pronounceable.js

// Luke Mitchell, 2016
// https://github.com/lukem512/pronounceable

var fs = require("fs");
var path = require("path");

var threshold = 0.001;

// Load probabilities from JSON files.
var tuples = JSON.parse(
  fs.readFileSync(path.resolve(__dirname, "data/tuples.json"), "utf8")
);
var triples = JSON.parse(
  fs.readFileSync(path.resolve(__dirname, "data/triples.json"), "utf8")
);

// Remove any non-alphabet characters
// and convert to lower case.
function clean(word) {
  return word.replace(/[^a-zA-Z]/g, "").toLowerCase();
}

// Make a percentage.
function percent(score, count) {
  return score / count * 100;
}

// Check for undefined probabilities.
function undef(w, i, depth, probs) {
  if (depth <= 1) return typeof probs[w[i]] === "undefined";
  if (typeof probs[w[i]] === "undefined") return true;
  return undef(w, i + 1, depth - 1, probs[w[i]]);
}

// Extract probabilities of word t 	uple.
function trainTuples(words) {
  var probs = {};
  var count = 0;

  words.forEach(function(w) {
    w = clean(w);

    for (var i = 0; i < w.length - 1; i++) {
      if (!probs[w[i]]) probs[w[i]] = {};
      if (!probs[w[i]][w[i + 1]]) probs[w[i]][w[i + 1]] = 1;
      else probs[w[i]][w[i + 1]]++;
      count++;
    }
  });

  Object.keys(probs).forEach(function(first) {
    Object.keys(probs[first]).forEach(function(second) {
      probs[first][second] = percent(probs[first][second], count);
    });
  });

  return probs;
}

// Extract probabilities of word triples.
function trainTriples(words) {
  var probs = {};
  var count = 0;

  words.forEach(function(w) {
    w = clean(w);

    for (var i = 0; i < w.length - 2; i++) {
      if (!probs[w[i]]) probs[w[i]] = {};
      if (!probs[w[i]][w[i + 1]]) probs[w[i]][w[i + 1]] = {};
      if (!probs[w[i]][w[i + 1]][w[i + 2]]) probs[w[i]][w[i + 1]][w[i + 2]] = 1;
      else probs[w[i]][w[i + 1]][w[i + 2]]++;
      count++;
    }
  });

  Object.keys(probs).forEach(function(first) {
    Object.keys(probs[first]).forEach(function(second) {
      Object.keys(probs[first][second]).forEach(function(third) {
        probs[first][second][third] = percent(
          probs[first][second][third],
          count
        );
      });
    });
  });

  return probs;
}

// Extract probabilities of word tuples and triples
// from a large list of words.
module.exports.train = function(filename, callback) {
  fs.readFile(filename, "utf8", function read(err, data) {
    if (err) throw err;

    var words = data.trim().split(/\s+/);
    var tuples = trainTuples(words);
    var triples = trainTriples(words);

    callback(tuples, triples);
  });
};

// Check whether a word is pronounceable using
// the word tuple probabilities.
module.exports.test = function(word) {
  var w = clean(word);

  switch (w.length) {
    case 1:
      break;

    case 2:
      for (var i = 0; i < w.length - 1; i++) {
        if (undef(w, i, 2, tuples)) return false;
        if (tuples[w[i]][w[i + 1]] < threshold) return false;
      }

    default:
      for (var i = 0; i < w.length - 2; i++) {
        if (undef(w, i, 3, triples)) return false;
        if (triples[w[i]][w[i + 1]][w[i + 2]] < threshold) return false;
      }
  }

  return true;
};

// Compute a normalised score for
// the pronounceability of the word.
module.exports.score = function(word) {
  var w = clean(word);
  var score = 0;

  switch (w.length) {
    case 1:
      return 1;

    case 2:
      for (var i = 0; i < w.length - 1; i++) {
        if (undef(w, i, 2, tuples)) {
          score = score + 0;
        } else {
          score = score + tuples[w[i]][w[i + 1]];
        }
      }

    default:
      for (var i = 0; i < w.length - 2; i++) {
        if (undef(w, i, 3, triples)) {
          score = score + 0;
        } else {
          score = score + triples[w[i]][w[i + 1]][w[i + 2]];
        }
      }
  }

  return score / w.length;
};