Laurent Miclet

What Is the Search Space of the Regular Inference

This paper revisits the theory of regular inference, in particular by extending the definition of structural completeness of a positive sample and by demonstrating two basic theorems. This framework enables to state the regular inference problem as a search through a boolean lattice built from the positive sample. Several properties of the search space are studied and generalization criteria are discussed. In this framework, the concept of border set is introduced, that is the set of the most general solutions excluding a negative sample. Finally, the complexity of regular language identification from both a theoritical and a practical point of view is discussed.

Grammatical Interference: Learning Syntax from Sentences

Learning grammatical structure using statistical decision-trees.- Inductive inference from positive data: from heuristic to characterizing methods.- Unions of identifiable families of languages.- Characteristic sets for polynomial grammatical inference.- Query learning of subsequential transducers.- Lexical categorization: Fitting template grammars by incremental MDL optimization.- Selection criteria for word trigger pairs in language modeling.- Clustering of sequences using a minimum grammar complexity criterion.- A note on grammatical inference of slender context-free languages.- Learning linear grammars from structural information.- Learning of context-sensitive language acceptors through regular inference and constraint induction.- Inducing constraint grammars.- Introducing statistical dependencies and structural constraints in variable-length sequence models.- A disagreement count scheme for inference of constrained Markov networks.- Using knowledge to improve N-Gram language modelling through the MGGI methodology.- Discrete sequence prediction with commented Markov models.- Learning k-piecewise testable languages from positive data.- Learning code regular and code linear languages.- Incremental regular inference.- An incremental interactive algorithm for regular grammar inference.- Inductive logic programming for discrete event systems.- Stochastic simple recurrent neural networks.- Inferring stochastic regular grammars with recurrent neural networks.- Maximum mutual information and conditional maximum likelihood estimations of stochastic regular syntax-directed translation schemes.- Grammatical inference using Tabu Search.- Using domain information during the learning of a subsequential transducer.- Identification of DFA: Data-dependent versus data-independent algorithms.

Approximative fast nearest-neighbour recognition

We analyze (on pseudo-randomly generated data) the errors committed with a log(N) algorithm, attempting to recognize the nearest neighbour among N vectors. We propose improvements in O(log(N)), and present an example of application on speech data.

Analogical Equations in Sequences: Definition and Resolution

We present a definition of analogy on sequences which is based on two principles: the definition of an analogy between the letters of an alphabet and the use of the edit distance between sequences. Our definition generalizes the algorithm given by Lepage and is compatible with another definition of analogy in sequences given by Yvon.

Speaker adaptation for phoneme recognition

A procedure for the adaptation of a phonetic recognition system to a new speaker, is described. The reference for each phoneme(cepstrum of the autoregressive model) is transformed linearly in order to fit optimally the utterances of a new speaker. In a first step, samples uttered by the new speaker are mapped onto corresponding utterances (reference) through a dynamic comparison. In a second step, the linear transformation is computed through canonical correlation analysis of the samples. Experimental simulations provided satisfactory results.

Applying Grammatical Inference in Learning a Language Model for Oral Dialogue

We present an application of the ECGI algorithm to the learning of a language model for Speech Recognition. Results are given on a real dialogue corpus. Integrating this technique in a Speech Recognizer is discussed.

Low bit rate transmission of speech by vector quantization of the spectrum

Abstract We present a speech transmission system operating at a low bit rate (about 1000 bits/s), which is based on the following principle. The speech signal is analysed using a LPC technique and, for each frame, the global parameters of the input signal (energy, pitch, voicing) on the one hand, and secondly the filter coefficients are separately coded. Studies have been mainly focused on the last point. We have compared several representation spaces (autocorrelation, cepstrum, LPC analysis without and with preemphasis), in order to choose the most suitable representation for a good vector quantization. The latter has been performed by a very simple algorithm, which we have compared to the “LBG” method. We have shown that, in our experimental conditions, the simplicity of our algorithm is an advantage, and that it performs a good vector quantization of the spectral space. The second part of the study is oriented towards the use of the codebook obtained as described above. We compute it from several speakers, for a total of about 30000 frames (about six minutes) of speech: it consists of more than 1500 vectors. We have studied how to obtain a fast coding of a vector with this codebook, losing the optimality of the nearest neighbour coding. We have shown that the distortion is only slightly increased by using clustering techniques on the codebook, leading to a hierarchical coding decision, which allows a very fast coding of any new vector. In conclusion, the simplification of the codebook construction (associated with a correct choice of the spectral representation space), and a fast (but suboptimal) method of coding with such a codebook lead to a system whose performances are only slightly degraded compared to reference spectral vector quantization systems for speech transmission.

Analogical Proportions in a Lattice of Sets of Alignments Built on the Common Subwords in a Finite Language

We define the locally maximal subwords and locally minimal superwords common to a finite set of words. We also define the corresponding sets of alignments. We give a partial order relation between such sets of alignments, as well as two operations between them. We show that the constructed family of sets of alignments has the lattice structure. The study of analogical proportion in lattices gives hints to use this structure as a machine learning basis, aiming at inducing a generalization of the set of words.

Analogie entre séquences : Définition, calcul et utilisation en apprentissage supervisé

This article is concerned with the learning by analogy in the world of sequences, based on the resolution of analogical equations. It presents a definition of an analogical relation, based on the edit distance and studies the solving of an analogical equation on sequences. It presents a construction with finite-state trandsducers which computes all the solutions of this equation, reducing the problem to that of solving analogical equations on a finite alphabet. It studies also what is analogy on alphabets and describes two algebraic structures which are compatible with the computation of solutions on sequences. Finally, it presents a direct suboptimal algorithm to compute a solution to an analogical equation on sequences.