« New Google Checkout Promo; New Google Trust Worries | Main | SearchCap: The Day In Search, Jan. 16, 2007 »
Jan. 16, 2007 at 3:03pm Eastern by Danny Sullivan
Disclosing When Queries Are Autocorrected
Google Auto-correcting Queries from Google Operating System notes that Google seems to have moved beyond suggesting spelling corrections and now automatically does them for you. Actually, I think this is the stemming feature kicking in. However, it highlights that it's useful for search engines to tell you exactly what they've modified, if they do so.
Years ago, Google would automatically change your spelling if it found no matches for what you looked for. That annoyed me, to the point I pushed for it to change in 2002. It annoyed several others in the search and research fields. Maybe it didn't annoy general users. But Google eventually dropped it, so that it was no longer making some decision on your behalf.
The automatic change under discussion today is different. In a search for oper labs, Google instead automatically seems to be searching for opera labs. What's wrong with that? Nothing, as long as you know the search has changed.
Other search engines have done this before. In 2001, I recall MSN Search changing spelling so that it flagged you when an auto-correction happened. I seem to remember the AllTheWeb service also experimenting with changing queries for you and disclosing if this was done, so people could override the decision. AltaVista did various types of automatic phrase searching behind the scenes beginning in 1998, such as I talked about here. You could tell this most of the time because it would show you somewhere exactly the query that was done. You had to hunt for the disclosure, but it was there.
I've got no problem with automatic changes happening, as long as it's clearly shown somehow what you searched for. If Google wants to auto-correct spelling, more power to them. But tell users if this is happening.
Want a good example of this? Search for incredibal on Yahoo. Look at the bottom of the page, and you'll see:
In order to get better results, we searched for more than what you typed.
To get an exact match for what you typed, use the plus sign: +incredibal
Nicely done. And that type of disclosure helped William "IncrediBill" Atchison understand why he disappeared from Yahoo last month when an automatically spelling change kicked in. Don't worry -- Yahoo got it all fixed in the end.
It's important to note that in the case at Google, something other that automatically spelling correction might be making the change. I think it's stemming.
Google does stemming, added in late 2003 and explained here. This is where a search where it seeks the word you searched for and variations of that word. It might be looking for [opera] as a stemming variation of [oper].
Indeed, I suspect that's the case. Consider this:
- opera labs: 1,390,000 matches
- oper labs: 31,900,000 matches
See how the second search is broader, brings back more matches than the first one? This suggests that Google must be looking doing more than changing [oper] to [opera], otherwise the counts would be the same. Instead, I think it is looking for both those words and giving ranking preference to those that are saying [opera labs].
Of course, we wouldn't have to guess if Google somehow indicated that stemming was happening. FYI, to override stemming, add a + symbol in front of words you won't want it to get variations of, such as:
Like The Story? Vote For It On Yahoo Buzz!
Send me the monthly search newsletter too! (Learn more about our newsletters and feeds)Subscribe To Our Search Feed!
Share & Bookmark This Story! By Danny Sullivan
Permalink
Reader Comments
Disclosure really is important, I don't know anyone, who doesn't work in search or development, other than people I showed, who knows you can add '+' to a query, and there's nowhere on that google 'oper labs' search which tells you or gives you the chance to change it back from opera. If you hit search a few times it does actually give you the oper search results properly.
Who's going to know that though?
I can't reproduce this on other searches so perhaps its just a test on a few techies to see how confused people get?
Comment by Gurtie| January 17, 2007 6:12 AM
This feature may work well somewhere down the road but there are just too many nouns that G just may not understand.
I'd say this is going to have some impact on their market share over time unless Google wants to undertake effort to educate their users about the more advanced search features such as using quotes. How many times will someone keep using Google when they are sure they typed in the correct query - don't see any results - but are sure their topic is not a figment of their imagination.
Sure, Google will keep all the tech-heads but they aren't the people clicking on Adwords, either.
On a tangential note, this adds to my curiosity when it comes to the specifics of broad matching in AdWords. I shy away from broad matching unless I have metrics that indicate otherwise. Simply because you do not know what decisions, as to relevance, are being made by the AdWords AI.
Count me in the -transparency please- crowd.
-T
Stemming is one possibility, though there may be others. Google came out with a number of patent applications that describe query refinements, including spelling corrections, which would also display of a number of pages in results that go with the refinement. Here's a snippet from one of those:
There's a method described in those patent applications, of defining a confidence level for different possible refinements above which it would display a refinement and associated results upon the original results page, and below which it would require someone to click upon the refinement to see results that go with it.