The inferred mode protocol uses contextual reasoning and local mediators to eliminate the need to access specic modes to perform draw, select, move and delete operations in a sketch interface. This thesis describe an observational experiment to understand the learn- ability, user preference and frequency of use of mode inferencing in a sketch appli- cation. Novel methodology is presented to study both quantitative and long term qualitative facets of mode inferencing. The experiment demonstrated that participants instructed in the in- terface features enjoyed fluid transitions between modes. As well, interaction techniques were not self-revealing: Participants who were not instructed in interaction techniques took longer to learn about inferred mode features and were more negative about the interaction techniques. Over multiple sketching sessions, as users develop expertise with the system, they combine inferred mode techniques to speed interaction, and frequently make use of scratch space on the display to retrain themselves and to tune their behaviors. Lastly, post- task interviews outline impediments to discoverability and how performance is affected by negative perceptions around computational intelligence. The results of this work inform the design of sketch interface techniques that incorporate noncommand features.

The inferred mode protocol uses contextual reasoning and local mediators to eliminate the need to access specic modes to perform draw, select, move and delete operations in a sketch interface. This thesis describe an observational experiment to understand the learn- ability, user preference and frequency of use of mode inferencing in a sketch appli- cation. Novel methodology is presented to study both quantitative and long term qualitative facets of mode inferencing. The experiment demonstrated that participants instructed in the in- terface features enjoyed fluid transitions between modes. As well, interaction techniques were not self-revealing: Participants who were not instructed in interaction techniques took longer to learn about inferred mode features and were more negative about the interaction techniques. Over multiple sketching sessions, as users develop expertise with the system, they combine inferred mode techniques to speed interaction, and frequently make use of scratch space on the display to retrain themselves and to tune their behaviors. Lastly, post- task interviews outline impediments to discoverability and how performance is affected by negative perceptions around computational intelligence. The results of this work inform the design of sketch interface techniques that incorporate noncommand features.