While the small group discussion is widely used in language classes, there is little empirical research on its efficacy. This research specifically focuses on novice level language students in order to understand the ways that they express agreement and disagreement in group interaction. This study utilizes the methodological framework of Conversation Analysis conducting a micro-analysis of student turn-taking practices and their embodied behavior. This research uncovered the fact that the novice level language learners utilized resources that are not generally considered when investigating agreement and disagreement. Nonverbal actions such as smiles and gaze shifts accomplished affiliative work mitigating disagreement turns. Facial expression, laughter, and gestures were often relied on to compensate for deficits in grammar and lexicon. A second finding of the research was that the students were able to accomplish significantly more as members of a group than they could as individuals. The multi-person context created a framework enabling members to participate. The students demonstrated a high level of collaboration, joining in word searches, successfully constructing collaborated completions, and frequently offering support to each other through receipt tokens, nods, and smiles. They proved to be each other's best resource. Another finding of the study was the importance of basic patterns of turns in effective group discussion. For example, in order for an argumentative sequence to emerge, a third response was expected: Turn 1, the claim; Turn 2, disagreement; and, Turn 3, defense, counterattack, or concession by the first speaker or a different speaker. For less skillful groups where topics were not well developed, only two-part sequences were utilized, not allowing subsequent and related talk to occur. Finally, this study contributes to research on the acquisition of disagreement strategies. Surprisingly, in expressing disagreement, these novice level language students employed a number of different means to express disagreement that were more often associated with advanced learners. For example, they delayed their disagreement turns, and they utilized accounts, exemplification, and elaboration when disagreeing. Though these students were not always able to express themselves fluently, they were nevertheless quite capable in expressing agreement and disagreement in the target language. / English

Since the inception of AI research, great strides have been made towards achieving the goal of extending natural language conversation as a medium of interaction with machines. Today, we find many Conversational Agents (CAs) situated in various aspects of our everyday life such as information access, education and entertainment. However, most of the existing work on CAs has focused on agents that support only one user in each interactive session. On the other hand, people organize themselves in groups such as teams of co-workers, family and networks of friends. With the mass-adoption of Internet based communication technologies for group interaction, there is an unprecedented opportunity for CAs to support interactive situations involving multiple human participants. Support provided by these CAs can make the functioning of some of these groups more efficient, enjoyable and rewarding to the participants. Through our work on supporting various Multi-Party Interactive Situations (MPIS), we have identified two problems that must be addressed in order to embed effective CAs in such situations. The first problem highlights the technical challenges involving the development of CAs in MPIS. Existing approaches for modeling agent behavior make assumptions that break down in multi-party interaction. As a step towards addressing this problem, this thesis contributes the Basilica software architecture that uses an event-driven approach to model conversation as an orchestration of triggering of conversational behaviors. This architecture alleviates the technical problems by providing a rich representational capability and the flexibility to address complex interaction dynamics. The second problem involves the choice of appropriate agent behaviors. In MPIS, agents must compete with human participants for attention in order to effectively deliver support and interventions. In this work, we follow a model of human group interaction developed by empirical research in small group communication. This model identifies two fundamental processes in human group interaction, i.e., Instrumental (Task-related) and Expressive (Social-Emotional). Behaviors that constitute this expressive process hold the key to managing and regulating user attention and serve other social functions in group interaction. This thesis describes two socially capable conversational agents that support users in collaborative learning and group decision making activities. Their social capabilities are composed of a set of behaviors based on the Social-Emotional interaction categories identified by work in small group communication. These agents demonstrate the generalizability of our methodology for designing and implementing social capabilities across two very different interactive situations. In addition to the implementation of these agents, the thesis presents a series of experiments and analysis conducted to investigate the effectiveness of these social capabilities. First and foremost, these experiments show significant benefits of the use of socially capable agents on task success and agent perception across the two different interactive situations listed above. Second, they investigate issues related to the appropriate use of these social capabilities specifically in terms of the amount and timing of the constituent social behaviors. Finally, these experiments provide an understanding of the underlying mechanism that explains the effects that social capabilities can achieve.

No description available.

AMI Meeting Corpus (AMI) -databasen används för att undersöka igenkännande av gruppaktivitet. AMI Meeting Corpus (AMI) -databasen ger forskare fjärrstyrda möten och naturliga möten i en kontorsmiljö; mötescenario i ett fyra personers stort kontorsrum. För attuppnågruppaktivitetsigenkänninganvändesbildsekvenserfrånvideosoch2-dimensionella audiospektrogram från AMI-databasen. Bildsekvenserna är RGB-färgade bilder och ljudspektrogram har en färgkanal. Bildsekvenserna producerades i batcher så att temporala funktioner kunde utvärderas tillsammans med ljudspektrogrammen. Det har visats att inkludering av temporala funktioner både under modellträning och sedan förutsäga beteende hos en aktivitet ökar valideringsnoggrannheten jämfört med modeller som endast använder rumsfunktioner[1]. Deep learning arkitekturer har implementerats för att känna igen olika mänskliga aktiviteter i AMI-kontorsmiljön med hjälp av extraherade data från the AMI-databas.Neurala nätverks modellerna byggdes med hjälp av KerasAPI tillsammans med TensorFlow biblioteket. Det finns olika typer av neurala nätverksarkitekturer. Arkitekturerna som undersöktes i detta projektet var Residual Neural Network, Visual GeometryGroup 16, Inception V3 och RCNN (LSTM). ImageNet-vikter har använts för att initialisera vikterna för Neurala nätverk basmodeller. ImageNet-vikterna tillhandahålls av Keras API och är optimerade för varje basmodell [2]. Basmodellerna använder ImageNet-vikter när de extraherar funktioner från inmatningsdata. Funktionsextraktionen med hjälp av ImageNet-vikter eller slumpmässiga vikter tillsammans med basmodellerna visade lovande resultat. Både Deep Learning användningen av täta skikt och LSTM spatio-temporala sekvens predikering implementerades framgångsrikt. / The Augmented Multi-party Interaction(AMI) Meeting Corpus database is used to investigate group activity recognition in an office environment. The AMI Meeting Corpus database provides researchers with remote controlled meetings and natural meetings in an office environment; meeting scenario in a four person sized office room. To achieve the group activity recognition video frames and 2-dimensional audio spectrograms were extracted from the AMI database. The video frames were RGB colored images and audio spectrograms had one color channel. The video frames were produced in batches so that temporal features could be evaluated together with the audio spectrogrames. It has been shown that including temporal features both during model training and then predicting the behavior of an activity increases the validation accuracy compared to models that only use spatial features [1]. Deep learning architectures have been implemented to recognize different human activities in the AMI office environment using the extracted data from the AMI database.The Neural Network models were built using the Keras API together with TensorFlow library. There are different types of Neural Network architectures. The architecture types that were investigated in this project were Residual Neural Network, Visual Geometry Group 16, Inception V3 and RCNN(Recurrent Neural Network). ImageNet weights have been used to initialize the weights for the Neural Network base models. ImageNet weights were provided by Keras API and was optimized for each base model[2]. The base models uses ImageNet weights when extracting features from the input data.The feature extraction using ImageNet weights or random weights together with the base models showed promising results. Both the Deep Learning using dense layers and the LSTM spatio-temporal sequence prediction were implemented successfully.