April 14, 2011

The following is a letter of feedback that i sent to google. I am posting it here because i hope that it may reach a wider audience than the google employee who opens the email. It points to a significant failure of recent origin in the google search algorithm, namely the forcing of two-word or multi-word phrases on searches for misspelled adjectives.

Here is what happened:

I searched for the word "bourgeois" but misspelled it. I was given the forced two-word result "bougeois pig." The result wasn not even about the use of that term in political protests of the 1960s, but listed many URLs relating to a bar in Chicago. I was not offered any URLS leading to pages on politics or history within the top ten or more.

I decided to take the time to fill out a google feadback form because this sort of problem has become more and more common in the past year.

When i went to fill out the form, i was given three areas in which to reply. Google's prended text appears in bold and mine in regular type.

"You searched for bourgeois pig."

I did NOT search for "bourgeois pig." I was searching for any site with helpful information about my topic, the word "bourgeois" -- perhaps wikipedia but preferably a scholarly political history site. My search consisted only of one word -- and "pig" was not in my search at all.

What information were you trying to find?

I wanted to learn more about the political term "bourgeois" with particular reference to a song by the African American singer Leadbelly, who wrote and performed a song called "Bourgeois Blues" in the 1940s.

What information did Google provide instead?

I mistyped the word as "bourgouis" and google forced (over-wrote my request) and led me to results concerning a bar in Chicago, called the "Bourgeois Pig." It did NOT offer to correct my spelling, but offered me only the choice between this bar and my misspelled word.

This is ridiculous. Google used to offer corrected spellings -- now it advertises a bar instead. The word "pig" was NOT in my search, and had nothing to do with what i was looking for. Google used to be smart. The past year, with these forced alternate search results, it has become so stupid and time-wasting that it verges on incomprehensible.

In order to get my corrected spelling, i would have had to have copied the bar name (assuming it was the correct spelling, which i have no way of knowing, since google did not offer to correct my SPELLING, only to correct my SEARCH, which, when you stop to think about it, is insultingly rude in a social sense, as well as supremely unhelpful in the context of search engine technology) -- and then delete the word "pig" and re-search. Stupid, stupid, stupid.

I well recall how, two years ago, google would have offered results for "bourgeois" -- but NEVER taken the error-prone tactic of adding some seemingly random second word to my search. If you don't understand the problem (and you don't, because i have seen this "add-a-word" issue come up as a routine error on mistyped or misspelled words), then ask someone higher than you at the company to explain it to you. Ask Larry Page -- seriously. Go ahead, ask him.

How can you satisfy a search request for an adjective by adding stupid additional nouns like "pig" to them?

Bourgeois is a political and historical term, and to add a weird random word like "pig" and then return results for the name of some obscure corner bar in Chicago -- plus a map to this bar, no less! -- and to tell me that i searched for this bar (!) as the result of a misspelled search for this common political term is a sign that something is seriously wrong with google's algorithm at this point.

My advice: stop ADDING WORDS. Just correct misspellings.

Here's how your error arose: Since "bourgeois" is an adjective, the new "helpful" algorithm looked for all nouns associated with that adjective and served up "pig" as the most common.

Now, let's say i created 150 web sites in which the adjective "bourgeois" was linked to the noun "sushi." And, to sweeten he deal, let's say i created a storefront called "bourgeois sushi" with a phone number, street address, and map. And then let's say i cross-linked all 150 sites in a semi-random, non-link-farmed manner.

Would a misspelled word search on "bourgeois" now produce the forced result of "bourgeois sushi"? Yes, it would.

And 150 links may be an over-estimate. It might take as few as 18, depending on the specific adjective that was misspelled and what the hash-pile of associated nouns turned up.

And that's what is wrong with the new word-forcing algo -- it is brainless. It runs on numbers, without reference to context.

The searching individual would be far better served if you would drop the force of a noun and let the adjective stand alone, rather than forcing the "most popular adjective-noun combination" when that combination has no relevance to the true context of the adjective.

You need more THINKERS at google and fewer programmers. The mind -- whether human intelligence or artificial intelligence -- operates by means of context. You COULD program context into all adjectival words (just pay a committee of librarians to help you) or you could simply abandon the misguided algo that forces the addition of a noun based on statistical link-popularity. Either method would return google to a greater functionality.

Again, ask Larry Page -- isn't artificial intelligence one of his interests? Ask him -- shouldn't adjectives be defined in terms of CONTEXT rather than a recently popular two-word catch-phrase? I think he would agree. At least, i should hope he would.