modest natural-language processing in javascript

_{by
Spencer Kelly and

many contributors}

save yourself from regex-whackamole🤞:

nlp(entireNovel).sentences().if('the #Adjective of times').out()
// "it was the blurst of times??"

move things around:

nlp('she sells seashells by the seashore.').sentences().toFutureTense().out()
// 'she will sell seashells...'

respond to text input:

if( doc.has('^simon says (shoot|fire) #Determiner lazer') ){
  fireLazer()
} else {
  dontFire()
}

compromise is not the cleverest.
but it is small, quick, and good-enough for a bunch of stuff.

<script src>

one javascript file

🙏

npm install compromise

86%

on the Penn treebank

IE9+

caniuse, youbetcha

⚡️ on the Client-side

<script src="https://unpkg.com/compromise@latest/builds/compromise.min.js"></script>
<script>
  var doc = nlp('dinosaur')

  var str = doc.nouns().toPlural().out('text')
  console.log(str)
  // 'dinosaurs'
</script>

🌋 Server-side!

var nlp = require('compromise')

var doc = nlp('London is calling')
doc.sentences().toNegative()
// 'London is not calling'

Get the hang of things:

Tutorial #1

_{Input → output}

Tutorial #2

_{Match & transform}

Tutorial #3

_{Making a bot}

Detailed docs:

API

Full Tagset

Plugins

Outputs

Match Syntax

Examples:

Part-of-Speech tagging _{nouns! verbs! adjectives!}	Named-entities _{people, places, organizations}	Number parsing _{seven hundred and fifty == 750}
Grammar-match _{like a regex for a sentence}	Verb conjugation _{all your base are belong}	Normalization _{case, whitespace, contractions..}

Plural/singular: - grab the noun-phrases, make em plural:

doc = nlp('a bottle of beer on the wall.')
doc.nouns(0).toPlural()
doc.out('text')
//'The bottles of beer on the wall.'

Number parsing: - parse written-out numbers, and change their form:

doc = nlp('ninety five thousand and fifty two')
doc.values().toNumber().out()
// '95052'

doc = nlp('the 23rd of December')
doc.values().add(2).toText()
doc.out('text')
// 'the twenty fifth of December'

Normalization: - handle looseness & variety of random text:

doc = nlp("the guest-singer's björk   at seven thirty.").normalize().out('text')
// 'The guest singer is Bjork at 7:30.'

Tense: - switch to/from conjugations of any verb

let doc = nlp('she sells seashells by the seashore.')
doc.sentences().toFutureTense().out('text')
//'she will sell seashells...'

doc.verbs().conjugate()
// [{ PastTense: 'sold',
//    Infinitive: 'sell',
//    Gerund: 'selling', ...
// }]

Contractions: - grab, expand and contract:

doc = nlp("we're not gonna take it, no we ain't gonna take it.")
doc.has('going') // true
doc.match('are not').length // == 2
doc.contractions().expand().out()
//'we are not going to take it, no we are not going to take it'

Named-entities: - get the people, places, organizations:

doc = nlp('the opera about richard nixon visiting china')
doc.topics().data()
// [
//   { text: 'richard nixon' },
//   { text: 'china' }
// ]

Custom lexicon: - make it do what you'd like:

var lexicon={
  'boston': 'MusicalGroup'
}
doc = nlp('i heard Boston\'s set in Chicago', lexicon)

//alternatively, fix it 'in-post':
doc.match('heard #Possessive set').terms(1).tag('MusicalGroup')

Handy outputs: - get sensible data:

doc = nlp('We like Roy! We like Roy!').sentences().out('array')
// ['We like Roy!', 'We like Roy!']

doc = nlp('Tony Hawk').out('html')
/*
<span>
  <span class="nl-Person nl-FirstName">Tony</span>
  <span>&nbsp;</span>
  <span class="nl-Person nl-LastName">Hawk</span>
</span>
*/

Plugins: - allow adding vocabulary, fixing errors, and setting context quickly:

var plugin = {
  tags:{
    Character:{
      isA: 'Noun'
    }
  },
  words:{
    itchy: 'Character',
    scratchy: 'Character'
  }
}
nlp.plugin(plugin)
nlp(`Couldn't Itchy share his pie with Scratchy?`).debug()
/*
   couldn't   - #Modal, #Verb
   itchy      - #Character, #Noun
   share      - #Infinitive, #Verb
   ...
*/