While rules and exemplars are usually viewed as opposites, this paper argues that they form end points of the same distribution. By representing both rules and exemplars as (partial) trees, we can take into account the fluid middle ground between the two extremes. This insight is the starting point for a new theory of language learning that is based on the following idea: If a language learner does not know which phrase‐structure trees should be assigned to initial sentences, s/he allows (...) (implicitly) for all possible trees and lets linguistic experience decide which is the “best” tree for each sentence. The best tree is obtained by maximizing “structural analogy” between a sentence and previous sentences, which is formalized by the most probable shortest combination of subtrees from all trees of previous sentences. Corpus‐based experiments with this model on the Penn Treebank and the Childes database indicate that it can learn both exemplar‐based and rule‐based aspects of language, ranging from phrasal verbs to auxiliary fronting. By having learned the syntactic structures of sentences, we have also learned the grammar implicit in these structures, which can in turn be used to produce new sentences. We show that our model mimicks children’s language development from item‐based constructions to abstract constructions, and that the model can simulate some of the errors made by children in producing complex questions. (shrink)

This paper deals with the problem of derivational redundancy in scientific explanation, i.e. the problem that there can be extremely many different explanatory derivations for a natural phenomenon while students and experts mostly come up with one and the same derivation for a phenomenon (modulo the order of applying laws). Given this agreement among humans, we need to have a story of how to select from the space of possible derivations of a phenomenon the derivation that humans come up with. (...) In this paper we argue that the problem of derivational redundancy can be solved by a new notion of “shortest derivation”, by which we mean the derivation that can be constructed by the fewest (and therefore largest) partial derivations of previously derived phenomena that function as “exemplars”. We show how the exemplar-based framework known as “Data-Oriented Parsing” or “DOP” can be employed to select the shortest derivation in scientific explanation. DOP’s shortest derivation of a phenomenon maximizes what is called the “derivational similarity” between a phenomenon and a corpus of exemplars. A preliminary investigation with exemplars from classical and fluid mechanics shows that the shortest derivation closely corresponds to the derivations that humans construct. Our approach also proposes a concrete solution to Kuhn’s problem of how we know on which exemplar a phenomenon can be modeled. We argue that humans model a phenomenon on the exemplar that is derivationally most similar to the phenomenon, i.e. the exemplar from which the largest subtree(s) can be used to derive the phenomenon. (shrink)