International Conference on Grammatical Inference 2018: Preface

1994

In this paper a new methodology for inferring regular grammars y from a set of positive and negative examples and a-priori knowledge (if available), that we have called active grammatical inference, is presented. The methodology is based on a combination of neural and symbolic techniques, with the interesting feature that the learning process can be guided by the validated previous results and/or by introducing constraints or (positive/negative) rules dynamically. This is possible thanks to a new neural learning scheme, in which the inserted symbolic knowledge, that is also preserved, controls the learning mechanism. The whole process is conceived as a sequence of learning cycles, each one including the steps of FSA insertion, neural training, FSA extraction, symbolic manipulation and validation. The insertion of symbolic information into a recurrent neural network is carried out through linear system solving. An example which allows to compare the active grammatical inference methodology with just the neural or symbolic approaches is also included.

Models based on Chomskyan grammar are not as pervasive in certain NLP applications as might be expected from their status in linguistics, comparing to, e.g., statistical n-gram models or vector space models. In contrast, here grammar inference is approached from the viewpoint of constructionist theories of language, which avoid the separation of morphology, syntax, semantics and pragmatics. We consider the utility of constructionist theories in NLP applications, present a computational framework for learning constructions, and discuss related methodological work.

Lecture Notes in Computer Science, 2008

Empirical grammatical inference systems are practical systems that learn structure from sequences, in contrast to theoretical grammatical inference systems, which prove learnability of certain classes of grammars. All current empirical grammatical inference evaluation methods are problematic, i.e. dependency on language experts, appropriateness and quality of an underlying grammar of the data, and influence of the parameters of the evaluation metrics. Here, we propose a modification of an evaluation method to reduce the ambiguity of results.

Pattern Recognition, 1984

A new and simple method for the inference of regular grammars or finite automata is presented. It is based on the analysis of the successive appearances of the terminal symbols in the learning strings. It is shown that for syntactic pattern recognition applications, this method is more efficient than other algorithms already proposed.

2008

Grammatical Inference deals with the problem of learning structural models, such as grammars, from different sort of data patterns, such as artificial languages, natural languages, biosequences, speech and so on. This article describes a new grammatical inference tool, Grammar-based Classifier System (GCS) dedicated to learn grammar from data. GCS is a new model of Learning Classifier Systems in which the population of classifiers has a form of a context-free grammar rule set in a Chomsky Normal Form. GCS has been proposed to address both regular language induction and the natural language grammar induction as well as learning formal grammar for DNA sequence. In all cases near-optimal solutions or better than reported in the literature were obtained.

Psychology [Miller 6, Suppes 561. There is, of course, a vast literature on pattern recognition [Uhr 661, but it has been exclusively concerned with pattern descriptions which are structurally simpler than grammars.

2002

In recent years there have been significant advances in the field of Unsupervised Grammar Inference (UGI) for Natural Languages such as English or Dutch. This paper presents a broad range of UGI implementations, where we can begin to see how the theory has been put in to practise. Several mature systems are emerging, built using complex models and capable of deriving natural language grammatical phenomena. The range of systems is classified into: models based on Categorial Grammar (GraSp, CLL, EMILE); Memory Based Learning models (FAMBL, RISE); Evolutionary computing models (ILM, LAgts); and string-pattern searches (ABL, GB). An objectively measurable statistical comparison of performance Of the systems reviewed is not yet feasible. However, their merits and shortfalls are discussed, as well as a look at what the future has in store for UGI.

Encyclopedia of Artificial Intelligence, 2009

Proceedings of the EACL 2009 Workshop on …, 2009

2003

The evaluation of grammar inference systems is clearly a non-trivial task, as it is possible to have more than one correct grammar for a given language. The ‘looks good to me’ approach, carried out by computational linguists analysing their own grammar inference system results, has prevailed for many years. This paper explores why this method has been so popular, in terms of its strengths, and also why it is no longer adequate as a reliable means to measuring performance. Corpus based methods, that can be performed automatically, are investigated to see how they can meet the needs of this difficult problem.