Google needs to learn to read :-)…
« previous post | next post »
John Lawler writes:
I recently had reason to ask the following question of Google
https://www.google.com/search?q=how+many+words+does+the+average+person+say+in+a+day
and the result turned up an old LL post, which is great, except the selection algorithm picked the wrong number as the answer, and even quoted the post you were complaining about as if it were true.
This should probably be brought to someone's attention, but it seems, what with the vast amounts of irony, hyperbole, bullshit, lying, and fact-checking on the net, this is not an isolated problem.
The obligatory screenshot:
Read the cited post ("An invented statistic returns", 2/22/2013) to see the context for what Google's algorithms decided was the answer to the question.
Lai Ka Yau said,
July 22, 2019 @ 9:02 pm
Here is what Google looked like when I performed the search: https://i.imgur.com/ePJhep4.png
I think that in my screenshot, it looks more like Google is treating the query as a question for a question-answering system and taking a quote from the LLog post as the answer to the question, compared to Prof Liberman's, where it only seems to be displaying the link and quote as a search result.
tangent said,
July 22, 2019 @ 11:58 pm
Yeah, understanding that you quote that bit many times because it's false is probably hard for language processing. But there's an easier bit flubbed where they take a question about "average person" and answer it with a supposed statistic for men while discarding the one right next to it for women.
tangent said,
July 23, 2019 @ 12:11 am
Now I wonder how much better the monkey brains do. If we give the 2013 post, along with a couple of others as filler, to a population of standard-issue college sophomores, and later ask them "as best you can recall, how many words are spoken per day by men and by women?" I'm betting most quote the bogus number. Repeated falsehoods can stick.
Bert said,
July 23, 2019 @ 1:47 am
(At least in German Google) this search turns up a "snippet", which extracts its information from Languagelog. Snippets suggest that they provide a definitive answer to the question instead of referring the user to a website. Screenshot: https://imgur.com/a/wqZxjQm
This is much more problematic I believe than the version shown in the post above. A number of snippets that I have seen turn out to be incorrect or at least give incomplete answers.
Andreas Johansson said,
July 23, 2019 @ 3:00 am
Regarding which sex talks more, I'm infected with two opposite stereotypes – one says that women are chatty and men (or at least "real men") are taciturn, the other says that men tend to be socially dominant (or perhaps just louder) and don't let women speak.
jaap said,
July 23, 2019 @ 6:14 am
Reminds me of the journalist who wrote an article about how Facebook does not have a customer support phone number, after which Google took his mobile phone number from the article's byline and presented it as Facebook's:
https://www.vice.com/en_us/article/zmpm43/google-thought-my-phone-number-was-facebooks-and-it-ruined-my-life
cM said,
July 23, 2019 @ 6:20 am
google seems to like snippets that are set apart somehow in the source page.
There's a semi-famous example of the same thing happening with onion caramelisation times, where his citation of an "obviously way off" example (formatted via <blockquote>) was the part google decided to show on the results page. And then even got that wrong.
The author was annoyed:
https://gizmodo.com/googles-algorithm-is-lying-to-you-about-onions-and-blam-1793057789
Cervantes said,
July 23, 2019 @ 8:01 am
I'm not sure what you're complaining about. I read the old post and while it does debunk the claim that women talk more than men, it also provides the true information that answers the question — it cites several credible studies regarding how many words people say in a day. So it seems to me a perfectly appropriate result.
J.W. Brewer said,
July 23, 2019 @ 9:00 am
Further to Andreas Johansson's point, I think it's rather nice that there are easily-available stereotype-based just-so stories that can plausibly account for either the males-talk-more possibility or the females-talk-more possibility. You'd think that might allow people just to go with the answer that ends up being supported by the evidence without undue cognitive dissonance. (It's perhaps important to note that the rival stereotypes are not conceptually incompatible — it's perfectly possible to hold both at once and accept that while both true they point in different directions and thus to some extent will predictably offset each other's anticipated effect, with which one one ends up having the stronger effect on the net outcome being an empirical question where one ought to remain open-minded and see what the data shows.)
stephen said,
July 23, 2019 @ 7:04 pm
Here's another problem. I wanted to know how to pronounce the last name of an actor named
Drew Haytaoglu. The results were no help at all. One of the results is how to pronounce the word "drew". Another is how to pronounce Chinese names. And several results don't even include the words Drew, Haytaoglu or Actor.
On another topic, why do so many search results come from newspapers in England, India or Australia?
I'm not in any of those places.
Lars said,
July 23, 2019 @ 7:18 pm
@Stephen: Well, it's Turkish in origin, isn't it? so try writing haytaoğlu instead? How the guy himself pronounces it is another matter, of course.
ALB said,
July 24, 2019 @ 12:37 am
@Lai Ka Yau: your description fits the results I am getting – https://i.imgur.com/31VVMcx.png, although I also receive an answer extracted from the LL post text and placed above in big, bold type so you can't miss it.
@tangent: it seems we may not be getting the same answer (see above), as the figure I am getting is the one cited for women (not men) in the original LL post.
Philip Taylor said,
July 24, 2019 @ 2:22 am
Stephen — No idea what question you asked Google, but I find that if I ask '"Haytaoglu" pronunciation' I get sensible answers. As to "why do …", you can restrict the search domain by adding (e.g.,) '+site:.fr' or what ever is the appropriate TLD for your country of residence.
tom davidson said,
July 28, 2019 @ 12:16 am
I queried Google translate for the term "clutch of of eggs" in Chinese. the result? 一个蛋 (one egg). I submitted a correction.