A PRODUCTION SYSTEM MODEL OF FIRST LANGUAGE ACQUISITION 
Pat Langley 
Department of Psychology 
Carnegie-Mellon University 
Pittsburgh, Pennsylvania USA 15213 
Abstract 
AMBER is a model of first language 
acquisition that improves its 
performance through a process of error 
recovery. The model is implemented in 
ACTG, an adaptive production system 
language. AMBER starts with the ability 
to say only one word at a time, but adds 
rules for inserting additional words in 
the correct order, based on comparisons 
between predicted and observed 
sentences. These insertion rules may be 
overly general and lead to errors of 
commission; in turn, these lead to more 
conservative rules with additional 
conditions. AMBER's learning mechanisms 
account for many of the developments 
observed in children's speech. 
Introduction 
The acquisition of language has been 
a popular topic among researchers in 
Artificial Intelligence. Impressive 
language learning programs have been 
developed by Siklossy \[i\], Hedrick \[2\], 
Anderson \[3\], Selfridge \[4\], and Berwick 
\[5\]. The generality and power of these 
systems vary greatly, but they share one 
characteristic: none of the programs 
provide a psychologically plausible 
model of children's language learning. 
In this paper I describe the 
beginnings of a more realistic model of 
first language acquisition. This model 
is called AMBER, an acronym for 
Acquisition Model Based on Error 
Recovery. As its name implies, the 
model simulates the incremental nature 
of the child's language learning 
process. AMBER is concerned with the 
production component of children's 
speech, since most of the reliable data 
relate to production rather than the 
understanding process. 
Below I summarize the major 
developments found during this period. 
After this, I present an overview of 
ACTG, the production system language in 
which the model is stated. Next I 
consider some assumptions about the 
child's linguistic knowledge at various 
stages during the learning process. 
After considering the initial and final 
stages of AMBER, I discuss the learning 
mechanisms leading to the transition 
process. Finally, I consider the 
limitations of the model and propose 
directions for future research. 
The Mayor Phenomena 
Children do not learn language in an 
all-or-none fashion. They begin their 
linguistic careers uttering one word at 
a time, and slowly evolve through a 
number of stages, each co~taining speech 
more like that of the adult than the one 
before. In this section I discuss the 
features of the three stages which AMBER 
attempts to explain. I discuss these 
stages in their order of occurrence, 
dealing only with the major phenomena in 
each case. 
The One-Word Stage 
Around the age of one year, most 
children begin to produce words in 
isolation, and continue this strategy 
for some months. Presumably tile child 
spends much of this period connecting 
particular words to particular concepts; 
once this has been done, he can produce 
these words under the appropriate 
circumstances. AMBER does not attempt 
to explain the word-learning process. 
Like Anderson's LAS \[3\], it assumes that 
links between words and concepts have 
already been established. 
Bloom \[6\] has examined this period in 
detail, with an eye to understanding the 
relation between the one-word stage and 
those which follow it. Early in this 
stage, successive one-word utterances 
seem entirely disconnected; the child 
randomly comments on anything that 
happens to be in the environment. 
Later, he begins to name in succession 
different aspects of the same event or 
object; words are still separated by 
noticeable pauses and no regular order 
can be detected, but conceptual 
continuity seems present. Moreover, 
this development occurs only a few 
months before the child begins to 
combine words into very simple 
sentences. AMBER's starting point lies 
somewhere within this later part of the 
one-word stage. 
-183 
Teleg \[.#phi c Speech 
Around the age of 18 months, the 
child begins to combine words into 
meaningful sequences. In order-based 
languages such as English, the child 
usually follows the adult order. 
Initially only pairs of words are 
produced, but these are followed by 
three-word and later by four-word 
utterances. The simple sentences 
occurring in this stage consist almost 
entirely of content words. Brown \[7\] 
has described speech during this period 
as telegraphic, since g rammatical 
morphemes such as tense endings and 
prepositions are absent, as they would 
be in a telegram. 
Brown has also noted that the 
majority of two-word utterances express 
a rather small set of pairwise semantic 
relations. AMBER assumes a small number 
of case relations such as agent, action, 
and possession from which Brown's 
pairwise relations can be derived. In 
addition, tile child uses a few function 
words like "there", "more", and 
"all-gone" to express simple forms of 
nomination, recurrence, and negation. 
AMBER attempts to learn the relative 
word orders for expressing these 
recurring relations. 
The Acquisition of Grammatical Morphemes 
Brown \[7\] has also studied the period 
from about 24 to 40 months, during which 
the child masters the grammatical 
morphemes which were absent during the 
previous stage. Brown pointed out that 
these morphemes modulate the major 
meanings of sentences which are 
expressed through content words. AMBER 
reflects this distinction by 
representing the information expressed 
by contents words and grammatical 
morphemes in different ways. These 
morphemes are learned gradually; the 
time between the initial production of a 
morpheme and its mastery (i.e., when it 
is correctly used in all required 
contexts) may be as long as 16 months. 
In addition, Brown has examined the 
order in which 14 English morphemes are 
acquired, and has found this order to be 
remarkably consistent across children. 
For example, present progressive 
(eating) and plural (dogs) were always 
learned quite early, while third person 
singular (eats) and copulas (is, are) 
took longer. He found that the 
syntactic and semantic complexities of 
the morphemes were highly correlated 
with their order of mastery. Since the 
current version of AMBER cannot deal 
with exceptions, I will consider only 
regular constructions in this paper. 
The ACTG Formalism 
AMBER is implemented in ACTG, an 
adaptive production system language. 
Below I present an overview of AC%G, 
beginning with a discussion of its 
propositional network. After this I 
consider the representation of 
procedures as productions. Finally, I 
examine ACTG's facilities for changing 
its own behavior through the creation of 
new productions. 
The Propositional Network 
ACTG stores its factual, declarative 
knowledge in a long-term propositional 
network. Individual facts are stored as 
propositions, which may be arbitrary 
list structures. As we will see in more 
detail below, AMBER incorporates two 
main types of propositions. One sort 
expresses a goal to say a particular 
word in a certain position. The second 
type of proposition expresses various 
kinds of relations, including facts like 
x possesses y, y is a *ball, and "ball" 
is the word for *ball (where concepts 
are preceded by "*" to distinguish them 
from their associated words). 
At any given time, some subset of the 
propositional network is active. Many 
of the active propositions have been 
recently added to the network by 
productions. Others, after lying 
dormant for a time, have been 
reactivated through their association 
(i.e., sharing of symbols) with other 
recently activated facts. AMBER uses 
this process of spreading activation 
primarily to retrieve information about 
the words associated with particular 
concepts. The level of activation for a 
proposition naturally decays over time, 
unless it is offset by other factors. 
The Production System 
ACTG represents procedural knowledge 
as a set of condition-action rules 
called productions. The conditions and 
actions of these rules can be quite 
general, since they may contain 
variables that match against arbitrary 
structures. When all the conditions of 
a production match against some portion 
of active memory, its actions may be 
carried out. These may interact with 
the environment, or add new propositions 
to the active part of the network. 
Structures matching variables in the 
conditions remain bound to these 
variables in the actions. After a 
production has been applied, the state 
of memory is reexamined and the system 
cycles. 
184- 
If two or more productions are found 
to be true, one must be selected in 
preference to the others. This decision 
is based on the relative strength ~ of 
the productions, and on the summed 
activations of the propositions matched 
by each. The product of these two 
numbers is computed, and the production 
with the highest value is selected. 
Since a single production can match 
against a set of propositions in 
different ways, ties may sometimes 
occur. In such cases, one of the 
matches is selected at random. 
The ACTG Learning Mechanisms 
ACTG incorporates a powerful set of 
mechanisms for modeling learning 
phenomena. The most basic of these is 
the designation process, which allows 
the creation of a new production as one 
of the actions of an existing rule. 
Variables bound in the conditions of the 
learning rule are passed to the 
offspring, making the new rule more 
specific than its creator. Most of 
AMBER'S learning heuristics rely on the 
designation process. 
A second mechanism leads to the 
strengthening of a production each time 
it is recreated. Since the strength of 
a rule plays an important role in the 
selection phase, productions which have 
been relearned many times will be 
preferred. On the other hand, the 
strength of a rule can be decreased if 
it leads to an error, lowering its 
chances for selection. 
The discovery of an error also leads 
to a call on the discrimination process. 
Here the recent firings of the 
responsible production are examined. If 
one or more propositions have been 
present at successful firings and absent 
at faulty ones, they are added as extra 
conditions on a new, more conservative 
version of the rule. Together with the 
strengthening and weakening processes, 
this mechanism gives ACTG the ability to 
recover from overgeneralizations. 
AMBER's Linguistic Knowledge 
Learning is the result of an 
interaction between a set of relatively 
general techniques for acquiring 
knowledge and the environment in which 
they find themselves. In this section I 
consider AMBER's representation of that 
environment. After this I examine the 
procedures the model assumes at the 
outset, as well as the form of the rules 
at which it eventually arrives. 
Representing Sentences 
Before AMBER can learn how to 
generate legal sentences, it must be 
exposed to examples of such sentences. 
One might represent a sentence as a 
simple list of words in the order they 
are said. However, though children 
learn to produce words in the correct 
order very early on, they also omit many 
words that an adult would include. For 
example, the utterance "Daddy ball" 
omits information about the action being 
carried out, as well as tense 
information. AMBER's representation of 
the sentences it hears reflects this 
ability to note order in the absence of 
information about adjacency. 
The model represents the occurrence 
of each morphem e as a separate 
proposition, each containing information 
about the speaker, the word being 
produced, and the relative order of 
occurrence. Thus, the fact that Mommy 
said the sentence "Daddy bounces the 
ball" would be stored as a set of seven 
propositions: (said 1 Mommy pause); 
(said 2 Mommy Daddy); (said 3 Mommy 
bounce); (said 4 Mommy s) ; (said 5 Mommy 
the); (said 6 Mommy ball); and (said 7 
Mommy pause). The first and last 
propositions act as delimitors which 
mark the beginning and end of the 
sentence. This representation, combined 
with ACTG's pattern-matching capability, 
allows the statement of learning rules 
which focus on relative word order but 
ignore adjacency information. The 
resulting production rules omit words, 
just as the child does. 
Representing Meaning 
Adults conversing with a child almost 
invariably discuss recent or ongoing 
events, so that the child can associate 
some event with every sentence he hears. 
The language acquisition process does 
not consist solely of learning to 
produce or parse legal word 
combinations; it consists of learning 
the mapping between meanings and words. 
Accordingly, AMBER is presented not with 
isolated sentences as its data, but with 
sentence/meaning pairs. 
AMBER represents the meaning of a 
sentence as a number of propositions, 
each incorporating one of a small set of 
relations. The most prevalent of these 
is the type relation, which connects 
tokens to the various concepts of which 
they are examples. There is no 
restriction on the number of type 
relations which may come off a token; 
-185--- 
thus, the propositions (token-i type 
red) and (token-i type ball) state that 
the object token-i is both red and a 
ball. Events are represented with 
relations such as a~, action, and 
object. The propositlons (event-i agent 
token-2), (event-i action token-3), and 
(event-i object token-4) represent an 
event with an agent, action, and object 
whose types have yet to be specified. 
AMBER's representation makes a strong 
distinction between the main meaning of 
a sentence as expressed through its 
content words and the modulations of 
this meaning as expressed through its 
grammatical morphemes. The model 
assumes that a type relation pointing to 
a particular concept (e.g., *ball) is 
present for every content word (e.g., 
ball) found in the associated sentence. 
Moreover, a word-for relation is assumed 
present to establish the connection 
between word and concept. The presence 
of these two relations tells AMBER when 
a word contributes to the major meaning 
of a sentence. 
Modulations on this meaning are 
represented by a different set of 
relations, such as number, 
time-of-action, possession, and so 
forth. Some of these relations connect 
tokens to various values, as in token-i 
number singular) and token-2 
time-of-action past). Others, as in 
(token-4 possesses token-5) and token-5 
in token-6), actually relate tokens. 
AMBER's Initial Performance System 
AMBER starts with the ability to 
produce single words in isolation. But 
even at this stage, the model draws on a 
set of general heuristics for generating 
utterances which will still be useful 
after its learning is complete. AMBER 
does not say words as soon asthey come 
to mind; first there is an active 
planning stage during which sequential 
goals are set. 
The model starts with rules for 
initializing and ending this planning 
phase, and for implementing its plans 
once they are complete (that is, 
actually saying the words in the planned 
order). The goals which result from the 
planning process look very like the data 
from which AMBER learns. The two-word 
utterance "Daddy ball" would be 
represented by the propositions (goal 1 
AMBER pause), (goal 2 AMBER Daddy), 
(goal 3 AMBER ball), and (goal 4 AMBER 
pause), in which the model is the 
speaker. 
At the outset, AMBER has only a 
single rule for inserting such goals in 
memory; stated in English for the sake 
of clarity and with its variables 
underlined, it is: 
If you have no goals yet, 
and you see vtoken with type v type, 
and vword is the word for vtype, 
then set up a goal to pause, 
followed by a goal to say vwo~d, 
followed by a goal to pause. 
This rule separates the goal utterance 
from others by initial and final pauses. 
Thus, even though successive words may 
describe different aspects of the same 
event, they will be separated by 
noticeable gaps just as Bloom observed. 
Only after additional rules have been 
formed for inserting sounds between the 
initial word and the pauses can 
multi-word utterances begin to occur. 
AMBER at Later Stages 
On the basis of comparisons between 
sentences it hears and those it 
predicts, AMBER creates and modifies 
rules for saying multiple words at a 
time. These rules lead to the insertion 
of new goals between existing ones. 
Thus, they are dependent on the innate 
rules described above for initializing 
the goal insertion process and for 
carrying out goals once they have been 
set. 
Imagine a situation in which AMBER 
sees Daddy bouncing a ball. Also 
suppose that the one-word rule we saw 
above happens to select "bounce" as the 
word that should be said. This would 
lead to three goals: (goal 1 AMBER 
pause), (goal 2 AMBER bounce), and (goal 
3 AMBER pause). After some experience 
with English, the model will have 
generated a rule like: 
If you have a goal to pause, 
followed by a goal to say vword2, 
and you have no intermediate goals, 
and vword2 is the word for vtype2, 
and vtoken2 is of type vtype2, 
and vtoken2 is the action of vevent, 
and vtokenl is the agent of vevent, 
and Vtokenl is of type vty~el, 
and vwordl is the word for vtypel, 
then insert a goal to say vwordl 
between the other goals-- 
This rule would add a goal to say the 
agent "Daddy" after the first pause and 
before "bounce", using the proposition 
(goal 1.5 AMBER Daddy). Similar rules 
lead to the production of two- and 
three-word sentences expressing the 
major relations described by Brown. 
-186 
Later, AMBER also acquires rules for 
inserting grammatical morphemes. Since 
most grammatical morphemes are adjacent 
to the word whose meaning they modulate, 
they are generally inserted directly 
before or after the content word with 
which they occur. For example, a rule 
for regular pluralization might be 
stated: 
If you have a goal to say vword, 
and vword is the word for vtype, 
and vtoken is of type vtype, 
and vtoken is the agent of vevent, 
and the number of vtoken is plural, 
then insert a goal to say S 
directly after vword 
This rule is specific to the agent of an 
event, but similar rules could be 
learned for objects and locations. Some 
morphemes express a relation between two 
content words, such as the prepositions 
"in" and "on" and the morpheme for 
possession. In these cases, the 
morpheme is inserted between the two 
related content words. 
The Acquisition Process 
For a system to learn from its 
mistakes, it must be able to compare its 
own actions to the desired ~ones, note 
the differences between them, and modify 
its behavior accordingly. In this 
section I describe AMBER's error 
correction mechanisms. First I examine 
the model's prediction mechanism and its 
relation to the goal structures 
mentioned earlier. Next I discuss 
AMBER's response to errors of omission, 
first for content words and then for 
grammatical morphemes. Finally, I 
consider errors of commission and the 
resulting call on the discrimination 
mechanism. 
The Equivalence of Goals and Predictions 
AMBER learns by comparing its 
predictions about what will be said in a 
given situation to what it actually 
hears. However, a learning system must 
do more than improve its ability to 
predict; it must also improve its 
ability to perform. AMBER accomplishes 
this by using the same productions for 
making predictions and for planning its 
speech acts. As we saw above, these 
rules add goal structures such as (goal 
3 AMBER bounce) when AMBER is the 
speaker. When another person is the 
speaker, the resulting structures, such 
as (goal 3 Mommy bounce) if Mommy is the 
speaker, are treated as predictions 
instead of goals. Learning occurs only 
when someone else is speaking and the 
system is in prediction mode, while 
sentences are produced only in 
performance mode. 
Correcting Content Word Omissions 
AMBER's transition from the one-word 
to the multi-word stage is primarily due 
to the actions of a single learning 
heuristic. This rule applies when a 
content word is heard between two other 
words (or pauses) but was not predicted 
there; the result is a new performance 
rule for inserting analogous words in 
analogous positions in the future. 
AMBER knows enough about the nature of 
language to generalize across the 
particular words and concepts involved. 
However, it retains information about 
case relations and shared tokens (e.g., 
two of the words may have described 
aspects of the same object). 
As an example, suppose AMBER sees 
Daddy bouncing a ball. The model 
predicts the one-word sentence "Daddy" 
(preceded and followed by pauses), while 
it actually hears "Daddy is bounce ing 
the ball" (again bounded by pauses). 
Since the grammatical morphemes "is", 
"ing", and "the" are not connected to 
concepts by word-for links, they are 
ignored by the current learning 
heuristic. However, the words "bounce" 
and "ball" each have associated concepts 
which occurred in the observed event. 
An insertion rule is created for each, 
the first inserting the action word 
after the agent word and before the 
final pause. The second is very 
similar, inserting the object word after 
the agent and before the pause. 
These rules give AMBER the abiiity to 
generate agent-action and agent-object 
combinations, but no more. The new 
rules cannot cooperate to produce 
agent-action-object combinations, for 
once one of the rules has fired, the 
conditions of the other are no longer 
met. But once this has happened, the 
system can learn additional insertion 
rules, such as that for inserting the 
object word between the action and the 
final pause. Yet even after this has 
occurred, the performance rules are 
dependent on the selection of the agent 
word as the initial goal. Additional 
insertion rules must be learned to deal 
with cases in which the action or object 
is the first goal to be inserted. 
- 187-- 
Correcting Morpheme Omissions 
As it is improving its ability to 
produce strings of content words, AMBER 
is also learning to insert grammatical 
morphemes. Some of the morphemes which 
modify a single token, such as tense and 
pluralization endings, occur after the 
words describing the token. Others, 
such as copulas (is, are, were) and 
articles (a, the), occur before the 
modified words. Separate learning 
heuristics are necessary for these two 
cases, but there forms are nearly 
identical. 
These learning rules are evoked when 
a particular morpheme is heard before or 
after a content word, but was not 
predicted in that position. The result 
is a performance rule which inserts the 
morpheme either before or after words 
playing similar roles in the future 
(e.g., "ing" after the word for the 
action). AMBER knows that the 
particular content word is irrelevant; 
however, the case relation filled by 
that word and the morpheme are retained. 
AMBER also knows that several content 
words may be used to describe the same 
object (e.g., "the big red ball s") , and 
that these words will occur together in 
any legal sentence. This is analogous 
to an assumption made by Anderson's LAS 
\[3\], which he has called the 9raph 
deformation condition. AMBER incorpor- 
ates this assumption into its morpheme 
insertion rules, ensuring that a 
morpheme will be inserted either before 
the earliest or after the latest content 
word describing a token (thus, "big the 
red s ball" would never be produced). 
The acquisition of relational 
morphemes, such as those expressing 
possession and location, is handled by a 
different rule. This heuristic is 
evoked in the same situations as the 
heuristic for content words, except that 
the unpredicted morpheme must not be 
associated with any concept via a 
word-for link. In addition, the objects 
described by the two correctly predicted 
words must be directly related (e.g., a 
token of milk is on a token of table). 
The resulting performance rule inserts 
the morpheme between the sets of words 
describing objects in the observed 
relation. Note that A~BER cannot 
acquire such relational morphemes until 
it can correctly predict the order of 
the words to be related. 
Correcting Errors of Commission 
Once AMBER has learned a number of 
rules for inserting goals, it can make a 
new sort of error: the model can 
incorrectly predict that a word will 
occur in a certain position. A single 
learning heuristic is sufficient to deal 
with all such errors of commission. Its 
condition is simple, but in addition to 
weakening the offending rule, its action 
calls on the discrimination mechanism to 
produce a more conservative variant. 
To reiterate, this technique compares 
the last successful application of a 
rule to the more recent faulty 
application in the hope of finding 
additional conditions to constrain it in 
the future. Thus, if AMBER predicted 
the sequence "ball red" when "red ball" 
was heard, the discrimination process 
would be evoked. Comparing this case to 
an earlier one in which "blue block" was 
correctly predicted, AMBER would note 
that "blue" is a color while "ball" is 
not. Thus, the new rule would insert 
one word before another describing the 
same object only if the former were a 
color like "blue" or "red" 
Although discrimination is useful in 
learning some content word orders, its 
major import lies with grammatical 
morphemes. Since the initial rules for 
nonrelational morphmemes are too 
general, their use quickly leads to 
wrong predictions. For instance, if the 
morpheme "ed" was incorrectly expected 
to follow an action word, AMBER would 
note that correct predictions of "ed" 
occurred only when the time of the 
action was past. Similarly, AMBER would 
quickly learn to insert "s" after the 
word for the agent only when its number 
was plural. 
The conditions under which some 
morphemes are applied can be more 
complicated. Thus, the morpheme "were" 
is inserted before the action word only 
when the time of the action is past, and 
when the nu~er of the agent associated 
with that action is plural. AMBER would 
be forced to learn these conditions in 
two stages, first creating a variant 
with one condition and later a version 
including both. 
As a result, the more complex the 
conditions under which a morpheme 
occurs, the longer AMBER will take to 
master its use. If one equates the 
188 
number of conditions with semantic 
complexity, then the discrimination 
process provides an elegant explanation 
of Brown's data on the order of 
acquisition for grammatical morphemes. 
Semantically more complex morphemes are 
mastered later because they require more 
conditions, and these conditions can be 
learned only one at a time. 
Suggestions for Future Research 
In summary, AMBER does a fine job of 
accounting for the major phenomena 
described at the beginning of the paper. 
However, the model makes a number of 
simplifying assumptions and stops 
improving after it has reached a certain 
level of expertise. In this section I 
suggest some directions in which AMBER 
should be extended. 
Learning Word/Concept Associations 
AMBER assumes that words and concepts 
are already connected through word-for 
links stored in long-term declarative 
memory. These connections play an 
important role in letting the system 
distinguish between content words and 
grammatical morphemes. At least some of 
these connections must be present before 
any ordering rules can be learned, but 
the model provides no explanation of 
their origin. Extending AMBER to let it 
make its own word/concept associations 
is clearly a direction for future work. 
Selecting a Representation 
AMBER relies heavily on the 
representational distinction between 
major meanings (expressed by type links) 
and modulations of those meanings 
(expressed by others). Unfortunately 
for the model, some languages express 
through content words what others 
express through grammatical morphemes. 
This suggests that the child does not 
start with a representation like 
AMBER's, though it may arrive at the 
same point as the result of experience 
with a particular language. Future 
research should consider how the child 
comes to treat some meanings as major 
and some as minor as a function of his 
native language. 
Dealing With Exceptions 
In its current version, AMBER cannot 
deal with irregular grammatical 
constructions. Some past forms, such as 
"ate", require a special word for past 
events, but this cannot be expressed in 
the current formalism for word/concept 
associations. Some plural forms r.equire 
different endings than most, such as 
"oxen". These can be expressed in 
production form, but no conditions exist 
to distinguish these situations from the 
majority. Future versions of AMBER 
should have extended representations 
which address these issues. 
Explaining Later Stage s 
AMBER's progression stops after it 
has mastered the grammatical morphemes. 
It never learns how to ask questions, or 
how to generate sentences with relative 
clauses. In fact, to present the model 
with other than simple declarative 
sentences would be an invitation to 
disaster. Future incarnations of the 
system should begin with the basic 
notions of recursion and transformation. 
Coupled with the existing learning 
mechanisms and the extensions discussed 
above, this should allow AMBER to 
progress far beyond its present level of 
expertise, and to become a true language 
user. 
Acknowledgements 
This research was supported by Grants 
SPI-7914852 and IST-7918266 from the 
National Science Foundation. I would 
like to thank John R. Anderson for 
useful discussions which led to many of 
the ideas presented in this paper. 

References 

\[i\] Siklossy, L. Natural language 
learning by computer. In H. A. Simon 
and L. Siklossy (eds.) , Representation 
and Meaning: Experiments with Infor- 
mation Processing Systems. Englewood 
Cliffs, N. J.: Prentice-Hall, 1972. 

\[2\] Hedrick, C. Dissertation, Carnegie- 
Mellon University, 1974. 

\[3\] Anderson, J. R. Induction of 
augmented transition networks. Cogni- 
tive Science, 1977, i, 125-157. 

\[4\] Selfridge, M. Dissertation, Yale 
University, 1979. 

\[5\] Berwick, R. Dissertation, Massa- 
chusetts Institute of Technology, 1980. 

\[6\] Bloom, L. One Word at a Time. The 
Hague: Mouton, 1976. 

\[7\] Brown, R. A First Language : The 
Early Stages. Cambridge, Massachusetts: 
Harvard University Press, 1973. 
