Grammar Induction, also known as Grammatical Inference, is the usual name given to the process of learning (inferring) a grammar from a set of sample strings, and, in view of the equivalences that may be established between grammars and automata, the task of learning an automaton from a set of sample strings may also be called grammatical inference.
Grammatical inference is usually formulated in terms of learning recognition or classification tasks from sets in which all strings are labelled as belonging to one or another class (language); however, tasks such as learning a finite-state machine that transducer (translates) strings from one language into strings from another language or learning a probabilistic finite-state machine that generates strings following a certain probability distribution may also be formulated as grammatical inference tasks.
Learning Recursive Transition Networks (It works by converting grammatically correct sentences into transition networks that are similar to finite state diagrams).Learning CFG using Version Spaces, Learning NPDA using Genetic Search and Learning Deterministic CFG using Connectionist Networks.
It should be also mentioned that there are different models of grammar Induction, such as learning from examples, learning using examples and queries, incremental VS non incremental learning, distribution free models of learning, learning under various distributional assumptions, Impossibility results, complexity results ans finally characterizations of representational and search biases of grammar induction algorithms.
Sources:
- Grammar Induction. Fecha de consulta: 19-05-08 from http://www.info.ucl.ac.be/~pdupont/pdupont/gram.html
- Grammar inference. Fecha de consulta: 19-05-08 from http://www.dlsi.ua.es/~mlf/nnafmc/pbook/node70.html
