Add some custom words to tokenizer in Spacy

Question

I have a sentence and would like to see expected tokens as following.

Sentence: "[x] works for [y] in [z]."
Tokens: ["[", "x", "]", "works", "for", "[", "y", "]", "in", "[", "z", "]", "."]
Expected: ["[x]", "works", "for", "[y]", "in", "[z]", "."]

How can I do this by a custom tokenizer function?

Answer 1

You can remove [ and ] from the tokenizer prefixes and suffixes so that the brackets are not split off from adjacent tokens:

import spacy
nlp = spacy.load('en_core_web_sm')

prefixes = list(nlp.Defaults.prefixes)
prefixes.remove('\\[')
prefix_regex = spacy.util.compile_prefix_regex(prefixes)
nlp.tokenizer.prefix_search = prefix_regex.search

suffixes = list(nlp.Defaults.suffixes)
suffixes.remove('\\]')
suffix_regex = spacy.util.compile_suffix_regex(suffixes)
nlp.tokenizer.suffix_search = suffix_regex.search

doc = nlp("[x] works for [y] in [z].")
print([t.text for t in doc])
# ['[x]', 'works', 'for', '[y]', 'in', '[z]', '.']

The relevant documentation is here:

https://spacy.io/usage/linguistic-features#native-tokenizer-additions