Stéphane Rauzy | Aix-Marseille University (original) (raw)

Papers by Stéphane Rauzy

Proceedings of the 20th ACM International Conference on Intelligent Virtual Agents

Text, Speech, and Dialogue

Schizophrenia Research, 2014

ABSTRACT The multiplication of annotation schemes and coding formats is a severe limitation for i... more ABSTRACT The multiplication of annotation schemes and coding formats is a severe limitation for interoperability. We propose in this pa-per an approach specifying the annotation scheme in terms of typed feature struc-tures, that are in a second step translated into XML schemas, from which data are encoded. This approach guarantees the fact that no information is lost when trans-lating one format into another.

La compréhension des mécanismes du langage nécessite de prendre en compte très précisément les in... more La compréhension des mécanismes du langage nécessite de prendre en compte très précisément les interactions entre les différents domaines ou modalités linguistiques, ce qui implique la constitution et le développement de ressources. Nous décrivons ici le CID (Corpus of Interactional Data), corpus audio-vidéo de 8 heures, en français, constitué au Laboratoire Parole et Langage (LPL). L'annotation multimodale du CID inclut la phonétique, la prosodie, la morphologie, la syntaxe, le discours et la mimogestualité. Les premiers résultats de nos études sur le CID permettent de confirmer l'intérêt d'une annotation multimodale pour mieux comprendre le fonctionnement du discours.

Computing Research Repository, 2008

Methodologically oriented, the present work sketches an approach for prosodic information retriev... more Methodologically oriented, the present work sketches an approach for prosodic information retrieval and speech segmentation, based on both symbolic and probabilistic information. We have recourse to probabilistic grammars, within which we implement a minimal hierarchical structure. Both the stages of probabilistic grammar building and its testing in prediction are explored and quantitatively and qualitatively evaluated.

Modélisation probabiliste de l'interface « forme ~ fonction » pour des phénomènes intonatifs Nous... more Modélisation probabiliste de l'interface « forme ~ fonction » pour des phénomènes intonatifs Nous explorons l'application des méthodes hybrides, reposant à la fois sur des représentations symboliques (phonologiques) et probabilistes dans la modélisation de l'interface « forme ~ fonction » pour des phénomènes intonatifs. À partir d'une représentation symbolique ancrée dans l'acoustique du signal et en accord avec les principes de la phonologie intonative, deux modèles d'enchaînement des catégories tonales sont dérivées moyennant les méthodes de grammaires probabilistes. Deux modèles probabilistes sont testés : le modèle des bigrammes et le modèle des patrons ; leur performance est ensuite évaluée à l'aide de la mesure d'entropie. Ces modèles sont enfin testés en prédiction.

Astrophysical letters & communications

Astronomy and Astrophysics

Astronomy and Astrophysics

Astron. Astrophys. 256, 1-9 (1992) ASTRONOMY AND ASTROPHYSICS Detecting non-Hubble velocity field... more Astron. Astrophys. 256, 1-9 (1992) ASTRONOMY AND ASTROPHYSICS Detecting non-Hubble velocity fields in the universe Stéphane Rauzy, Marc Lachièze-Rey, and RN Henriksen* Service d'Astrophysique, CEN Saclay, F-91 191 Gif sur Yvette Cedex, France Received July 16, ...

Astrophysical letters & communications

In this paper we derive a Tully Fisher relation from measured I band photometry and Hα rotation c... more In this paper we derive a Tully Fisher relation from measured I band photometry and Hα rotation curves of a large survey of southern sky spiral galaxies, obtained in by deprojecting and folding the raw Hα data of . We calibrate the relation by combining several of the largest clusters in the survey, using an iterative maximum likelihood procedure to account for observational selection effects and Malmquist bias. We also incorporate a simple model for the line of sight depth of each cluster. Our results indicate a Tully Fisher relation of intrinsic dispersion ∼ 0.3 mag, corresponding to a distance error dispersion of 13%. Application of this relation to mapping the large scale velocity field is underway.

Astronomy and Astrophysics

We present a method, based on the properties of wavelet transforms, for inferring an irrotational... more We present a method, based on the properties of wavelet transforms, for inferring an irrotational velocity field from its observed radial component. The method is comparable in its objective to POTENT but it offers in addition a natural smoothing that does not lose information, and a separation of the scales. This has required us to conduct for the first time

This paper completes the statistical modeling of the Hubble ow when a Tully-Fisher type relation ... more This paper completes the statistical modeling of the Hubble ow when a Tully-Fisher type relation is used for estimating the absolute magnitude M a p + b from a line width distance indicator p. Our investigation is performed with the aim of providing us with a full understanding of statistical biases due to selection e ects in observation, regardless of peculiar velocities of galaxies. We show that unbiased H -statistics can be obtained by means of the maximum likelihood method as long as the statistical model can be de ned. We focus on the statistical models related to the Direct, resp. Inverse, Tully-Fisher relation, when selection e ects on distance, resp. on p, are present. It turns out that the use of the Inverse relation should be preferred, according to robustness criteria. The formal results are ensured by simulations with samples which are randomly generated according to usual characteristics.

Astronomy and Astrophysics

We provide the mathematical framework which elucidates the way of using a Tully-Fisher (TF) like ... more We provide the mathematical framework which elucidates the way of using a Tully-Fisher (TF) like relation in the determination of the Hubble constant H0, as well as for distances of galaxies. The first step toward the comprehension of this problem is to define a statistical model which accounts for the (linear) correlation between the absolute magnitude M and the line width distance estimator p of galaxies, as it is observed. Herein, we assume that M = a.p + b − ζ, where ζ is a random variable of zero mean describing an intrinsic scatter, regardless of measurement errors. The second step is to understand that the calibration of this law is not unique, since it depends on the statistical model used for describing the distribution of variables (involved in the calculations). With this in mind, the methods related to the so-called Direct and Inverse TF Relations (herein DTF and ITF) are interpreted as maximum likelihood statistics. We show that, as long as the same model is used for the calibration of the TF relation and for the determination of H0, we obtain a coherent Hubble's constant. In other words, the H0 estimates are not model dependent, while the TF relation coefficients are. The choice of the model is motivated by reasons of robustness of statistics, it depends on selection effects in observation which are present in the sample. For example, if p-selection effects are absent then it is more convenient to use a (newly defined) robust statistic, herein denoted by ITF ⋆ . This statistic does not require hypotheses on the luminosity distribution function and on the spatial distribution of sources, and it is still valid when the sample is not complete. Similarly, the general above results apply also to distance estimates of galaxies. The difference on the distance estimates when using either the ITF or the DTF model is only due to random fluctuations. It is interesting to point out that the DTF estimate does not depend on the luminosity distribution of sources. Both statistics show a correction for a bias, inadequately believed to be of Malmquist type. The repercussion of measurement errors, and additional selection effects are also analyzed.

We present a new method for fitting peculiar velocity models to complete flux limited magnitude-r... more We present a new method for fitting peculiar velocity models to complete flux limited magnitude-redshifts catalogues, using the luminosity function of the sources as a distance indicator. The method is characterized by its robustness. In particular, no assumptions are made concerning the spatial distribution of sources and their luminosity function. Moreover the inclusion of additional observables, such for example the one carrying the Tully-Fisher information,is straightforward. As an illustration of the method, the predicted IRAS peculiar velocity model is herein tested using the fluxes of the IRAS 1.2 Jy sample as the distance indicator. The results suggest that this model, while successful in reproducing locally the cosmic flow, fails to describe the kinematics on larger scales.

Astronomy and Astrophysics

We present a new method of bias correction for deriving reliable Tully-Fisher distances in a magn... more We present a new method of bias correction for deriving reliable Tully-Fisher distances in a magnitude or diameter selected sample of field galaxies. This normalized distance method (NDM) is first presented in a theoretical way using the formalism of bayesian statistics and is further applied to three different Tully-Fisher samples corresponding to three different passbands in B, I and r. Constraints imposed by the method, and influences of underlying assumptions and measurement errors are discussed in detail. A main feature of the methodology is to extract an unbiased subsample from the parent sample. We show that by taking into account all the components influencing the bias at a given distance, the NDM method allows to extend the unbiased range and to analyze peculiar velocities of galaxies within a sphere out to 8000 km s(-1) around the Local Group. Finally, thanks to this tool, we show first evidences for both frontside and backside large amplitude infall toward the Perseus-Pis...

Proceedings of the International Astronomical Union

The S-tree method is used to investigate the problem of the mutual motion of the substructures of... more The S-tree method is used to investigate the problem of the mutual motion of the substructures of the Local Group galaxies, as well as of the velocity of the Local Group itself. The different degree of mutual influence between the Local Group galaxies is estimated which will affect the star forming and other individual properties of the galaxies. The existence of statistically significant bulk flow of the Milky Way subsystem is shown via 3D reconstruction procedure, which uses the information on the radial velocities of the galaxies, but not on their distances. The Local Group velocity in the CMB frame is also estimated: VLG -> CMB = 668 km s-1 pointing toward galactic coordinates l = 265 and b = 42.

Astronomy and Astrophysics

This paper completes the statistical modeling of the Hubble flow when a Tully-Fisher type relatio... more This paper completes the statistical modeling of the Hubble flow when a Tully-Fisher type relation is used for estimating the absolute magnitude M=~ap+b from a line width distance indicator p. Our investigation is performed with the aim of providing us with a full understanding of statistical biases due to selection effects in observation, regardless of peculiar velocities of galaxies. We show that unbiased H_0_-statistics can be obtained by means of the maximum likelihood method as long as the statistical model can be defined. We focus on the statistical models related to the Direct, resp. Inverse, Tully-Fisher relation, when selection effects on distance, resp. on p, are present. It turns out that the use of the Inverse relation should be preferred, according to robustness criteria. The formal results are ensured by simulations with samples which are randomly generated according to usual characteristics.

Proceedings of the 20th ACM International Conference on Intelligent Virtual Agents

Text, Speech, and Dialogue

Schizophrenia Research, 2014

ABSTRACT The multiplication of annotation schemes and coding formats is a severe limitation for i... more ABSTRACT The multiplication of annotation schemes and coding formats is a severe limitation for interoperability. We propose in this pa-per an approach specifying the annotation scheme in terms of typed feature struc-tures, that are in a second step translated into XML schemas, from which data are encoded. This approach guarantees the fact that no information is lost when trans-lating one format into another.

La compréhension des mécanismes du langage nécessite de prendre en compte très précisément les in... more La compréhension des mécanismes du langage nécessite de prendre en compte très précisément les interactions entre les différents domaines ou modalités linguistiques, ce qui implique la constitution et le développement de ressources. Nous décrivons ici le CID (Corpus of Interactional Data), corpus audio-vidéo de 8 heures, en français, constitué au Laboratoire Parole et Langage (LPL). L'annotation multimodale du CID inclut la phonétique, la prosodie, la morphologie, la syntaxe, le discours et la mimogestualité. Les premiers résultats de nos études sur le CID permettent de confirmer l'intérêt d'une annotation multimodale pour mieux comprendre le fonctionnement du discours.

Computing Research Repository, 2008

Methodologically oriented, the present work sketches an approach for prosodic information retriev... more Methodologically oriented, the present work sketches an approach for prosodic information retrieval and speech segmentation, based on both symbolic and probabilistic information. We have recourse to probabilistic grammars, within which we implement a minimal hierarchical structure. Both the stages of probabilistic grammar building and its testing in prediction are explored and quantitatively and qualitatively evaluated.

Modélisation probabiliste de l'interface « forme ~ fonction » pour des phénomènes intonatifs Nous... more Modélisation probabiliste de l'interface « forme ~ fonction » pour des phénomènes intonatifs Nous explorons l'application des méthodes hybrides, reposant à la fois sur des représentations symboliques (phonologiques) et probabilistes dans la modélisation de l'interface « forme ~ fonction » pour des phénomènes intonatifs. À partir d'une représentation symbolique ancrée dans l'acoustique du signal et en accord avec les principes de la phonologie intonative, deux modèles d'enchaînement des catégories tonales sont dérivées moyennant les méthodes de grammaires probabilistes. Deux modèles probabilistes sont testés : le modèle des bigrammes et le modèle des patrons ; leur performance est ensuite évaluée à l'aide de la mesure d'entropie. Ces modèles sont enfin testés en prédiction.

Astrophysical letters & communications

Astronomy and Astrophysics

Astronomy and Astrophysics

Astron. Astrophys. 256, 1-9 (1992) ASTRONOMY AND ASTROPHYSICS Detecting non-Hubble velocity field... more Astron. Astrophys. 256, 1-9 (1992) ASTRONOMY AND ASTROPHYSICS Detecting non-Hubble velocity fields in the universe Stéphane Rauzy, Marc Lachièze-Rey, and RN Henriksen* Service d'Astrophysique, CEN Saclay, F-91 191 Gif sur Yvette Cedex, France Received July 16, ...

Astrophysical letters & communications

In this paper we derive a Tully Fisher relation from measured I band photometry and Hα rotation c... more In this paper we derive a Tully Fisher relation from measured I band photometry and Hα rotation curves of a large survey of southern sky spiral galaxies, obtained in by deprojecting and folding the raw Hα data of . We calibrate the relation by combining several of the largest clusters in the survey, using an iterative maximum likelihood procedure to account for observational selection effects and Malmquist bias. We also incorporate a simple model for the line of sight depth of each cluster. Our results indicate a Tully Fisher relation of intrinsic dispersion ∼ 0.3 mag, corresponding to a distance error dispersion of 13%. Application of this relation to mapping the large scale velocity field is underway.

Astronomy and Astrophysics

We present a method, based on the properties of wavelet transforms, for inferring an irrotational... more We present a method, based on the properties of wavelet transforms, for inferring an irrotational velocity field from its observed radial component. The method is comparable in its objective to POTENT but it offers in addition a natural smoothing that does not lose information, and a separation of the scales. This has required us to conduct for the first time

This paper completes the statistical modeling of the Hubble ow when a Tully-Fisher type relation ... more This paper completes the statistical modeling of the Hubble ow when a Tully-Fisher type relation is used for estimating the absolute magnitude M a p + b from a line width distance indicator p. Our investigation is performed with the aim of providing us with a full understanding of statistical biases due to selection e ects in observation, regardless of peculiar velocities of galaxies. We show that unbiased H -statistics can be obtained by means of the maximum likelihood method as long as the statistical model can be de ned. We focus on the statistical models related to the Direct, resp. Inverse, Tully-Fisher relation, when selection e ects on distance, resp. on p, are present. It turns out that the use of the Inverse relation should be preferred, according to robustness criteria. The formal results are ensured by simulations with samples which are randomly generated according to usual characteristics.

Astronomy and Astrophysics

We provide the mathematical framework which elucidates the way of using a Tully-Fisher (TF) like ... more We provide the mathematical framework which elucidates the way of using a Tully-Fisher (TF) like relation in the determination of the Hubble constant H0, as well as for distances of galaxies. The first step toward the comprehension of this problem is to define a statistical model which accounts for the (linear) correlation between the absolute magnitude M and the line width distance estimator p of galaxies, as it is observed. Herein, we assume that M = a.p + b − ζ, where ζ is a random variable of zero mean describing an intrinsic scatter, regardless of measurement errors. The second step is to understand that the calibration of this law is not unique, since it depends on the statistical model used for describing the distribution of variables (involved in the calculations). With this in mind, the methods related to the so-called Direct and Inverse TF Relations (herein DTF and ITF) are interpreted as maximum likelihood statistics. We show that, as long as the same model is used for the calibration of the TF relation and for the determination of H0, we obtain a coherent Hubble's constant. In other words, the H0 estimates are not model dependent, while the TF relation coefficients are. The choice of the model is motivated by reasons of robustness of statistics, it depends on selection effects in observation which are present in the sample. For example, if p-selection effects are absent then it is more convenient to use a (newly defined) robust statistic, herein denoted by ITF ⋆ . This statistic does not require hypotheses on the luminosity distribution function and on the spatial distribution of sources, and it is still valid when the sample is not complete. Similarly, the general above results apply also to distance estimates of galaxies. The difference on the distance estimates when using either the ITF or the DTF model is only due to random fluctuations. It is interesting to point out that the DTF estimate does not depend on the luminosity distribution of sources. Both statistics show a correction for a bias, inadequately believed to be of Malmquist type. The repercussion of measurement errors, and additional selection effects are also analyzed.

We present a new method for fitting peculiar velocity models to complete flux limited magnitude-r... more We present a new method for fitting peculiar velocity models to complete flux limited magnitude-redshifts catalogues, using the luminosity function of the sources as a distance indicator. The method is characterized by its robustness. In particular, no assumptions are made concerning the spatial distribution of sources and their luminosity function. Moreover the inclusion of additional observables, such for example the one carrying the Tully-Fisher information,is straightforward. As an illustration of the method, the predicted IRAS peculiar velocity model is herein tested using the fluxes of the IRAS 1.2 Jy sample as the distance indicator. The results suggest that this model, while successful in reproducing locally the cosmic flow, fails to describe the kinematics on larger scales.

Astronomy and Astrophysics

We present a new method of bias correction for deriving reliable Tully-Fisher distances in a magn... more We present a new method of bias correction for deriving reliable Tully-Fisher distances in a magnitude or diameter selected sample of field galaxies. This normalized distance method (NDM) is first presented in a theoretical way using the formalism of bayesian statistics and is further applied to three different Tully-Fisher samples corresponding to three different passbands in B, I and r. Constraints imposed by the method, and influences of underlying assumptions and measurement errors are discussed in detail. A main feature of the methodology is to extract an unbiased subsample from the parent sample. We show that by taking into account all the components influencing the bias at a given distance, the NDM method allows to extend the unbiased range and to analyze peculiar velocities of galaxies within a sphere out to 8000 km s(-1) around the Local Group. Finally, thanks to this tool, we show first evidences for both frontside and backside large amplitude infall toward the Perseus-Pis...

Proceedings of the International Astronomical Union

The S-tree method is used to investigate the problem of the mutual motion of the substructures of... more The S-tree method is used to investigate the problem of the mutual motion of the substructures of the Local Group galaxies, as well as of the velocity of the Local Group itself. The different degree of mutual influence between the Local Group galaxies is estimated which will affect the star forming and other individual properties of the galaxies. The existence of statistically significant bulk flow of the Milky Way subsystem is shown via 3D reconstruction procedure, which uses the information on the radial velocities of the galaxies, but not on their distances. The Local Group velocity in the CMB frame is also estimated: VLG -> CMB = 668 km s-1 pointing toward galactic coordinates l = 265 and b = 42.

Astronomy and Astrophysics

This paper completes the statistical modeling of the Hubble flow when a Tully-Fisher type relatio... more This paper completes the statistical modeling of the Hubble flow when a Tully-Fisher type relation is used for estimating the absolute magnitude M=~ap+b from a line width distance indicator p. Our investigation is performed with the aim of providing us with a full understanding of statistical biases due to selection effects in observation, regardless of peculiar velocities of galaxies. We show that unbiased H_0_-statistics can be obtained by means of the maximum likelihood method as long as the statistical model can be defined. We focus on the statistical models related to the Direct, resp. Inverse, Tully-Fisher relation, when selection effects on distance, resp. on p, are present. It turns out that the use of the Inverse relation should be preferred, according to robustness criteria. The formal results are ensured by simulations with samples which are randomly generated according to usual characteristics.