| Popis: |
In the past decades, technological advance has led to a revival of natural language dia- logue systems (or conversational agents). Now, conversational agents are used for profes- sional purposes such as customer support, marketing, e-learning, and tutoring purposes, just to mention a few. They also have found their way into everyday life, e.g. as speech- based personal assistants on mobile devices or as freely available text-based information retrieval systems in the web. However, there are still many unexplored application areas for conversational agents. Personal coaching, for example, is a promising field. Coaching has been successfully applied across many areas, for example in human resource devel- opment, team-building, and improvement of various individual skills such as leadership, communication or sales, and to support humans with rational decision-making. Never- theless, personal coaching by a human coach or counsellor is very expensive and therefore only available to a limited circle of individuals in organizations. So far, conversational agents have not been extensively used for personal coaching purposes. This work intends to explore the potential of conversational agents for simulated hu- man coaching or consulting. Therefore, we have developed VPINO, a text-based dialogue system, intended for holding structured coaching conversations in German language. This work provides a detailed description of VPINO. A general maxim of many modern coach- ing approaches is to regard the client as the expert for the relevant problem, rather than seeing the coach as the expert for the client’s problem. VPINO aims to open up a new perspective and help the “expert” develop his/her own solution. Instead of giving ad- vice or pushing the client into any certain direction, VPINO uses the technique of Socratic questioning, i.e. targeted questions to support clients to reflect on their goal accomplish- ment process. VPINO is an expert on structuring the user’s implicit knowledge and merely assists the user like a human coach. The intelligence or value of our dialogue system is not in its ability in detailed understanding what the user says and will therefore not pro- vide answers to particular questions. The key is that, nevertheless, VPINO understands yet enough to keep control over the conversation, in a way that will not be noticed by the user. To do so, the system determines the dialogue act of each user utterance, i.e. the function of that utterance for the conversation. A suitable response is selected from a set of preformu- lated utterances, based on a model of potential dialogue act sequences. VPINO has been developed for application in the particular scenarios of (1) training transfer coaching for communication skills and (2) rational decision support. The effectiveness and usefulness of VPINO in both scenarios are evaluated in a series of user studies. For the scenario of training transfer coaching, we conducted a field experimental study that evaluates the effectiveness of coaching, right after participants received an online com- munication training. Training transfer, the application of newly acquired skills in everyday work, requires a maximum of support and reflection. Our results suggest that a computer- based coaching with VPINO can effectively provide this support and improve the partici- pants communication skills. Moreover, we explored the influence of the user’s personality and behaviour when talking to the computer based coach. Whereas users that are more “open to new experience” particularly benefit from chat coaching with VPINO, more scep- tical and less cooperative users do not profit from using VPINO. For the scenario of rational decision coaching, VPINO was evaluated in a series of user studies. VPINO guides the users through a process of structured decision-making on a distinctly rational basis, in the spirit of Benjamin Franklin’s method “Pros and Cons”. Par- ticipants were free to talk about a decision problem of their own choice. Whereas VPINO could successfully help participants with rational problems, the conversation was less ef- fective with more emotionally stressful problems. However, participants with a generally less structured decision making approach particularly benefit from using VPINO. Partic- ipants that are generally more open to receiving support from an artificial coach evalu- ated the usefulness of the conversation higher than sceptical users. However, a high level of motivation and cooperativeness is required for fruitful conversation. In general, user expectations, behaviour, and willingness to cooperate have an influence on how users in- teract with the virtual coach. The use of chat language greatly varies between users in many dimensions, e.g. grammatical and syntactical correctness, style, length of answers, and politeness. This thesis shows that a natural language dialogue system can serve as a highly avail- able, cheap, low threshold solution to support a large target group, where a human per- sonal coach is not available. |
