YGingras^.net

It happens all the time. I am about to start a new project and I can't find a good name. I could postpone the decisions until publication. With my completion rate, that would save me a lot of thinking. Still, the name of a project, a programming one at least, is scattered all over the place from the start: you create a tree in your cvs/git/svn, you create a package/namespace/module in your programming environment and you write the project's name in the documentation (at least you intend to). You can change the name later but you save a lot of work if you start with the right one.

There was a time when you could pick a random word from the dictionary and be done with it. Today, it seems, all the real words are taken, even vellication. Here I use a widely accepted definition of "taken": is the .com owned by someone else. There are other root domains (.net, .org, .info) but the squatters bought the most of the dictionary there too. If we don't limit ourself to the dictionary, why should we after all, combinatorics saves the day. There are just to many ways to put letters together. Squatters can buy all the tree letter acronyms but that's as far as they can get.

My first idea was to use googleability. Googleability, you'll recall, is the quality of being searchable, and found, on Google. Words like "sun", "act" and "run" have low googleability. They are in the 0.1% of three letter words with the most Google hits. At the other end of the spectrum, you get words like "bxu" and "yzo" that no one seem to put on his website (thought this self referential sentence invalidates its claim). If your friend tells you about YZO's new BXU, you can google it and you will find the right website. Googleability works but I wasn't the first to think about it. Almost all pronounceable three and four letter words with high googleability have a registered .com. Exploring longer words is a challenge. Google doesn't like scripts. You have to fake a browser and to limit the query rate. I didn't. I seriously considered finding a new Internet service provider when I noticed that Google had banned me.

What is pronounceable? Some people can pronounce "w00t" and "pwned" but most can't. If you limit yourself to alternation of vowels and consonants you will get pronounceable words but you will miss really good ones. On the other hand, allowing a sequence of more than one consonant will yield words like "evzupricb". I was looking for a quality more subtle than just being pronounceable. Some words are warm, others are frigid. Anglo-saxon and Norse words, for example, are more visceral than Latin ones. The battles against the Orcs are not fought in Polytanie, they are fought in Azeroth. How could I capture this warmth, this evasive quality with the taste of raw meat?

Actually, that's quite simple. A lively text will contain more warm words than frigid ones. Extracting the warmth is just a matter of measuring the probability of going from one letter to the other, including the probability that a given letter will end the word. A probabilist engine trained on the King James Bible will produce a lot of old English sounding words. One last change is required though: the probability of going from "l" to "l" will be way to high and you'll end up with words like "alllerit". The full transition probability with a two letter prefix could be large but only recording the transitions that you actually see generates a transition table under 100k. And it makes sense to do so. In the King James Bible, the only thing that can follow "pr" is a vowels. Furthermore, "pr" will never be at the end of a word, unlike "ps", which is last 80% of the time.

Now that we have the rules to generate as many vivid words as we want, we just need to keep trying until we find one for which the .com is still available. It is good? Judge for yourself, as I write this, those words are available in at least two of the four main top level domains: abongs, abuted, affere, againg, amigeth, andreph, becomoses, becured, beffer, behosest, brisraell, challed, chout, egived, egyptayed, fatieford, fifies, flooses, gaing, humple, israndee, istion, joyether, judan, judgenst, kethich, morthers, moshall, offeruiled, prieven, reoph, rient, rought, sayints, shaarding, shaith, shalleak, sheaveth, speople, spild, tathild, therecaust, theried, thits, togypt, waying, whing, whings, whous, womenty. Nothing was hand picked, just random generation. We get names but we also get verbs, adverbs and adjectives. I wonder if this is how they create artificial languages.

Changing the training set for the weekly top 100 on Project Gutenberg produces a contemporary lexicon: adays, amalls, anned, appeance, appene, bectinty, cassion, descess, droort, eliked, exterid, flonst, hered, honsir, humety, immang, intanch, jused, losed, maketter, mathere, matinswer, opecous, othersuld, pears, pened, plorn, preathilly, prowhich, reard, reentich, retcheing, scrove, shoul, soothred, soubjes, spilly, squoully, therigh, throppen, unlip, vidded, wanked, whallon, wherents, whicars, whide, wouldishon, woute, youthered. Not as exotic but still credible.

I still don't have a magic trick to pick a project name but at least I can now limit the search space. I let my computer looking for words are taken during the night and I contemplate are few pages of candidates in the morning. My name generator even found its own name. Here is Yould. Now I need to find good texts for the training set.