A self-organizing spatial vocabulary
Summary written by: Yuval Aharoni (9/5/98)
The main problem
The main issue discussed in this article, is the
problem of how 'agents' within a society
learn the specific language of their society.
The paper begins by introducing Chomskey's theory
of universal grammar as one of the
possible explanations for this phenomena, It
then suggests a different approach to the problem based on a process of
self-organization, and finally describes an Alife experiment
demonstrating this second approach by showing
how a "spatial vocabulary" self-emerges within a society of artificial
agents.
Chomskey's theory of universal
grammar
Observing the complexity and richness of the human
language led Noam Chomskey to suggest the following mechanisms as the means
to language learning and usage:
-
The human brain contains complex hard-coded mechanisms implementing a "universal
grammar" capability.
-
The actual learning of a new language, reduces to setting the values
of a certain number of parameters within the universal grammar, thus
concretizing it with accordance to the actual language used in the society.
This theory is widely accepted, and many experimental results support it.
specifically, it has been shown that the human brain has a few specific
areas (not found in other mammals) which are physically detached from what
are thought to be the general problem-solving and intelligence areas, which
play an important role in language production and understanding.
Another conclusion following from this theory, is that language itself
becomes more complex via a process of direct evolution of the above mechanisms.
The self organization alternative
The alternative suggested in this paper states that
complex universal mechanisms are not necessary to explain human language:
-
Each agent could in fact implement a simple language
production/understanding mechanism.
-
Each agent can only control his own language, which
will gradually change in response to local interactions with nearby agents.
-
This state of many local interactions, with a strong
survival pressure to 'improve' the language will cause a coherent complex
language to self-emerge within the system.
An important consequence of this approach is that
language will become richer and more complex via 'learning' during the
life of the agents and not evolution of any hard-coded brain mechanisms.
An A-Life experiment
The plausibility of self-organization is demonstrated
in the paper by a simple experiment:
-
A 2 dimensional grid is defined as an artificial
world.
-
A set of agents with (x,y) coordinates are scattered
around.
-
The agents are capable of 'talking' to each other
conveying as meanings their own names, and spatial 2-dimensional relations.
-
Couples of agents are randomly chosen, and forced
to have dialogs in which each agent, with the language he 'knows' thus
far tries to describe a 3rd agent to the other one.
-
In the initial stages of the experiment the agents
are allowed to 'point' on objects, in a meta-language they all share, to
give the different meanings a chance to propagate through the society.
-
In the end of each dialog, each agent performs changes
in his set of word-meaning associations according to the 'success' of his
dialog. (Thus the urge for better language is directly simulated).
Monitoring the behavior of this environment, under
varying sets of parameters, indeed revealed the formation of coherent all-embracing
vocabulary within a few hundreds generations.
Summary and remarks
The paper manages to show that self-organization
could occur in some aspects of human language. However, viewing it as contrary
to chomsky's ideas is probably too far-fetched.
A better approach would be, in my opinion, to
view this process of self organization as part of the parameter-setting
suggested by chomskey.