Learning languages from positive examples with dependencies

We investigate learning dependency grammar from positive data, in Gold's identification in the limit model. Examples are dependency trees. For this, we introduce reversible lexical dependency grammars which generate a significant class of languages. We have demonstrated that reversible dependency languages are learnable. We provide a-time, in the example size, algorithm. Our objective is to contribute to design and the understanding of formal process of language acquisition. For this, dependency trees play an important role because they naturally appear in every tree phrase structure. From Tesniére (Tesnì ere , 1959) seminal study , and from ideas of Mel ' ˘ cuk (Mel ' ˘ cuk, 1988) , we propose a two tier communication process between two speakers , see Figure 1. Jean transmits a sentence to Marie. At the first stage , Jean generates a structural sentence , like the following dependency tree the rabbit runs fast Then , Jean transforms it into a linear phrase , the rabbit runs fast , and send it to Marie.