While the formal and natural language requirements specification process was examined unidirectionally (i.e. only from the specifier's perspective) in the first two funding phases (see diagram below), this sub-project focuses on two-way communication in the third funding phase. The aim is, on the one hand, to resolve ambiguities in the specification cooperatively in a personalized dialogue and, on the other hand, to convey knowledge about a service configuration to the customer in natural language in order to clarify the scope of what is feasible at an early stage. This results in an informed end user who, through targeted inquiries and queries, receives information on which requirements can be implemented and can decide for themselves whether a given service configuration is acceptable to them or whether they need to modify their original requirements depending on the task template. In the third funding phase, we place the end user even further at the center of our considerations.

To this end, we are not only creating the necessary framework in the area of User-Centered Dialogue Planning and Design, but also investigating the question of how highly individual conversations with the end user can be controlled in a targeted manner without defining rules in advance, thus enabling a flexible dialogue structure in a multimodal environment in a user-friendly way. Therefore, active learning and comparable techniques are to be researched in order to learn from user feedback which possible steps (e.g. queries, examples as explanations, suggestions) are generally possible and how these can be implemented.

In the Automated Completion of Task Templates work area, the knowledge from the machine-readable configuration process is recorded in templates, then interpreted in natural language and subsequently flows into dialog planning. The aim is to be able to ask the user specific questions that make the instantiation values of the task templates more accurate or more complete. To this end, procedures must be developed that translate configuration parameters into statements that people can understand in order to ultimately be able to ask specific questions.

With the Service Configuration Transparency workspace, we make the configuration processes carried out more comprehensible for the end user. In this way, end users should not have to try things out to find out whether their expectations of the service configurations have been met, but rather gain an early understanding of the feasibility of requirements. To this end, we are researching how user-typical formulations based on the requirement descriptions from the second funding phase are suitable for training a learning NLG approach and which techniques can be used to generate text in the question-answer dialog with the user that also provide the necessary flexibility.

More information about SFB901 (extern) >

< Back to overview