TheseriesofworkshopsonMachineLearningforMultimodalInteraction(MLMI) celebratesthisyearits?fthanniversary.Onthisoccasion, anumberofinnovations havebeenintroducedin the reviewingandpublicationprocedures, while keeping the focus onthe samescienti?c topics. For the ?rst time, the reviewing process has been adapted in order to p- parethe proceedings in time for the workshop, held on September 8-10,2008, in Utrecht, The Netherlands. The 47 submissions received by the Program C- mittee were ?rst reviewed by three PC members each, and then advocated by an Area Chair. Overall, 12 oral presentations (ca. 25% of all submissions) and 15 poster presentations were selected. Authors were given one month to revise their papers according to the reviews, and the ?nal versions were brie?y checked by the two Program Co-chairs. Both types of presentation have been give equal space in the present proceedings. The 32 papers gathered in this volume cover a wide range of topics - lated to human-human communication modeling and processing, as well as to human-computer interaction, using several communication modalities. A sign- icant number of papers focus on the analysis of non-verbal communication cues, such as the expression of emotions, laughter, face turning, or gestures, which demonstrates a growing interest for social signal processing. Yet, another large set of papers targets the analysis of communicative content, with a focus on the abstractionofinformationfrommeetingsintheformofsummaries, actionitems, ordialogueacts.OthertopicspresentedatMLMI2008includeaudio-visualscene analysis, speech processing, interactive systems and applica

This book contains a selection of revised papers from the 4th Workshop on Machine Learning for Multimodal Interaction (MLMI 2007), which took place in Brno, Czech Republic, during June 28-30, 2007. As in the previous editions of the MLMI series, the 26 chapters of this book cover a large area of topics, from multimodal processing and human-computer interaction to video, audio, speech and language processing. The application of machine learning techniques to problems arising in these ?elds and the design and analysis of software s- portingmultimodalhuman-humanandhuman-computerinteractionarethetwo overarching themes of this post-workshop book. The MLMI 2007 workshop featured 18 oral presentations-two invited talks, 14 regular talks and two special session talks-and 42 poster presentations. The participants were not only related to the sponsoring projects, AMI/AMIDA (http://www.amiproject.org) and IM2 (http://www.im2.ch), but also to other largeresearchprojects onmultimodalprocessingand multimedia browsing,such as CALO and CHIL. Local universities were well represented, as well as other European, US and Japanese universities, research institutions and private c- panies, from a dozen countries overall.

Bringing together experts in multimodal signal processing, this book provides a detailed introduction to the area, with a focus on the analysis, recognition and interpretation of human communication. The technology described has powerful applications. For instance, automatic analysis of the outputs of cameras and microphones in a meeting can make sense of what is happening - who spoke, what they said, whether there was an active discussion and who was dominant in it. These analyses are layered to move from basic interpretations of the signals to richer semantic information. The book covers the necessary analyses in a tutorial manner, going from basic ideas to recent research results. It includes chapters on advanced speech processing and computer vision technologies, language understanding, interaction modeling and abstraction, as well as meeting support technology. This guide connects fundamental research with a wide range of prototype applications to support and analyze group interactions in meetings.