Some Background for Experience Based Language Acquisition
From the abstract:
The future of human-computer interaction (HCI) may well depend upon the ability of humans and computers to communicate using natural language. While much has been done in the field of computational linguistics to bridge the gap between computers and natural language, traditional top-down approaches usually do not attempt to ground the context of language in sensory perception. Thus an “apple” to a computer is only logically defined in terms of other words, not sensory information.
This paper investigates how a computer might be enabled to acquire some form of natural language that is grounded in perception using a bottom-up approach. A model is proposed for establishing a hierarchical conceptual system based on objects, schemata, and event representations. This model will be grounded in a series of short movies that will be analyzed with a Java-based computer vision system. Cross-situational and probabilistic learning are discussed as approaches for integrating a lexicon into the newly formed conceptual system.
del.icio.us
Digg
Facebook
StumbleUpon
Comments
Re: Some Background for Experience Based Language Acquisition
Thanks for posting and keep up the good work. I look forward to reading more of your dissertation -- particularly the background and the details of your own approach.
Of course, the introduction only raises many more questions than it answers :) For which parts of the system have you settled on initial algorithms, for which parts of the system do you have vague plans, for which parts is everything still wide open?
Re: Some Background for Experience Based Language Acquisition
The dissertation is progressing well. I've written about 50 pages and now I'm in the trenches coding. It the this phase that is actually letting me know what works and what doesn't, but so far, so good. I'll definitely post more of the disseration and some code as things progress.