IMPLEMENTATION OF GROUNDED LANGUAGE ACQUISITION IN ERNEST

Diksha Gupta

MOTIVATION

Broadly speaking there are currently two main schools of language acquisition theories:

● Generative grammar as proposed by Chomsky.

It proposes that there is an innate universal grammar, which forms the scaffolding for learning any specific language. As the phonemes, words associated to a specific language are learnt they are mapped onto the innate grammar to produce meaningful utterances. Owing to its innate nature, such a grammar is not ontogenetically learnt.

● Usage­-based theory as proposed by Tomasello

This theory proposes that language acquisition occurs through contextual language use. It attributes language acquisition to infants’ cognitive abilities of pattern­finding, categorizing, and intention inference along with their innate predisposition towards social communication. The individual words are learnt by abstracting the repeated co­occurrence of words and their contextual object/action in parents’ speech or any verbal stimuli.. And the grammar emerges from patterns of use of multi­unit utterances.

The lack of convincing evidence in favour of universal grammar has convinced me in favour of Tomasello’s theory of usage­based language acquisition. However, most of the current computational models of language acquisition neglect the contextual semantics associated with words and define them symbolically for the artificial agent. Therefore, I intend to simulate sensory­motor grounded language acquisition in an artificial agent.

PROPOSAL

I propose to stimulate grounded language acquisition in ERNEST (an artificial agent that implements emergent learning phenomena). The primary aim being, implementation of phenomenon by which ontological language acquisition occurs in infants. This model shall ground language in motor acts and sensory percept of touch/vision. The agent will learn and ground language while interacting with the environment.

This project might be relevant as:

● Unlike most of the current models of language which are not grounded, this agent in this model shall have lexical grounding.

● The agent Ernest, with an added ability of grounded language acquisition will be a comprehensive model of early developmental learning in infants.

A Brief Introduction to ERNEST

ERNEST is an artificial agent simulated by Georgeon, Ritter. It implements the emergent learning phenomena as has been theorized in natural organisms. It is capable of self­organising behavior due to intrinsic motivations, dynamical hierarchical learning, action­directed reinforcement and its ability to autonomously simulate course of action through inhibition. The name, Ernest itself is an acronym for Evolutionist Pragmatic Self­Oriented Learning in a Constructivist and Bottom­up Approach.

METHODOLOGY

Ernest will be made to move through a simple rectangular grid. It will be trained with different sentences describing its actions as it turns/moves/senses the grid. Through statistical weighting of the co­occurrence of its sensory/motor state and the words used in the sentences, it will learn the words and associated semantics.

REFERENCES

● Cowie, Fiona, "Innateness and Language", The Stanford Encyclopedia of Philosophy (Summer 2010 Edition), Edward N. Zalta (ed.), URL = .

● Tomasello, Michael. "First steps toward a usage­based theory of language acquisition." Cognitive Linguistics 11.1/2 (2000): 61­82.

● Olivier L. Georgeon, Frank E. Ritter (2011). An Intrinsically­Motivated Schema Mechanism to

Model and Simulate Emergent Cognition. Cognitive Systems Research, doi:10.1016/j.cogsys.2011.07.003

● Roy, Deb. "Towards Visually­Grounded Spoken Language Acquisition."Proceedings of the 4thIEEE International Conference on Multimodal Interfaces. IEEE Computer Society, 2002.

● Roy, Deb. "Grounded spoken language acquisition: Experiments in word learning." Multimedia,

IEEE Transactions on 5.2 (2003): 197­209. ● http://e­ernest.blogspot.in/