Three Learnable Models for the Description of Language [chapter]

Alexander Clark
2010 Lecture Notes in Computer Science  
Learnability is a vital property of formal grammars: representation classes should be defined in such a way that they are learnable. One way to build learnable representations is by making them objective or empiricist: the structure of the representation should be based on the structure of the language. Rather than defining a function from representation to language we should start by defining a function from the language to the representation: following this strategy gives classes of
more » ... tions that are easy to learn. We illustrate this approach with three classes, defined in analogy to the lowest three levels of the Chomsky hierarchy. First, we recall the canonical deterministic finite automaton, where the states of the automaton correspond to the right congruence classes of the language. Secondly, we define context free grammars where the non-terminals of the grammar correspond to the syntactic congruence classes, and where the productions are defined by the syntactic monoid; finally we define a residuated lattice structure from the Galois connection between strings and contexts, which we call the syntactic concept lattice, and base a representation on this, which allows us to define a class of languages that includes some non-context free languages, many context-free languages and all regular languages. All three classes are efficiently learnable under suitable learning paradigms.
doi:10.1007/978-3-642-13089-2_2 fatcat:r5y7op6cn5gk5nvpu3owjbiqiy