tnt-tagger

A statistical part-of-speech tagger.

This is an implementation of Thorsten Brants' TnT parser. TnT, which stands for Trigrams'n'Tags, is "an efficient statistical part-of-speech tagger", and its implementation is described in the article TnT -- A Statistical Part-of-Speech Tagger.

In fact, tnt-tagger is a port of Python's NLTK implementation of said parser.

This is currently a work in progress. Future work includes refactoring code to make it more Javascript-like (for now, it feels a bit artificial due to the direct translation from Python).

Installation

$ npm install tnt-tagger

Usage

const TnT = require('./index');
const {Sentence,Token} = require('cetem-publico');

const ts = [new Sentence(1, [
    new Token('Jersei', {pos: 'N'      }) ,
    new Token('atinge', {pos: 'V'      }) ,
    new Token('média',  {pos: 'N'      }) ,
    new Token('de',     {pos: 'PREP'   }) ,
    new Token('Cr$',    {pos: 'CUR'    }) ,
    new Token('1,4',    {pos: 'NUM'    }) ,
    new Token('milhão', {pos: 'N'      }) ,
    new Token('em',     {pos: 'PREP|+' }) ,
    new Token('a',      {pos: 'ART'    }) ,
    new Token('venda',  {pos: 'N'      }) ,
    new Token('de',     {pos: 'PREP|+' }) ,
    new Token('a',      {pos: 'ART'    }) ,
    new Token('Pinhal', {pos: 'NPROP'  }) ,
    new Token('em',     {pos: 'PREP'   }) ,
    new Token('São',    {pos: 'NPROP'  }) ,
    new Token('Paulo',  {pos: 'NPROP'  })
  ])];

let corpus = {
  sentences:  function*(){
    n = 1;

    for(i=0; i<n; i++){
      yield ts[i];
    }
  }
};





let s = new Sentence(1, [
    new Token('Jersei'),
    new Token('atinge'),
    new Token('média' ),
    new Token('de'    ),
    new Token('Cr$'   ),
    new Token('1,4'   ),
    new Token('milhão'),
    new Token('em'    ),
    new Token('a'     ),
    new Token('venda' ),
    new Token('de'    ),
    new Token('a'     ),
    new Token('Pinhal'),
    new Token('em'    ),
    new Token('São'   ),
    new Token('Paulo' ),
  ]);

let t = new TnT();

t.train(corpus)
  .then(() => t.tag(s))
  .then(console.log);

Methods

TODO

Acknowledgements

Thanks to Thorsten Brants for the original version of this algorithm, and to NLTK's team for the implementation in which this module is based on.

Bugs and stuff

Open a GitHub issue or, preferably, send me a pull request.

License

MIT