The research that we present here are related to a study for the design and implementation of a follow-up survey of students via an interactive voice response (IVR) using VoiceXML, a W3C standard language. We present a corpus of questions and answers obtained in natural language, and we validate scientific hypotheses concerning the use of modes of interaction (voice versus direct manipulation). Then, we explain how we passed from a mechanism of exogenous traces (with a monitoring system performed by external tools recordings) to an endogenous mechanism (with a monitoring system made from within the IVR) to provide tools and instruments more adapted to the evaluation of multimodal applications that use speech and gesture (telephone keypad or mouse click on hyperlink). The trace mechanism for telephone interactions using VoiceXML presented here increases the quality of the evaluation of human-machine telephone interactions, because these traces are automatically recorded and reusable. Furthermore, we show that it is possible to get instant statistics (histograms and graphs made in real time, in PHP) using the method presented here. Thus, we have shown that pedagogical surveys, which traditionally are laborious, complex to implement and very time consuming can be facilitated through the methods and tools we recommend.

The main characteristic of intelligent devices that compose our environment is their capability to perceive and collect relevant information (context awareness) in order to assist users in their daily tasks. However, these tasks evolve frequently and require dynamic and evolutionary systems (context-aware systems) to improve intelligent devices skills according to user's context. Some context-aware systems are described in the literature, but most of them have extremely tight coupling between the semantic used in the application and sensors used to obtained the data for this semantic interpretation. The objective of our research is to study and implement a proactive approach able to use existing sensors and to create dynamically human-machine conversational situations when needed. The system presented in this paper is named X-CAMPUS (eXtensible Conversational Agent for Multichannel Proactive Ubiquitous Services). It aims to assist user in his/her daily tasks thanks to its ability to perceive the state of the environment and interact effectively according to the user's needs. In this paper we describe our approach for proactive intelligent assistance and we illustrate it through some scenarios showing that according to a given multi-parameters context, our X-CAMPUS agent notifies the user via personalized messages (e.g., suggestion of restaurants according to menus and users' preferences) across the most appropriate channel (instant messaging, e-mail or SMS) and the most appropriate modality (text, gesture or voice). Then, we discuss our quantitative results, based on four principal hypotheses in order to evaluate our system's capability to manage many users simultaneously with different contextual information. We argue and we show that the proactive assistance is very relevant in complex situations with various criteria to take into account (user's profile, location, task, etc.).