Author Topic: Is language really rule-based? [Details inside] (Read 52074 times)

Daniel · « **on:** January 02, 2014, 05:59:36 PM »

In a very specific sense, I'm wondering if languages follow "rules":

Is language strictly rule-based?

It is very clear that there are patterns in languages. That's not in question here.

What I'm wondering is whether those patterns are strict rules, without exceptions, and whether those rules fully explain all forms in the language.

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.

Another way of phrasing this is to ask whether a linguist, with complete success, would be able to explain everything about a language using a logical system.

To give some examples, there seem to be instances where language change reveals illogical language use. That is, the use is internally inconsistent for the language. See these discussions:
"Do you miss not having a job?"
Hadn't have...
Both of these forms are in a very literal sense illogical. 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.
In the first case, English doesn't have double negation and the semantics of "miss" are already negative, so the form is apparently a contradiction. In the second, the same auxiliary is used twice in the sentence vacuously.

There are also some idioms:
"I could care less!"

Sure, we could argue that these are just "more specific rules" (eg, all the discussion about the Elsewhere Principle, which I happen to like), but I'm not sure that quite captures it.

So we could describe everthing just descriptively-- note the patterns, note the exceptions, continue until (in theory) we've 'finished' the whole language. But does that count as rules?

If there are rules, then there should be limits to what is possible. So, is that true? Are there limits to what a language can do? Is there evidence that something is impossible thus showing that language can't do it?
Chomsky likes to cite the sentence "Is the man who is tall happy?" to show that sub-aux inversion follows non-linear rules, but I find it hard to know that it's impossible for it to instead follow linear rules. Is language so constrained? Or is it just useful? Do the rules just end up being used because the work out for communication? Or are there inherent limits on the system?

(Of course I'm now brought up individual languages and also universal considerations, so feel free to discuss both.)

jkpate · « **Reply #1 on:** January 03, 2014, 02:37:58 AM »

Quote from: djr33 on January 02, 2014, 05:59:36 PM

So we could describe everthing just descriptively-- note the patterns, note the exceptions, continue until (in theory) we've 'finished' the whole language. But does that count as rules?

Why wouldn't this count as rules? For example, maybe there's one rule for the general pattern, and lots of rules for the exceptions.

Quote from: djr33 on January 02, 2014, 05:59:36 PM

If there are rules, then there should be limits to what is possible.

Why does the existence of rules mean that there are limits to what is possible? Indeed, what do you mean by "what is possible"? There are NP-hard problems that can be stated quite simply in terms of rules.

Corybobory · « **Reply #2 on:** January 03, 2014, 05:52:39 AM »

I like the word 'conventions' rather than rules, and I think they emerge and are passed on by people who interpret them to have meaning, and therefore there is no biological component at all other than the limits put on our world by having an embodied mind. People forget these rules, make mistakes, interprets its usage differently, or even just slightly differently, and that is what results in a speech community having variation in how these conventions are played out. So 'hard' rules might just be something that is unlikely to evolve!

Daniel · « **Reply #3 on:** January 03, 2014, 10:02:49 AM »

Quote from: jkpate

Why wouldn't this count as rules? For example, maybe there's one rule for the general pattern, and lots of rules for the exceptions.

It could, but so what? If "rule" just means "describable" is it still interesting? Note that a completely arbitrary system is describable by listing everything out. Is that also rule-based?
I'm happy to discuss the deep question of what it means to be rule-based and where we'd draw the line. I'm not presupposing an answer to that question.

Quote

Why does the existence of rules mean that there are limits to what is possible? Indeed, what do you mean by "what is possible"? There are NP-hard problems that can be stated quite simply in terms of rules.

Within the system, there should be limits on what could be a possible sentence, either synchronically or (immediately) diachronically. Grammaticality seems to prove the first, but with random changes occuring diachronically, it's hard to see what the rules really accomplish. If any form is a possible change, then is it conceptually helpful to think of these as "rules" rather than just, as Cory said, conventions?
More broadly (at a universal level), there should be some limits on what a possible language would look like. But I'm entirely unconvinced. Some languages wouldn't be useful, but I don't see how any conceivable language would be impossible for human communication.

I was once told (very early on in my studies of linguistics!) that Chomsky's theories suggest that if aliens were to come to Earth they would be unable to learn human languages because it is a uniquely human property with UG. Let's assume that's true-- human languages follow human language rules. This means there must be some limits to the form of languages and not just that we are able to understand patterns and so forth. Because aliens can also (presumably) understand patterns and could possibly then learn to speak English. As I said, I'm completely unconvinced by this. I think humans just have great parsing (and generative) capabilities and end up stumbling upon systems that look like what we know as languages. I don't think there are any inherent limits on the system.

To rephrase my question, perhaps: what's the point in calling the patterns "rules"? Is it helpful? It it meaningful?

(Again I realize I'm mixing universal and language-specific rules into a single question here. Feel free to respond to either or both. I'm not even sure what a "rule" is anyway, so it's hard to ask these questions. I'm just confused about why we assume languages are rule-based.)

Quote from: Cory

I like the word 'conventions' rather than rules, and I think they emerge and are passed on by people who interpret them to have meaning, and therefore there is no biological component at all other than the limits put on our world by having an embodied mind. People forget these rules, make mistakes, interprets its usage differently, or even just slightly differently, and that is what results in a speech community having variation in how these conventions are played out. So 'hard' rules might just be something that is unlikely to evolve!

Very well said. That's the null hypothesis. And without any compelling evidence against it, I'm inclined to believe it. Humans use patterns mostly reliably, with conventions that are reliable enough for others to follow our intentions. But there are no inherent rules either within or across languages.

In other words: human languages are a near* optimal way to convey ideas. As humans we have an ability to encode and parse information into linear strings, and this would like resemble the same process as done by any other equally intelligent species for communication, though some details might vary based on human (or other) experience. Systemically, it's all about the form conveying information, and not about any arbitrary rules.

(*No idea how near-- might be pretty far or pretty close, as long as it's better than many other not-as-useful systems out there, which is certainly true. There are widely infinitely many systems that are worse than English for conveying ideas (such as everyone saying only "aaaaa"), so I'm happy to assume human languages are approaching the limit of what would be optimal, though I have no idea how close they are.)

Corybobory · « **Reply #4 on:** January 03, 2014, 10:45:29 AM »

^There's another analogy between linguistic and biological evolution for us then, how 'optimal' something is for use in its environment. Lingustic items get used and understood and passed on if they're good enough, or, not selected against - how 'perfect' are a butterfly's wings or an elephants nose, or the human spine? They get the job done!

Daniel · « **Reply #5 on:** January 03, 2014, 10:56:24 AM »

Right. That's all I mean. But I also mean that, interestingly also still parallel to evolution, coincidental similarities aren't unexpected-- squid eyes and human eyes are very very similar in structure yet are unrelated genetically. As a very reasonable null hypothesis (yours above, and my current intuition) I think this is how language works-- languages have evolved to be useful within the limits of how combinatorial systems work, NOT some special human system that would be incomprehensible to equally intelligent aliens!
Human languages (and their rules) should be expected developments given what they need to do-- just like wings are an expected development given that some animals need to fly.

jkpate · « **Reply #6 on:** January 03, 2014, 05:22:20 PM »

Quote from: djr33 on January 03, 2014, 10:02:49 AM

Quote from: jkpate
Why wouldn't this count as rules? For example, maybe there's one rule for the general pattern, and lots of rules for the exceptions.
It could, but so what? If "rule" just means "describable" is it still interesting? Note that a completely arbitrary system is describable by listing everything out. Is that also rule-based?
I'm happy to discuss the deep question of what it means to be rule-based and where we'd draw the line. I'm not presupposing an answer to that question.

Haha, I think you just wrote out my view of the issue. Simply listing everything of an arbitrary system is rule based, because a list of rules is a list of... rules. It's a boring system, but there's no rule (hah!) that rule-based systems must be cool.

In general, we know that if your system has one set of rules, such as the simply-typed lambda calculus, you can describe anything, while if it has another, like a finite state automaton, you are much more limited in what you can describe. So the question of whether language is rule-based is completely uninteresting, as far as I can see. However, what the rules look like is very interesting.

And most of the time, people are actually arguing about what the rules look like when they argue over whether language is "rule-based." Depending on the debate, "rule-based" can mean:

a small number of rules
deterministic rules (always does the same thing with the same input)
non-probabilistic rules (this is different from deterministic!)
rules that refer only to discretely-valued variables
rules that refer to unobserved variables
rules that build unobserved hierarchical structure
rules that obligatorily apply
and even more

All of these potential qualities of rules are orthogonal to the question of whether a dataset has been generated by rules. Bayesian models, the currently trendy "alternative" to Generativism, clearly and unambiguously use rules (Bayesian modelling just applies probability theory to the rules). Connectionist models are the other primary alternative. Unfortunately, the term "connectionist" is more sociological than technical, and encompasses a range of models, so I can't demonstrate that all connectionist models are rule-based in this short post, but I think they are. Some of these models actually perform inference in a Bayesian model, and so are rule-based in the same way; they are just built in a way that obscures the Bayesian model. I'm happy to talk about other connectionist models and the kinds of rules they implicitly consider if people are interested.

Quote from: djr33 on January 03, 2014, 10:02:49 AM

Quote
Why does the existence of rules mean that there are limits to what is possible? Indeed, what do you mean by "what is possible"? There are NP-hard problems that can be stated quite simply in terms of rules.
Within the system, there should be limits on what could be a possible sentence, either synchronically or (immediately) diachronically. Grammaticality seems to prove the first, but with random changes occuring diachronically, it's hard to see what the rules really accomplish. If any form is a possible change, then is it conceptually helpful to think of these as "rules" rather than just, as Cory said, conventions?
More broadly (at a universal level), there should be some limits on what a possible language would look like. But I'm entirely unconvinced. Some languages wouldn't be useful, but I don't see how any conceivable language would be impossible for human communication.

Sure, but this is orthogonal to the question of whether language is rule-based. How is "convention" different from "rule"? To my mind, the only difference is that a "convention" is probably social and is more likely to be a rule that does not obligatorily apply. Even in her post, Cory later refers to the conventions as "rules" and then more specifically as "hard rules." I think this is another case of arguing about what the rules look like, not whether there are rules in the first place.

Quote from: djr33 on January 03, 2014, 10:02:49 AM

I was once told (very early on in my studies of linguistics!) that Chomsky's theories suggest that if aliens were to come to Earth they would be unable to learn human languages because it is a uniquely human property with UG. Let's assume that's true-- human languages follow human language rules. This means there must be some limits to the form of languages and not just that we are able to understand patterns and so forth. Because aliens can also (presumably) understand patterns and could possibly then learn to speak English. As I said, I'm completely unconvinced by this. I think humans just have great parsing (and generative) capabilities and end up stumbling upon systems that look like what we know as languages. I don't think there are any inherent limits on the system.

To rephrase my question, perhaps: what's the point in calling the patterns "rules"? Is it helpful? It it meaningful?

I think it is potentially confusing to call the patterns "rules," because, in practice, people mean so many different things by the term "rule." So when you are talking about your rules, don't simply say that they are rules, but provide more detail (e.g. "we are using context free grammar rules with probabilities," "we are using maximum entropy constraints with negative weights", &c.)

Daniel · « **Reply #7 on:** January 03, 2014, 06:36:08 PM »

Do patterns necessarily imply rules?
For example, tidal patterns exist, but are there rules?

I wonder (only wonder) whether it is somehow possible for language to exist as a kind of approximation process where meaning is sort of encoded in what we say, just enough that it works, but without anything deterministic, probabilistic or really "rule-based" in any sense-- it's unconstrained but near something that is consistent. In the same sense that pixels more or less represent an image, the language more or less encodes meaning.

However, that odd thought aside, you're probably right: language is inherently rule-based, if we properly define what "rules" are.

But then I'm not sure what question I'm trying to ask. I do think there's a question to be asked still. One would certainly be something like whether rules are probabilistic or deterministic, but that's a little too narrow for what I'm thinking.

I guess I'd like to bring up the idea that (reportedly) Chomsky would say aliens couldn't (potentially) understand human language. This suggests "rules" that are part of language beyond what is necessary for a communication system by random chance.

Is more going on in human languages than just a linearization of meaning? Is it fair to say that human language has properties, or just that human language is a linearization of meaning that then entails whatever properties we observe?

Is English anything more than the application of those properties to a lexicon?

Is there any inherent weighting given to hierarchical structures over linear ones? That is, is there some underlying system that requires hierarchical (not linear) rules, or are the hierarchical rules just more useful?

I can easily imagine a language in which the following is possible:
"Is the man who tall is happy?"
But it seems utterly stupid to use such a system because it's so irrelevant to the real world. (Admittedly that could be bias from my only-human brain.)

The point is... as linguists, is there anything more to be done than descriptive work? Are there true underlying rules to be discovered that will tell us something about language? Or is the best we can do just discovering the forms of languages and also perhaps the forms of meaning encoding?

To again phrase the question inversely:
Are there certain things we can eliminate as possibilities from a systemic (not functional) perspective? Are some languages impossible as systems for humans? [Feel free to constrain this to imaginable systems of equal or lesser complexity than existing languages. I don't really care that humans can't, for example, process 1,000,000-word sentences.]
Are there any times when it would be reasonable for a linguist to call something "illogical" or an inherent error? Either within or across languages.

If there are no constraints, I'd like to argue there are no (interesting) rules.

Guijarro · « **Reply #8 on:** January 05, 2014, 04:45:24 AM »

From my point of view, this debate is "nominalist" (as in the Middle Ages).

What are rules, conventions, patterns, etc.? Are they synonymous, or are they?

If things would act according to rules, these rules had to emerge from God. But, as I can't believe in that other "term" having a real referent, I can't believe in rules either.

Another thing is to be able to find patterns in aspects of our environment, to try to sort the conventional from the unconventional ones, and to make a describing effort accing the result a set of (grammatical) rules.

As the Spanish poet said:

En este mundo traidor (in this treacherous world)
nada es verdad ni es mentira (nothing is true, nothing is false)
todo depende del color (everything depends on the colour)
del cristal con que se mira (of the glass through which you are looking)

I do think that Chomsky's glass is very convenient, but others might prefer other colours.

Daniel · « **Reply #9 on:** January 05, 2014, 12:08:28 PM »

Quote

What are rules, conventions, patterns, etc.? Are they synonymous, or are they?

If things would act according to rules, these rules had to emerge from God. But, as I can't believe in that other "term" having a real referent, I can't believe in rules either.

But "God" is just one way to understand the universe. I had a conversation with a psychologist / cognitive scientist and we thought that maybe mathematics is the basics of everything-- the universe runs on math (physics, etc.).
Certainly you could argue that everything is just an illusion and that there is nothing to explain because nothing can be explained, but that's a boring position, and it doesn't give any motivation for scientific study.
But if you do accept at least that math "exists" and explains things, then I think the rest can relatively comfortably follow. There should be some "fact of the matter" to questions like "how does language work?" assuming the questions are properly defined (that is, in what sense we mean "language" and "work").

Quote

I do think that Chomsky's glass is very convenient, but others might prefer other colours.

This is a worrying statement and why, I think, many people don't see linguists as scientists. Theories aren't something to play around with when they happen to work for you. Theories should be proposed as genuine answers to problems, although we should always be open to being wrong. Chomsky, as far as I can tell, seems to believe that a lot of his ideas are Right, although I'm sure he'd be flexible about many of the details. To say it's just another glass to look through is worrying: that means the ideas are essentially meaningless, something like doing a crossword puzzle-- intellectually stimulating but not actually about the real world, not science.

It would be a scientist's worst nightmare to find out that nothing is to be explained. Ironically a true scientist would accept that if the evidence pointed to it (if such a certainty about uncertainty were possible-- "everything is uncertain" is a paradox after all).

At the same time, that might be freeing-- the best descriptive method is whatever works, because there is no "fact of the matter" or "answer", no absolute rules out there for how language behaves. Then we can all stop arguing and just use whatever theory feels convenient-- a linguist's true job would be to describe languages, discuss the data, nothing more. That wouldn't be terrible. I like languages and data.

But I do hope there is a "fact of the matter" to discover. If there are no underlying rules of some kind, then linguists as scientists are wasting their time. Of course I don't mean to imply that necessarily then the form of language has rules-- maybe it has none; maybe the rules exist at another level, such as how the mind works. Then we're not really linguistic scientists after all, then, but instead just a type of cognitive (neuro?) scientist instead.

MalFet · « **Reply #10 on:** January 05, 2014, 02:26:30 PM »

Quote from: Guijarro on January 05, 2014, 04:45:24 AM

From my point of view, this debate is "nominalist" (as in the Middle Ages).

What are rules, conventions, patterns, etc.? Are they synonymous, or are they?

If things would act according to rules, these rules had to emerge from God. But, as I can't believe in that other "term" having a real referent, I can't believe in rules either.

Another thing is to be able to find patterns in aspects of our environment, to try to sort the conventional from the unconventional ones, and to make a describing effort accing the result a set of (grammatical) rules.

I couldn't have said it better, even with a thousand monkeys and a thousand typewriters at my disposal. There are some interesting new challenges to nominalism, though. I'm not aware of any that deal with "rules" specifically, but I was surprised to learn that Plato is making a comeback!

Daniel · « **Reply #11 on:** January 05, 2014, 02:36:06 PM »

As a genuine question, if that is all the case, then what is the point, and why would you believe in the scientific method? Is it just a game?

MalFet · « **Reply #12 on:** January 05, 2014, 04:17:06 PM »

Quote from: djr33 on January 05, 2014, 02:36:06 PM

As a genuine question, if that is all the case, then what is the point, and why would you believe in the scientific method? Is it just a game?

Is this addressed to me? I'm baffled by these dichotomies you keep cooking up. If the world is not anchored by Platonic forms, it must all be a game? Good gravy, man!

Daniel · « **Reply #13 on:** January 05, 2014, 04:28:42 PM »

Simple questions:

1. Is there a fact of the matter? Is there, upon sufficient knowledge, an indisputable correct answer to (well-formed) questions? [Are Chomsky's theories necessarily either right or wrong, rather than just an arbitrary perspective, a "glass"?]

2. If not, what is the goal of science?

And, fine, perhaps "game" is the wrong word, but... "puzzle"? If there is no inherent "right answer" then all we're doing is entertaining ourselves via "scientific" puzzles-- we're not approaching actual knowledge because such knowledge does not exist.

In short, I would think that a scientist would want there to be a fact of the matter. If there is no god, it makes very little sense to be religious (and religious people should give up their faith). If there is no truth, I can't help but think the same applies to scientists.

Now, certainly, there might be other reasons to continue studying. For example, engineering is a useful field and does not rely on "truth", just effectiveness.

Quote

Is this addressed to me?

You and Guijarro. You seemed to agree with his post, although you didn't specifically quote the section on "glasses", which I think follows from the rest.

Again, my question was a serious one. You act surprised by my apparently ridiculous question and answer with rhetorical questions. But... what's the answer?

How can science not be a game if it isn't possible to achieve knowledge?

Falsifiability is the essence of science. Without it, if we can just alter it by using a different "glass", then I fail to see the point of science.

jkpate · « **Reply #14 on:** January 05, 2014, 04:58:27 PM »

Wait, where did God come into the picture? and why would the rules have "had to emerge from God"?

Also, this turn in the discussion reminds me of the relativity of wrong by Isaac Asimov, which views theories as not "right" or "wrong" but simply more or less wrong, or more or less complete.

Author Topic: Is language really rule-based? [Details inside] (Read 52074 times)

Daniel

Is language really rule-based? [Details inside]

jkpate

Re: Is language really rule-based? [Details inside]

Corybobory

Re: Is language really rule-based? [Details inside]

Daniel

Re: Is language really rule-based? [Details inside]

Corybobory

Re: Is language really rule-based? [Details inside]

Daniel

Re: Is language really rule-based? [Details inside]

jkpate

Re: Is language really rule-based? [Details inside]

Daniel

Re: Is language really rule-based? [Details inside]

Guijarro

Re: Is language really rule-based? [Details inside]

Daniel

Re: Is language really rule-based? [Details inside]

MalFet

Re: Is language really rule-based? [Details inside]

Daniel

Re: Is language really rule-based? [Details inside]

MalFet

Re: Is language really rule-based? [Details inside]

Daniel

Re: Is language really rule-based? [Details inside]

jkpate

Re: Is language really rule-based? [Details inside]