Re: The Poverty of the Stimulus

From: Stevan Harnad (
Date: Sat May 25 1996 - 20:31:34 BST

> From: "Susie (Suzanne Jane Petrie)" <>
> Date: Tue, 16 Apr 1996 16:26:36 GMT
> In order to communicate, the majority of human beings use spoken
> language to do so. With the aid of `grammar', languages follow
> specified rules, so that its speakers can share this means of
> communication to their advantage. However, every different language
> has a different kind of grammar. Grammatical devices are used in
> different ways around the world. Although these rules of spoken
> language are unsimilar, it is possible to say one meaning in all
> languages. Infact, `anything you find in one language can also be
> found in every other language, perhaps at a more abstract level of
> representation', (Chomsky). Spoken language has its differences but
> we cannot deny the similarities. All have phrase structure rules,
> rules of linear order, tense and auxilaries. These properties may
> not convey exactly the same significance or relevance in every
> language but they no doubt occur somewhere.

Two different issues here: (1) You can say anything that you can say in
one language in any other language (though not necessarily with the same
number of words or the same grammatical structure). (2) All languages'
grammars are just special cases of Universal Grammar (UG). They differ
in some parameters, like slightly different settings of the dials on a
complicated device, but the device is the same.

> Grammar is a different component in every language but Universal
> Grammar (UG) is one for every. UG is the way the mind logically
> thinks. `It specifies the allowable mental representations (stands
> for) and operations that all languages are confined to use',
> (Pinker). The thoughts everyone experiences are guided by the UG
> facility.

Not the CONTENT of the thoughts, though, just the grammatical FORM is
dictated by UG.

> By consequence, language is a means of communicating it.
> As Skinner said, ` thinking is a form of verbal behaviour. Verbal
> behaviour is a prime manifestation of "thought"'. It is thought
> that every child is born with UG; the ability to acquire (have)
> language is species specific. Even without environmental cues,
> `children learn languages governed by highly subtle and abstract
> principles'(Chomsky). UG cannot be learned from environmental
> examples; it seems to be a mental structure which is innate (born
> with). If environmental exposure was its origin, there appears to
> be very little or no evidence as to how a child knows wrong rules
> from right ones.

UG turns out to be too complex, and the data the child encounters too
simple, to allow the child to learn UG by trial and error. Negative
cases especially are missing: The child neither hears nor tries enough
cases to figure out the rules of UG by trial and error.

> This is where the issue of poverty of the stimulus arises. Before
> this subject goes any further, what exactly is poverty of the
> stimulus? Well, to say the least, by the age of four , children are
> excellent speakers of language, using complex structures extremely
> well. However, the big question is, how do they gather such an
> ability with so few examples? They may receive an abundance of
> positive things to say correctly but how do they know not to say
> incorrect phrases?

Exactly. It's the negative evidence that is missing.

> It's amazing how a child of four comes out with
> new correct sentences never heard of before. Where does the
> guidance come from? `Children have no manual to consult', (Pinker
> and Bloom). A child never gets negative examples , only positive
> ones, therfore these are too impoverished for him or her to learn UG
> rules from. Instead,'children exploit special ways in which their
> mothers talk to them by making unself conscious errors', (Pinker).
> The point with poverty of the stimulus is that if UG was learn't
> through environmental guidance, there just wouldn't be enough data
> for the child to learn from. One would have thought more evidence
> was needed. So, it seems the child already has UG to begin with ,
> hence, mistakes are rarely made and has no need for corrections.


> Even so, a child's exposure to positive evidence is by no way
> lacking. Positive evidence is the information available to a child
> about which groups of words are correct grammatical sentences of a
> particular language. Basically, a four year old child only produces
> or hears positive examples of UG from others such as parents or
> other carers. It is strongly suggested that, 90% of the time,
> children speak correctly, (Pinker). Infact, studies carryed out by
> Pinker, Stromswold, Crain and Marcus et al, found that for every
> rule that was observed, children of three obeyed them the majority
> of the time. Acquisition of UG may be thought to be innate but it is
> unable to strive alone. Only in cooperation with environmental cues
> will it succeed. Most children have access to spoken language.
> They need some form of linguistic input to acquire a language. If
> grammatical abilities are of biological origin they are ` too
> formalised and systematic to generate sound speech, words and
> grammatical constructions on their own' (Pinker).

> Grammatical rules
> are not all straight forward and compliant; they often stray from
> their proper places. For one, there are exceptions to every rule.
> In the case of language, there are no doubt more than one.
> Deviations of these kind probably answer up to the environmental
> exposure everyone faces. It must be said, rules, especially rules
> of language (grammar) change with experience and practice.
> Subsequently, language develops both with the innate UG device and
> the environmental exposure of the language.

I'm not sure what the last paragraph said, or meant...

> All these positive examples of the correct way to `speak' seem to be
> ample evidence for a child to have the facility of UG. On the other
> hand, `children do not need to hear a full language; as long as they
> are in a community with other children and some source of individual
> words, they will invent one on their own, often in a single
> generation', (Pinker). Children of such a young age adapt extremely
> easily to the environment they live in. Case studies by Bickerton
> found examples of the above situation. There have been children
> that grew up in plantations and slave colonies who were actually
> exposed to crude pidgin (type of language several different native
> speakers use) but infact grew up to speak genuinely new languages.
> These are better known as expressive `creoles' with their own
> complex grammar, (Bickerton, 1984, see also chapter by Newport and
> Gleitman).
> Even though the sheer existence of UG puts the child's ability of
> learning a language to a great advantage, he or she needs to hear an
> existing language to learn it. Naturally sounding sentences in
> colloquial speech (every day talk) are classed as `grammatical',
> positive evidence. These may include such utterances as slang
> (Pinker 1994a); for example, `where are you going'? and, `to the
> shops'. Grammatical samples drive language acquisition. Even
> supposedly ungrammatical examples are mean't to guide our language
> formation, for instance, `he be coming soon'. Plus , many people
> consider the grammar of working class folk to be poor, whereas, in
> fact, it is scientifically complex in differing ways, depending on
> the language.

And it does not violate UG; just the "style" grammars that particular
languages have, and that may differ from social group to social group
and may change across time as a "wrong" form gets accepted.

> Children do experience lots of positive evidence from their mother
> tongue language. In many respects, this should make up for the
> impoverished negative evidence.

No, it cannot. If you were trying to learn what a "Zarzong" was and I
kept pointing to things and saying: That's a Zarzong, that's a zarzong.
And I never pointed to anything that WASN'T a Zarzong, how would you
ever figure out what a Zarzong was? How would you know what to point to
and what NOT to point to?

> Maybe it is because there is so
> much positive exposure to be guided by, there is no reason for
> completely `wrong' utterances to occur; they have never been heard
> in the first place.

But that's just the same as saying people knew which things to point to
as Zarzongs without needing to learn it! They alreday knew.

> However, the child still knows what not to say
> and what to say. This all miraculously happens without being
> formally instucted but growing up in normal environmental conditions
> hearing spoken utterances. The best part of these being produced by
> the child's parents.
> Coming from the other direction, negative evidence for UG is
> somewhat under nourished; this is why the child's language
> acquisition device is so impoverished by incorrect grammatical
> utterances. There is plenty of stimuli for correct, well formed
> grammatical sentences, but absolutely none, if any, for bad ones.
> One would have expected to hear both aspects of a language in order
> to learn which is the better one to produce.

More than that: If we didn't already know the rules, we'd have to figure
them out by trial and error. We hear some positive cases and some
negative cases, guess (consciously or unconsciously) that the rule is
"big words first, little words last," then we TRY it, and see whether we
got feedback that what we had produced was a positive case, a
grammatical sentence. If we got feedback that it was NOT, then we'd ave
to try another rule.

But if we never got this feedback, yet we never got anything wrong, the
only explanation must be that we already had the rules built into our
heads in the first place.

> Funnily enough,
> though, the child's brain never gets stimulation of such a negative
> kind. In general terms, negative of UG consists of information
> available to the child about sentences that are not proper. From a
> young age, the majority of childrens' efforts to speak are not
> corrected. How does this explain their practically perfect use of
> language by the time they reach four years of age? It comes across
> as a mystery as to how a child this age simply knows what can't be
> said if external experience is the only cause for acquiring it.

Not just that they're not corrected, but they never make UG mistakes!

> In a paper written by Pinker, different hypotheses for negative
> evidence were suggested. The way a child acquires a language
> (English especially) could take four possible routes. For a start,
> it should be acknowledged that childrens' language differs from
> adults'. The first theory mentioned claims the child's hypothesis
> language (H) (basis for reasoning of what the child thinks is an
> appropriate language which can be understood) is separate in parts
> from the language later to be acquired (to have), referred to as the
> target language (T). A second theory hints that the child's
> language and the (t) language actually intersect; the child
> supposedly only uses some correct grammatical sentences. Thirdly,
> another explanation says the the child's (H) language is infact a
> subset or mini version of the target language (T), where the child
> has succeeded to speak some good English. However, all these
> explanations for acquiring a language surrendered to evidence
> suggesting otherwise. It was more likely to be incorrect to develop
> a language in the above ways just from hearing positive evidence of
> UG without hearing wrong examples. A fourth theory described the
> child's hypothesis (H) language as being a superset (used to its
> full extent) of the target language (T); for instance, stems of
> irregular verbs may be used in the past tense(we broke it, we went,
> we breaked it, we goed). Similarly, this was discovered impossible.
> Negative evidence of UG would be needed in order to hear
> corrections. In seldom an environmental situation, the fact that
> negative evidence is non existent would proove impossible for a
> child to produce a superset of the target language (T). The child
> would guess too large a language which the world would never
> correct. Obviously, there are limits as to how much a child can
> possibly say because there would be things no one could understand.

The only way to narrow it down to all and only the grammatically correct
ones is to get enough negative evidence (which does not happen) or to
have it in the head already, as UG.

> In attempts to research behind the reasoning of UG acquisition,
> experiments have been carryed out on young children. MacWhinney and
> Snow (1985,1990,1991) experimented with young children, where they
> were asked to answer sentences and describe pictures. The aim of
> these practicals was to get the children to indicate what sounded
> `silly' to them. In this light, an apparent existence of UG could
> be conveyed. It was also discovered that, within the first year of
> life, children can control speech muscles and are sensitive to
> phonetic (vocal differences in meanings of words) distinctions used
> in their parents language. Indeed, this ability is produced before
> the comprehension of words so learning doesn't necessarily rely on
> sound and meaning of a language.
> In general terms, childrens' language development takes the same
> path world wide. By the first year, their first words are similar.
> Basic necessities of life are appreciated as very important and this
> shows in their initial use of the language; for example, utterances
> concerning food, clothing and people are heard. Another study found
> that young children rapidly adapt this basic language into a more
> complicated one. Childrens' two or three word utterances resembled
> examples of longer sentences expressing a complete and more
> complicated idea. Three children were studied; even though they
> didn't ever produce a sentence as complex as `mum gave John lunch in
> the kitchen', they did mutter strings containing all these
> components aswell as in the correct order (Brown 1973, Bloom 1970,
> Pinker 1984). Once again, this ability comes with the
> experience of positive UG use but no negative use.
> Be that as it may, there still has to be some source of negative
> evidence. A language cannot survive without it. Gold and Pinker
> claim that if the child has no negative evidence and works out a
> rule that generates a superset of a language, he or she will have no
> way of knowing that he or she is wrong, (Gold 1967, Pinker 1979,
> 1989). The environmental constraints don't allow the child access
> to negative evidence of UG so there must be an underlying biological
> answer. There has to exist, a mechanism that either avoids
> generating too large a language or that can survive for generations.

Not quite sure what that last bit meant, but having UG in your head
already, with only the dials to be set by experience, is what makes it
all possible.

> Skinner's behaviourist's claim that language learning depends on
> parents' reinforcement of childrens' grammatical errors was tested
> by Roger Brown and Camille Hanlon (1970). Adults responses to
> grammatically well formed and ill formed phrases were divided into
> approvals (eg: `Yes good') and those of unapprovals (eg: `No that's
> not right'). Results illustrated that there was no correlation
> between reinforcement and language learning. Parents didn't
> differentially express approval or disapproval to their child's
> utterances. Instead, approval depended on whether the child's
> statement was true. Infact, parents don't understand their
> children's well formed sentences any better than their badly formed
> ones.

But in many ways the experiments were beside the point, because the
child never made UG-violating mistakes anyway, whether ocrrected or not.
(Most correction has to do with the particular grammar -- a grammatical
style, really -- of a given language or a subgroup of language speakers.
Correction is not the source of the universal capacity to produce
grammatical language in teh first place.)

> The reason why negative evidence does'nt answer the UG acquisition
> problem is because the amount of sentences a child is exposed to is
> finite (limited). Utterances of sentences are heard not the whole
> sentence, which can be deffective in many ways, for example slurred
> speech, slips of the tongue and mispronounciations. Plus, children
> don't have explicit guidance from a speech community seeing as they
> make few errors. Adults tend not to follow systematic correction
> because in reality, they are happy the child is speaking. Anyway,
> childrens' language intake is limited by their short attention span
> and their restricted short term capacity. These cognitive
> limitations have to affect complex utterances heard. The only
> reliable source of linguistic data has to be simple sentence
> utterances for the child. So, all in all, the information that is
> available to guide the process of UG acquisition is severely
> impoverished; it cannot be driven by the environmental existence of
> data, (Norbert Heinstein).

The problem is not that the evidence is finite: Finite evidence would be
enough to teach you UG if there were enough of it, both positive and
negative evidence. But the negative evidence does not occur, and it all
happens very fast, and the child is already speaking.

> Therefore it must be assumed that the child uses simple well formed
> sentences excluding `negative evidence' (information that a sentence
> is ill formed) to acquire UG. Children must come with a rich innate
> structure that guides language acquisition and grammar construction,
> otherwise how do they have such knowledge as not to say one thing
> but another? Children come biologically equiped to vocalise,
> produce and recognise particular classes of speech sounds (Mehler,
> 1981, C.J Darwin, 1987). It is thought that the `capacity to
> acquire a language is species specific and genetically transmitted
> which depends on that acquired by maturation and socialisation'
> (Chomsky). This biological view has strong reasoning for being
> maintained. Environmental aid may support the innate structure of
> UG acquisition but nature is the number one factor. In conclusion,
> the `poverty of the stimulus' purely exists due to environment's
> unresolved constraints.

Your grip on what the poverty of the stimulus really means seems to
loosen at times. The first two sentences in the above paragraph
contradict one another: Could you explain it now, in a kid-sib style
single paragraph?

This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:24:16 GMT