No depth-charge channel is too noisy to be confused by
« previous post | next post »
Yuhan Zhang, Rachel Ryskin & Edward Gibson, "A noisy-channel approach to depth-charge illusions." Cognition, March 2023:
The “depth-charge” sentence, No head injury is too trivial to be ignored, is often interpreted as “no matter how trivial head injuries are, we should not ignore them” while the literal meaning is the opposite – “we should ignore them”. Four decades of research have failed to resolve the source of this entrenched semantic illusion. Here we adopt the noisy-channel framework for language comprehension to provide a potential explanation. We hypothesize that depth-charge sentences result from inferences whereby comprehenders derive the interpretation by weighing the plausibility of possible readings of the depth-charge sentences against the likelihood of plausible sentences being produced with errors. In four experiments, we find that (1) the more plausible the intended meaning of the depth-charge sentence is, the more likely the sentence is to be misinterpreted; and (2) the higher the likelihood of our hypothesized noise operations, the more likely depth-charge sentences are to be misinterpreted. These results suggest that misinterpretation is affected by both world knowledge and the distance between the depth-charge sentence and a plausible alternative, which is consistent with the noisy-channel framework.
Yuhan Zhang discusses the paper in a thread on Twitter.
Speaking of depth, I'm definitely out of mine when it comes to noisy-channel frameworks. But it isn't the case that I'm not so ignorant as to fail to recognize that this paper is not too unimportant for Language Log not to pay no attention to it.
(Hey, ChatGPT — betcha can't make sense out of that!)
Jerry H said,
March 19, 2023 @ 10:20 pm
Weird. I literally cannot figure out how to read the example sentence in what the quote says is the "literal" way.
I parse it as "(There is) no head injury (which) is too trivial to be ignored."
Which, as far as I can tell, means that any head injury, no matter how trivial, should not be ignored. Which the article says is the opposite of how I should be reading the sentence. I just can't seem to wrap my head around it. Even after reading the twitter thread, I can't seem to make it read any other way.
I guess I can read it to say you should not ignore the lack of a head injury, but I don't think that's the intended literal meaning either.
AntC said,
March 20, 2023 @ 1:40 am
(I've no wish to re-litigate such a fertile source of ambiguity, but …)
I wouldn't be so sure anybody could claim
There's a negative quantifier "No …". There's a co-ordinated comparison "too X to be Y" including a passive with a deleted Agent.
Unwinding what the "literal" and "compositional" "implies" is hazardous. Adding "actually" seems to me to be substituting asseveration for analysis.
I'd prefer to say (1) is just plain meaningless/failing to convey anything coherent. Those darned Linguisticians should just stop forcing innocent monkey brains into making any commitment.
If someone's banged their head, get them to E.D. and stop pontificating. (Indeed any words at all are wasting precious time.)
Jamie said,
March 20, 2023 @ 2:36 am
I agree with AntC. The sentence is completely idiomatic — I can’t see any way of parsing it that makes sense. Therefore, it actually means what people think it means
DJL said,
March 20, 2023 @ 4:34 am
Oh, I did feed this sort of sentences to ChatGPT once upon a time:
https://leiterreports.typepad.com/blog/2021/10/ai-chatbot-part-2.html
https://3quarksdaily.com/3quarksdaily/2023/03/chatgpt-hasnt-learned-any-language-and-it-also-doesnt-display-general-intelligence-but-you-can-ask-it-to-complete-your-sentences.html
Aristotle Pagaltzis said,
March 20, 2023 @ 4:49 am
Yes, but it’s even weirder than that. Because the way it gets to its meaning is by implying as a given that only sufficiently trivial head injuries deserve attention. On that basis it then claims that head injuries so trivial as to meet this standard for receiving attention do not exist. Therefore practically speaking, all head injuries should be ignored – not just serious and even less serious ones, but all of them.
Truly a marvelously confused construction.
Annie Gottlieb said,
March 20, 2023 @ 7:21 am
Here’s a diabolical brain twister I just saw on Twitter: I can’t even figure out whether it says one thing or the opposite!
BluePaintedRaven
@raven_painted
You will never cease to be anything less than impressive to me.
J.W. Brewer said,
March 20, 2023 @ 7:25 am
Imagine the following dialogue:
Q. What should we do about head injuries?
A. Ignore them.
Q. But what about the really trivial ones?
A. Ignore even the really trivial ones.
That seems on pragmatic grounds like an extremely odd position, or set of positions, to take (given certain presuppositions about the speaker/writer and their goals), but the sentences are, I should hope, understandable and indeed unambiguous. The sentence in question merely (on the presumably unintended literal parse) sums up the bottom line of the dialogue.
I am baffled by seemingly fluent people who are unable to parse English to a degree that they purport to be unable to understand the literal parse of the sentence in question even after they've been primed by being told that the literal parse conveys a meaning that's weird/counterintuitive/nonsensical on pragmatic grounds. It's good on balance that our brains can subconsciously jump past a literal parse to a likely intended meaning when the literal parse is sufficiently weird or unexpected, but it's not good on balance to be unable to see the weirdness if one backs up and focuses on it.
I am also baffled by the motivation to strain to come up with an alternative parse that can rescue the sentence rather than simply accept that native speakers sometimes get fuddled (due to the limitations of their "poor monkey brains," to use Prof. Liberman's formulation) and fail to use the wording that would conventionally express their intended meaning.
Cervantes said,
March 20, 2023 @ 7:49 am
Well, "No head injury is so trivial as to be ignored" doesn't seem problematic. But the problem sentence should read "No head injury is too trivial to respond to." But as the first sentence makes sense, and seems very similar in construction to the second, that's how we read the problem sentence. Semantic plausibility triumphs.
KeithB said,
March 20, 2023 @ 8:08 am
"I don't know half of you half as well as I should like; and I like less than half of you half as well as you deserve.”
Bilbo Baggins.
Though of course, this sentence was probably carefully crafted.
MG said,
March 20, 2023 @ 8:43 am
Sorry…
Me: I read this blog post today. [post body (in between lines of "—")] Can you make sense of that final paragraph?
ChatGPT: Yes, I can make sense of that final paragraph. The author is acknowledging that the concept of noisy-channel frameworks is beyond their understanding, but they still recognize the importance of the research paper on depth-charge illusions. They also make a playful remark challenging ChatGPT to make sense of their writing.
J.W. Brewer said,
March 20, 2023 @ 8:49 am
Conveniently enough, the very first sense of "trivial" given at wiktionary is "ignorable; of little significance or value." So one can abstract the problem with the problem sentence from head injuries – the usual way in which "ignore" and "trivial" interact due to their ordinary semantics is that the more trivial a given instance of X is, the stronger the case for ignoring that instance of X. The intended meaning of the problem sentence is presumably that head injuries are so serious that the default tendency to treat trivial instances* as ignorable needs to be overridden. By contrast, the problem sentence seems to presuppose that triviality and ignorability are negatively rather than positively correlated, which may be the key conceptual weirdness (or simple error) that leads to the weird/erroneous output.
*The intended meaning is in any event not to be taken overliterally because there are plenty of real-world instances of bumps, bruises, cuts, scrapes etc. to the human head that the medical profession does not in practice take seriously. Presumably the dodge is to not consider those "injuries" in the first place because the working definition of "injury" has some sort of materiality (or non-triviality) element buried in it.
Cervantes said,
March 20, 2023 @ 9:24 am
Yes, I think "head injury" in this context is probably intended to mean a concussion or suspected concussion. But that's a different quibble.
Victor Mair said,
March 20, 2023 @ 9:58 am
@Annie Gottlieb
"You will never cease to be anything less than impressive to me."
You will always be impressive to be.
Bloix said,
March 20, 2023 @ 10:22 am
1) No head injury is too trivial to be ignored.
– That injury is too trivial to be ignored.
2) No head injury is too serious to be ignored.
– That injury is too serious to be ignored.
Seems to me that (1) is wrong, not because it means "we should ignore them," but that "too trivial to be ignored" is nonsensical. Put another way, the problem is not that (1) means "we should ignore them. It means "we should not ignore them," but for a reason that implies that a less trivial injury would be ignorable, which is makes no sense.
Bloix said,
March 20, 2023 @ 10:54 am
Oh, dear. Look at all my typos!
Let me try again – the problem with "No head injury is too trivial to be ignored" is not that it has a literal meaning the opposite of what is intended.
The problem is that the literal meaning, although it literally states more or less what is intended (that even the most trivial head injury must be examined), it also necessarily implies something that is not intended (that some more serious injuries need not be examined). I.e., our statement logically implies that "The statement 'some head injuries are too trivial to be ignored' is false." This version appears to imply the possibility that some head injuries are serious enough to be ignored.
So the problem is that the statement seems to posit a world in which some serious injuries can be ignored but no trivial ones may be ignored.
But you can only get to this result by rigorously imposing a chain of logic in a manner that's not the way we usually interpret language. In ordinary usage, the problem would not rise to the level of consciousness.
Gregory Kusnick said,
March 20, 2023 @ 11:12 am
Victor's reading of Annie's "never cease" sentence may be what was intended (or maybe not), but what it actually says is "You will always be less than impressive to me" — the opposite of the (presumptive) intended meaning.
(Without knowing the context, it's conceivable this is intentional irony.)
J.W. Brewer said,
March 20, 2023 @ 12:04 pm
I'm not sure what Bloix means by "nonsensical." Nonsensical a la "colorless green ideas sleep furiously", or something less than that? Because I don't think sentences that contain logical fallacies and/or have embedded in them presuppositions that are contrary to most people's assumptions about the world are necessarily nonsensical in a useful sense of that word, although they well may be production mistakes. Or sometimes they're just evidence that the speaker/writer is (like many human beings) prone to logical fallacies and/or (like at least some human beings) has an idiosyncratic understanding of reality.
Part of the relevant context of the problem sentence here is a cultural assumption that instructions/advice given by medical personnel to other medical personnel in a professional setting will in fact be logical and empirically well-grounded. If the context were otherwise, there might be less of an incentive to "fix" the problem by imputing a sensible probably-intended meaning to the production-error sentence.
To use legal jargon, when you conclude that something oddly-phrased (either locally or when read in broader context) in a contract or similar document is a "scrivener's error" you interpret the document based on what differently-worded sentence appears to have been intended to have been put in the document rather than the sentence actually found in the document, but that's not IMHO quite the same thing as pretending that the "error" sentence "means" something other than what it means. Rather, you interpret it as if it means something other than that, precisely because you have classified it as a production error (or, in some cases, a judge formally "reforms" the document to in effect change its actual wording to achieve the desired result).
Victor Mair said,
March 20, 2023 @ 1:14 pm
@Gregory Kusnick
It's a headscratcher for sure, and I thought about it for quite a while.
"never cease" = "always" no problem
"less than impressive" = "not impressive; unimpressive" no problem
But what is the meaning / function / purpose of "anything"?
Here are a couple of sentences for comparison.
Cf.
"This was never anything less than brutal."
in context
https://www.theguardian.com/sport/2014/apr/05/ulster-saracens-heineken-cup-match-report
"Anything less than perfection could tarnish it".
in context
https://www.nytimes.com/2003/08/05/business/bulgari-builds-on-brand-in-many-ways.html
Neal Goldfarb said,
March 20, 2023 @ 1:45 pm
Thanks to MG for getting ChatGPT to take up the challenged I laid down in the final paragraph of my post.
ChatGPT was spot-on in its interpretation of the first and third sentence.
Its reading of the second sentence presents a more difficult problem. I can't say whether it got the semantics right, because I don't know what the sentence's literal meaning is. In writing the sentence, I wanted to compose the Mother of All Misnegations, which I tried to accomplish by stringing together an unseemly number of embeddings and negations, with no intended interpretation in mind. Unsurprisingly, ChatGPT failed to pick up on that, which I take to mean that the data it was trained on didn't include anything by me.
OTOH, in a looser sense, ChatGPT's interpretation isn't that far off, in that I did think it was important for notice of the paper to be taken on Language Log, given LLog's longstanding interest in misnegation.
Xtifr said,
March 20, 2023 @ 3:53 pm
Hmm, seems like the "no injury…" sentence might be a double depth-charge! Underneath the obvious misinterpretation ("no injury is trivial enough…) is a second, semi-obvious misinterpretation (the paper's "we should ignore them"), which seems to assume the sentence is a simple misnegation.
In fact, if you look at the template, "no head injury is too X to be ignored.", you can see that it's true for all X. No head injury should be ignored; no modifier can change that. Therefore, the original sentence is true! (Albeit silly.)
There is something like misnegation going on ("too trivial" vs. "trivial enough"), but, unusually, it doesn't apply to the whole sentence. Hence, a second, hidden depth charge.
Can you have a triple depth charge, I wonder?
Philip Anderson said,
March 20, 2023 @ 4:00 pm
I would understand a trivial head injury as meaning one that could be ignored, so “no head injury is too trivial” would mean “no head injury is to be ignored”.
We can join these statements with an “and”, as “no head injury is too trivial and no head injury is to be ignored” but if we merge them an “or” is logical, i.e. “no head injury is too trivial or to be ignored“. But what would “no head injury is too trivial and to be ignored“ mean? Much the some I think. And if neither “or” nor “and” is used?
AntC said,
March 20, 2023 @ 7:37 pm
@JWB I am baffled by seemingly fluent people who are unable to parse English to a degree that they purport to be unable to understand the literal parse of the sentence in question …
I am baffled by why researchers think that any behaviours wrt these sort of pathological sentences provide any reliable findings about speaker competencies.
To JWB's point: I didn't say I was "unable". I did say (in effect) normal conversation isn't supposed to involve solving cryptic crosswords. (There's probably a Gricean norm to that effect.) If you produced that sentence in conversation I wouldn't try to make sense of it; I'd ask you to rephrase. Since it seems to involve a matter of life and death, you should put in better effort to be understood.
Rather than all that legal flummery about 'scrivener's error', if the author of the sentence was unavailable for asking, I'd ignore it and use my own judgment.
Peter Grubtal said,
March 21, 2023 @ 5:16 am
Such sentences remind me of those optical conundrums like the cube frame viewed obliquely, where when you look first the eye identifies what it thinks is the vertical closest to you, but if you blink and look again it seems to be the back vertical.
What the OP calls the "literal meaning", I had to blink several times to arrive at.
Does this have any general implications, in particular for UG? Where negatives (sometimes implicit) are involved it sometimes arises that an utterance is difficult to resolve, and I think this can apply in other languages as well.
Amongst the ancients, wasn't there a cheap backhander: "you will be a famous poet when Homer is forgotten"?
Taylor, Philip said,
March 21, 2023 @ 5:45 am
I respectfully disagree. Annie did not write "You will never cease to be less than impressive to me", she wrote "You will never cease to be anything less than impressive to me". We can replace "less than" by "other than" to see that she meant "You will never cease to be anything other than impressive to me", or more simply "You will always be impressive to me".
Bloix said,
March 21, 2023 @ 8:22 am
JW Brewer- I mean nonsensical in a sense that we use all the time. Edmund Lear wrote a book of limericks called The Book of Nonsense which included things like,
There was an Old Man with a beard,
Who said, "It is just as I feared!—
Two Owls and a Hen,
Four Larks and a Wren,
Have all built their nests in my beard!"
This is not like "colorless green ideas", which is meaningless – you can't draw a picture of it or even imagine a meaning for it. You can easily draw a picture of Lear's Old Man, and of course, Lear did. But it's nonsense, because it describes a situation that can't exist in our world. We use the word nonsense far more frequently to describe a purported situation or event that can be imagined but can't exist than to describe a collection of words that is grammatically correct but meaningless. (A recent expression for this kind of meaninglessness is "word salad.")
The sentence under discussion – no head injury too trivial – is nonsense because it posits a world in which the more serious the injury, the more it is to be ignored. That's not the world we live in, so the sentence makes an assertion about the world that's as nonsensical as a world in which birds build nests in men's beards.
Nathan said,
March 21, 2023 @ 10:58 am
Taylor, Philip: What the person Annie quoted "meant" is anyone's guess.
At issue is what they wrote.
Trying to read "literally", I get an odd number of negations of impressive:
"You are less than impressive" = NOT impressive
"You are anything less than impressive" = NOT impressive (no extra negation here)
"You cease to be anything less than impressive" = impressive
"You will never cease to be anything less than impressive" = NOT impressive
The moral of the story, as usual, is to avoid piling up the negations, which are implicit in a wide array of English constructions.
Taylor, Philip said,
March 21, 2023 @ 11:28 am
Well, Nathan, I do not know what you meant by what you wrote, but I believe that I can reasonably infer your meaning. In just the same way, I believe that I can reasonably infer what Annie meant by "You will never cease to be anything less than impressive to me". The fact that you infer something different to me tells us more about our respective thought processes than it does about what Annie meant …
Gregory Kusnick said,
March 21, 2023 @ 11:33 am
Sorry, Philip, but I don't see how "anything other than impressive" can reasonably be taken to mean "impressive" when the former explicitly excludes the latter.
Taylor, Philip said,
March 21, 2023 @ 11:52 am
Well, I think that I understand your rationale for that analysis, Gregory, so I think that I must have been mistaken — the problem appears to be more complex than I originally thought.
If we take as a starting point "You will never be anything less than impressive to me", it is clear that this is intended as a compliment — adding "cease to" inverts the sense, so with that added I now read it as a cleverly disguised put-down.
Jonathan Smith said,
March 21, 2023 @ 6:57 pm
"You will never cease to be anything less than impressive to me" isn't coherent; "anything" + adj requires negative polarity context surely.
Cf. no such thing as "there's always anything interesting on LL"
Arthur Baker said,
March 21, 2023 @ 10:28 pm
@J.W. Brewer
"I am baffled by seemingly fluent people who are unable to parse English to a degree that they purport to be unable to understand the literal parse of the sentence in question even after they've been primed by being told that the literal parse conveys a meaning that's weird/counterintuitive/nonsensical on pragmatic grounds."
And I am baffled by commenters who are baffled by others' inability to parse, but nevertheless write a 52-word sentence with multiple-level sub-clauses that took me about 15 minutes to parse.
DaveO said,
March 23, 2023 @ 11:03 am
I didn't get the "unintended" meaning of the example sentence, I'm probably too literal-minded.
The first example I ever heard of this type of construction was on an SNL skit with Edwin Newman, "You can never put too much water into a nuclear reactor."
ie. It's OK to put a lot of water in, OR, it's never a good idea to put a lot in. :D
Taylor, Philip said,
March 23, 2023 @ 12:36 pm
I don't see both meanings in the text as quoted, Dave. For me, "You can never put too much water into a nuclear reactor" can mean only "It's OK to put a lot of water in" — for the sentence to mean "[I]t's never a good idea to put a lot in", it would have to be re-cast using "should" where the original has "can" (i.e., You should never put too much water into a nuclear reactor").
Michael Watts said,
March 24, 2023 @ 7:37 pm
I'd have to disagree with the original assertion about the literal meaning of the sentence. You can parse it perfectly well as saying that "for head injuries, it is never true that they are too trivial to ignore". And, if you parse the sentence that way, it is quite obviously true: nothing is ever too trivial to ignore for the simple reason that the concept "too trivial to ignore" is ill-founded. Since it is never the case that things become more difficult or less advisable to ignore as they become more trivial — this is part of the meaning of "trivial" — it is true of all things that those things are not too trivial to ignore. And since it's true of all things, it must be true of all head injuries.
This type of claim is not common outside of mathematical proofs (though it is common inside them), but there's no difficulty in parsing the sentence that way.
That is not an example of this type of construction. It's an example of ambiguity between deontic modality ("you are not allowed to put too much water into a reactor") and alethic modality ("you are not able to put too much water into a reactor"). Both are expressed by the modal verb can.
This construction is just a mistake on the part of the speaker. The deontic and alethic uses of can are both allowed by the language.
JPL said,
March 24, 2023 @ 10:34 pm
The speaker's problem with the expression of what they intended to express centers on the choice between 'too' and 'so'. What (I am guessing) bubbled up in their conscious awareness were 'trivial' and 'ignore', and then, as often happens, they didn't pay attention to the negations, which can show up in different ways.
What you have is a situation understood in terms of two scales or gradients of parallel direction with an inferential relation between them: 1) one from "significant" to "trivial", and the other from "address" to "ignore" (or some equivalents), with the "eye" at the source end and an ill-defined dividing line for the transition between the one and the other. 'so' includes all positions on the gradient from the source to the dividing line and just over it, while 'too' indicates a point beyond the dividing line. So one could express the speaker's intended meaning in several different ways that are equivalent wrt descriptive accuracy. 1) If you want to keep 'trivial' and 'ignore', you could say, "No head injury is so trivial that it can be ignored". 2) If you want to keep 'too', then you can have "No head injury is too trivial to address". There are several other possibilities. The "too" expression was probably more salient (accessible) for the speaker than the expression with "so".
I would suggest that the authors not use the term "literal meaning" for the conventional interpretation of the speaker's expression. We are not dealing with metaphor or "figurative" language. What you have is simply 1) the speaker's intended proposition (meaning intended to be expressed by his sentence); 2) the proposition you would get from a critical application of the rules of expression of the language being used (norms), available to all speakers (the "literal" interpretation of the problematic sentence, not equivalent to the speaker's intention); 3) and then, as we have seen here on LLog, the speaker comes up with a proposition (expressed as (1) or (2) above) that is essentially equivalent to the one the speaker intended in spite of the problems of expression, so the speaker says, "Yeah, that's what I meant!". I don't think "illusion" is involved; just conversational repair. The proposition is not given in the perceived sentence; you have to "do" it. Also, when you want to express something, mind the negations.
Michael Watts said,
March 24, 2023 @ 11:36 pm
You can stay closer to the original statement by correcting it to "No head injury is trivial enough to ignore". "S enough to V" is just as idiomatic as "too S to V", and expresses that the S and should-V scales move with each other instead of moving against each other.