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.

Cerebral Gaseous Microemboli are Detectable During Continuous Venovenous Hemodialysis in Critically Ill Patients: An Observational Pilot Study

BACKGROUND Continuous venovenous hemodialysis (CVVHD) may generate microemboli that cross the pulmonary circulation and reach the brain. The aim of the present study was to quantify (load per time interval) and qualify (gaseous vs. solid) cerebral microemboli (CME), detected as high-intensity transient signals, using transcranial Doppler ultrasound. MATERIALS AND METHODS Twenty intensive care unit (ICU group) patients requiring CVVHD were examined. CME were recorded in both middle cerebral arteries for 30 minutes during CVVHD and a CVVHD-free interval. Twenty additional patients, hospitalized for orthopedic surgery, served as a non-ICU control group. Statistical analyses were performed using the Mann-Whitney U test or the Wilcoxon matched-pairs signed-rank test, followed by Bonferroni corrections for multiple comparisons. RESULTS In the non-ICU group, 48 (14.5-169.5) (median [range]) gaseous CME were detected. In the ICU group, the 67.5 (14.5-588.5) gaseous CME detected during the CVVHD-free interval increased 5-fold to 344.5 (59-1019) during CVVHD (P<0.001). The number of solid CME was low in all groups (non-ICU group: 2 [0-5.5]; ICU group CVVHD-free interval: 1.5 [0-14.25]; ICU group during CVVHD: 7 [3-27.75]). CONCLUSIONS This observational pilot study shows that CVVHD was associated with a higher gaseous but not solid CME burden in critically ill patients. Although the differentiation between gaseous and solid CME remains challenging, our finding may support the hypothesis of microbubble generation in the CVVHD circuit and its transpulmonary translocation toward the intracranial circulation. Importantly, the impact of gaseous and solid CME generated during CVVHD on brain integrity of critically ill patients currently remains unknown and is highly debated.

Hypotension and hypovolemia during hemodialysis: is the usual suspect innocent?

Hypotension during intermittent hemodialysis is common, and has been attributed to acute volume shifts, shifts in osmolarity, electrolyte imbalance, temperature changes, altered vasoregulation, and sheer hypovolemia. Although hypovolemia may intuitively seem a likely cause for hypotension in intensive care patients, its role in the pathogenesis of intradialytic hypotension may be overestimated.