Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with
or
.
Download ZIP
natural language processing in the browser
JavaScript Other
branch: master

README.md

No training, no prolog.

a Natural-Language-Processing library in Javascript, small-enough for the browser, and quick-enough to run on keypress :two_men_holding_hands:

it does tons of clever things. it's smaller than jquery, and scores 86% on the Penn treebank.

nlp.pos('she sells seashells by the seashore').to_past().text()
//she sold seashells by the seashore

npm version

Check it out

Video

Justification

If the 80-20 rule applies for most things, the ''94-6 rule'' applies when working with language - by Zipfs law:

The top 10 words account for 25% of used language.

The top 100 words account for 50% of used language.

The top 50,000 words account for 95% of used language.

On the Penn treebank, for example, this is possible:

  • just a 1 thousand word lexicon: 45% accuracy
  • ... then falling back to nouns: 70% accuracy
  • ... then some suffix regexes: 74% accuracy
  • ... then some sentence-level postprocessing: 81% accuracy

The process is to get some curated data, find the patterns, and list the exceptions. Bada bing, bada boom. In this way a satisfactory NLP library can be built with breathtaking lightness.

Namely, it can be run right on the user's computer instead of a server.

Client-side

<script src="https://rawgit.com/spencermountain/nlp_compromise/master/client_side/nlp.min.js"> </script>
<script>
  nlp.noun("dinosaur").pluralize()
  //dinosaurs
</script>

or, use the angular module

Server-side

$ npm install nlp_compromise

nlp = require("nlp_compromise")
nlp.syllables("hamburger")
//[ 'ham', 'bur', 'ger' ]

API

Sentence methods

  var s= nlp.pos("Tony Danza is dancing").sentences[0]

  s.tense()
  //present

  s.text()
  //"Tony Danza is dancing"

  s.to_past().text()
  //Tony Danza was dancing

  s.to_present().text()
  //Tony Danza is dancing

  s.to_future().text()
  //Tony Danza will be dancing

  s.negate().text()
  //Tony Danza is not dancing

  s.tags()
  //[ 'NNP', 'CP', 'VB' ]

  s.entities()
  //[{text:"Tony Danza"...}]

  s.people()
  //[{text:"Tony Danza"...}]

  s.nouns()
  //[{text:"Tony Danza"...}]

  s.adjectives()
  //[]

  s.adverbs()
  //[]

  s.verbs()
  //[{text:"dancing"}]

  s.values()
  //[]

as sugar, these methods can be called on multiple sentences from the nlp.pos() object too, like:

nlp.pos("Tony is cool. Jen is happy.").people()
//[{text:"Tony"}, {text:"Jen"}]

Noun methods:

nlp.noun("earthquakes").singularize()
//earthquake

nlp.noun("earthquake").pluralize()
//earthquakes

nlp.noun('veggie burger').is_plural()
//false

nlp.noun('tony danza').is_person()
//true
nlp.noun('Tony J. Danza elementary school').is_person()
//false
nlp.noun('SS Tony danza').is_person()
//false

nlp.noun('hour').article()
//an

nlp.noun('mayors of toronto').conjugate()
//{ plural: 'mayors of toronto', singular: 'mayor of toronto' }

nlp.noun("tooth").pronoun()
//it
nlp.noun("teeth").pronoun()
//they
nlp.noun("Tony Hawk").pronoun()
//"he"
nlp.noun("Nancy Hawk").pronoun()
//"she"

var he = nlp.pos("Tony Danza is great. He lives in L.A.").sentences[1].tokens[0]
he.analysis.reference_to()
//{text:"Tony Danza"...}

Verb methods:

nlp.verb("walked").conjugate()
//{ infinitive: 'walk',
//  present: 'walks',
//  past: 'walked',
//  gerund: 'walking'}
nlp.verb('swimming').to_past()
//swam
nlp.verb('swimming').to_present()
//swims
nlp.verb('swimming').to_future()
//will swim

Adjective methods:

nlp.adjective("quick").conjugate()
//  { comparative: 'quicker',
//    superlative: 'quickest',
//    adverb: 'quickly',
//    noun: 'quickness'}

Adverb methods

nlp.adverb("quickly").conjugate()
//  { adjective: 'quick'}

Part-of-speech tagging

86% on the Penn treebank

nlp.pos("Tony Hawk walked quickly to the store.").tags()
// [ [ 'NNP', 'VBD', 'RB', 'IN', 'DT', 'NN' ] ]

nlp.pos("they would swim").tags()
// [ [ 'PRP', 'MD', 'VBP' ] ]
nlp.pos("the obviously good swim").tags()
// [ [ 'DT', 'RB', 'JJ', 'NN' ] ]

Named-Entity recognition

nlp.spot("joe carter loves toronto")
// [{text:"joe carter"...}, {text:"toronto"...}]

Sentence segmentation

nlp.sentences("Hi Dr. Miller the price is 4.59 for the U.C.L.A. Ph.Ds.").length
//1

nlp.tokenize("she sells sea-shells").length
//3

Syllable hyphenization

70% on the moby hyphenization corpus 0.5k

nlp.syllables("hamburger")
//[ 'ham', 'bur', 'ger' ]

US-UK Localization

nlp.americanize("favourite")
//favorite
nlp.britishize("synthesized")
//synthesised

N-gram

str= "She sells seashells by the seashore. The shells she sells are surely seashells."
nlp.ngram(str, {min_count:1, max_size:5})
// [{ word: 'she sells', count: 2, size: 2 },
// ...
options.min_count // throws away seldom-repeated grams. defaults to 1
options.max_gram // prevents the result from becoming gigantic. defaults to 5

Date parsing

nlp.value("I married April for the 2nd time on June 5th 1998 ").date()
// { text: 'June 5th 1998',
//   from: { year: '1998', month: '06', day: '05' },
//   to: {} }

Number parsing

nlp.value("two thousand five hundred and sixty").number()
//2560
nlp.value("ten and a half million").number()
//15000000

Unicode Normalisation

a hugely-ignorant, and widely subjective transliteration of latin, cryllic, greek unicode characters to english ascii.

nlp.normalize("Björk")
//Bjork

and for fun,

nlp.denormalize("The quick brown fox jumps over the lazy dog", {percentage:50})
// The ɋӈїck brown fox juӎÞs over tӊe laζy dog

Details

Tags

  "verb":
    "VB" : "verb, generic (eat)"
    "VBD" : "past-tense verb (ate)"
    "VBN" : "past-participle verb (eaten)"
    "VBP" : "infinitive verb (eat)"
    "VBZ" : "present-tense verb (eats, swims)"
    "VBF" : "future-tense verb (will eat)"
    "CP" : "copula (is, was, were)"
    "VBG" : "gerund verb (eating,winning)"
  "adjective":
    "JJ" : "adjective, generic (big, nice)"
    "JJR" : "comparative adjective (bigger, cooler)"
    "JJS" : "superlative adjective (biggest, fattest)"
  "adverb":
    "RB" : "adverb (quickly, softly)"
    "RBR" : "comparative adverb (faster, cooler)"
    "RBS" : "superlative adverb (fastest (driving), coolest (looking))"
  "noun":
    "NN" : "noun, singular (dog, rain)"
    "NNP" : "singular proper noun (Edinburgh, skateboard)"
    "NNPA" : "noun, acronym (FBI)"
    "NNAB" : "noun, abbreviation (jr.)"
    "NNPS" : "plural proper noun (Smiths)"
    "NNS" : "plural noun (dogs, foxes)"
    "NNO" : "possessive noun (spencer's, sam's)"
    "NG" : "gerund noun (eating,winning" : "but used grammatically as a noun)"
    "PRP" : "personal pronoun (I,you,she)"
    "PP" : "possessive pronoun (my,one's)"
  "glue":
    "FW" : "foreign word (mon dieu, voila)"
    "IN" : "preposition (of,in,by)"
    "MD" : "modal verb (can,should)"
    "CC" : "co-ordating conjunction (and,but,or)"
    "DT" : "determiner (the,some)"
    "UH" : "interjection (oh, oops)"
    "EX" : "existential there (there)"
  "value":
    "CD" : "cardinal value, generic (one, two, june 5th)"
    "DA" : "date (june 5th, 1998)"
    "NU" : "number (89, half-million)"

Lexicon

Because the library can conjugate all sorts of forms, it only needs to store one grammatical form. The lexicon was built using the American National Corpus, then intersected with the regex rule-list. For example, it lists only 300 verbs, then blasts-out their 1200+ derived forms.

Contractions

It puts a 'silent token' into the phrase for contractions. Otherwise a meaningful part-of-speech could be neglected.

nlp.pos("i'm good.")
 [{
   text:"i'm",
   normalised:"i",
   pos:"PRP"
 },
 {
   text:"",
   normalised:"am",
   pos:"CP"
 },
 {
   text:"good.",
   normalised:"good",
   pos:"JJ"
 }]

Tokenization

Neighbouring words with the same part of speech are merged together, unless there is punctuation, different capitalisation, or some special cases.

nlp.pos("tony hawk won").tags()
//tony hawk   NN
//won   VB

To turn this off:

nlp.pos("tony hawk won", {dont_combine:true}).tags()
//tony   NN
//hawk   NN
//won   VB

Phrasal Verbs

'beef up' is one verb, and not some direction of beefing.

Licence

MIT

Something went wrong with that request. Please try again.