Re: The Poverty of the Stimulus

From: Stevan Harnad (
Date: Sat May 25 1996 - 21:17:16 BST

> Date: Thu, 18 Apr 1996 23:04:21 +0100 (BST)
> From: "E.J.Fletcher" <>
> Evidence suggests that all languages are founded around a 'structure'
> which is innate and common to all humans; as such, mankind possesses
> some form of linguistic system which enables distinction between
> 'grammatical' and 'ungrammatical' phrases from birth. (The term
> 'grammatical', in this context, refers to whether a vernacular phrase
> conforms to these innate, and universal, rules of grammar).

And hence whether native speakers would judge it as grammatical or not.

> "It holds that beyond the peripheral processing mechanisms, there are
> innate ideas and principles of various kinds that determine the form of
> the acquired knowledge, in what may be, a rather restricted, and highly
> organised way. A condition for innate mechanisms to become activated is
> that appropriate stimulation be presented". Chomsky (1965).

But remember, this is neither a definition nor an assumption of
Chomsky's: It's a conclusion he drew from the evidence, and the only
available theoretical explanation for it.

> Therefore it is thought that, from birth, the infant is equipped with
> a fixed, comprehensive 'database' of universal grammatical rules, from
> which are extracted those which are appropriate to his language.
> Chomsky terms the 'database' "Universal Grammar" (UG), and the
> selection process, which occurs on exposure to a language, "parameter
> setting". Evidence for the existence of this innate structure is
> supported by the theory's ability to explain how language develops
> under the "poverty of the stimulus".
> The stimulus is, in effect, the information that the child can
> extract from the surrounding environment concerning language. The child
> might gain information about grammatical structure through a process of
> listening and attempted interaction, in which trial-and-error, and
> feedback from others in the environment, would allow learning to occur.
> This would require the child to be exposed to both grammatical and
> ungrammatical phrases (or negative evidence). The distinction between
> correct and incorrect syntax could then be made. However, the child
> never hears or produces fundamentally ungrammatical phrases. This
> makes the feedback and trial-and-error methods void, as they rely on
> both positive and negative stimuli. Infringements of universal grammar
> would therefore have to exist for universal grammar to be learnt in
> this way. As such the stimulus is 'impoverished', or incomplete.


> Then what explanation can be given for the child's ability to
> learn language? The only possible explanation is that the underlying
> principles of language exist within the mind before birth. "What the
> stimulus lacks is produced by the organism from its inner (genetically
> endowed) resources" Chomsky (1980). Approaches to language which do not
> consider this to be the case e.g. empiricism, must provide evidence
> which shows the existence of both positive and negative stimuli within
> the child's environment; and in doing so, show the poverty of the
> stimulus to be a fallacy.

Correct again.

> Lauria's 1975 belief that such an internal
> system of grammatical explanation could not exist, as "Universal
> Grammar makes a postulate out of a problem", is thus ill- founded.
> Although UG is based on something that has not been proven , evidence
> may be presented in the form of the poverty of the stimulus, to suggest
> that such a faculty exists.

Scientific theories are not proven; only mathematical theorems are
proven. Theories are based on evidence. UG is based on evidence. It does
not have to be observed in the brain or the genes in order to exist (or
in order for us to be confident that it exists). It need merely explain
the available evidence and have no better rival. (It is behaviouristic
to think that unobserved or even unobservable structures and processes
cannot exist; if hypothesising that they exist allows us to explain
evidence that would otherwise be inexplicable, short of proof, that's as
good a reason as any for accepting the hypothesis is true.)

> "...What we can try to do is to find
> properties for which it is very implausible to assume that everyone has
> had relevant experience" Chomsky (1983). Ironically, Lauria
> continues by saying "...all further study in the area can lead us
> nowhere". This indeed may be accurate, as UG, whose existence is
> suggested through the poor nature of the stimulus, appears to fulfill
> all the criteria for language acquisition. However if the stimulus
> (i.e. the total of all the linguistic information which the child
> receives) was shown to be less impoverished (i.e. the stimulus was
> shown to provide enough information for learning to occur solely
> through exposure to the language, without a inherent linguistic
> structure) Chomsky's theory of Universal Grammar would be made
> redundant. This is not the case: the stimulus may be shown to be poor.

Correct; to reject UG, you must either show that, though UG is the right
structure, the child DOES learn it from positive and negative evidence
after all (this has not been shown), or you must show that, though UG is
unlearnable, something else, instead of UG, something that will also
explain our grammatical ability, exists, and is simpler, and learnable
(this too has not been successfully shown by anyone yet).

> The linguistic information the child receives from social interaction
> is limited. How can such a limited input be manipulated to form such a
> developed array of linguistic skills? Furthermore, the accuracy of the
> stimulus may also vary, as the child may be exposed to errors made by
> the speech community. Surely the child would be tempted to generalise
> mistakes and over apply rules? Confusion over rules should occur where
> the child over generalises. A child exposed to both "the cat seems to
> be asleep" and "the cat seems asleep" would presumably believe that "to
> be" may be dropped from the phrase. However, should this be applied to
> the sentence "the cat seems to be sleeping", the result would be
> ungrammatical. i.e. "the cat seems sleeping".
> Distinctions between
> nouns and verbs may also be expected to become confused in a similar
> way. e.g. when presented with the following stimuli,
> 'Andrew hates flies.' 'Andrew hates dogs.' 'Andrew flies',
> the child should consider 'Andrew dogs' as grammatically correct. It is
> extremely unreasonable therefore to assume that every child will make
> these mistakes and proceed to have them corrected by the speech
> community. Children do not receive negative evidence as they do not
> make the initial errors.

i.e., they never make these errors in the first place; so they already
have UG.

> If the stimulus is impoverished in this
> manner, the most plausible explanation for language acquisition points
> to the existence of a pre-programmed linguistic system in the mind;
> whereby some form of mental process eliminates what seem to be
> 'logical' grammatical rules. The child later uses this innate source of
> knowledge as a foundation from which to expand his linguistic capacity.
> Trial-and-error, and feedback from others, allow innate grammatical
> rules to be utilised with newly learnt rules. For example, the child
> may learn, through trial-and-error, that phrases can be linked by the
> word 'and'. The word "and" did not pre-exist within the mind, it had to
> be learnt. However the grammatical structure in which the learnt word
> becomes applied is inherent.

Correct; the innate rules of UG we never learn by trial and error. What
we learn is which "dial-setting" of UG our own native language happens
to be, and of course we learn particular "style-grammar" whose violation
is social and cultural, but not a violation of UG (like "I ain't").

> The stimulus may be shown to be impoverished in several ways. Firstly,
> the amount of input (linguistic information) the child receives from
> the speech community is too impoverished to explain the child's vast
> linguistic capacity. Similarly, how can an adult's ability to use
> words in an infinite number of ways, be explained by the mimicry and
> correction of a limited number of phrases experienced within his
> lifetime? This would again rely heavily on the existence of negative
> evidence. It seems that if the acquisition of grammar occurred in this
> fashion another problem would immediately arise: who in the speech
> community could have gleaned enough information to provide adequate
> correction for others?

Lack of negative evidence IS a problem. Finiteness is not; given enough
time and positive and negative evidence, we could eventually figure it
out after all, that's what Chomsky and his fellow-linguists are doing
across the years! But a child of four can't do it in so short a time!

Feedback would be no problem, because all adults know whether any
sentence does or does not violate UG, as long as the child generates the
errors -- but the child doesn't.

What adult speakers (aside form linguists) could not do would be to
produce the negative evidence (errors) FOR the child, if the child did not
produce it on its own.

> It would seem almost impossible for the child to
> receive information in this manner. Assuming that 'negative evidence'
> is unobtainable, what prevents the child from creating too large a
> language? After all it would be assumed that without negative stimuli,
> i.e. without some form of correctional restraint, the child would
> assume ungrammatical phrases to be correct. For the same reason that
> ungrammatical phrases are never heard, the child never produces an
> ungrammatical phrase. i.e. an internal system of grammatical rules must
> presumably exist. It would be expected therefore that errors in speech
> rarely occur.

Plenty of errors of many kinds occur, but not violations of UG.

The over-extension/under-extension problem is quite simple. Consider
any rule: I'll use rules for pattern-drawing instead of grammar:
Patterns have to be blue or square. You don't know if that's right or
not. So you draw and see a lot of blue patterns, but no square ones. So
you assume the rule is right. But maybe "blue or square" is wrong,
because if anyone ever drew a square pattern that was not blue, it
would be judged incorrect ("ungrammatical") by everyone; but no one
ever does, so you don't know. That's over-extension. Under-extension is
if the "blue or square" rule is correct, but because you only draw and
see blue patterns, you assume the rule is that all patterns must be
blue. So you would have thought a pattern that was square and not blue
was wrong (if you ever drew or saw it), yet it would not be.

To learn such rules by trial and error, you have to see or draw the
correct and the incorrect patterns and get feedback till you can adjust
the rule so it's neither over- nor under-extended. If you do it right
without needing the feedback, you must know the rule already.

> Pinker (1984) suggests that "errors occur between 0.1%-8%
> of the time in three year olds...the opportunity for making errors is
> more than 90% of the time". Furthermore if correction of such errors
> was solely dependent on reaction from the primary care givers, learning
> would be an arduous process. Feedback in the form of parental reaction
> has been shown to be ambiguous, and difficult for the child to
> understand. Active verbal interaction and feedback from others in the
> environment also appear to be of little importance, as a child who was
> unable to speak was shown to be able to understand complicated
> sentences without difficulty. Stromswold (1994).

True, but the main point is that the critical errors never get made!
The negative cases never happen, because an invisible hand (UG) is
already guiding the child.

> Secondly the quality of the grammatical information presented to
> the child is impoverished, not in the sense that it is degenerate (i.e.
> the stimulus is poorly formed containing errors such as unfinished
> phrases, impromptu pauses, and slips of the tongue) but that the
> stimulus itself is of poor quality. i.e. it does not contain all the
> information needed for all the principles of grammar to be acquired by
> the child. Thus even motherese, simplified 'baby talk' is impoverished:
> and perhaps is even harder for the child to comprehend than the
> language from which it is derived. Motherese may neglect grammatical
> rules and impoverish the stimulus for language acquisition still
> further, despite being grammatically correct 99.93% of the time. This
> form of parent-child communication still fails to provide a stimulus
> which is rich enough to provide an adequate base for language
> acquisition.

The problem is that the real errors never get made by either mother or
child, so the pattern cannot be learnt -- unless it's already built in
in the first place.

> Surprisingly bi-lingual language acquisition in children occurs with
> little confusion. A distinction between two languages is created in the
> absence of information to instruct the child to keep their varying
> grammatical structures separate. The stimulus is impoverished as it
> does not provide enough information for the acquisition of grammar to
> occur. One would naturally expect the child to generalise the
> grammatical rule within each of the separate language. It would also be
> reasonable to assume that the generalisations produced in one language
> would be drawn across into the other language. However once again this
> does not occur. Children have the capacity to learn the nuances of
> grammatical language without instruction or environmental input. This
> suggests an inherent grammatical system. Even severely retarded
> children, whose ability to acquire language from the speech community
> is by nature impaired, show the ability to utilise inborn grammatical
> rules.

Neither the capacity to learn two languages at once nor the language
learning capacity of the severely retarded child is critical: what is
critical is the lack of UG-errors and error correction.

> In conclusion the stimulus may be shown to be impoverished,
> in that the examples which the child would need to observe to allow him
> to learn grammatical rules are incomplete. i.e. negative evidence is
> missing. For the stimulus to be sufficient, negative and positive
> evidence is required. Evidence suggests that such ungrammatical phrases
> do not exist as the speech community do not produce them. Furthermore,
> why is it that the child does not over generalise, or indeed over
> generate rules?


> The poverty of the stimulus naturally calls for an explanation of
> how the grammatical rules of language are acquired. Chomsky provides an
> answer to this through his assumption that "what the organism lacks is
> produced from its inner (genetically endowed) resources". The poverty
> of the stimulus has thus been instrumental in providing strong evidence
> for Universal Grammar.

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