Joseph Peters graduated as an English major from Dartmouth College in 1999. Since then he has been in a variety of jobs involving some combination of web design and writing, including doing some web design for the Howard Dean campaign. His interests are quite varied but include chat bots, the Turing test, the history of near misses of the NKS discovery, and the structure and use of language.
Project: Learning the Past Tense of English Verbs, Part Deux
In brief: selecting simple rules that will distinguish between different types of verbs and conjugate them accordingly. Turing once suggested that one of the best tests of a human-like intelligence would be the proper use of language. This theory has come to be known as the “Turing test.” In 1982, cognitive scientists McClelland and Rumelhart took an important step toward this ideal by designing a network system that could learn and express the past tenses of English verbs. I will repeat this experiment, using a simple system suggested in the NKS book, to observe what rules are generated by the system when it encounters unfamiliar verbs. The process is described below.
- Build a list of 200 verbs represented in their present and past forms. Store the associated phonemes.
- Build a word-comparison program to generate symbols for each category of verb.
- Expose the system to a new set of verbs, observing which rules are generated and their success rates.
Favorite Two-Color, Radius-2 Rule
Rule chosen: 2121212121
My favorite CA is rule 2121212121–not just for its near-palindrome title, but for its appearance as well.