I guess Google did exercise their "right to address such requests individually", because "did you mean" is still offered for less straightforward searches:

http://img452.imageshack.us/img452/9717/heinventedfq4.png

I have noticed they also removed some things that were very popular in Serbian. Extremely vulgar language well, now it is part of history...

Can't reproduce this Roger, plus I'm getting the new design... are you perhaps searching from a country-specific Google and they didn't roll this out there (at least yet)?

http://blogoscoped.com/files/she-inveented.png

Actually, even though I can't reproduce it, the lack of a spelling correction for that is a bug too, as "inveented" should be "invented" and the user would certainly like to know...

Interestingly, I see the "correction" for

["she inveented"]

but not for

[she inveented]

"(Note: no matter what Google tells you, algorithms are always influenced by those who design, write & test them – including their preferences)"

While that is true, I doubt all the biases you see in the search results were influenced by the engineers creating the algorithms. In other words, there is general influence on a large scale with the general direction of results, but that does not translate into a bias influence of "he invented" versus "she invented".

I'm going to open a can of worms here,
but I'm thinking that the way Google suggests is based on an algorithm that READS WEB PAGES.

So, for example, there are far more websites with the phrase "he invented" (including a host of non-scientific publications; stories, books, etc.) than there are with the phrase "she invented".

I doubt Google specifically programmed that in, or that they can be blamed. Society as a whole has for a long time had trouble acknowledging women achievers as much as they do.

Not only that, but it's also the difference of 1 letter, which Google seems to correct more than 2 letters "wrong".

> there is general influence on a large scale with the
> general direction of results, but that does not
> translate into a bias influence of "he invented"
> versus "she invented".

Agreed!

Further thoughts:

http://sethf.com/infothought/blog/archives/001203.html

I no longer get "did you mean" when I search for [she inveented]. I think the Googlies are reading this blog and extending their exceptions list by the hour.

If I search for ["she invented it"] I still get "did you mean":
http://img245.imageshack.us/img245/7560/heinvented2up1.png

I'm using google.com, not a country-specific site. Right now I'm getting the old interface, although at times I get the new one. Just depends which data centre I connect to, I guess, although it's annoying and I wish the interface would stabilise.

"Note: no matter what Google tells you, algorithms are always influenced by those who design, write & test them – including their preferences"

wrong,

its not google's fault that in past most of the inventors, professors, discoverers (and more) were men.

Google's algorithm innocently and naturally thinks that a female and inventor doesn't fit together (whatsoever the reason is due to past) so it suggests "he" without knowing it can be socially offensive.

Now that YouTube also gives "do you mean" hints, we find the same suggestions there:
http://img360.imageshack.us/img360/7494/heinvented3gu1.png

ORB: I did not suggest that Google was biased in the he vs she spellchecking case. It was a reply to Google's statement that they're creating completely unbiased algorithms, from which I quoted.

Seems to be an exclusion list for particular word combinations, and not a general "don't correct she to he" change to the algorithm.

Proof of this is searching for "she is the best" offeres a correction of "did you mean 'he is the best'"....

Unless of course the large number of sites reporting on the 'she invented' phrase ironically increased the number of hits enough that google no longer attempts to correct it.

Also:
she wins the medal
she wins the game
she wins the fame
she reaches the top
she received honours
she led them to victory..

and on and on.. all get corrected to 'he'. And if the number of hits are anything to go by, chances are it is the correction you want, regardless of how needlessly politically correct you want to pretend be.

> And if the number of hits are anything to go
> by, chances are it is the correction you want

A larger quantity of pages for a certain spelling does not mean it's what the user wants this to be corrected to – it may be one of the many factors you need to take into account, but by its own, it would be a useless algorithm for a spellchecker ("their" would be corrected to "the", "he" would be corrected to "we" etc.). Hence, you have no indicator more people mean "he wins the medal" when they type "she wins the medal". This doesn't have anything to do with political correctness but a lot with usability.

Here's a more likely misspelling, and it shows the bad usability again: when you enter ["we wins the medal"], Google returns 0 results yet doesn't offer a spellchecker, even though the grammar is clearly false (it's "we win the..." not "we wins the..."). Yet, when you enter ["she wins the medal"], even though it returns at least 2 results (2 more than "we wins..."), Google offers a spellchecking to "he wins the medal".

This may or may not have to do more with "she" vs "he" ratio taken on its own than the individual quantity of the phrase search perhaps – it's almost impossible to tell from the outside. What's not so hard to tell is that it's silly for a spellchecker to suggest to correct "she" into "he". Again, first and foremost, bad usability.

Google's algorithms are human-written and hence prone to false assumptions about the usefulness of certain approaches, thresholds, quantities needed etc. This may be one of those cases. Maybe it's such an exotic bug that it's not worth it to change the algorithm, but it might also be that you can fix this on its own without affecting the usefulness of the other spelling corrections.

Just spotted an official response from Google in regards to their spellchecker and these spellcheckings, courtesy of Chicago Tribune via the Google Operating System blog:

<<Google develops its own spell-checking algorithms based on sophisticated machine learning methods, using cues from aggregated user input, Web documents, and many other sources. The algorithm provides a 'best-guess' alternative suggestion that we think might improve the search results, and is completely generated without human input. It can be thought of as a suggestion offer, rather than a definitive answer.>>
http://newsblogs.chicagotribune.com/news_columnistswischlist/2007/05/is_google_sexis.html
http://googlesystem.blogspot.com/2007/05/googles-spell-checker-reveals.html

(Small logical incorrectness, but I guess we know what they mean: "Google ... spell-checking algorithms based on ... user input ... generated without human input" = "users aren't humans")