Author Topic: Is language really rule-based?/ "Do you miss not having a job?"(2014)*revisited*  (Read 2493 times)

Offline waive15

  • Linguist
  • ***
  • Posts: 64
Hi,

The discussion warmed my heart. 

Daniel(2014) /Is language really rule-based?/:
"In a very specific sense, I'm wondering if languages follow "rules". Is language strictly rule-based?"

waive15(2020):
Yes, "it" is.

jkpate(2014) /"Do you miss not having a job?"/:
"I don't want to derail the thread into the philosophy of logic/mathematics, but it's worth remembering that, formally speaking, there is no monolithic notion of "logical." A logic is just a mathematical object, and statements that may be a theorem of one logic will not be of another. In particular, not( not( P ) ) does not entail P in Intuitionist logic. So blanket statements that "X is logical" or "X is not logical" are meaningless; you have to define the logic you are using. Maybe language really is logical, once we get the right system of logic."

Immanuel Kant:
Every natural science contains as much truth as much mathematics it contains.


 
« Last Edit: May 16, 2020, 05:45:42 PM by waive15 »

Offline Daniel

  • Administrator
  • Experienced Linguist
  • *****
  • Posts: 2036
  • Country: us
    • English
You're quoting me out of context:
http://linguistforum.com/outside-of-the-box/is-language-really-rule-based-(details-inside)/
And also discussion here: http://linguistforum.com/english/'do-you-miss-not-having-a-job'/

What I said was that, as you did quote, I was wondering about rules in a specific sense. I went on to explain several paragraphs what I meant by that. Here's one important part:
Quote
In other words, are there exceptions to rules? If so, they aren't really rules. They're just patterns. And patterns exist everywhere, in all behaviors. So language isn't really logical then.
Regardless, my position was for the sake of discussion, not a specific mathematical proposal.

One of my professors cited Pānini as having proposed that "there are no exceptions, only more specific rules", although I've never been able to track down the exact source for that phrasing in the quotation. Regardless, that's one way to look at it. Generalizations apply across all contexts, except where a narrower rule applies more specifically instead. That's one way of including exceptions.

Here's where things get puzzling:
Quote
It's not a question of whether, for example, a language does or does not have "double negation" which is said (by prescriptivists) to be objectively illogical. It's a question of whether these forms follow rules within the languages themselves.

So the question I was wondering about, and it's a tough one, is how exceptions work. Do rules apply generally? Do they apply to only a subset of constructions to which they could in theory apply? Do they apply but allow other more specific rules to override them as exceptions? Do they always apply but let speakers ignore them via pragmatics?

In the end, What is a rule? And how can't it be broken?

I take a rule as something more absolute than an observable pattern. What does it mean to say our theory involves rules, not just patterns?

Anyway, since then I've written an article about some of this, including some of the data. I didn't really address the question about "rules" except to point out that these types of examples are difficult to deal with from a traditional theoretical perspective:
http://hdl.handle.net/2142/102155

To be clear, I don't really disagree with you (nor jkpate!). But you have not yet answered what to do about the weird cases with "exceptions". If you want to maintain that there are "rules", then how do you formulate them? It's not easy.
Welcome to Linguist Forum! If you have any questions, please ask.

Offline Rock100

  • Linguist
  • ***
  • Posts: 68
> Do they apply but allow other more specific rules to override them as exceptions?
I believe this is the best option. That is how context-sensitive grammars work. This is the highest reasonable structure for describing (formal) languages nowadays I believe. The next level is when (almost??? not sure) everything is possible and there is nothing to study.
> In the end, What is a rule?
One might think of the rules as about a set of entities (there are 5 of something like that) that allow to generate the valid sentences of the language or to solve the reverse task – to determine if a sentence belongs to the language. One applies the rules of the grammar to the sentence recursively and if there is a sequence of the rules that is able to “process” the whole sentence than the sentence belongs to the language.
> And how can't it be broken?
I believe the rules either apply or not. If one generates the sentences of a language with the rules than the sentence belongs to the language by definition. If one has a sentence and there is no a sequence of the rules of the language that may be applied to it than the sentence does not belong to the language. If one has a sentence, he knows that it belongs to the language but there are no rules in the grammar that apply to it that one needs to add a new rule/rules to the grammar. And vice versa if grammar can produce sentences that do not belong to the language one needs to correct the rules of the grammar. So, I believe it is easier to think the rules are not broken but they do not apply to a sentence.
But I doubt one may create a finite grammar even for a subset of a real language that takes into account the semantics.
Such (algebraic) approach may probably help a little in understanding some aspects of languages but it looks like it is nearly totally useless in practice – the amount of entities that make up the grammar of a real language appears to be enormous. I wish I could read more about it in a popular language somewhere.
Just my rather dilettantish point of view, I am not a professional or even an English speaker.

Offline Daniel

  • Administrator
  • Experienced Linguist
  • *****
  • Posts: 2036
  • Country: us
    • English
Quote
"Rules" are just as imaginary as "observable patterns".
Then what are they? They must reflect some kind of psychological reality, and they must also be helpful in predicting linguistic behavior. Different theories approach "rules" (and "patterns") differently. I'm asking about nuance, not whether we need something, and I don't have a definitive answer.

Quote
Scientific method is PRAGMATIC.
Pragmatics, in Linguistics, is the study of language usage in context. It's not related to the sense of "be pragmatic in your decisions". Thus Pragmatics (usage in context) can add or change meanings in an utterance from what they literally mean via Semantics.

As for the rest, we seem to have some different perspectives about language, but regardless I think the "right" theory is the one that properly unifies various perspectives and answers a variety of different questions, not just the ones that one theory-maker thinks of at the beginning. In Linguistics, it really is hard to figure out what that overall-theory should look like!

Similar to what you've said jokingly above, it seems that the biggest weakness in any theory (like popular theories of syntax, or your theory that language is just mathematics) is the data. How to account for the data... :)
Welcome to Linguist Forum! If you have any questions, please ask.

Offline waive15

  • Linguist
  • ***
  • Posts: 64
Hi, Daniel,

I agree.

"No need to give up a good theory just because it isn't true." :)


P.S.
The last post I wrote was a babble. It is gone.

Offline Daniel

  • Administrator
  • Experienced Linguist
  • *****
  • Posts: 2036
  • Country: us
    • English
Rock100,
Quote
The next level is when (almost??? not sure) everything is possible and there is nothing to study.
That's the kind of question that interests me: can we really delimit natural language to identify specific things that are impossible? I have yet to see conclusive evidence of that, but I enjoy looking for it.

Quote
One might think of the rules as about a set of entities (there are 5 of something like that) that allow to generate the valid sentences of the language or to solve the reverse task – to determine if a sentence belongs to the language. One applies the rules of the grammar to the sentence recursively and if there is a sequence of the rules that is able to “process” the whole sentence than the sentence belongs to the language.
Sure, and that's what we do with the toy grammars in a class. But they're never complete.
Quote
I believe the rules either apply or not. If one generates the sentences of a language with the rules than the sentence belongs to the language by definition. If one has a sentence and there is no a sequence of the rules of the language that may be applied to it than the sentence does not belong to the language. If one has a sentence, he knows that it belongs to the language but there are no rules in the grammar that apply to it that one needs to add a new rule/rules to the grammar. And vice versa if grammar can produce sentences that do not belong to the language one needs to correct the rules of the grammar. So, I believe it is easier to think the rules are not broken but they do not apply to a sentence.
But I doubt one may create a finite grammar even for a subset of a real language that takes into account the semantics.
Such (algebraic) approach may probably help a little in understanding some aspects of languages but it looks like it is nearly totally useless in practice – the amount of entities that make up the grammar of a real language appears to be enormous. I wish I could read more about it in a popular language somewhere.
What you're saying makes sense abstractly, but the same problem from above applies: the data! You need to show that this approach really works for all data. And that's no easy task.

Quote
Just my rather dilettantish point of view, I am not a professional...
I do this full time, and all I can say with certainty is that I'm not sure. It's complicated. That's why I asked the original question. Perhaps better phrasing would simply be: How do rules apply?
Welcome to Linguist Forum! If you have any questions, please ask.

Offline Rock100

  • Linguist
  • ***
  • Posts: 68
> What you're saying makes sense abstractly, but the same problem
> from above applies: the data! You need to show that this approach
> really works for all data. And that's no easy task.
Yeah, this is true. But to be honest it is not just uneasy task it is impossible in the present technologies task. Here go some elaborations of enormous notion above. To be able to deal with the entities that describe a real language grammar one would need to store the data (you are talking about) in a memory. Let us suppose we have the most efficient memory ever – one atom per bit of the information. I believe enormous in this case will be one third of the atoms that make up the Earth. So, the top-down approach I consider will never work (at least in the nearest future) – nobody allows me to use such resources. So I believe this “abstract” view may help to understand something but is totally useless in solving real practical tasks. I also believe that the linguistics as the science uses more pragmatic (with it’s drawbacks, of course) bottom-up approach. You, linguists, do have some working pieces of the system that really work (at least they do pay your bills) but you will never get the working system as the whole (due to the same reason of the lack of resources – the approaches are equivalent). I am not a professional, I am just a kind of a sympathetic one. I do like the subject but I begin to start to understand that from the practical point of view we are a kind of out of the main stream. It looks like for practical stuff the artificial intelligence (AI) gets much better results. The probability theory wins another battlefield. It looks like it is fun to know some fundamentals but if one needs to sell a real product he has to use different approaches that are probably not that exact but require fewer resources. There is the thing I know for sure: in the formal languages world – compilers (the translating grammars from one very formal/simple representation/language to another) – people get rid of grammar helping tools in favor of manual coding. So, the theory appeared to be a kind of disappointing. But I have not heard about using an AI as a translating grammar so far too.

> How do rules apply?
If one has an unambiguous grammar (the set of the rules) then he tries the rules one by one in sequence until one of them works. This is for parsing. I have never used formal grammars for generating the sentences (parsing and translation only) so I will not invent something but it shall be easy too. For example, here is a variant of the English grammar (very simplified and probably incomplete):
<S>: <Exclamation><subject>
<S>:<adverb><aux verb><subject><predicate>
<S>:<subject><predicate><adjunct>
<subject>: <pronoun>|<article><noun>
<adjunct>: <noun>
<predicate>: <verb>|<compound verb>
<verb>: go|goes|drink|kill
<noun>: elephant|home
<article>: <empty>|a|an|the
If we try the sentence (S) “an elephant goes home” than the third starting rule (S) applies. The algorithm is the following: you take s1, find <exclamation>, check the rule(s) for exclamations, etc. For the third rule <subject>-><pronoun> -- fails, go to the alternative (|) <subject><article> --> an, Ok, so far <subject><article><noun> an elephant – perfect! We have <article><noune><predicate> --> find verb goes, and so on. This is how I speak English indeed.

Offline Daniel

  • Administrator
  • Experienced Linguist
  • *****
  • Posts: 2036
  • Country: us
    • English
Quote
Yeah, this is true. But to be honest it is not just uneasy task it is impossible in the present technologies task. Here go some elaborations of enormous notion above. To be able to deal with the entities that describe a real language grammar one would need to store the data (you are talking about) in a memory. Let us suppose we have the most efficient memory ever – one atom per bit of the information. I believe enormous in this case will be one third of the atoms that make up the Earth. So, the top-down approach I consider will never work (at least in the nearest future) – nobody allows me to use such resources. So I believe this “abstract” view may help to understand something but is totally useless in solving real practical tasks. I also believe that the linguistics as the science uses more pragmatic (with it’s drawbacks, of course) bottom-up approach. You, linguists, do have some working pieces of the system that really work (at least they do pay your bills) but you will never get the working system as the whole (due to the same reason of the lack of resources – the approaches are equivalent). I am not a professional, I am just a kind of a sympathetic one. I do like the subject but I begin to start to understand that from the practical point of view we are a kind of out of the main stream. It looks like for practical stuff the artificial intelligence (AI) gets much better results. The probability theory wins another battlefield. It looks like it is fun to know some fundamentals but if one needs to sell a real product he has to use different approaches that are probably not that exact but require fewer resources. There is the thing I know for sure: in the formal languages world – compilers (the translating grammars from one very formal/simple representation/language to another) – people get rid of grammar helping tools in favor of manual coding. So, the theory appeared to be a kind of disappointing. But I have not heard about using an AI as a translating grammar so far too.
Just two replies:
1. The task isn't impossible. Human brains do it with ease. If it was hard to understand astronomical scales, I'd understand. Hard to understand black holes and quasars. But actually it seems like we understand them better than we understand ourselves. We don't know how language works, but we keep using it every day, easily and happily.
2. The irony is that while appealing to a mathematical solution you're actually describing ideas more along the lines of Cognitive Linguistics with a connectionist approach to the data, rather than abstracted formal rules. Chomsky's revolution was about taking a formal, mathematical approach to understanding the rules of language. You're saying we should approach it mathematically, yet in the end you're drifting toward the now competing theory where language is treated as an emergent phenomenon of connectionism in the brain, not a listable set of specific rules!

Quote
If one has an unambiguous grammar (the set of the rules) then he tries the rules one by one in sequence until one of them works. This is for parsing. I have never used formal grammars for generating the sentences (parsing and translation only) so I will not invent something but it shall be easy too. For example, here is a variant of the English grammar (very simplified and probably incomplete):
Sure, but that's the easy part! The hard part is getting it to work with the weird data. I have my students write mini computational grammars that work like that, and they work well, but they run into problems when you try to explain nuances or "exceptions". Claiming that rules apply logically but only treating the most general cases is not a scientific argument at all. That was my point to begin with.
Welcome to Linguist Forum! If you have any questions, please ask.

Offline Rock100

  • Linguist
  • ***
  • Posts: 68
> The task isn't impossible. Human brains do it with ease.
Absolutely right. But they do it a different way. We have tried to solve the task analytically and failed. The formal grammars appeared to be of very limited use. This was my point.

> The irony is that while appealing to a mathematical solution you're
> actually describing ideas more along the lines of Cognitive Linguistics with
> a connectionist approach to the data, rather than abstracted formal rules.
> Chomsky's revolution was about taking a formal, mathematical approach to
> understanding the rules of language. You're saying we should approach it
> mathematically, yet in the end you're drifting toward the now competing
> theory where language is treated as an emergent phenomenon of connectionism
> in the brain, not a listable set of specific rules!
Yeah, it is a shame that tossing a coin approach works better.

> I have my students write mini computational grammars that work like that,
> and they work well, but they run into problems when you try to explain nuances
> or "exceptions". Claiming that rules apply logically but only treating the most
> general cases is not a scientific argument at all. That was my point to begin with.
If I understand you right, you need a facility to describe a rule in a context. A kind of that in a concrete context the rule changes to something different. I believe this may be done with a context-sensitive grammar:
<pre_condition1><rule><post_condition1>:= …
<pre_condition2><rule>:= …
<pre_condition2><rule><post_condition1>:= …
etc.
Context-sensitive grammars are a way out of my head; I do not understand them and cannot figure out how they may work in a real application.

Offline Daniel

  • Administrator
  • Experienced Linguist
  • *****
  • Posts: 2036
  • Country: us
    • English
The problem is that connectionism doesn't lend itself to abstraction as formalism. You can't approach it in the way you're going at the top of your answer and end up with a rule like you're proposing at the end. And if that's true, then it's not really clear if there are rules anymore at all, just emergent patterns.
Welcome to Linguist Forum! If you have any questions, please ask.

Offline Rock100

  • Linguist
  • ***
  • Posts: 68
> The problem is that connectionism doesn't lend itself to abstraction
> as formalism. You can't approach it in the way you're going at the top
> of your answer and end up with a rule like you're proposing at the end.
No, no, the moaning about “it will never work in general” and the proposed solution for the exceptions handling are unrelated. Formal grammars do their job within the proper environments and the exceptions handling proposal was for such environments only.
By the way, in real at least software projects (where nobody uses context-sensitive grammars at all) the context-dependent features are added with some additional non-grammar helpers like tables, dictionaries, etc. And if you add the stack data structure to a banal finite state machine to handle it’s states you will get a context-free capable processor which will be practically enough for a lot of real context-free applications. So the exception handling solution as above may be not a universal one but enough for some purposes.

Offline Daniel

  • Administrator
  • Experienced Linguist
  • *****
  • Posts: 2036
  • Country: us
    • English
Quote
No, no, the moaning about “it will never work in general” and the proposed solution for the exceptions handling are unrelated.
Such claims without evidence are, well, without evidence. Make a perfect grammar, and show it to us. Until then, the objection is valid, because in practice unusual data can derail an analysis, and it often does. Those obsessed with generalizations only tend to work themselves into a corner where they can't explain less typical data.
Quote
By the way, in real at least software projects (where nobody uses context-sensitive grammars at all) the context-dependent features are added with some additional non-grammar helpers like tables, dictionaries, etc. And if you add the stack data structure to a banal finite state machine to handle it’s states you will get a context-free capable processor which will be practically enough for a lot of real context-free applications. So the exception handling solution as above may be not a universal one but enough for some purposes.
I'm losing track of what you're claiming. You're jumping back and forth between connectionist (e.g. neural network) approaches and formal grammars. I responded in a post above, but you did not address that, instead talking about ideas we've already covered.
Welcome to Linguist Forum! If you have any questions, please ask.

Offline panini

  • Linguist
  • ***
  • Posts: 195
Panini (the older one) never said anything as verbose as "there are no exceptions, only more specific rules". Actually, his treatment of "exceptions" was to state a general rule, vidhi, then state the exception, niyama.

Before embarking on a discussion of a question like "Is language strictly rule-based?", you have to define basic terms at least well enough that there is agreement about what "language" is and what a "rule" is. Also, what it means to be "based" on something. In the standard conception of "rule" in generative grammar, a rule is a mapping from representational string to representational string. In models prevalent up to the early 70's, all parts of a grammar are based on rules. In current models of grammar, only certain theories of phonology have rules (I don't understand morphology, so it's unclear what people are thinking there). So, yes or no, depending on which parts of grammar, and which theories. You can modify the definition of "rule" so that Optimality Theory constraints are "rules". So, you have to be more specific about what a "rule" is.

Also, "language" is a huge thing. Linguists study the structure of language, relying on a concept of "grammar", although what a "grammar" is is rather unclear, if you sum up the various things that people think grammars do. Some people think that beliefs about the real world are part of grammar, or the cognitive principles that make it hard to identify low-amplitude sounds; i.e. any form of behavior that has to do with "language". Other people have a more mathematical view, where a grammar says for example what the basic elements of a language are and how they can be combined – this is kind of like the theory of physics, which you can't really directly see instantiated in the ordinary world, because it isolates underlying causal principles and describes each in an ideal way that is separate from other ideal descriptions of other ideal principles.

I don't know of any linguists who adamantly deny human free will: whatever these rule things are that underlie this language thing, people free are rules those break to are, perhaps at some communicative cost, or perhaps just freely.

Offline Daniel

  • Administrator
  • Experienced Linguist
  • *****
  • Posts: 2036
  • Country: us
    • English
Quote
Panini (the older one) never said anything as verbose as "there are no exceptions, only more specific rules". Actually, his treatment of "exceptions" was to state a general rule, vidhi, then state the exception, niyama.
My undergrad semantics professor said this in class, and I've been casually searching for the quotation ever since, but without success, so it's relevant to hear you say this. He might have been phrasing it as a pedagogical tool (perhaps related to using a declarative programming language to represent linguistic rules), but I remember it clearly. I've thought it might instead connect to Kiparsky's "Elsewhere principle", but again I haven't found an explicit statement like that, but just the general ideas. Thank you for your input on this.

Quote
I don't know of any linguists who adamantly deny human free will: whatever these rule things are that underlie this language thing, people free are rules those break to are, perhaps at some communicative cost, or perhaps just freely.
What do you mean by this exactly? I'm not taking a prescriptive position. Are you saying that any generalizations in a grammar can be freely broken by speakers? I suppose that's possible, but then adds another complexity to the analysis. Given the very real consistency with which we speak, however, I'm tempted to say that, allowing for some exceptions like speech errors or (very gradual) change over the lifespan of a speaker, we don't generally break our own rules. If we did, I suppose it would look much more like a very literal interpretation of Exemplar Theory, where generalizations are emergent and ever-changing.

As for the rest, I'll just let you jump into the conversation with the others, but I'll follow along. I agree with what you wrote.
Welcome to Linguist Forum! If you have any questions, please ask.

Offline Rock100

  • Linguist
  • ***
  • Posts: 68
> Such claims without evidence are, well, without evidence. Make a perfect
> grammar, and show it to us. Until then, the objection is valid, because in
> practice unusual data can derail an analysis, and it often does. Those obsessed
> with generalizations only tend to work themselves into a corner where they
> can't explain less typical data.
Formal approach does not require me to do that – it, IMO, deals with the proofs of existence, correctness and things like that. And I cannot really make such a grammar indeed. But I probably know which directions are worth to try. It is not that much but I do feel it is enough for Linguist's Lounge > Outside of the box forum branch. I do understand that the list of alternatives in a context-free grammar like
<rule>:= <main rule description>
<rule>:= <exception 1 from the rule>
<rule>:= <exception 2 from the rule>

may be not enough for you so I proposed a more general solution which I believe the final approach indeed because the next level grammars – unrestricted grammars – have nothing to study, they just describe what they see exactly. But it is the whole everything indeed. Unrestricted grammars require a full-featured Turing machine for lexical and syntactical analyses. A full-featured Turing machine is proven to be able to describe everything the human mind can think of. So, the language is the grammar (it is by definition in math and it looks like many linguists are agree and appeal to the existence of Twas brillig, and the slithy toves / Did gyre and gimble in the wabe; /All mimsy were the borogoves, /And the mome raths outgrabe) and it can be described with a grammar (at least by a totally useless unrestricted one). So, IMO if you want to continue to improve your studying of languages with the formal approach you are either switch to context-sensitive grammars (and become the sixth person in the world who really understands them) or invent something revolutionary different.
Personally I prefer to stay pragmatic and do not try to develop a universal solution (I know it is impossible) but solve the concrete task separately. Again, it is not scientific at all but I believe is a kind of a fun for this section of the forum.

> I'm losing track of what you're claiming. You're jumping back and forth
> between connectionist
You are absolutely right. I know this sin of mine. That was a kind of illustration for the pragmatic approach I use in real applications – instead of switching to the next more complex and expensive level I try to hack the lower and more simple solution to fit for the task it was not originally supposed for. Without an elaboration it was the real “a, well, without evidence” – I will remember the phrase and will use it as the euphemism for the “bullshit”. But anyway, that my passage about using non-grammar helpers was a kind of unrelated indeed, sorry.