4

I am tinkering with the NSLinguisticTagger.

Identifying basic word types like noun, verb, prepositions works really well.

However the recognition of person names NSLinguisticTagPersonalName hardly works in my tests (iOS8). Places NSLinguisticTagPlaceName also seem to work pretty well, however most of the times also person names are categorised as places.

Here's my basic setup (using NSLinguisticTagSchemeNameTypeOrLexicalClass)

    var tagger:NSLinguisticTagger = NSLinguisticTagger(tagSchemes: NSLinguisticTagger.availableTagSchemesForLanguage("en") , options: 3)
    tagger.string = entryString
    tagger.enumerateTagsInRange(NSMakeRange(0, entryString.length), scheme: NSLinguisticTagSchemeNameTypeOrLexicalClass, options: (NSLinguisticTaggerOptions.OmitWhitespace | NSLinguisticTaggerOptions.JoinNames), usingBlock: {
        tag,tokenRange,sentenceRange,_ in
        let token = entryString.substringWithRange(tokenRange)
        println("[\(tag)] \(token) \(tokenRange)")

Example 1

 "Meeting with John in Paris"

  Evaluation

 [Verb] Meeting
 [Preposition] with
 [Noun] John
 [Preposition] in
 [PlaceName] Paris

Example 2

 "Meeting with John"

  Evaluation

 [Verb] Meeting (0,7)
 [Preposition] with (8,4)
 [PlaceName] John (13,4)

Any idea how I could improve the matching for person names?

Also I'd be interested to know how a Name would need to appear to be recognized. (I assumed e.g. a preposition like "with" would be a good indicator … apparently this isn't enough). I'd appreciate any ideas or additional insights on this. It's an exciting field.

Bernd
  • 11,133
  • 11
  • 65
  • 98
  • 1
    Heya @Bernd! I saw this question has been here for a while without an answer. I wanted to PM you, but Stack won't allow it, so I'm commenting here instead, hoping you'll see this. I'm tinkering with NSLinguisticTagger myself right now, and I have the exact same issue as you. It makes little sense that "Sam Smith is from Hong Kong" gets recognized properly (both Sam Smith as PersonalName and HongCong as PlaceName) but "My name is Sam Smith" doesn't. Any experience you could share would be greatly appreciated! Thank you so much and have an amazing day! – borg123 Feb 15 '16 at 11:31
  • @borg123 My current take on this is to simply avoid using PlaceName and PersonName tags from the NSLinguisticTagger. Those tags just don’t work well enough. Instead I would rather use the preposition tag as clues and create your own tagger based on those. – Bernd Feb 15 '16 at 14:54

1 Answers1

3

Apparently the correct answer is: "wait a few years for Apple to improve NSLinguisticTagger in Swift 4"

Here is the Swift 4 code written and executed in Xcode 9 (beta): 

let entryString = "Meeting with John"

let schemes = NSLinguisticTagger.availableTagSchemes(forLanguage: "en")
let options: NSLinguisticTagger.Options = [
    .omitWhitespace, .omitPunctuation, .joinNames
]

let tagger = NSLinguisticTagger(tagSchemes: schemes, options: Int(options.rawValue))
tagger.string = entryString

let rangeOfEntireEntryString = NSRange(location: 0, length: entryString.utf16.count)

tagger.enumerateTags(
    in: rangeOfEntireEntryString,
    scheme: .nameTypeOrLexicalClass,
    options: options)
{ (tag, tokenRange, sentenceRange, _) in
    guard let tag = tag?.rawValue else { return }
    let token = (entryString as NSString).substring(with: tokenRange)
    print("[\(tag)] \(token) \(tokenRange)")
}

and here are results with your first example string:

let entryString = "Meeting with John in Paris"

[Noun] Meeting {0, 7}
[Preposition] with {8, 4}
[PersonalName] John {13, 4}
[Preposition] in {18, 2}
[PlaceName] Paris {21, 5}

and your second example string:

let entryString = "Meeting with John"

[Noun] Meeting {0, 7}
[Preposition] with {8, 4}
[PersonalName] John {13, 4}
Jeffrey Fulton
  • 4,146
  • 1
  • 17
  • 17