Google phrase correction

More than just a spellchecker

[andye]

Maybe this is already widely known, but if you Google for

dante "diving comedy"

then it gives you

Did you mean: dante "divine comedy"?

which seems interesting to me, because the query wasn't misspelt.

It seems to know that Dante wrote a work by that name, because if you just search for "diving comedy" (without Dante) then it doesn't suggest the correction.

So maybe Google Sets results are feeding into the spellchecker?

Andy.

[ScottM]

Not sure of a spellchecker, probably more about the current database they have of common words.

Although, spellchecking may be a part of it.

[jeremy goodrich]

Hm, they probably do an internal look up in their 'phrase' table.

Then, seeing that the search is a close match for the 'misspelt' phrase, they offer the suggestion, as the suggested spelling probably has the bigger search volume over all.

Google will do this the opposite for uncommon words, it will suggest that you are spelling wrong, and suggest another word, because the word you used wasn't in their 'phrase table' even though it is a real English word. :)

Pretty nice feature, imho. Works well most of the time.

[vitaplease]

The phrase correction also has some geo-targetting in it.

If I do a search for my domain in Google.nl, no suggestion comes.

If I do a search for my domain in Google.com it gets a suggestion towards another US domain, spelled with one letter difference.

(Seemingly Google thinks only US companies are international or English oriented...)

[GoogleGuy]

The spellchecker is context-sensitive, so it's much more than just a set of suggested corrections.

[vitaplease]

Googleguy,

what do I have to do, to let Google not suggest that other domain (company), spelled with one letter difference (the problem is they do similar stuff).

Both domains are non-existing words in any language.

[albert]

I have the same situation Vitaplease mentioned: my company's name (artificial word) differs just in one letter from a common word in jugoslawian language.

So it's not nice if somebody types in my correct company name, and then Google comes up asking "did you mean <jugoslawian word>".

My company's site is listed at #1, anyway.

But i would prefer if there was no spellchecker suggestion to my company's name.

What can I do?

<edit>typo</edit>

[vitaplease]

Albert,

in your situation Google suggests a "real" word.

In my situation it suggests another "unreal" word (the other domain).

Google seems to be arbitrary in their suggestion, because the (geo-)suggestion only works from my domain to the other and not vice-versa.

I have a feeling that some years back, when Google started the suggestion tool, they took some kind of inventory of dictionary words together with existing domain names. As of now, they have not updated this inventory of "words" to include the newer(domain) words.

[albert]

Vitaplease,

you're right, in my situation it's "real" vs. "unreal".

But Googleguy mentioned context sensitivity. If you have just one word typed in it's hard to presume any context.

For the "real" word the context is jugoslawian, for my "unreal" word english and german (language of my company's site).

[GoogleGuy]

Hey, I would write to user support and mention that it's a bad spelling correction. They'll pass it on to the engineers, who will want to test if it's a weird corner case, or if it usually improves quality for the average user. Drop us a line.

[albert]

GoogleGuy, thank you for your hint. I'll try this.

[Dynamoo]

My main site (take a wild guess at the name) used to be picked up as a mis-spelling of "dynamo", but no longer. There must come a threshold where it's no longer considered a typo.

[takagi]

Just did a search on news.google.com for

sars "new diagnostic tool"

and Google replied with

Did you mean to search for: cars "new diagnostic tool"

Other strange results when searching for "writeble" in different countries. In English both writable (298,000 hits) and writeable (108,000 hits) are correct.

writable
google.com (USA)
google.co.jp (Japan)
google.co.kr (S.Korea)
google.co.uk (UK)

writeable
google.de (Germany)
google.fr (France)
google.be (Belgium - French mode)

<no hints>
google.nl (The Netherlands)
google.be (Belgium - Dutch mode)
www.google.com/intl/ar/ (Arabic)

In German, French, Korean & Japanese neither word exist. Still it shows one of the two. You would expect that if 'writeable' is advised by google.de that searching in pages written in German would result in more pages for 'writeable' than 'writable'. Wrong!
writeable 1530
writable 2710