Interaction techniques for ambiguity resolution in recognition-based interfaces

Jennifer MAnkoff, Scott E. Hudson, Gregory D. Abowd [Georgia Institute of Technology Carnegie Mellon University]

Summary:
Paper briefly presents a survey of existing error correction techniques in the user interface. These 'mediation' techniques most commonly fall in to one of the two strategies, repetition and choice. Paper calls them 'mediators' because they are mediate between the user and computer to specify the correct interpretation of the user's input. Mediation techniques then serve to resolve ambiguity of possible interpretations of user input, helping to determine which of those potential interpretations is the one user intended.
Paper lists programs that resolve ambiguity and says that they do not go ahead of choice or repetition techniques. So the purpose of this paper is to introduce four new mediation techniques other than choice and repetition.
1. Adding alternatives to choice mediators: Extending the basic n-best list choice mediator, by enabling users to control filtering and specify individual characters (ex: selecting only some part of an suggested word from the list), allow users to specify length (ex: user can feed mediator to indicate that users intended word is longer) and allow users to escapce (in case of totally wrong start)
2. Occlusion in choice mediators: Problem is that mediator interactions and user inputs can overlap. What paper suggest is: The mediator repositions all interactors that intersect with its suggestions so as to remove any occlusion.
3. Target Ambiguity: Target ambiguity arises when the target of the user's input is uncertain. For example selecting word by circling and it can be addressed by treating the mouse an area instead of a point. The magnified area is interactive and users can click on interactors inside it just as an unmagnified portion of the screen.
4. Errors of Omission: This error occurs when some or all of the user's inpout is not interpreted at all by the recognizer. Paper address this with guieded re-recognition. The user can re initiate re-recognition by specifiying a segment that should have been recognized. By doing so, the user is giving system an important new information.

Discussion:
Paper address problems of common recognition systems and makes a good separation of mediators from recognizers. Also each proposed mediator offer and address a problem in mediating techniques. As all recognizers are error prone, the methods described are useful while designing a recognition system. However, first method (adding alternatives) makes a more important point than others because other techniques seem to be specific for some domains, especially in Occlusion and Target Ambiguity.