Tagger#

Tagger annotates text with part-of-speech, lemma, language/script, named entities, sentiment, and custom labels from gazetteers.

#Construction

const tagger = new NaturalLanguage.Tagger([
  "lexicalClass",   // part of speech
  "nameType",       // person / place / organization
  "lemma",
  "sentimentScore"
])
tagger.setText("Tim Cook visited Beijing yesterday. I love this city!")

Pass only the schemes you actually need — they control which analyses the tagger performs.

#Tag schemes

type TagScheme =
  | "tokenType"              // word / punctuation / whitespace / other
  | "lexicalClass"           // Noun, Verb, Adjective, ...
  | "nameType"               // PersonalName, PlaceName, OrganizationName
  | "nameTypeOrLexicalClass"
  | "lemma"
  | "language"
  | "script"
  | "sentimentScore"         // numeric string in [-1, 1]

#Methods

#setText(text: string): void

Attach text. All subsequent queries operate on it.

#setLanguage(language: Language, range?: StringRange): void

Override the language hint for the whole text or a sub-range.

#tag(at, unit, scheme): { tag: string | null, range } | null

Look up the tag at a single UTF-16 offset.

#tags(range, unit, scheme, options?): { tag: string | null, range }[]

Enumerate all tags in a range. Pass null as range to scan the whole text.

options (TaggerOptions) lets you skip tokens you don't care about:

• omitWords / omitPunctuation / omitWhitespace / omitOther
• joinNames (treat "John Smith" as a single token)
• joinContractions (treat "don't" as a single token)

#tagHypotheses(at, unit, scheme, maximumCount): { hypotheses, range }

Top-K candidate tags at a position, as { "Noun": 0.7, "Verb": 0.2, ... }.

#enumerateNamedEntities(range?, options?): NamedEntityResult[]

Convenience over tags(..., "word", "nameType", ...): returns person / place / organization mentions only, with joinNames enabled so multi-token names appear as one entity.

#sentimentScore(range?): number | null

Score sentiment in the range [-1, 1] (more negative → more positive). Requires the tagger to have been constructed with "sentimentScore". Scoring starts from range.location at paragraph granularity.

#setGazetteers(gazetteers, scheme): void

Attach one or more Gazetteer to a scheme. Gazetteer hits take precedence over the system model for that scheme. See Gazetteer.

#Examples

#Named entities

const tagger = new NaturalLanguage.Tagger(["nameType"])
tagger.setText("Tim Cook visited Beijing yesterday.")
console.log(tagger.enumerateNamedEntities())
// [
//   { entity: "personalName", text: "Tim Cook", range: { location: 0,  length: 8 } },
//   { entity: "placeName",    text: "Beijing",  range: { location: 17, length: 7 } }
// ]

#Part-of-speech

const tagger = new NaturalLanguage.Tagger(["lexicalClass"])
tagger.setText("Apple released new products.")
const tags = tagger.tags(null, "word", "lexicalClass", {
  omitWhitespace: true,
  omitPunctuation: true
})
// → [ { tag: "Noun", range }, { tag: "Verb", range }, { tag: "Adjective", range }, { tag: "Noun", range } ]

#Sentiment

const tagger = new NaturalLanguage.Tagger(["sentimentScore"])
tagger.setText("I love this beautiful sunny day!")
const score = tagger.sentimentScore()   // → ~0.9 (positive)

#Tag hypotheses

const r = tagger.tagHypotheses(0, "word", "lexicalClass", 3)
// {
//   hypotheses: { "Noun": 0.72, "ProperNoun": 0.18, "Adjective": 0.05 },
//   range: { location: 0, length: 5 }
// }