intelligently handling input. In our paper we deal with th e other end of the pipeline presentations . 3) Explicit Models i n Intelligent Interface Desig n James A . Mason and Jack L . Edwards . AI Managemen t & Development Corp . Abstract : This paper provide s support for the followin g points regarding the desig n of intelligent interfaces (o r more generally, intelligen t dialog systems) : 1 ) Constraints of human dialo g processing, including limite d attention span and partia l understanding, should b e dealt with explicitly in design of intelligent dialog systems ; 2) An intelligent dialo g system should be a partl y autonomous agent, thoug h accountable to its users ; 3) A n intelligent dialog syste m should be designed t o incorporate explicit activ e models of tasks, users, th e system itself and the proces s of dialog with the user ; 4) Models should be built an d maintained both for types o f tasks, users and dialo g processes and for specifi c instances for tasks, users an d dialogs . In particular, a n intelligent dialog syste m should model users in term s of their individual huma n characteristics ; 5) Th e configuration of model s maintained by an intelligen t interface should be sel f referencing . That is, th e system's model of itself , which is a component of th e model configuration, shoul d be able to refer to the mode l configuration as a whole, an d should be able to access th e assertions in tha t configuration ; and 6) Dialo g processes should be modele d from a first-party (system ) view and the system' s second-party view of the user , rather than from a third-party (observer's) view . 4)Communicating Wit h High-Level Plan s Jeffrey Bonar . Learnin g Research & Developmen t Center, University o f Pittsburg h Abstract : We discuss ou r experience with an interfac e that gives users the ability t o directly represent an d manipulate goals at severa l levels of detail . Th e interface is built into Bridge , a tutorial environment fo r novice programmer s [Bonar88] . The name come s from our intended "bridge " between novice and exper t conceptions of programming . In order to understan d student designs and partia l programs, Bridge provide s languages that allow a student to talk about his o r her high-level designs an d partial work . We call th e vocabulary of these language s plans . Plans are bundles o f knowledge about the standar d subtasks in a domain , designed and organized base d on a typical user's point o f view . Many intelligent interface s monitor low-level use r actions, attempting to infe r higher level plans . Thes e inferences are typicall y implemented with partia l matching schemes, based on a plan catalog (see, fo r example, [Johnson86]) . Th e inferences allow the system t o complete user actions, correc t errors, or provide tutoria l assistance . This approach t o inference of user intentions i s quite difficult . We propose a different approach, designe d to more effectively an d accurately capture use r intentions . Our approach gives users a very high-level plan language . By "high-level" we mean a language that is informal , vague, contains much implici t information, and is designe d to represent goals of interest to a particular class of users . In particular, the pla n language makes assumption s about the user's backgroun d knowledge and overal l intentions . This is consisten t with our interest in providin g interfaces to professional s and domain experts who hav e no programming experience . We focus on users who ar e experts in a particular tas k domain and are using a computer to extend o r augment that expertise . Ou r system must take such a user' s specification and derive a n implementation using th e primitives provided by a standard computer system . In the rest of the article w e begin by developing a framework for approachin g intelligent interfaces . I n particular, we discuss th e dilemma of a very high-leve l programming languag e intended for use by expert s who are not programmers . 5) Graphica l Knowledge-Based Structur e Editor s Marilyn Stelzner and Alle n Cypher . IntelliCor p Abstract : AI has mad e significant advances in th e design of graphica l knowledge-based interfaces . One type of knowledge-base d interface, the graphical mode l editor, extends the "toolki t notion of interface" , exemplified by Bill Budge' s Pinball Constructor Set 1 . I n these interfaces, "the desire d operations are done simply b y moving the appropriate icon s onto the screen an d connecting them together . Connecting the icons is th e equivalent of writing a program . . . There are n o hidden operations, no synta x of command names to learn . " 2) The end user can quickly , and with minimal training , build up a description of a complex structural model, ru n SIGCHI Bulletin July 1988 Volume 20 Number 1

/lp/association-for-computing-machinery/communicating-with-high-level-plans-JjKNszEToD