Explorations in Language Learning by Integrating Analogy and Probability

We present a new approach to language learning which 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 can be formalized by the most probable shortest combination of subtrees from all trees of previous sentences. It is shown that this model can learn both exemplar-based and rule-based aspects of language, ranging from constructions and phrasal verbs to auxiliary fronting. We support our theoretical results by corpus-based experiments on the Penn Treebank and the CHILDES database. Our model suggests a rapprochement between nativism and empiricism: it starts from the principle of structural dependency; all else is analogy.