The text search tokenizer considers only the characters \\\f\v\t\r\n\'~`!@#$%^&*(-=+[]{}|;:"<>,. /? as token delimiters. As such, words adjacent to a curly quote (or an emdash, etc) won't be recognized as a stopword or indexed under the correct term. The set of recognized delimiters needs to be expanded and made unicode-aware.

Original ticket description:

I find that directional quotes can affect the ranking of search terms in the $text search.

This is a problem because our ranking is supposed to look at words, and should ignore symbols.

Steps to reproduce:

I have the following document*:

{
  "_id" : {
    "chapter" : "23",
    "bookname" : "Ezekiel",
    "verse" : "2"
    },
  "text" : "“Son of man, there were two women who were daughters of the same mother."
}

(note that the text character that appears at the beginning of the string in the "text" field is not a standard quote I get from shift-")

When I run the following query:

> db.bible.runCommand("text", { search : "Israel \"Son of man\""} )

the text gets a score of "score" : 0.5833333333333334

But, when I remove that weird quite and re-insert the document under a different _id, it gets a score of "score" : 1.1666666666666667

I would expect them both to get the same score, and for that score to be greater than 1, since it matches one of my search strings.

• Sorry for the heavily Christian text in the example, I found a random bible verse generating JSON API and just ran with it today.