Researches on language(categorization) acquisition with robot-human interactions
Lakoff’s “Women, Fire, and Dangerous Things” mentions difficulty of solving the problems related to categorization. This material reminds me that parents-babies interaction is considered to be one of the essential factors for babies to get their mother tongue by developmental psychologists, and robots are recently often used for researchers to try to find and/or construct a mechanism of how human being learns concepts and/or words through interactions with its parents(or instructors around robots).
In these kind of researches, reseachers try to struggle categorization problems by using specific properties of robots, such as sensory-motor systems(It represents some kind of embodied knowledge. Also, it can represent not only categorization of objects but also categorization of verbs and it is expected to avoid some kind of symbol-grounding problem), and interaction between instructors(parents) and robots(children) with probabilistic learning models(HMM, etc).
Results of these research are still limited. Most of these researches could be valid under a limited “toy” environment, and the robots could get only limited categories and words. However, in my opinion, this approach may be one of useful and interesting methods to find how human being learns concepts, categorization and words, and the results will be able to be applied to categorization problems. Also, human-agent interaction scheme could be applied not only to robots but also to web-agents.
One sample of related researches: Iwahashi, N. (2006). “Robots that learn language — Developmental Approach to Human-Machine Conversations” Proc. of Int. Workshop on Emergence and Evolution of Linguistic Communication, pp.142-179. http://www.slt.atr.jp/~niwaha/Publication/Iwahashi06-EELC.pdf
(There are a lot of other researches, but I have no time to check. I think there is a more concise review article that covers this kind of researches.)