| Abstract | This paper shows how machine learning can help in analyzing andunderstanding historical change. Using data from the Canadian censusof 1901, we discover the influences on bilingualism in Canada at be-ginning of the last century. The discovered theories partly agree with,and partly complement the existing views of historians on this ques-tion. Our approach, based around a decision tree, not only infers the-ories directly from data but also evaluates existing theories and revisesthem to improve their consistency with the data. One novel aspect ofthis work is the use of confidence intervals to determine which factorsare both statistically and practically significant, and thus contributeappreciably to the overall accuracy of the theory. When inducing a decision tree directly from data, confidence intervals determine whennew tests should be added. If an existing theory is being evaluated, confidence intervals also determine when old tests should be replacedor deleted to improve the theory. Our aim is to minimize the changesmade to an existing theory to accommodate the new data. To thisend, we propose a semantic measure of similarity between trees anddemonstrate how this can be used to limit the changes made. |
|---|
