Time-lapse pressure tomography for characterizing CO2 plume evolution in a deep saline aquifer

A time-lapse pressure tomography inversion approach is applied to characterize the CO2 plume development in a virtual deep saline aquifer. Deep CO2 injection leads to flow properties of the mixed-phase, which vary depending on the CO2 saturation. Analogous to the crossed ray paths of a seismic tomographic experiment, pressure tomography creates streamline patterns by injecting brine prior to CO2 injection or by injecting small amounts of CO2 into the two-phase (brine and CO2) system at different depths. In a first step, the introduced pressure responses at observation locations are utilized for a computationally rapid and efficient eikonal equation based inversion to reconstruct the heterogeneity of the subsurface with diffusivity (D) tomograms. Information about the plume shape can be derived by comparing D-tomograms of the aquifer at different times. In a second step, the aquifer is subdivided into two zones of constant values of hydraulic conductivity (K) and specific storage (Ss) through a clustering approach. For the CO2 plume, mixed-phase K and Ss values are estimated by minimizing the difference between calculated and “true” pressure responses using a single-phase flow simulator to reduce the computing complexity. Finally, the estimated flow property is converted to gas saturation by a single-phase proxy, which represents an integrated value of the plume. This novel approach is tested first with a doublet well configuration, and it reveals a great potential of pressure tomography based concepts for characterizing and monitoring deep aquifers, as well as the evolution of a CO2 plume. Still, field-testing will be required for better assessing the applicability of this approach.

Balanced bilinguals favor lexical processing in their opaque language and conversion system in their shallow language

Referred to as orthographic depth, the degree of consistency of grapheme/phoneme correspondences varies across languages from high in shallow orthographies to low in deep orthographies. The present study investigates the impact of orthographic depth on reading route by analyzing evoked potentials to words in a deep (French) and shallow (German) language presented to highly proficient bilinguals. ERP analyses to German and French words revealed significant topographic modulations 240-280ms post-stimulus onset, indicative of distinct brain networks engaged in reading over this time window. Source estimations revealed that these effects stemmed from modulations of left insular, inferior frontal and dorsolateral regions (German>French) previously associated to phonological processing. Our results show that reading in a shallow language was associated to a stronger engagement of phonological pathways than reading in a deep language. Thus, the lexical pathways favored in word reading are reinforced by phonological networks more strongly in the shallow than deep orthography.

Shallow Dialogue Processing Using Machine Learning Algorithms (or not)

This paper presents a shallow dialogue analysis model, aimed at human-human dialogues in the context of staff or business meetings. Four components of the model are defined, and several machine learning techniques are used to extract features from dialogue transcripts: maximum entropy classifiers for dialogue acts, latent semantic analysis for topic segmentation, or decision tree classifiers for discourse markers. A rule-based approach is proposed for solving cross-modal references to meeting documents. The methods are trained and evaluated thanks to a common data set and annotation format. The integration of the components into an automated shallow dialogue parser opens the way to multimodal meeting processing and retrieval applications.