There is no language instinct (2014)
aeon.co
aeon.co
Check out the Piraha channls, where they also have "hum speech", "yell speech", "musical speech". http://itre.cis.upenn.edu/~myl/languagelog/archives/003175.h...
humans may have linguistic "knowledge" that they're born with, they may not... but this article is beside the point.
Chomsky and his ilk have made very strong, very specific claims about language in the past. They have claimed that certain combinations of linguistic parameters are impossible, and that certain features are shared by all languages. They have actively denied the wide variety of linguistic behavior until the evidence piles up enough that they change the topic.
Followers of this brand of linguistics have resisted empirical tests of these claims, and some of them have academically attacked those who disagree. And yet these claims have been chipped away anyway until all that's left is a really vague claim that most humans can communicate using recursion.
Your second paragraph seems at odds with your first. If these parameters exist to be set, that's knowledge. Knowing that word order is a thing that can vary is knowledge. Chomsky says we are born with this knowledge. Some who disagree with him say we aren't born with it, and that we have to learn it from experience.
In the part where you hedge and say "they may not...", you're surprisingly agreeing with this article more than Chomsky. Chomsky would say "Humans absolutely have linguistic knowledge they are born with. Anyone who disagrees is a behaviorist charlatan."
Chomsky's linguistics was indeed a way better model of learning than behaviorism, but that was 60 years ago and we need a better model now.
(edit: Note though that "All humans have no innate language knowledge" is still orders of magnitude more unlikely than "All humans have some common particular set of innate language knowledge" - the latter just requires the highly unlikely even that the random drawing of every single human's innate language knowledge came out to some particular value, but the former requires that the random drawing of all of humanity's innate language knowledge came out to the particular set of preconceptions that you consider to be "no preconceptions", and then that the random drawing of every single human's innate knowledge also came out to that value. Think likelihood of discovering an animal species all of whose living members have exactly the same colour of fur at 24-bit granularity, versus the likelihood of discovering one all of whose living members have #F45080 fur.)
I could say that because he has made that claim. You call it obviously false, he calls it obviously true.
Mentally impaired humans do not necessarily contradict Chomsky's view: I think he'd say that they have an internal language ("I-language") that they are not capable of expressing in the real world.
The idea that language is just a probability distribution over ways we can learn to communicate is the contrary position to universal grammar. If you're saying that's what Chomsky actually meant, you're basically moving his goalposts to the other side of the field.
For instance, it makes the absurd comparison between the supposed innate talent for language and a spiders web-weaving - spiders weave webs without examples, so why not kids? Spiders are about as poor a stand-in for a human as you'll find in the macroscopic world. Also, even if something is instinctual, it's hardly reasonable to assume that instinct will always overcome any situation, regardless of trauma, especially in a animals as impressionable as humans. Finally, just because an animal has an instinct and/or innate aptitude doesn't mean that practice is irrelevant. The girl that was held in a cellar until she was 13 only shows that either the instinct isn't absolute (no surprise there), or even just that the instinct still requires exercise.
Then there's the flatly false statement that dna doesn't have the storage capacity to store a universal grammar. Oh really? The human genomes codes for on the order of a gigabyte of data. AFAIH, there's strong evidence that there's not a lot of evolutionary pressure to keep genomes small, IIRC based on the observation that even small, competing microbes have dramatically varying sizes of genomes (when you'd expect smaller-genomed competitors to have won if it really mattered). For some context: A compressed wordlist containing 350000 english words is less than a megabyte. In general, grammars tend to be much smaller, especially if you consider that it's not claimed that a full grammar is stored, but at most the "basics" that can get rehydrated based on learnt language. Another poster in this thread claimed his 500 line python program can generate a good approximation of human language given enough input data. I wouldn't be surprised if a few kilobytes would suffice to describe the components from which any grammar can be built - much smaller than the full grammar itself.
There's another absurd misunderstanding of evolution in which the article manages to suggest that because evolution is necessarily continuous in the space of genomes, it thus must be continuous in the space of outcomes (i.e. language is so large a leap, it couldn't have evolved). I'm sure many creationists feel the same way about the eye - not to mention the fact that the article then proceeds to knock down some straw-men about when that language ability might have evolved according to chomsky (which is basically saying that because chomsky apparently doesn't know how language evolved, his other theory must be false. Really?)
I have no idea about the details here, but this article is about as unconvincing as you can get. He may well be right - I'm not a linguist, and to be frank I'm not entirely sure why the claim even matters (not a great sign for a scientific hypothesis). But I'm sure this article isn't making a lot of sense.
yes - e.g. BNF
You're misunderstanding. It's not that it couldn't have evolved, it's that it couldn't have evolved so quickly and with no evidence of a mass extinction.
The eye had millions and millions of years to evolve, and there are examples of different stages of eye evolution throughout the animal kingdom. There is one confirmed example of a language-using species, and on an evolutionary timescale it got language basically instantaneously.
The implication is that language is a learned skill like agriculture, not a biological entity like the eye.
Theorems like EPP and much of 80s control theory (the interplay between recursive structure and what determiner phrases, e.g. pronouns like "he" or demonstratives like "that", refer to), while occasionally being painfully overstretched, do seem to exhibit far better than chance predictive power and had their physics-like moments of conspicuous gaps in predicted taxonomies being neatly filled later. (Check "empty categories" or the rules mentioned under https://en.wikipedia.org/wiki/PRO_%28linguistics%29.)
I'm fairly disappointed comparing this to the usual quality of Aeon's articles.
Innateness of language is not a theory because it lacks what is the hallmark of any scientific theory: the ability to produce falsifiable predictions. People argue against ridiculous straw-men because there is nothing better that we could test. Chomsky proposed innateness to challenge the behaviourists who simply assumed that language is learned without even trying to proof it. Back in the day that was an honorable goal but half a century later we have to acknowledge that Chomsky and his followers have failed to offer something better, they just replaced one unfounded belief by another.
But Chomsky has switched his theoretical perspective so much and so many times that it's hard to tell what his current framework actually is. And increasingly, how relevant it is to contemporary research. It seems like you're advocating the "principles and parameters" approach, but AFAIK Chomsky hasn't advocated that in something like 20 years. P&P is where Chomsky lost me—with the argument that dropping pronouns in English and French amounts to a mental switch that languages flip on or off. Whereas subsequent research immediately showed that pronoun-dropping is just a continuum where English is at one extreme. And lots of Chomsky's arguments, like the poverty of the stimulus, just plain look bad.
Stephen Pinker has completely dominated popular understanding of linguistics with "The Language Instinct", which has influenced how a generation of ordinary people think about language and linguistics. And Chomsky's program dominated both academic and popular understanding before that. But linguistics is a strongly multidisciplinary and multiparadigmatic field, and this is an article aimed at a popular audience that helps make that more plain.
[1]: https://gist.github.com/karpathy/d4dee566867f8291f086
[2]: http://karpathy.github.io/2015/05/21/rnn-effectiveness/
Gold's result basically says that it's impossible for a learner to learn a non-finite language (which means anything from regular languages and above) without access to an infinite number of examples, both positive and negative, of the language, or failing that, access to an oracle that knows the language perfectly and can confirm or reject the learner's evolving model of the language. In some versions, the oracle must respond to errors with a negative counter-example, teaching the learner why its model is wrong ("if the sentence you just formed was correct then you could form this other sentence, which is obviously wrong").
As far as I know, Gold's result still stands. So I think you may be overestimating the power of the python script you link to.
I should say that I'm aware of Karpathy's project, and it's really cool (or awesome if you prefer), but reproducing text is not the same as learning language.
In other words, if you expect any program to learn language, you better have infinite data to train it with and also an infinite amount of time. I stress that this doesn't only apply to human languages. You can try teaching a network to learn the syntax of Java, for example. You'll still need infinite resources, or an Oracle.
++ Edited to remove unfortunate scare quotes and also to add: if you could machine-learn a language, even "just" a CFG, we wouldn't need to write parsers for programming language compilers anymore. It would suffice to present a machine learning algorithm with some examples (a lot of them probably) and the algorithm would form a model of the language, that we could then use as a parser.
This is far from the state of the art today, however.
__________________
[1] http://web.mit.edu/~6.863/www/spring2009/readings/gold67limi...
Learning a foreign language can be as challenging as learning to program or building a software system. It takes a lot of dedication and practice. Involves using a lot of memory and pattern recognition, and all the "intellectual" capacities that are needed to handle a large engineering project.
> [Chomsky] concluded that [children] must be born with a rudimentary body of grammatical knowledge – a ‘Universal Grammar’ – written into the human DNA.
> ... Chomsky is plain wrong
> What is in dispute is the claim that knowledge of language itself – the language software – is something that each human child is born with.
> Our brains really are ‘language-ready’ in the following limited sense: they have the right sort of working memory to process sentence-level syntax, and an unusually large prefrontal cortex that gives us the associative learning capacity to use symbols in the first place.
> And of course, language doesn’t need to be spoken. ..linguistic meaning can be conveyed in multiple ways: in speech, by gestured signs, on the printed page or computer screen. It does not depend upon a particular medium for its expression.
> As it happens, cognitive neuroscience research from the past two decades or so has begun to lift the veil on where language is processed in the brain. The short answer is that it is everywhere.
> Why is it that today, only humans have language, the most complex of animal behaviours?
> [Early humans'] new ecological situation would have led, inexorably, to changes in human behaviour. ... This allows us to picture the emergence of language as a gradual process from many overlapping tendencies.
> We see this instinct [(cooperation)] at work in human infants as they attempt to acquire their mother tongue. Children have far more sophisticated learning capacities than Chomsky foresaw. They are able to deploy sophisticated intention-recognition abilities from a young age, perhaps as early as nine months old, in order to begin to figure out the communicative purposes of the adults around them. And this is, ultimately, an outcome of our co‑operative minds.
We can imagine an amoeba having a rudimentary ability to detect whether its surroundings are hostile or not and employ an instinctual response to evade it. It's not much of a leap from there to say humans have a social instinct, and that our ability to evade hostility has become strongly correlated with our ability to identify behaviors that are appropriate in context.
And language is, in an abstract sense, just this: choosing appropriate actions. Maybe they're words. Maybe they're intonations. Or other social norms. Nobody learns 'just' language as a child, they learn a whole way of life. They learn "swearing is wrong," not simply "this is the proper syntax for swearing." And they feel its wrongness when they swear. You can't know what swear words mean if you can't feel that, and you probably won't be able to survive very well as a human if you can't determine what behavior is appropriate or not in your situation.
> But recent research in neurobiology suggests that human DNA just doesn’t have anything like the coding power needed to do this
How could the author arrive at this conclusion without knowing what the universal grammar which needs encoding is? And beyond that, what if such a universal grammar is simply an artifact of existing structures accounted for in DNA?
And language is probably a product of singing. Checkout "The Master and His Emissary: The Divided Brain and the Making of the Western World". Singing is used to share feelings in larger groups then ordinary grooming permit. Language is a systematization of singing. That is why music evokes feeling response more readily then words in ordinary speech, speech evolved later.
This is why I believe that if there are any genetic encodings for language (whether either specific to language alone or non-specific but used also by language, if you get my drift) they must have occurred with twins who were then able to communicate more effectively with each other relative to their peers and so they survived and so did their offspring and their genes spread throughout the gene pool. I know it's a wacky idea but it gets around the rebuttal of "presumably had no one to talk to."
And sure, it's an unlikely event but we're talking hundreds of thousands of years here, thousands of generations, billions of people.
Ctrl+F Everett