re = util.compile_suffix_regex(nlp_model.Defaults.suffixes)
        custom_infixes = ['\.\.\.+',
                          '(?<=[0-9])-(?=[0-9])',
                          # '(?<=[0-9]+),(?=[0-9]+)',
                          '[0-9]+(,[0-9]+)+',
                          u'[\[\]!&:,()\*—–\/-]']

        infix_re = util.compile_infix_regex(custom_infixes)

        # infix_re = re.compile(ur'[\[\]!&:,()]')

        tokenizer = Tokenizer(nlp_model.vocab,
                              nlp_model.Defaults.tokenizer_exceptions,
                              prefix_re.search,
                              suffix_re.search,
                              infix_re.finditer,
                              token_match=None)
        return lambda text: tokenizer(text)

    nlp = load('en', create_make_doc=custom_tokenizer)

    sentence = "The 8 and 10-county definitions are not used for the greater Southern California Megaregion."
    context = [word.text for word in nlp(sentence)]
    assert context == [u'The', u'8', u'and', u'10', u'-', u'county', u'definitions', u'are', u'not', u'used',
                       u'for',
                       u'the', u'greater', u'Southern', u'California', u'Megaregion', u'.']

    # the trailing '-' may cause Assertion Error
    sentence = "The 8- and 10-county definitions are not used for the greater Southern California Megaregion."
    context = [word.text for word in nlp(sentence)]
    assert context == [u'The', u'8', u'-', u'and', u'10', u'-', u'county', u'definitions', u'are', u'not', u'used',
                       u'for',
                       u'the', u'greater', u'Southern', u'California', u'Megaregion', u'.']

Your Environment

• Operating System: Ubuntu 16.04
• Python Version Used: python 2.7
• spaCy Version Used: master head
• Environment Information:

[Vimos]

I have found a solution for this bug, I'd like to push a pull request.

diff --git a/spacy/tokenizer.pyx b/spacy/tokenizer.pyx
index 276f0ef..799e4bd 100644
--- a/spacy/tokenizer.pyx
+++ b/spacy/tokenizer.pyx
@@ -312,7 +312,8 @@ cdef class Tokenizer:
 
                         start = infix_end
                     span = string[start:]
-                    tokens.push_back(self.vocab.get(tokens.mem, span), False)
+                    if span:
+                        tokens.push_back(self.vocab.get(tokens.mem, span), False)
         cdef vector[const LexemeC*].reverse_iterator it = suffixes.rbegin()
         while it != suffixes.rend():
             lexeme = deref(it)

[honnibal]

Thanks, this looks good!

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.