TokensRegex not merging entity mentions - stanford-nlp

Question

I have noticed that the Stanford default entity mention tags don't merge with my new tags. For example; I want to find and tag telephone numbers. [phone] before my tags get run, Standford tags "+1" as "NUMBER" and "234-567-8901" also as "NUMBER" however, they remain as separate tags. I have tried creating my own TokensRegex rule to find these and mark them as "TELEPHONE_NUMBER". They are successfully changed to the new tag, but they remain separate tags.
I have also noticed that this happens when tagging text and digits. Say if I wanted to tag "my number is 234-567-8901" then it would tag "my number is" as one tag and "234-567-8901" as a separate tag. I have tried using the B-TELEPHONE_NUMBER and I-TELEPHONE_NUMBER tags as mention in the documentation and it does remove the B and I, but they remain separate tags. I have noticed this behavior only happens with Stanford's tags. I have tested it with DATE and it does the same thing
So, my question is, how can I get Stanford's tags to merge with mine? If they can't be merged, is there a way to delete that tag and then add my own tag?
EDIT
here is another example of the tags not merging together.
ner = { type: "CLASS", value: "edu.stanford.nlp.ling.CoreAnnotations$NamedEntityTagAnnotation" }
$CORONA = "/((c[A-za-z]*?r[A-za-z]*?a)(.*?v[A-za-z]*?r[A-za-z]*?s)?)/"
$COVID = "/(((c[A-za-z]*?v[A-za-z]*?d)(.*?19)?|(c[A-za-z]*?d.*?19)))/"
{ pattern: ( (([/c.*?[vd].*/]) ([/19/]))|([/c.*?a/] [/v.*?s/])|([ $CORONA|$COVID ]) ),
action: (Annotate($0, ner, "CORONAVIRUS")) }
This snippet is suppose to catch covid 19
(([/c.*?[vd].*/]) ([/19/]))
However, 19 is tagged as a NUMBER before this rule is run. it tags covid as CORONAVIRUS and changes 19's tag from NUMBER to CORONAVIRUS. It's suppose to then combine the two into one entity mention, but it doesn't.
When it comes across corona virus, which is also separated into two tokens, it combines the two into one entity mention.
([/c.*?a/] [/v.*?s/])
EDIT 2
I decided to make a small pipeline to replicate the issue. I based it off the example found here https://stanfordnlp.github.io/CoreNLP/tokensregex.html I made modification to more closely replicate the one used in our pipeline. here is the function
public static void main(String[] args) throws ClassNotFoundException {
// set properties
Properties props = new Properties();
props.setProperty("annotators", "tokenize,ssplit,pos,lemma,ner");
props.setProperty("ner.additional.tokensregex.rules", "basic_ner.rules");
// build pipeline
StanfordCoreNLP pipeline = new StanfordCoreNLP(props);
// annotate
Annotation ann = new Annotation("We had to close our business due to covid 19. " +
"There will be a big announcement by Apple Inc today at 5:00pm. " +
"She has worked at Miller Corp. for 5 years.");
pipeline.annotate(ann);
// show results
System.out.println("---");
System.out.println("tokens\n");
for (CoreMap sentence : ann.get(CoreAnnotations.SentencesAnnotation.class)) {
for (CoreMap mention : sentence.get(CoreAnnotations.MentionsAnnotation.class)){
System.out.println("mention: " + mention);
}
System.out.println("---");
for (CoreLabel token : sentence.get(CoreAnnotations.TokensAnnotation.class)) {
System.out.println(token.word() + "\t" + token.ner());
}
System.out.println("");
}
System.out.println("---");
System.out.println("matched expressions\n");
for (CoreMap me : ann.get(MyMatchedExpressionAnnotation.class)) {
System.out.println(me);
}
}
Here is the output
---
tokens
mention: covid
mention: 19
---
We O
had O
to O
close O
our O
business O
due O
to O
covid CORONAVIRUS
19 CORONAVIRUS
. O
mention: Apple Inc
mention: today at 5:00pm.
---
There O
will O
be O
a O
big O
announcement O
by O
Apple COMPANY
Inc COMPANY
today DATE
at DATE
5:00 DATE
pm DATE
. DATE
mention: Miller Corp.
mention: 5 years
mention: She
---
She O
has O
worked O
at O
Miller COMPANY
Corp. COMPANY
for O
5 DURATION
years DURATION
. O
---
As you can see the tags like COMPANY are merged into one mention when next to each other. However, like I mentioned above, covid 19 remains separate. This happens when a portion of the tokens already have a tag. In this case 19 was NUMBER.