Learning from positive data is a center goal in grammatical inference. Some language classes have been characterized in order to allow its learning from text. There are two different approaches to this topic: (i) reducing the new classes to well known ones, and (ii) designing new learning algorithms for the new classes. In this work we will use reduction techniques to define new classes of even linear languages which can be inferred from positive data only. We will center our attention to inferable classes based on local testability features. So, the learning processes for such classes of even linear languages can be performed by using algorithms for locally testable regular languages.