Previous work on learning regular languages from exemplary training sequences showed that Long Short- Term Memory (LSTM) outperforms traditional recurrent neural networks (RNNs). Here we demonstrate LSTM's superior performance on context free language (CFL) benchmarks for recurrent neural networks (RNNs), and show that it works even better than previous hardwired or highly specialized architectures.

DCTG-GP is a genetic programming system that uses definite clause translation grammars. A DCTG is a logical version of an attribute grammar that supports the definition of context–free languages, and it allows semantic information associated with a language to be easily accomodated by the grammar. This is useful in genetic programming for defining the interpreter of a target language, or incorporating both syntactic and semantic problem–specific contraints into the evolutionary search. The DCTG-GP system improves on other grammar–based GP systems by permitting non–trivial semantic aspects of the language to be defined with the grammar. It also automatically analyzes grammar rules in order to determine their minimal depth and termination characteristics, which are required when generating random program trees of varied shapes and sizes. An application using DCTG-GP is described.

Determinism plays an important role in grammatical inference.

Information extraction (IE) addresses the problem of extracting speci c information from a collection of documents. Much of the previous work for IE from structured documents formatted in HTML or XML uses techniques for IE from strings, such as grammar and automata induction. However, such documents have a tree structure. Hence it is natural to investigate methods that are able to recognise and exploit this tree structure. We do this by exploring the use of tree automata for IE in structured documents. Experimental results on benchmark data sets show that our approach compares favorably with previous approaches.

Learning from positive data constitutes an important topic in Grammatical Inference since it is believed that the acquisition of grammar by children only needs syntactically correct (i.e. positive) instances. However, classical learning models provide no way to avoid the problem of over-generalization. In order to overcome this problem, we use here a learning model from simple examples, where the notion of simplicity is defined with the help of Kolmogorov complexity. We show that a general and natural heuristic which allows learning from simple positive examples can be developed in this model. Our main result is that the class of regular languages is probably exactly learnable from simple positive examples.

Inferring an appropriate DTD or XML Schema Definition (XSD) for a given collection of XML documents essentially reduces to learning deterministic regular expressions from sets of positive example words. Unfortunately, there is no algorithm capable of learning the complete class of deterministic regular expressions from positive examples only, as we will show. The regular expressions occurring in practical DTDs and XSDs, however, are such that every alphabet symbol occurs only a small number of times. As such, in practice it suffices to learn the subclass of regular expressions in which each alphabet symbol occurs at most k times, for some small k. We refer to such expressions as k-occurrence regular expressions (k-OREs for short). Motivated by this observation, we provide a probabilistic algorithm that learns k-OREs for increasing values of k, and selects the one that best describes the sample based on a Minimum Description Length argument. The effectiveness of the method is empirically validated both on real world and synthetic data. Furthermore, the method is shown to be conservative over the simpler classes of expressions considered in previous work.

Abstract. The World-Wide Web contains a wealth of semistructured information sources that often give partial/overlapping views on the same domains, such as real estate listings or book prices. These partial sources could be used more effectively if integrated into a single view; however, since they are typically formatted in diverse ways for human viewing, extracting their data for integration is a difficult challenge. Existing learning systems for this task generally use hardcoded ad hoc heuristics, are restricted in the domains and structures they can recognize, and/or require manual training. We describe a principled method for automatically generating extraction wrappers using grammatical inference that can recognize general structures and does not rely on manually-labelled examples. Domain-specific knowledge is explicitly separated out in the form of declarative rules. The method is demonstrated in a test setting by extracting real estate listings from web pages and integrating them into an interactive data visualization tool based on dynamic queries. 1

The use of genetic programming for probabilistic pattern matching is investigated. A stochastic regular expression language is used. The language features a statistically sound semantics, as well as a syntax that promotes efficient manipulation by genetic programming operators. An algorithm for efficient string recognition based on approaches in conventional regular language recognition is used. When attempting to recognize a particular test string, the recognition algorithm computes the probabilities of generating that string and all its prefixes with the given stochastic regular expression. To promote efficiency, intermediate computed probabilities that exceed a given cut-off value will pre-empt particular interpretation paths, and hence prune unconstructive interpretation. A few experiments in recognizing stochastic regular languages are discussed. Application of the technology in bioinformatics is in progress.

Graph grammars combine the relational aspect of graphs with the iterative and recursive aspects of string grammars, and thus represent an important next step in our ability to discover knowledge from data. In this paper we describe an approach to learning node replacement graph grammars. This approach is based on previous research in frequent isomorphic subgraphs discovery. We extend the search for frequent subgraphs by checking for overlap among the instances of the subgraphs in the input graph. If subgraphs overlap by one node we propose a node replacement grammar production. We also can infer a hierarchy of productions by compressing portions of a graph described by a production and then infer new productions on the compressed graph. We validate this approach in experiments where we generate graphs from known grammars and measure how well our system infers the original grammar from the generated graph. We also describe results on several real-world tasks from chemical mining to XML schema induction. We briefly discuss other grammar inference systems indicating that our study extends classes of learnable graph grammars.

Information extraction from HTML pages has been conventionally treated as plain text documents extended with HTML tags. However, the growing maturity and correct usage of HTML/XHTML formats open an opportunity to treat Web pages as trees, to mine the rich structural context in the trees and to learn accurate extraction rules. In this paper, we generalize the notion of delimiter developed for the string information extraction to tree documents.