Is there a way to write a rule containing "OR" in the Matcher ?

[lio-data]

Hi,
I am using the Matcher to mark some text. I have quite a few complex rules and I would need to generalize it. Here is one of them :
[[
{"IS_SENT_START": true}, {"IS_SENT_START": false, "OP": "{0,9}"}, {"LOWER": ":"},
{"OP": "?"}, {"LIKE_NUM": true}, {"_": {"is_unit": true}}, {"OP": "?"},{"TEXT": {"REGEX": "\n"}}
]]

I would want to have the Matcher use LIKE_NUM or another custom attribute "._.is_cn". is_cn is simply a custom extension that I put on complicated text describing numbers likes 64x7 or 7/2. My pipeline recognize these complex representation of numbers.

So to some extent I would like the Matcher to be able to use this pattern :
[[
{"IS_SENT_START": true}, {"IS_SENT_START": false, "OP": "{0,9}"}, {"LOWER": ":"},
{"OP": "?"}, {"LIKE_NUM": true} OR {"": {"is_cn": true}}, {"": {"is_unit": true}}, {"OP": "?"},{"TEXT": {"REGEX": "\n"}}
]]

I have not seen any way to do that...

Currently the only I see to achieve my goal is either

• to duplicate all my patterns and replace the {"LIKE_NUM": true} by {"_": {"is_cn": true}} but this is just not great... my list of patterns will become unreadable !
• to create another custom boolean token extension which I put as False unless LIKE_NUM or is_cn is True and then use that extension for the Matcher. This is currently my favorite option but it is not great either. LIKE_NUM is not an writable attribute of token as otherwise I would have simply put LIKE_NUM = True for all token where ._.is_cn = True

Any better idea ?

[svlandeg]

Hey @lio-data,

This is perhaps a naive suggestion, but when you're assigning things to be ._.is_cn in your custom code, can't you also label all numbers as such, so that you don't need the LIKE_NUM part anymore? For English, you can find its definition here.

One other alternative that is maybe a little more involved, but would result in nicer matcher patterns is this: you can influence the definition of the LIKE_NUM attribute directly by going in and writing new language defaults and replacing the lex_attr_getters with your own implementation. For more information on that, see here: https://spacy.io/usage/linguistic-features#language-data

[lio-data]

@svlandeg thank you for the reply. Your suggestion is in fact maybe the best option. I am considering using ._.is_cn to mark all kind of numbers. My other alternative is to create a function/class to create the rules for the matcher. Thank you !