[HN Gopher] Translating natural language to first-order logic fo...
___________________________________________________________________
Translating natural language to first-order logic for logical
fallacy detection
Author : ColinWright
Score : 142 points
Date : 2025-03-04 17:36 UTC (5 hours ago)
(HTM) web link (arxiv.org)
(TXT) w3m dump (arxiv.org)
| rahimnathwani wrote:
| The paper links the code repo:
| https://github.com/lovishchopra/NL2FOL
|
| But I don't see a pretrained model in there, so I'm not sure what
| to pass as `your_nli_model_name`: python3
| src/nl_to_fol.py --model_name <your_model_name> --nli_model_name
| <your_nli_model_name> --run_name <run_name> --dataset --length
| janalsncm wrote:
| If we check the script, it seems to support open ai models and
| llama
| https://github.com/lovishchopra/NL2FOL/blob/main/src/nl_to_f...
|
| It would have been a lot cooler if this was set up as a
| pretrained model using RL to translate.
| rahimnathwani wrote:
| That's for --model_name, not --nli_model_name:
|
| https://github.com/lovishchopra/NL2FOL/blob/4635a81f216da2ad.
| .. nli_tokenizer =
| AutoTokenizer.from_pretrained(args.nli_model_name)
| nli_model = AutoModelForSequenceClassification.from_pretraine
| d(args.nli_model_name)
| FloorEgg wrote:
| Not familiar with FOL as a formalism, and would love to see this
| in action. I feel like it's a big part of the solution to
| propaganda.
|
| The other part seems to be values obfuscation, and I wonder if
| this would help with that too.
|
| If Joe says that nails are bad, it can mean very different things
| if Joe builds houses for a living and prefers screws, or if Joe
| is anti development and thinks everyone should live in mud huts.
|
| Propaganda will often cast a whole narrative that can be
| logically consistent, but entirely misrepresents a person or
| people's values (their motivations and the patterns that explain
| their actions), and there will be logical fallacies at the
| boundaries of the narrative.
|
| We need systems that can detect logical fallacies, as well as
| value system inconsistencies.
| andrewdb wrote:
| A prompt that I like to use for this:
|
| ---
|
| Intake the following block of text and then formulate it as a
| steelmanned deductive argument. Use the format of premises and
| conclusion. After the argument, list possible fallacies in the
| argument. DO NOT fact check - simply analyze the logic. do not
| search.
|
| After the fallacies list, show the following:
|
| 1. Evaluate Argument Strength: Assess the strength of each
| premise and the overall argument.
|
| 2. Provide Counterarguments: Suggest possible counterarguments
| to the premises and conclusion.
|
| 3. Highlight Assumptions: Identify any underlying assumptions
| that need examination.
|
| 4. Suggest Improvements: Recommend ways to strengthen the
| argument's logical structure.
|
| 5. Test with Scenarios: Apply the argument to various scenarios
| to see how it holds up.
|
| 6. Analyze Relevance: Check the relevance and connection
| between each premise and the conclusion.
|
| Format the argument in the following manner:
|
| Premise N: Premise N Text
|
| ETC
|
| Conclusion:
|
| Conclusion text
|
| [The block of text to evaluate]
| FloorEgg wrote:
| Nice prompt, I've been doing something similar but not this
| robust. I'll give this a spin.
|
| Thanks again!
| heyitsguay wrote:
| Humans aren't rational actors who get tricked into embracing
| propaganda by subtle logical fallacies. This will be of no more
| help than fact checking.
|
| It's a neat project on its own, tbc, I just have very low
| expectations of broader impact.
| FloorEgg wrote:
| I disagree with your first point. People are far more
| rational than you are making them out to be, it's just that
| they are rational within their own value system, not yours.
|
| Also today's propaganda is capable of adapting itself to each
| audience member's value system to make it more palatable, and
| then gradually nudge the audience towards the desired
| narrative/beliefs/values. The systems that distribute the
| propaganda are already analyzing people's values and using
| that information to manipulate people. I think that
| information asymmetry is part of the problem. I could be
| wrong, but I think flipping that dynamic around so the public
| can see the true values of the subjects of propaganda may
| help neutralize a lot of propaganda.
|
| As far as what impact this specific project will have, I have
| no idea. You may be right. I'm curious about its limitations
| and how it can be applied.
| kubb wrote:
| I thought so too, but recently so many people dropped or
| adapted their core beliefs to be able to support and defend
| people in power that they really love that it made me
| change my mind. Now I think that value systems are
| malleable and are formed by whatever makes us feel good.
| And the logical consistency on top is very optional.
| FloorEgg wrote:
| Or maybe these people you speak up assume that the people
| in power have values aligned with their own, and if there
| was an unbiased system that highlights value
| discrepancies using formal logic, they might not "love"
| those people as much anymore.
|
| What I assume you might be missing is that you are
| looking at the world through a different lens than these
| other people. Both you and they are consuming propaganda
| and can't detect it as propaganda because it aligns with
| your values. However it subtly nudges your values in a
| direction over time.
|
| I agree that people's values and core beliefs are
| malleable, but in the same way a tree trunk is. It may
| seem like these people have changed a lot and you
| haven't, but I think it's more likely that you've changed
| too, and that they've changed less than you think.
|
| No one is immune to propaganda, which is why anything
| that can help disarm it interests me.
| kubb wrote:
| You touched on many points, but one thing to consider:
| people seek out information that confirms their worldview
| and actively protect themselves against anything that can
| harm their feelings for their idols.
|
| John Doe isn't trained in logic and can adjust any of his
| premises if it means he can continue to admire his
| favorite celebrity. It's a combination of flawed
| reasoning and premise flexibility.
|
| Not to mention, any fact can be endlessly challenged and
| questioned even if it's agreed upon and largely
| incontestable.
| naasking wrote:
| Humans are not fully rational, but they're more rational than
| many assume. For instance, many thought the illusory truth
| effect showed that people are biased towards believing things
| they hear many times over, which is great for propagandists,
| but it turns out this is only true when they are in a "high
| quality information" environment. This is quite rational!
| They _should_ update towards believing repeated statements
| when the environment they 're in has shown itself to be
| reliable. When the environment they're in has shown itself to
| be unreliable, the illusory truth effect basically
| disappears.
|
| [1] https://x.com/ROrchinik/status/1885820697160859951
| kennysoona wrote:
| How does that explain conservatives doubling down on
| whatever they hear even if it's obviously false? I guess
| because they wrongly consider some "low quality
| information" environments "high quality information"
| environments?
| naasking wrote:
| Not everything can be reduced to this one cognitive
| phenomenon. The behaviour you describe stems from:
| confirmation bias, and the backfire effect/identity-
| protective cognition. Also this isn't exclusive to
| conservatives:
|
| Science Denial Across the Political Divide: Liberals and
| Conservatives Are Similarly Motivated to Deny Attitude-
| Inconsistent Science, https://journals.sagepub.com/doi/ab
| s/10.1177/194855061773150...
| kennysoona wrote:
| > Also this isn't exclusive to conservatives:
|
| You link aside, I think the obvious evidence is that that
| behavior is significantly more common in conservatives.
| Literally the most basic of facts get denied in bulk. I
| don't understand how you could make any argument that any
| other major political affiliation engages in the same
| behavior to a comparable extent.
| naasking wrote:
| > You link aside, I think the obvious evidence is that
| that behavior is significantly more common in
| conservatives. Literally the most basic of facts get
| denied in bulk.
|
| This is modulated by who is currently in power.
| Conservatives were worse when they lost and Biden was in
| power. Democrats are ramping up the crazy now that
| they're the underdogs.
|
| > I don't understand how you could make any argument that
| any other major political affiliation engages in the same
| behavior to a comparable extent.
|
| Go check out X and Bluesky and how many people are
| denying Trump was legitimately elected, and how they are
| convinced Musk tampered with the voting machines.
|
| As for denying basic facts, there's a whole host of basic
| scientific facts that people who lean left deny
| wholesale, eg. heritability of behaviours, personality
| and other characteristics, differences between groups,
| denying certain features of sex and the sexes, etc.
|
| I won't claim that the problem is equal on both sides,
| for many reasons I won't belabour here, but it's not
| nearly as wide a margin as you're implying. Part of the
| reason it seems so one-sided to you is my-side bias + the
| biased coverage the other side gets.
| kennysoona wrote:
| > This is modulated by who is currently in power.
| Conservatives were worse when they lost and Biden was in
| power. Democrats are ramping up the crazy now that
| they're the underdogs.
|
| That isn't remotely true. Conservatives have been
| consistently in the lead, and there are studies showing
| how much more prone to believing misinformation they are.
|
| > Go check out X and Bluesky and how many people are
| denying Trump was legitimately elected, and how they are
| convinced Musk tampered with the voting machines.
|
| There's at least reasoned arguments for that. That isn't
| the same thing as rejecting useing masks during a
| pandemic.
|
| > it's not nearly as wide a margin as you're implying.
|
| It really is, but we clearly disagree.
|
| > Part of the reason it seems so one-sided to you is my-
| side bias + the biased coverage the other side gets.
|
| You shouldn't make assumptions about how or where I get
| my news. I don't think coverage bias applies at all in
| influencing my conclusion based on how I get my news.
| aeturnum wrote:
| I think you're envisioning this in a pessimistic way.
|
| I totally agree that the end conclusion "this statement is
| fallacious" is pretty useless. But I assume that a working
| process would also yield the chain of judgements (A is right,
| B is right, C is wrong, etc). I think that would be VERY
| useful.
|
| People who become captured by propaganda and lies generally
| are not sold on 100% of the propaganda. There are certain
| elements they care more about and others they can ignore. A
| way to deprogram people through conversation is to just ask
| them to explain things about their views and ask them to
| reconcile them with reality. The reconciliation is painful
| for them and that pain keeps people "in" irrational beliefs -
| but it's also how people find their way out. Once they no
| longer associate themselves with the conspiracy, they can
| discard beliefs associated with it...provided they can think
| through them.
|
| I think being able to automatically decompose a fact check
| into the elements of what "is true" and "is false" in a
| statement would be HUGE. An essential tool in helping people
| escape from information swamps.
| naasking wrote:
| This won't "solve" propaganda or misinfo IMO. Checking logical
| consistency and eliminating fallacies still wouldn't address
| the selective presentation or omission of facts, for instance,
| and the notion that it could avoid misrepresenting a person or
| their values assumes that someone has already accurately and
| fully captured a detailed description of a person's values. But
| that's the whole problem!
|
| This is just the formal specification problem all over again.
| Verifying software against a spec is good and useful, but
| verification doesn't tell you whether the spec itself correctly
| captured the desired objective, it can only tell you whether
| the spec is logically consistent and that you implemented it
| faithfully.
| janalsncm wrote:
| Maybe. One problem we have now is that fact checking is a lot
| more expensive than bullshitting. If we had a program that
| could bring things closer to parity it would be nice.
|
| But also, a lot of propaganda isn't _false_ per se but simply
| blown out of proportion, or underproportioned in cases of
| inconvenient truths. The truth is a distribution of events, and
| editors continuously choose how to skew that distribution.
|
| (One of my very interesting possessions is an old Chinese
| state-owned newspaper. As far as I could tell, their main tool
| wasn't lying, but simply omission.)
|
| For example, if you wanted to push a narrative that e.g. pit
| bulls are the most dangerous problem in America, you would just
| post a nonstop stream of pit bull attack videos. It taps into
| cognitive biases people have which aren't propositional logic
| statements.
|
| More broadly, the world is stochastic, at least in the way we
| experience it. So our brains have to make sense of that, which
| is an opportunity for narratives to creep in.
| drdeca wrote:
| You mention the world being (at least subjectively)
| stochastic. This brings to mind the idea that a model of
| probability rather than just logic, might be more beneficial?
|
| The example you gave of focusing excessively on some topic in
| order to make it seem like a bigger deal...
|
| hm, is there a way we could formalize such things in a way
| like how formal fallacies are formalized?
|
| It seems more difficult than classifying common formal
| fallacies.
|
| But, might it be possible?
| janalsncm wrote:
| The problem is that "news" is not a random sampling of all
| events. It is biased by the very fact that someone has to
| decide the event is notable.
|
| And even if you were to witness a random sampling of all
| events via some kind of clockwork orange mechanism, your
| brain has saliency biases as well.
|
| You might find the wiki page on cognitive bias interesting
| https://en.m.wikipedia.org/wiki/Cognitive_bias
| FloorEgg wrote:
| So maybe the solution is to have these FOL capabilities close
| to the user and far from the information source.
|
| FOL values analysis of information streams, that manifest as
| user interface for configuring the algorithms that decide
| what information is surfaced to you in media.
|
| This is why I said this sort of thing might be part of a
| solution. The whole solution would involve other significant
| parts.
| mirekrusin wrote:
| You can require that factual statements require source
| reference.
|
| Statement that "pit bulls are the most dangerous problem in
| America" requires source data (ie. cause of death or serious
| injuries in 2024 in USA).
|
| Publications can be signed by authorities (ie. university or
| government body).
|
| IMHO sooner or later we will (have to) end up with system
| like that.
|
| Every information will be signed and level of trust will be
| automatically established based on your preference who you
| trust.
| drdeca wrote:
| I don't think it would take you long to learn FOL, and I think
| it is a good formalism to have some familiarity with.
|
| It's pretty much the default modern formulation of general-
| purpose formal logic.
| RGamma wrote:
| It would work as well the internet bringing us more
| enlightenment. Besides, points of contention tend to form
| around ethics, whose axioms are unprovable if non-cognitivism
| is true (and we have no reason to believe it isn't).
| jfengel wrote:
| You know First Order Logic. It's just ordinary logic; it's the
| default thing people think of when they say "logic".
|
| But it's also not very useful for human reasoning. It's good
| for math and logic puzzles and bad at anything else. It's bad
| at time, at belief, at negation. None of those things act like
| you expect them to.
| thomastjeffery wrote:
| The problem isn't values obfuscation. The problem is that many
| people, especially conservatives, _do not care about values_.
| Instead, they care about _virtues_.
|
| People who approach politics from a virtue ethics perspective
| are vulnerable to propaganda because logic and value have no
| bearing whatsoever on their decision to accept or reject a
| narrative.
|
| You can't think critically _for_ someone else. They must do it
| on their own.
| ColinWright wrote:
| It was Gottfried Leibniz who envisaged the end of philosophic
| disputes, replacing argument with calculation.
|
| _" if controversies were to arise, there would be no more need
| of disputation between two philosophers than between two
| calculators. For it would suffice for them to take their pencils
| in their hands and to sit down at the abacus, and say to each
| other (and if they so wish also to a friend called to help): Let
| us calculate."_
| soulofmischief wrote:
| If only. Ethics are reached via consensus. Two calculators can
| indeed produce different results if the axioms supporting them
| differ.
|
| And good luck calculating some of these axioms, such as "Why is
| it my duty not to kill someone?" You could argue, "Well in the
| end, a society enabling such behavior at scale would be no
| society at all," to which one might reply, "I have no interest
| in letting others do as I do.", and you can't calculate away
| violent sociopaths. The rest of us derive our principles from
| functioning mammalian emotional circuits, but at some level we
| rest our case on subjective axioms.
| glenstein wrote:
| >Ethics are reached via consensus
|
| This is probably too big a topic for a whole side-branch on
| this, but modern meta-ethics teaches a range of possible
| approaches. Some notions of ethics are relativist, and are
| about the fact that moral norms are produced by some given
| society. But under some constructions that's just a
| procedural truism rather than a position on the content or
| the nature of morality itself.
|
| Then you have moral realism, a perfectly respected position,
| which can encompass things like utilitariansim and other
| ism's. And this might seem silly derail, and I'm trying not
| to, but this is important at the end of the day, because
| "ethics is reached via consensus" can mean a lot of things
| that cash out with completely different practical
| implications. It's the difference between, for instance,
| deciding we need to be consensus oriented and vote, or be
| research oriented and concerned with deepening our scientific
| understanding of things like insect consciousness and whether
| the physical effects of sleep deprivation fall under the
| traditional definition of torture.
|
| >And good luck calculating some of these axioms
|
| Not wrong, they can easily get computationally intractable.
| So I think one has to account to some degree for uncertainty.
| Here again, I worry that the intended upshot is supposed to
| be that we simply give up or treat the project of moral
| understanding like a cosmically impossible non-starter. I
| like to think there's a middle ground between where we
| presently stand and the hypothetical future where we've got
| perfect knowledge.
| kennysoona wrote:
| Those axioms can still be evaluated, quantified and compared,
| and eventually calculated.
| yifanl wrote:
| Based on what criteria? A set of meta-axioms?
| kennysoona wrote:
| No, no meta axioms. Just by quantifying whatever we can
| as much as possible to be as objective as possible.
| Mr-Frog wrote:
| My gut still tells me this relies on a human-defined
| optimization metric.
| soulofmischief wrote:
| "as possible" means an incomplete system that still
| relies on assumed axioms.
| kennysoona wrote:
| Maybe. I don't think so though. I think everything can be
| quantified and qualified.
| lo_zamoyski wrote:
| > Ethics are reached via consensus.
|
| Absolutely not! This is cultural relativism, and frankly, it
| would be circular: how exactly are we converging on a
| consensus if not from some preexisting sense of the good?
|
| The only defensible objective basis for the good is the
| nature of a thing and what actualizes the potentials
| determined by that nature, thus actualizing the thing as the
| kind of thing it is. Morality, only possible for things that
| have the capacity to comprehend their options for action
| (intellect) and choose freely among them (will) on the basis
| of that understanding, therefore concerns the question of
| whether an act performed by a thing furthers or frustrates
| the actualization of that thing.
|
| By cutting off my arm for no proportionate reason, I do an
| immoral thing, because it is my nature to have that arm, but
| if I have gangrene in that arm that threatens my life, then
| removing the gangrene with the undesirable side effect of
| losing an arm is morally justifiable, even if the loss of the
| arm is not good per se.
|
| Murdering a human being is gravely immoral, because it
| directly contradicts my nature as a social human being in a
| very profound and profoundly self-destructive way. However,
| killing a would-be murderer in defense of my life or that of
| another is a morally very good deed; it is in accord with my
| social nature, and indeed can be said to actualize it more
| fully in some respect.
|
| > The rest of us derive our principles from functioning
| mammalian emotional circuits
|
| Please refrain from making such silly pseudoscientific and
| pseudophilosophical statements.
|
| That being said, calculation is insufficient, because such
| calculation is formal: it explicitly excludes the conceptual
| content of propositions. But concepts are the material
| "carriers" of comprehension of what things are. We can also
| analyze concepts. Now, we can say that we can calculate a
| formal deduction according to formal rules, but we cannot
| calculate a concept or its analytical products. This is the
| produce of abstraction from concreta. Formal systems abstract
| from these. They are blind to conceptual content, on purpose.
| And having used a formalism to derive a conclusion, we must
| interpret the result, that is, we must reassign concepts to
| symbols that stand in for them. So formal systems are useful
| tools, but they are tools.
| kazinator wrote:
| "reached via" is not the same thing as "derived from".
| franktankbank wrote:
| Although, demanding even a shred of self-consistency goes a
| long way in short circuiting bad argumentation.
| dmos62 wrote:
| I have a pet theory that most inefficiency is about self-
| consistency (or lack thereof), whether that's in human-human
| or human-machine communications (e.g. program code).
| dmos62 wrote:
| I wonder if anyone else thought that that's how most of the
| world worked when they were a kid. I thought that most people
| would reason through everything, and if they couldn't, they
| would take it home as sort of homework and finish it there.
| bloomingkales wrote:
| Well, we can have AI do what we do but it will never be tied to
| an emotion. You can feel a lot just adding 2+2 (maybe someone
| held a gun to your head once). What does philosophy say about
| philosophy without emotion? What use is it to us without our
| human context? The philosophy of a tiger is not relevant to me
| mostly because I don't feel most of the things a tiger feels.
| languagehacker wrote:
| It sounds like the data set they use is designed to teach what
| logical fallacies are, which makes sense that it would do fine
| with it. I doubt this would do well against real-world language
| with things like structural ambiguity, anaphoric resolution, and
| dubious intent.
| MortyWaves wrote:
| Trolls that knowingly engage in bad arguments with flawed logic
| are going to be in shambles.
| mike_hearn wrote:
| Love the idea in theory and would like such a tool to exist, but
| the use cases they present aren't convincing. This would be
| useful in much more specific cases like drafting contracts, laws
| or technical documentation: places where unusually precise
| language without corner cases is mutually desired by everyone,
| and the set of fallacies that occur is small and specific.
|
| This paper doesn't target such use cases. Instead it's trying to
| tackle "pop misinformation" type claims, mostly related to
| climate change. Unfortunately the Logic and LogicClimate datasets
| that the paper are using as a benchmark have serious problems
| that should disqualify them from being considered a benchmark. If
| we check the paper that introduced them, Jin et al open by
| asserting that _" She is the best because she is better than
| anyone else"_ is an example of circular reasoning. It's actually
| a tautology. Then they try again with _" Global warming doesn't
| exist because the earth is not getting warmer"_ which is also not
| circular reasoning, it's another tautological restatement (you
| may say it's false, but disagreement over facts isn't a
| disagreement over logic - if either clause is true so is the
| other). Circular reasoning often involves a mis-definition and
| would be something like this real-world example from a few years
| ago:
|
| 1. A positive test is means you have COVID.
|
| 2. Having COVID is defined as having a positive test.
|
| Their second example is _" Extreme weather-related deaths in the
| U.S. have decreased by more than 98% over the last 100 years ...
| Global warming saves lives"_ which they classed as "false
| causality" (they mean non-sequitur). My experience has been that
| climate skeptics are surprisingly logical so this would be an odd
| statement for them to make, and indeed if we check the original
| Washington Times op-ed then we find Jin et al are engaging in
| malicious quoting. It actually says:
|
| _> "Contrary to sensational media reports, extreme weather-
| related deaths in the U.S. have decreased more than 98% over the
| last 100 years. Twenty times as many people die from cold as from
| heat, according to a worldwide review of 74 million temperature-
| related deaths by Dr. Antonio Gasparrini and a team of
| physicians. Global warming saves lives."_
|
| The saves lives claim is based on cold being more dangerous than
| heat. Warmer weather = fewer deaths from cold isn't a logical
| fallacy, which is why they had to delete that part to make their
| example. It might sound like a weird or disingenuous argument to
| you, but it's logical in the sense that an SMT solver would
| approve of it. If you disagree it's probably due to prior beliefs
| e.g. that perhaps extreme weather has increased even as society
| got orders of magnitude better at reducing the impacts, or
| perhaps the positive effects of warmer air on the elderly are
| offset by other effects of climate change, or that the future
| will be different to the past due to compounding effects. Such
| rebuttals aren't identifications of a logical fallacy though,
| just of different priors that could maybe be addressed with
| additional rounds of debate.
| _cs2017_ wrote:
| Out of curiosity, what fraction of readers do you think would
| understand the points you're making? And what fraction of
| readers do you think would blame you for taking the wrong side
| of some debate?
| mike_hearn wrote:
| Readers here? No idea but I'd love to know. What's your
| estimate?
| analog31 wrote:
| Doesn't Goedel's Theorem forbid building a logic checker?
| drdeca wrote:
| No.
|
| Godel's theorem forbids something that in general tells you
| whether a statement is true or not. (As in, a method which
| would work for every possible statement within a system.) It
| certainly doesn't preclude something that checks if a proof is
| correct, any more than it precludes checking whether some
| calculation is done correctly (I.e. it doesn't preclude it at
| all) .
|
| It says that there are statements which the proof system
| doesn't have a proof of the statement being true nor of it
| being false. It doesn't mean you can't have a proof system.
| bubblyworld wrote:
| Not Godel's theorem, but inference for first-order logic is
| undecidable in general for other reasons. You can still get
| pretty far with heuristics though. Don't let perfect be the
| enemy of good =P
| nico wrote:
| Is this just another form of the same concept behind smart
| contracts?
| zozbot234 wrote:
| I'm pretty sure that the semantics of natural language are a lot
| more complex than can be accounted for by these seemingly very
| ad-hoc translations into comparatively straightforward FOL
| formulas, as are given in this paper. A common approach for the
| understanding of NL semantics from a strictly formal POV is
| Montague semantics https://en.wikipedia.org/wiki/Montague_grammar
| https://plato.stanford.edu/entries/montague-semantics/ - even a
| cursory look at these references is enough to clarify the level
| of complexity that's involved. Very loosely speaking one
| generally has to work with multiple "modalities" at the same time
| each of which, when understood from the POV of ordinary FOL,
| introduces its own separate notion of abstract "possible worlds"
| (representing, e.g. an agent's set of beliefs) and ways in which
| these "worlds" can relate to one another. More complex cases will
| usually degenerate in some sort of very generic "game semantics"
| https://en.wikipedia.org/wiki/Game_semantics
| https://plato.stanford.edu/entries/logic-games/ where any given
| use of natural language is merely seen as a "game" (in the
| abstract strategic, game-theoretical sense) with its own set of
| possibly very ad-hoc 'rules'. The philosopher Ludwig Wittgenstein
| https://en.wikipedia.org/wiki/Ludwig_Wittgenstein
| https://plato.stanford.edu/entries/wittgenstein/ gave quite a
| good description of both of these approaches (from a very naive
| approach based on a supposedly straightforward translation to
| some kind of abstract logic, to a far richer one based on notions
| of strategies and games) to a "formal" understanding of natural
| language, throughout his extensive philosophical inquiry.
|
| Which is to say, I'm not sure how this paper's results are
| generally expected to be all that useful in practice.
| lapcat wrote:
| See also for example V.H. Dudman on the interpretation of "If"
| sentences:
| https://www.scribd.com/document/478756656/Dudman-1984-Condit...
| cs702 wrote:
| It could be useful for domains in which all or at least many
| problems are solvable (i.e., they can be stated and satisfied)
| with first-order logic.
|
| It could also be useful as a lower-level component of general-
| purpose systems that internally rely on chains of thought
| computed by sub-component LLMs.
| xhevahir wrote:
| It wouldn't be useful if, as the parent comment is saying, it
| won't do a decent job of translating natural language.
| cs702 wrote:
| Ah, got it. Thanks!
| tgv wrote:
| I've worked on classical NLP models for quite some time, and
| this indeed looks way too simple to be of any practical use. If
| you mention Montague, I'm going to refer you to "Pedro owns a
| donkey," the poster kid sentence for Discourse Representation
| Theory. That's 1980s work, and for simple sentences it's
| already complicated beyond what the OP article suggests, and
| fails on anything remotely complex. I think it goes 2nd order
| the moment a complement is introduced (I think that ...).
|
| And even if you can translate a sentence into a predicate, you
| haven't begun understanding what lies behind all those
| predicates. E.g., "Zelensky is ready to work under Trump's
| 'strong leadership' after 'regrettable' showdown." What good
| does it do to have that in FOP?
|
| [1]
| https://plato.stanford.edu/archIves/sum2011/entries/discours...
| zozbot234 wrote:
| It looks like classic models of NLP semantics mostly punt on
| the "logical" point of view precisely due to these
| difficulties, and focus mostly on the more surface level
| problem of describing how each word of the source text
| correlates with a deeper description of the "meaning" of the
| text as a whole. So it is simply assumed that the meaning of
| the text as a whole must be derived compositionally from the
| meaning of each part (usually described by a somewhat ad-hoc
| "frame" structure), but exactly what that entails in a
| "logical" sense is left unspecified. UMR (Universal Meaning
| Representations) seems to be a typical example of such a
| system https://github.com/umr4nlp/umr-
| guidelines/blob/master/guidel... The expected use case seems
| to be something like building a common intermediate language
| for an automated translation system; individual meaning
| elements can then be "mapped" in a useful way, even across
| different languages, but there's not much interest apparently
| in "inferring" further knowledge from what one already has,
| or even on verifying that any given inference is valid (as
| proposed by OP).
| WaxProlix wrote:
| Even beyond that you have a ton of pragmatics post-grice to
| deal with. Computing implicatures is complex and requires a
| _lot_ of knowledge about context etc. The truth value of a
| statement and the 'truth value' of a speech act are pretty
| different things - not sure it's really feasible to convert
| between them.
| a-dub wrote:
| would be interesting if they had adversarial/null llms
| attempting the noisy nlp reductions as well. then one could
| make arguments about the sturdiness of the noisy bit.
| thomastjeffery wrote:
| Text that is written in Natural Language is _open to
| interpretation_. There are many formal statements that can be
| said to interpret a given Natural Language text. Can we
| determine which formal representation is correct? What about
| most useful?
|
| The obvious answer to these questions is, "no". There is no
| such thing as a _conclusive interpretation_. If there was, then
| Natural Language wouldn 't be ambiguous in the first place!
|
| So we're all doomed to constantly misinterpret each other
| forever, right? No? We humans use Natural Language all the
| time, and usually figure out what the other person actually
| means!? How do we do it? Are we all just really good at
| guessing?
|
| No, we have something better: context.
|
| Context exists both in and around Natural Language text.
| Context determines _which_ formal meaning is used to interpret
| the text. If we don 't know which context is appropriate, there
| may be clues in the text itself that help us construct one that
| is useful or correct.
|
| ---
|
| I've been trying to work out an approach to language processing
| that interprets text into logical formalisms (arbitrary
| meaning). I call them "Stories". A Story is an arbitrary
| interpretation of text. A Story is never conclusive: instead it
| is used as arbitrary context to interpret the next text. I call
| this process "Backstory".
|
| We could even do the process backwards, and "write" an
| arbitrary formalism (meaning) in the same language/style/voice
| as a previously interpreted Story.
|
| Given enough example instances of Story, we should be able to
| read and write to each other through explicitly shared context.
| I call this process "Empathizing". I call my idea the Story
| Empathizer.
|
| I'm definitely out of my depth when it comes to the details,
| though...
| pixelpoet wrote:
| Oh man, where was this back in the 90s arguing with proto-trolls
| on IRC and usenet who shamelessly moved goalposts, stawmanned,
| appealed to authority, resorted to ad hominem, ...
|
| Imagine if you could click on a stupid internet discussion thread
| and make it give you a Lean proof of each argument where possible
| :D This thing would be hated even more than, say, vaccines, by
| the same sorts of people who deliberately choose to not
| understand things.
| shortrounddev2 wrote:
| I believe this is something Immanuel Kant tried to do in the 18th
| century
| qgin wrote:
| I don't know how much potential this has to solve propaganda /
| bad faith arguments because you can just say "that logic program
| is biased" and handwave the entire thing away.
|
| But you could imagine a role for this in arbitration or legal
| settings.
| biggoodwolf wrote:
| 0 in all three cases
| giardini wrote:
| Prolog has always had DCGs (Definite Clause Grammars) that allow
| you to write rules that resemble natural language grammar
| structures to parse and generate English sentences:
|
| https://www.metalevel.at/prolog/dcg
| svnt wrote:
| This is already something that e.g. Claude 3.7 Sonnet appears to
| be able to do very well, with the added benefit of explaining why
| if you let it -- what is the benefit of this model?:
|
| > "Sometimes flu vaccines don't work; therefore vaccines are
| useless." - Hasty generalization
|
| > "Every time I wash my car, it rains. Me washing my car has a
| definite effect on the weather." - Post hoc, ergo propter hoc
|
| > "Everyone should like coffee: 95% of teachers do!" - Appeal to
| popularity and hasty generalization
|
| > "I don't want to give up my car, so I don't think I can support
| fighting climate change." - False dilemma
| mannykannot wrote:
| It would take more subtle examples, embedded within what is
| mostly fallacy-free text, to evaluate the absolute and relative
| utilities of the two approaches to the problem - or, to put it
| another way, we should not hastily generalize from their
| performance on a few straightforwardly fallacious sentences.
| EigenLord wrote:
| This is very cool and definitely a step in the right direction,
| however, the question remains where exactly this formalizing
| module should be placed in the stack. As an external api, it's
| clear that the model is not "thinking" in these logical terms, it
| just provides a translation step. I'd argue it would be better
| placed during inference test-time compute (as seen in these so-
| called reasoning models). Better yet, this formalizing step would
| happen at a lower level entirely, internal to the model, but that
| would probably require totally new architectures.
| Geee wrote:
| Yes, this is exactly what I've been dreaming about. It might
| finally be possible to beat the bullshit asymmetry law, i.e.
| Brandolini's law: "The amount of energy needed to refute bullshit
| is an order of magnitude bigger than to produce it."
|
| If LLMs can debunk bullshit as easily as it's generated, the
| world will instantly turn into a better place.
|
| Bad ideas which sound good are the root of all evil.
| tiberius_p wrote:
| First order logic can only detect formal logic fallacies.
| Informal logic fallacies like ad hominem, strawman, red herring,
| etc. are cast in language. They can't me defined and resolved
| mathematically. The model should be fine tuned with examples of
| these informal fallacies and counter-arguments to them. Even so
| it won't be able to detect them in all cases, but it will at
| least have some knowledge about them and how to reply to them.
| This knowledge could be further be refined with in context
| learning and other prompt engineering strategies.
| jfengel wrote:
| I would expect a true logical fallacy detector to take any
| natural text and spit out "unsupported assumption, unsupported
| assumption" over and over and over.
| sergix wrote:
| https://xkcd.com/2610/
___________________________________________________________________
(page generated 2025-03-04 23:00 UTC)