Dumb mag buys grammar goof spin spot fraud
« previous post | next post »
A SpinSpotter tool — a plugin for the Firefox browser — has been announced in a credulous article by Jon Fine in Business Week. It will (its inventors claim) scan the text of web pages that you view, and identify passages of untrustworthy spinspeak. Our experts at Language Log's research laboratory have run it through our secret multi-million-dollar bullshit detector, and we got a strong positive. Having written several times before on Language Log about people who publish claims about language, and mention the passive voice, when they are completely unable to tell an active clause from a passive clause, I was delighted to see one more instance. Look at this description, from Jon Fine's description of SpinSpotter, detailing the "tenets" (i.e., diagnostics) that enable SpinSpotter to spot spin:
The tenets are: reporter's voice (adjectives used by a journalist that go beyond the supporting evidence in the article); passive voice (example: a story says "bombs land" without stating which party is responsible for them); a biased source (a quoted source's partisanship is not clearly identified); disregarded context (a political rally's attendance is reported to be "massive," but would it have been so huge had the surviving members of the Beatles not played?); and lack of balance (a news story on a controversial topic gives much more credence to one side's claims).
Bombs land is of course an active clause. Passive clauses always have a participial form of the verb, in almost all cases (setting aside "concealed passives" like "This needs looking at") a past participle. The past particple of land has the form landed. So quite independently of the absurdity of an algorithm running on raw text being able to spot things as subtle as strength of supporting evidence or balance on controversial topics, the inventors of this crucially linguistic tool (or the people who wrote their press release) don't know even the most elementary things about English grammar. Caveat downloader.
Hat tip to Fernando Pereira. And as a commenter remarks below, don't miss the delightful piece of more detailed demolition at Headsup, together with Mark Liberman's investigations elsewhere on Language Log. Jon Fine really has been sold a pup here. How do these print journalists get away with being so naive and trusting, in a world that they surely know is swimming with hustlers and press agents? Don't answer; that's a rhetorical question, a sort of written substitute for eye rolling and head tossing.
WindowlessMonad said,
September 9, 2008 @ 5:05 am
See http://headsuptheblog.blogspot.com/2008/09/kings-camelopard-or.html for Fev's delightful demolition of this nonsense.
Rod Whiteley said,
September 9, 2008 @ 6:02 am
I see no claim that it will scan text. The Get Started page says Spinoculars will "alert you to when spin has been identified on the page", using passive voice to conceal who or what does the identifying.
As far as I can tell, you only see the alerts that you and other users have created. References to "democracy" on the SpinSpotter site make it look like the intention is political. So it's really a tool for political activists to add their own spin to web pages they dislike.
Peter Howard said,
September 9, 2008 @ 6:59 am
Spinspotter's own definition of passive voice (on the 'rules' page) is worth quoting in full:
Er, that's not passive voice, either.
Whomever?
Mark Liberman said,
September 9, 2008 @ 7:18 am
Rod Whiteley: I see no claim that it will scan text.
You may need new glasses. Jon Fine's article says: "When turned on in a user's Web browser's toolbar, Spinoculars scans Web pages and spots certain potential indicators of bias" (emphasis added). The same claim is explicit on the company's web site, and in the NYT article by Claire Cain Miller.
But I downloaded SpinSpotter and tried it, and the most striking thing about its algorithm for spotting "spin" is not that it implements the common error of interpreting "passive voice" to mean "vague about agency". The most striking thing is that it does nothing at all.
See "SpinSpotter unspun" for details and discussion.
[I've moved this (originally rather long) comment to that post, in keeping with our comments policy.]
AFedchuck said,
September 9, 2008 @ 7:25 am
Given that the examples in their rules are all of 'liberal bias' (i.e. anti McCain)and that the front page picture of a site with bias is the NYT, it's clear that this site has a distinct political agenda.
It seems to me that their contribution is essentially to compile their users' views on an article and possibly give an overall score.
(As an aside an interesting statistic is provided on their site: 66% of Americans think the media is one-sided. I wonder if they all think it is biased the same way…)
Mark Liberman said,
September 9, 2008 @ 7:52 am
Further evidence that SpinSpotter is demoware: when I copy Jon Fine's article to my own computer and browse it at the new location, or even when I browse the printable version on the Business Week site, the two spin-spots go away. This is evidence, obviously, that the spotted spin is not being generated by algorithmically scanning the text in the browser, but by checking the URI in the company's database.
It looks like misunderstanding the passive voice is the least of this enterprise's credibility problems. I speculate that the SpinSpotter management team are aiming their efforts at credulous venture capitalists, and that Jon Fine has swallowed the bait.
It's a bit surprising that the download wouldn't even contain a list of keyword-sequences to check by string matching, or some other simple-minded algorithm along the lines of the usage advice that some word processes give you. To start promoting a piece of software that does absoutely nothing whatsoever would be unusually brazen — maybe there's a bug in the initial download that prevents part of it from running, or at least prevented it from running on my machine?
It's hard to believe that (what appear to be) reputable people could be associated with promoting (what appears to be) such a crock, but there you are.
Mark Liberman said,
September 9, 2008 @ 8:44 am
More on the SpinSpotter scam-spotting: I've tried the download in FireFox 3.X on three different machines, under OSX (10.4.11), Windows (Vista), and Linux (Ubuntu 7.04), with identical results.
Karen said,
September 9, 2008 @ 8:49 am
And even if it worked as advertised, see Fred Vultee's analysis of why it wouldn't spot spin at headsup: the blog.
Rod Whiteley said,
September 9, 2008 @ 8:57 am
The markers are probably associated with the URL in their database, so if you change the URL, the markers will not appear.
Perhaps the database is the important part. It will gather data on those parts of articles that readers consider spun (spinful?), and that data could be sold or donated to worthy causes. The privacy policy seems to allow this, if the data does not identify individual users.
James Wimberley said,
September 9, 2008 @ 9:00 am
Pullum: "… our secret multi-million-dollar bullshit detector…"
Weren't double hyphens frowned on by pundits? I'm entirely with you on this one; multiple hyphens are natural in such common constructions as "three-year-old". When and where did common sense break out?
Arnold Zwicky said,
September 9, 2008 @ 10:15 am
Larry Horn on ADS-L this morning:
"X dropped bombs (on Y)"? I guess they could have called it "unaccusative voice", but that might have been a bit obscure for some readers. Presumably the "passive" character stems from the fact that both true agentless passives ("The boat was sunk", "Mistakes were made") and unaccusatives ("The boat sank", "Bombs land on X") are devices to suppress the agent and thus to duck the assignment of responsibility, especially to oneself.
Stephen said,
September 9, 2008 @ 10:19 am
If, as seems likely, the software is just comparing the website to a human-made database of spin, that leaves the question of why the parts of Jon Fine's article that are flagged aren't the most obviously spun parts. Surely a person trying to pick out examples of spin in the article would have done better than that.
By the way, @Mark, it seems like you might be stretching the bounds of your own comments policy here. Wouldn't this long series of comments be better moved to its own post, rather than tagged onto a post that was originally mainly about a misused grammatical term?
[(myl) Yes. I'll move the material to a new post when I have time.
Since I'm an editor of the blog, the software lets me edit comments, and what started out as a short note got away from me. Unfortunately, the 20 minutes or so that I spent on this was about 15 more than I could spare, this morning.]
Ryan Denzer-King said,
September 9, 2008 @ 11:29 am
While clearly not a passive sentence, I find the company's "example" interesting, since it does demonstrate the desire to have a highly agentive (and probably specific) subject. Bombs aren't capable of doing anything of their own volition, so even though "bombs land" isn't passive, whoever wrote that description perceived it that way because he (or she) wants to understand the conscious action behind the bombs landing, and doesn't like inanimate "bombs" as the gramatical subject.
Chris said,
September 9, 2008 @ 12:50 pm
This is particularly disappointing since sentiment analysis is a legitimate field of NLP research. As multiple tools are already available for doing exactly this, curious that they didn't simply license something like Lingpipe for a beta release.
John said,
September 9, 2008 @ 2:23 pm
Mark –
It also doesn't work if you simply add an anchor ("…/blah.html#whatever") to the URL. What were they thinking?
Dan T. said,
September 9, 2008 @ 3:24 pm
@Mark Liberman, "This might be an unusual type of demoware, though, one that is released for general use in the hope that enough people will submit their proposed spin-spots to give the company enough free training data to actually develop some of the technology that they pretended to have in the first place."
…so it's a software version of stone soup!
marie-lucie said,
September 9, 2008 @ 3:38 pm
both true agentless passives ("The boat was sunk", "Mistakes were made") and unaccusatives ("The boat sank", "Bombs land on X") are devices to suppress the agent
"Suppressing the agent" seems to imply that every action is caused by an agent which could be named but is not, either through a conscious decision not to mention it (often because it is semantically predictable), or through inability to name the agent because it is unknown or too vague. "Unaccusative" verbs (i.e. verbs without a direct object) do not "suppress an agent" because there is not necessarily an agent to be suppressed. "The boat sank" states an event which can be due to a variety of factors which may or may not have anything to do with a conscious agent, unlike "the boat was sunk" which implies the existence of a responsible agent, whether known or not. It is our knowledge of the world, not that of English structure, which tells us that "Bombs landed on X" implies that someone was dropping bombs (not landing them), but that there is no necessary implication of an agent in "Newton first thought of the theory of gravity when an apple landed on his head as he sat under an apple tree."
dr pepper said,
September 9, 2008 @ 7:04 pm
In the parable of Stone Soup
1) The con is usually (there being many versions, including one where it's a nail) a response to the lack of hospitality shown by the locals.
2) The scammers get fed, nothing more.
3) The local sense of community is strengthened.
None of that seems to apply here.
Clayton Burns said,
September 9, 2008 @ 9:58 pm
If SpinSpotter worked it would be able to detect the lies of the governess in "The Turn of the Screw." But even if it detected evidence of any of her lies, would it be able to draw conclusions by comparing sets of facts that do not seem to add up?
For example, the governess was able to give a detailed account of the death of Peter Quint, meaning that she had to have an informant that she would have mined in her usual insistent way for a description of Quint's appearance and mannerisms. Therefore, it makes no sense to say that because the governess was able to give a description of Quint that he must actually have appeared to her as a ghost.
Besides, there are hints in the novella about a portrait of Quint that the governess could have consulted. If her lies could have been detected by surface features they would have been uncovered long ago, but you still see approving references in books to the idea that Quint must have been a "real" ghost for the above reason.
A lecturer on James noted recently that the descriptions of the ghost on the tower were too "real" for him to have been a figment of the governess's imagination. So if someone were a really skilled liar, that person would liberally dispense clues to potential deception that would then be overturned so as to con the closest of observers.
All human readers of "The Turn of the Screw" have been deceived. There is no sound interpretation of that novella. If a SpinSpotter set out to disclose the lies of Henry James in "T.S.", the Spotter would get hopelessly entangled in ambiguity, if it could "see" anything at all.
Which leaves us with a far more important question. Why have Linguistics professors failed to penetrate the lies of the governess?
Peter said,
September 10, 2008 @ 5:04 pm
" Bombs aren't capable of doing anything of their own volition" , says Ryan Denzer-King.
This statement is not true. Smart weapons have as much volition as their creators wish to give them, up to and including the levels of autonomous agency enjoyed by human soldiers. It is not for nothing that the main sponsor of AI research has always been and is still the military.
dr pepper said,
September 10, 2008 @ 10:27 pm
"Bomb Number Three: what are you doing?"
"Fullfilling my purpose."
— Dark Star
Jorge said,
September 11, 2008 @ 8:39 am
"I am not malfunctioning! I'm a conscientious objector."
other one spoon said,
September 28, 2011 @ 9:33 am
I feel sorry that none of the comments mentioned Prof. Pullum's magnificent (presumably intentional?) crash blossom.
"Dumb mag buys grammar goof spin spot fraud"
I was doing fine up to "Dumb mag buys" – the first natural interpretation of that beginning is that an unintelligent (not speechless) magazine (not Margaret) believes (not purchases) something. But then "grammar goof" makes you think that what the magazine believed was a grammatical mistake, and mistakes in and of themselves don't assert anything. Maybe the magazine believed that something WAS a mistake, when in fact it wasn't? That would be consistent with Language Log, and with the description of the magazine as "dumb." Maybe the rest of the headline will shed more light – but instead, "goof spin spot fraud" sheds only darkness.