Font Size:
Language acquisition: speech acts in multi-agent system
Last modified: 2011-06-08
Abstract
Language acquisition is the process by which humans acquire the capacity
to perceive, produce and use words to understand and communicate.
A major concern in understanding language acquisition is how these
capacities are acquired by infants from what appears to be a very small
amount of input information. A range of theories of language acquisition
has been created in order to explain this apparent problem.
On the other hand, agents in multi-agent systems communicate with their
own language, which is already implemented and it is difficult to change
it. Because of this, we can not move agent from one system to another
with different language and expect it will be able to communicate
properly in this new environment.
Because of this, we decided to implement Skinner`s Behavioral theory [3]
into an agent in multi-agent system. Under Behaviorism, it was argued
that language is learned solely through a form of operant conditioning,
namely reinforcement and punishment as Skinner suggested. Reinforcement
learning is defined as learning by trial-and-error from performance
feedback from the environment or an external evaluator [2]. The agent has
absolutely no prior knowledge of what action to take, and has to discover
(or explore) which actions yield the highest reward.
In the first stage of our experiment, we are focusing just on speech acts.
Speech act is an utterance defined in terms of a speaker's intentions
and the effects it has on a listener. According to Austin [1], there are
three main types of acts:
* locutionary acts - the actual utterance and its ostensible meaning,
* illocutionary act - the semantic of the utterance, its real, intended meaning,
* perlocutionary act – actual effect (intended or not).
For simplicity of the experiment, language used in agents is composed of
simple mathematics operators (add, multiply, etc), numbers and few
letters (used as variables). We are expecting that agent with no a priori
knowledge of the speech acts in the system will be able to learn them and
be able to use a proper speech act to achieve demanding effect. Therefore,
the agent moved to the different multi-agent system will learn new
speech acts just from communicating with other agents in the system.
[1] J.L. Austin, A.R. White, and J.O. Urmson, How to Do Things with Words, Harvard University Press, 1963.
[2] A.P. Engelbrecht, Computational Intelligence, Chichester, UK: John Wiley & Sons, Ltd, 2007.
[3] B.F. Skinner, Verbal behavior., East Norwalk, CT, US: Appleton-Century-Crofts, 1957.
to perceive, produce and use words to understand and communicate.
A major concern in understanding language acquisition is how these
capacities are acquired by infants from what appears to be a very small
amount of input information. A range of theories of language acquisition
has been created in order to explain this apparent problem.
On the other hand, agents in multi-agent systems communicate with their
own language, which is already implemented and it is difficult to change
it. Because of this, we can not move agent from one system to another
with different language and expect it will be able to communicate
properly in this new environment.
Because of this, we decided to implement Skinner`s Behavioral theory [3]
into an agent in multi-agent system. Under Behaviorism, it was argued
that language is learned solely through a form of operant conditioning,
namely reinforcement and punishment as Skinner suggested. Reinforcement
learning is defined as learning by trial-and-error from performance
feedback from the environment or an external evaluator [2]. The agent has
absolutely no prior knowledge of what action to take, and has to discover
(or explore) which actions yield the highest reward.
In the first stage of our experiment, we are focusing just on speech acts.
Speech act is an utterance defined in terms of a speaker's intentions
and the effects it has on a listener. According to Austin [1], there are
three main types of acts:
* locutionary acts - the actual utterance and its ostensible meaning,
* illocutionary act - the semantic of the utterance, its real, intended meaning,
* perlocutionary act – actual effect (intended or not).
For simplicity of the experiment, language used in agents is composed of
simple mathematics operators (add, multiply, etc), numbers and few
letters (used as variables). We are expecting that agent with no a priori
knowledge of the speech acts in the system will be able to learn them and
be able to use a proper speech act to achieve demanding effect. Therefore,
the agent moved to the different multi-agent system will learn new
speech acts just from communicating with other agents in the system.
[1] J.L. Austin, A.R. White, and J.O. Urmson, How to Do Things with Words, Harvard University Press, 1963.
[2] A.P. Engelbrecht, Computational Intelligence, Chichester, UK: John Wiley & Sons, Ltd, 2007.
[3] B.F. Skinner, Verbal behavior., East Norwalk, CT, US: Appleton-Century-Crofts, 1957.