Sunday, November 10, 2013

pretending to be black

I just came across this beautiful example of misleading without technically lying. Sleazebag politician Dave Wilson sent out flyers with pictures of black people saying "please vote for our friend and neighbor", and said he was endorsed by Ron Wilson. You hear the name Ron Wilson, you think of a prominent black politician. But he meant his white no-name cousin. Successfully having convinced people that he's black, he won the election, in the actual world, in 2013.

The juxtaposition of the pictures with the text is actually technically a lie, I would argue. It depends on spelling out the conventions for juxtaposing pictures with text, but the referent of "our" simply must be the people in the picture. There is no other possible reading. So I believe that the pamphlet logically implies that the candidate is the "friend and neighbor" of the people in the picture, which is false. But obviously we need to work out clear rules for juxtaposing pictures so I can see how he could get away with claiming that it's not a lie.

Now for the "Ron Wilson" thing.  First of all, saying you're endorsed by someone is not relevant to the Question Under Discussion (does this person stand for your values?) unless that person is a political figure. That's one way he misled people into thinking that he was referring to the black political figure. And using a name of a famous person in a public pamphlet to refer to your no-name cousin is obviously going to make people think you're referring to the famous person, because a proper name obviously refers to the most salient individual with the name. I bet if they asked people who they thought "Ron Wilson" was referring to, the only answer that people would come up with would be the famous black politician, so it's not even remotely ambiguous in this context.

Yet technically the "Ron Wilson" thing is not lying, I would say. One might reasonably feel that this level of misleadingness ought to be illegal though. How would one craft the law? One way to put it might be in terms of ambiguity: If your statement (slash juxtaposition of pictures with text) unambiguously implies something false, then it's illegal. And you could somehow run experiments or use theoretical arguments to show that the false thing is unambiguously implied.

Tuesday, October 8, 2013

Mitt Romney's false Depictive Sincerity Implicature


In a debate with Obama last year, Mitt Romney said, “Right now, the [Congressional Budget Office] says up to 20 million people will lose their insurance as Obamacare goes into effect next year.”



As PolitiFact pointed out, in fact, the Congressional Budget Office had calculated 5 different estimates:

  • an increase of 3 million
  • a decrease of 5 million (baseline)
  • a decrease of 10 million
  • a decrease of 12 million
  • a decrease of 20 million

The "baseline" or "most likely" estimate was the 5 million people would go off the insurance that they currently had (which I would not describe as "losing", but that's beside the point at the moment).

This connects to some of the research that I have done on scalar terms, starting with exclusives like only, merely, exclusively, sole, mere, and exclusive (official link here; PDF here), and then moving to the superlative modifiers at least and at most (here is our more theoretical paper and here is an experimental paper we did -- this is the one that has most to do with the Mitt Romney quote).  At most has been argued to be quite similar to Mitt Romney's up to, although this has been challenged. I will take at most and up to to be interchangeable for the purpose of the present discussion.


Let me back up a little bit. You might think that the following two sentences are equivalent, i.e. true under exactly the same circumstances:

Liz had more than 3 beers  <-->  Liz had at least 3 beers
Similarly, you might think that the following two sentences are equivalent:
Liz had fewer than 3 beers <--> Liz had at most 3 beers
But there are subtle differences between comparative modifiers like "more than" and "less than" and superlative modifiers like "at least" and "at most". One group of researchers did an experiment in which they gave participants a two sentence argument, where the first sentence was supposed to be the premise and the second sentence was supposed to be the conclusion, and they asked their participants to judge whether it was a good argument.  100% of their participants judged the following to be a good argument:
Liz had 3 beers. Therefore, Liz had more than 2 beers. (100%)
But only 50% of their participants judged the following sentence to be a good argument:
Liz had 3 beers. Therefore, Liz had at least 3 beers. (50%)
The comparative modifier "fewer than" behaved pretty much like its cousin "more than"; 93% judged the following to be valid:
Liz had 3 beers. Therefore, Liz had fewer than 4 beers. (93%)
But they had trouble with the superlative modifier "at most"; 61% judged the following to be a valid argument:
Liz had 3 beers. Therefore, Liz had at most 3 beers.
Why this difference? Researchers all seem to agree that superlative modifiers ("at least", "at most") imply that the speaker does not know how many objects of the relevant kind there are, while comparative modifiers do not have this implication. But there are two different schools of thought as to where this ignorance implication comes from:

  • According to some researchers, ignorance is an entailment, and the sentence is FALSE if the speaker knows how many objects of the relevant kind there are.
  • According to other researchers (myself included), ignorance is a conversational implicature, so the sentence may be ODD when the speaker knows how many objects of the relevant kind there are, but it's not necessarily FALSE.
My collaborator Thomas Brochhagen (University of Düsseldorf) and I did an experiment in order to find out what would happen when people are looking at a picture, of, say, 4 bananas, and it's clear to everyone how many bananas there are. Would they say that the sentence "There are at least 4 bananas in the picture" is FALSE? Here's an example stimulus from the experiment:



There are at least 4 bananas in the picture.
[ ] True
[ ] False

We found pretty much what I expected: If there are 4 bananas in the picture, then "There are at least 4 bananas in the picture" is reliably judged true (100% of the time). Same for "at most 4", "more than 3", "fewer than 5", and "at least 3".

But there was a twist.

Looking at a picture of 4 bananas, people were very likely to say that "There are at most 5 bananas in the picture" was FALSE.

Think about that. Looking at that picture with 4 bananas, would you say it is true or false that there are at most 5 bananas in it? 


There are at most 5 bananas in the picture.
[ ] True
[ ] False

A little less than half of our participants said it was true (44%), and a little more than half said false.

Well, this was surprising. None of the existing theories could explain that. We came up with an explanation along the following lines:
  • The sentence "There are at most 5 bananas in the picture" HIGHLIGHTS the possibility that there are 5 bananas in the picture.
  • The highlighted possibility is false.
  • There is a norm of conversation -- one of Grice's maxims -- which says that one shouldn't highlight possibilities that one knows to be false (or considers highly unlikely, as in Mitt Romney's case). We donned this norm the "Maxim of Depictive Sincerity".
So what Mitt Romney did was to highlight a false possibility. In other words, the "channel" through which Mitt Romney communicated a falsehood was what we can refer to as a Depictive Sincerity Implicature.

So we should add a notch on the Truth-O-Meter for depictive sincerity implicatures.


Tuesday, July 30, 2013

Those complex words "yes" and "no"

Here seems to be a pretty clear case of a lie: Someone asks you if the National Security Agency "collects any type of data on millions or hundreds of millions of Americans", and you are director of national intelligence, and you say "no", as James Clapper did on March 12. Edward Snowden's leaks have now shown that to be false. It's a very naughty lie, since it was done at an intelligence hearing, where lying entails perjury.

Yet James Clapper is somehow getting away with it. How? He claims that he thought that he was answering a different question. He thought he was being asked whether this data was being looked at, or used. Sneaky! Let's think about this with linguistics lingo.

The truth or falsity of an utterance consisting entirely of the particle "no" depends on the Question Under Discussion (QUD). In some cases, it's hard to determine what the QUD is, because it hasn't been made explicit, but in this case, the QUD was explicit: "Does the NSA collect any type of data on millions or hundreds of millions of Americans?" So "no" is clearly equivalent to "The NSA does not collect any type of data on millions or hundreds of millions of Americans," however that is interpreted. Clapper claims that what he meant to communicate was that NSA does not use any type of data on millions or hundreds of millions of Americans, because that's the negative answer to what he thought the QUD was; by "collect", he thought that the questioner (Senator Ron Wyden, D-Oregon) meant "use". It is almost entirely impossible to believe that this kind of misinterpretation is possible, but that's another issue.

To put it in different terms, Clapper claims he thought that his answer, "no", was relevant to a different question. One way of interpreting Grice's Maxim of Relevance (which just says "Be relevant") is, "Make sure that your utterance addresses the QUD." Since Clapper's "no" addressed a different QUD from the one that was really in play, one could possibly analyze this case as a false relevance implicature

A different case that I would definitely like to analyze in this way is the one where Senator Carl Levin asks Goldman Sachs CFO David Vinar if he feels bad that Goldman Sachs misled their customers and he responds that it was "unfortunate to have on email."
Senator Carl Levin: And when you heard that your employees, in these e-mails, when looking at these deals said, God, what a shitty deal, God what a piece of crap – when you hear your own employees or read about those in the e-mails, do you feel anything?  
Goldman Sachs CFO David Vinar: I think that’s very unfortunate to have on e-mail. 
Ha! Vinar is pretending to answer the question of whether he feels any twinge of moral regret, when what he's literally saying is just that it was a practical slip-up. So Vinar is not addressing the QUD. In Grice's terms, Vinar is "quietly and unostentatiously violating" the Maxim of Relevance, and when one violates a maxim in this manner, one is "liable to mislead".

But going back to the case at hand, with Clapper's "no", I am not entirely satisfied with the "false relevance implicature" analysis, because it seems that in this case, the statement ITSELF is really false, whereas in the Vinar case, the statement itself was true, and he was "dodging the question" -- answering a question other than the QUD. With "no", you can't really get away with saying something irrelevant; you can't not address the QUD. In the Vinar case, there were two different questions: The question he was asked and the question he answered. In the Clapper case, there were two different interpretations of the same question.

So although the Clapper case relies heavily on the notion of relevance, I would not analyze it as a false relevance implicature, unlike the Vinar case.

I do not pretend to have made this issue any clearer. Rather the opposite. But I hope to have convinced the reader that the cheekiness in the last sentence of the Slate article is not entirely merited:
Mr. Clapper’s participation in any public discussion of the limits of data mining will be of no value, since we are going to have to parse his meanings of complex words like “yes” and “no.”
The words "yes" and "no" actually are pretty complex.

Tuesday, May 14, 2013

Rachel Maddow ANGRY!!!

This is exciting: Rachel Maddow is totally p*****d off about PolitFact!

So, Martina Navratilova claimed that you can get fired for being gay or perceived as gay in 29 states. PolitiFact found out that there are 29 states that lack specific protections regarding sexuality. So (according to Rachel Maddow) the statement is TRUE, TRUE, TRUE!!!!!!

Yet PolitiFact rated the statement as "Half True". Which shows, according to Maddow, that "the very important concept of fact-checking has become pointless at a time in our country when we really need it to mean something". According to Maddow, PolitiFact "decided it didn't seem seemly or whatever to call it true, so then they researched unrelated information about other things besides states, like some companies decided they don't wanna discriminate, doesn't that count for something?? No! Because that is not the statement you were fact-checking. The statement you were supposed to be fact-checking is true, and until somebody figures out how to sue you in order to retrieve the word "fact" from the dark and airless hole you have stuffed it into, PolitiFact, then no, it is not OK for you to make this stuff up. You are truly terrible."

This is exciting because it shows how important it is that we have a more consistent and objective classification scheme for the kinds of statements that are subject to fact-checking, so we don't devolve into angry finger-pointing and name-calling like this. Semantics and pragmatics to the rescue!

(I guess it's good television, and maybe we don't want to take that away from the world, but maybe there's a way to have good television that doesn't involve angry ape grunts.)

It seems that PolitiFact was addressing a possible alternative interpretation of Navratilova's statement when they were looking into these details. Rachel Maddow's rant is passionately founded upon the presumption that there is only one possible interpretation of the statement, and that the statement can only either be true or false. Well, once you pick an interpretation, I agree that "what is said" (setting implicatures and such aside) can only be true or false, but there is such a thing as ambiguity, and even if Navratilova's statement was not particularly ambiguous, PolitiFact is not "terrible" for looking into alternative interpretations.

I think that what Martina Navratilova intended to convey was true, and what she intended to convey is possibly how most people would interpret the statement (although I wouldn't have known to verify it in quite that way myself), so the alternative possible interpretations that PolitiFact implicitly looked into are probably less likely interpretations. But it is not fair to say that they "researched unrelated information".

What PolitiFact should have said was that Martina Navratilova's statement can be interpreted in several ways:

1. In 29 states, state law does not prohibit firing people for being homosexual or perceived as such.
2. In 29 states, there are some jobs that you can get fired from for being homosexual or perceived as such (an "existential" interpretation -- "there are some")
3. In 29 states, no matter what job you have or where you work, throughout the entire state, you can be fired for being homosexual or perceived as such (a "universal" interpretation -- for all jobs)
4. In 29 states, it is the rule rather than the exception that you can be fired for being homosexual or perceived as such (a "generic" interpretation -- this is typically or usually the case)

Interpretation #1 is clearly the case. (You might think we don't need a distinction between interpretations #1 and #2, but they are logically independent. One might imagine that there are other protections that people can use which cover all the cases, so laws specifically prohibiting this kind of discrimination would be redundant. Then interpretation #1 would be true and interpretations 2-4 would be false.) 

Interpretation #2 also seems to be quite true: In 29 states, some jobs are like that.

Interpretation #3 ("all jobs are like that") is what PolitiFact showed to be false.

And they may have succeeded in casting doubt on Interpretation #4, the generic interpretation, because the gender-related discrimination laws may make it so that in most cases, it is actually illegal to fire someone if you merely perceive that they are gay.

I think that Interpretation #4 is actually a fairly reasonable interpretation. I don't know if I would have gone all the way down to "Half True" on the Truth-O-Meter, but I think there is a somewhat reasonable interpretation of Navratilova's statement that might not be the case.

So I'm in the strange position of defending PolitiFact here, even though this whole blog was founded upon a critique of PolitiFact. But the larger point is that we need to be able to use words like ambiguity in the public discourse, and discuss multiple possible interpretations, so we don't have to resort to angry name-calling.

Tuesday, April 9, 2013

Relevance violation

Previously I analyzed the following statement from Obama as a case of ambiguity: 

"There are millions of women all across the country who rely on Planned Parenthood for not just contraceptive care; they rely on it for mammograms, for cervical cancer screenings." 

(Contrary to what this might suggest, Planned Parenthood does not provide mammograms, although it does provide access to them.)

I said: "Obama's statement could be interpreted in two ways. Either it means that Planned Parenthood enables people to get mammograms (in which case it is true), or it means that Planned Parenthood provides people with mammograms (in which case it is false). The latter interpretation might be somewhat more likely, and therefore the statement is misleading."

But it might not have been a MERE case of ambiguity. It could also have been a "quiet and unostentatious" violation of Grice's Maxim of Relevance ("Be relevant."). 

(Grice distinguished between several different ways that a maxim could be violated, including 'flouting' -- making it perfectly obvious that you are violating it -- and 'quietly and unostentatiously violating' the maxim. In the latter case, says Grice, "one is liable to mislead." The cases I have discussed so far have all been 'quiet and unostentatious' violations.)

It depends a little bit on what the Question Under Discussion is in this case. One could argue that the Question Under Discussion was such that, if Obama's statement was relevant to it, then his statement must have meant that Planned Parenthood provides people with mammograms. In that case, it would be a quiet and unostentatious violation of Relevance. I'm not sure if this is really a great example of that.

Wednesday, April 3, 2013

Paul Ryan and the Bowles-Simpson Debt Commission


Paul Ryan was accused of lying by ThinkProgress.org for uttering the following statement (henceforth "the statement"):
[Obama] created a bipartisan debt commission [the Bowles-Simpson commission]. They came back with an urgent report. He thanked them, sent them on their way, and then did exactly nothing. 
(According to ThinkProgress.org, the reason that the statement was a lie is that “Ryan was instrumental in sabotaging the commission, leading the other House Republicans in voting against the plan.”) 

CNN analyzed it as "misleading". I concur. But I would add, "due to a conversational implicature".


It passes all of the tests for a conversational implicature.

Ryan’s statement implies several interconnected things that can be argued to be false. We can concentrate on this one (henceforth "the implicatum"):
Obama’s inaction prevented the Bowles-Simpson plan from being adopted. 
The statement is consistent with the negation of the implicatum; i.e., the inference is cancellable. The following is not a contradiction:
[Obama] created a bipartisan debt commission [the Bowles-Simpson commission]. They came back with an urgent report. He thanked them, sent them on their way, and then did exactly nothing. [This much is actually true.] But Obama’s inaction did not prevent the Bowles-Simpson plan from being adopted [because he didn’t have the votes necessary to pass it].
Someone who said this would not be contradicting themselves. So the inference is cancellable.

Another property of conversational implicatures is that they can be reinforced; that is, it doesn't sound redundant to assert the implicatum after asserting the statement.
[Obama] created a bipartisan debt commission [the Bowles-Simpson commission]. They came back with an urgent report. He thanked them, sent them on their way, and then did exactly nothing. Obama’s inaction prevented the Bowles-Simpson plan from being adopted.
If Paul Ryan said this, it would not sound like he was repeating himself. This further supports the conclusion that the inference is a conversational implicature.

So it was not lying to imply that Obama's inaction prevented the Bowles-Simpson plan from being adopted. What Paul Ryan actually said (maybe not literally, but figuratively) could all be argued to be true. He did create a bipartisan debt commission. They did come back with an urgent report. He did thank them. He did, I suppose, send them on their way. And he did not do anything about it.

It is the implicature that is false. True statement, false implicature => misleading.

How does this implicature arise? That is a more interesting question.

It has to be some kind of quantity implicature. Grice's Maxim of Quantity says:

1. Make your contribution as informative as is required (for the current purposes of the exchange).
2. Do not make your contribution more informative than is required.

There is a clear sense that Paul Ryan is leaving out information that is required for the purposes of the exchange, violating Quantity-1.  He tells a story, and the story has a point. The point is that Obama failed in this instance. Any other information relevant to whether or not Obama failed in this instance should be mentioned. The quantity implicature is that there is no other information relevant to this point.

Interesting note: The definitions for all of the Truth-O-Meter's shades of grey include the possibility of a Quantity-1 violation. 
  • Mostly True statements "need additional clarification or information". 
  • Half True statements "leave out important details or take things out of context". 
  • Mostly False statements "ignore critical facts that would give a different impression".
Maybe they have some way of deciding whether something is a mild Quantity-1 violation or a severe one.

Monday, April 1, 2013

Unverifiable yet misleading

There's something I left out of the discussion about the claim that 40% of gun sales are done without a permit. That claim was not merely unverifiable, it was also misleading. It was misleading because the speaker was acting as if he had a solid source for the information. So this is a straight-up violation of Grice's second maxim of Quality. Grice's maxim of Quality ("Try to make your contribution one that is true") had two sub-maxims:

1. Do not say what you believe to be false.
2. Do not say that for which you lack adequate evidence.

A cooperative dialogue participant will adhere to both of these maxims. The conversational implicature that people believe and have good evidence for what they say comes about through the assumption that people are being cooperative.

Saying that 40% of the gun sales are done without a permit is saying something for which you lack adequate evidence. It's a violation of Quality-2, as we say in the biz. So although the status of the claim is unverifiable, it is also misleading in the sense that it has a false implicature.

So I have to revise my shades of grey:

  • Ambiguous (true on one interpretation, false on another*)
  • Unverifiable (more research needed)
    • presented as such
    • carrying the false Quality-2 implicature of adequate evidence
  • True but misleading (carrying a false implicature)
*I think I might have to add a category for "true on one interpretation, unverifiable on another", but I will do that in a separate step.

Thursday, March 28, 2013

Critique of Politifact's Truth-o-meter

Politifact.org uses the following ratings in its Truth-o-meter:

True – The statement is accurate and there’s nothing significant missing. 

Mostly True – The statement is accurate but needs clarification or additional information. 

Half True – The statement is partially accurate but leaves out important details or takes things out of context.

Mostly False – The statement contains some element of truth but ignores critical facts that would give a different impression. 

False – The statement is not accurate. 

Pants on Fire – The statement is not accurate and makes a ridiculous claim. 

The reason for having these gradations is that "- especially in politics - truth is not black or white". 

I agree with this. And I appreciate the research that PolitiFact does. It is all I need in order to gain clarity for myself on things that are said in the media. But these categories don't seem to be helping much. 

How can you decide between Mostly True and Mostly False? What is the difference between needing clarification or additional information and ignoring critical facts that would give a different impression? When Romney was accused of being "director of a company that stole millions from Medicare", the claim was judged as "Mostly True": "needing additional clarification". According to PolitiFact, "His firm bought the company that was later found guilty of fraud, but Romney was not running the show while crimes were being committed."  Why doesn't that count as ignoring critical facts that would give a different impression? These criteria seem rather bogus.

Furthermore, some of the examples that are "Mostly False" are merely unknown, such as the claim that 40% of gun sales are done without a permit -- a number that is based on old research. We don't really know whether it contains some element of truth, and it's not like critical facts are being ignored here. We have uncertainty about the facts in this case.

Some of the cases are ambiguous, and true under one interpretation and false under another, such as the claim that there are twice as many gun stores as McDonald's -- it depends on what counts as a "gun store". No critical facts are being ignored here; it's just a matter of what the statement means. We have uncertainty about the interpretation of the statement in this case.

The interesting cases for a linguist such as myself are the ones where the statement itself is true, but it carries a conversational implicature that is false. A conversational implicature is an inference that a hearer can draw based on the fact that a speaker said something, crucially using the assumption that the speaker is adhering to certain norms of conversation. Read Grice (1975), "Logic and Conversation" if you don't know what I'm talking about. This is a class of misleading statements that should have its own category label, something like "True statements with false implicatures".

Here's an example with a conversational implicature.  There was a Republican attack ad that said that Chuck Hagel advocates nuclear disarmament for the US, and wants to "back down", despite a growing nuclear threat from our scary enemies. FactChecker.org fact-checked that claim, and found that indeed, Chuck Hagel had co-authored a document for an organization whose mission includes a reduction in nuclear stockpiles worldwide. So yes, at least in a sense, the claim is true. But they also point out that the document does not advocate unilateral disarmament, and says in fact that this would be a "less good option". The ad suggests that Chuck Hagel would be in favor of unilateral disarmament, and that is not true. So technically, what the ad says is true, but it implies something that is not true through a conversational implicature. You could say it's true, but misleading.

But the term "misleading" is still too vague. Here is another example of somebody saying something misleading. Obama said that "there are millions of women all across the country who rely on Planned Parenthood for not just contraceptive care; they rely on it for mammograms, for cervical cancer screenings." FactChecker.org assessed the statement as Half True, saying "The critics have a point that Planned Parenthood does not actually provide mammograms; instead, the organization refers patients who need them offsite. Supporters of Planned Parenthood also have a point that the organization serves as an important link between some female patients and the mammograms a doctor determines they need." I would analyze this case as follows. Obama's statement could be interpreted in two ways. Either it means that Planned Parenthood enables people to get mammograms (in which case it is true), or it means that Planned Parenthood provides people with mammograms (in which case it is false). The latter interpretation might be somewhat more likely, and therefore the statement is misleading.

But Obama's statement is misleading in a different way from the Republican attack ad. In Obama's case, the statement is ambiguous, and true or false depending on how you interpret it. In the case of the attack ad, it's relatively clear what the statement is saying, and that it is true, but there is a conversational implicature going along with it that is false. So "misleading" is too vague because it doesn't distinguish between these two cases.

So I would propose the following "shades of grey" between True and False:

  • Unverifiable (needs further research)
  • Ambiguous (true on one interpretation, false on another)
  • Technically true but misleading (due to a false implicature)