标题: Applying the Act-Function-Phase Model to Aviation Documentation [打印本页] 作者: 航空 时间: 2010-8-13 22:08:37 标题: Applying the Act-Function-Phase Model to Aviation Documentation
作者: 航空 时间: 2010-8-13 22:08:51
Applying the Act-Function-Phase Model to Aviation Documentation David G. Novick EURISCO 4 avenue Edouard Belin 31400 Toulouse, France +33(0)562 173838 novickQonecert.fr ABSTRACT The act-function-phase model systematically relates the acts of the dialogue at time-of-use to the acts of the dialogue between author and users at time-of-development. We show how this kind of model of communicative action can be applied to the interactions described and embodied in a flight crew operating manual for a commercial aircraft. We claim that the model’s abstraction provides basis for co-evolutionary design of procedures and their corresponding documentation. Keywords Dialogue acts, aircraft procedures 1. INTRODUCTION This research, in its broadest terms, aims at improving the methodology of system development so that issues of documentation are not pushed to the end of the development cycle, where the documentation may end up having to address issues unresolved during earlier phases of development. We developed the act-function-phase (AFP) model of interaction to aid authors of documentation by providing (1) a way of reasoning formally about the effectiveness of procedures or instructions, and (2) an explicit representation of what the user is doing in a way that makes it possible to create heuristics or even formal rules that connect content and expression. The model is based on concepts from the fields of the human- Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage, and that copies bear this notice and the full citation on the first page. To copy otherwise. to republish, to post on servers. or to redistribute to lists. requires prior .s&ic pem&.sion and/or a.fee 01998 ACM 1-58113-00&W9WOOO9 $5.00 Said Tazi Laboratory for Information Science University of Toulouse 1 31042 Toulouse Cedex, France +33(0)561 633564 tazi Q univ-tlsel .fr computer-interaction and computational dialogue. Our goal in this paper is to show how this new theory of documentation development could eventually be applied in practice to documentation, particularly for safety-critical systems. An emerging research community has been looking at building documentation more or less automatically, typically from formal specifications of the underlying system. A number of prototype systems have been created that automatically praluce documentation in English or other languages that reflect a formal, abstract representation of the system or its interface [e.g., 5, 6, 8, 91. This approach is particularly appropriate for safety-critical systems and large documents subject to revision. The work reported here extends this line of research. The AFP model is a formal representation for contextualized interaction through an interface. The model is distinctive in that it accounts explicitly for contexts at both time-of-use and timeof- understanding. An introduction to the discourse theory underlying the AFP model is presented in [4]. In this paper, we show the model’s application to the domain of commercial aviation as part of an on-going project to study how to account for cockpit procedures during development of systems and their documentation for future Airbus aircraft. In particular, we are studying the flight crew operating manual (FCOM), the principal aircraft documentation furnished to the flightcrew by the aircraft builder and/or the airline. Use of the model is intended to improve the preparation and content of procedures in FCOMs for future aircraft. This paper will briefly introduce the model and present an example of its application to an FCOM procedure for using the navigation interface for an Airbus A340 aircraft. The example will demonstraE explicit representation of both domain and meta acts across contexts of use, thus providing a basis for linking the contexts at design time. We thus claim that the model’s abstraction provides basis for co-redesign of procedures and their corresponding documentation. 2. THE MODEL In this aviation setting, the act-function-phase model of interaction represents and relates differences in (1) the dialogues among the crew and a&aft on the flight-deck to 243 (2) the dialogues between the author(s) of the flight crew operating manual (FCOM) and its users. The model has three components: acts, functions and phases. Our presentation of the model’s components will use some established terms from area of dialogue models and will define a couple of new terms for extending these models to documentation. Here are our definitions of a number of terms from the literature that should help to make the model clearer: An agent is something that has beliefs, processes information and can achieve goals by interacting with other agents. In the cockpit environment, agents include the members of the crew and the aircraft itself. An act is something performed by an agent, either the user or the system, that has meaning for the recipient of the act. A domain act involves doing something that involves the agent’s main goals, like flying an aircraft. So the word “domain” here means areas of real-world knowledge, goals and accomplishment. A meta act involves doing something that involves the interaction itself rather than the agent’s real-world, domain goals. Asking someone to speak more loudly is a typical meta act. The word “meta” here means about or relating to the interaction rather than to the underlying tasks. A function is something’s “job” (i.e., goal-directed activity). For example, a function of documentation is to help users operate the documented system. L. 1. Acts The first component consists of dialogue acts, which are an abstract way of characterizing interaction. Dialogue acts are an extension of the well-known speech-acmt odel of conversation [1,7], that combine both domain acts and meta acts [2, 3, lo]. The speech-act model captures what people “do” or achieve when they say something. The model applies this notion of communicative action for humans and systems to general dialogue interaction, which is viewed as a multi-layered composition that contains both the domain acts that accomplish things in the world of the parties’ nominal goals and the meta acts that accomplish things in the sphere of the communication itself. 2.2. Functions The second component consists of the task functions intended to he achieved through the system and its documentation in the FCOM. Broadly speaking, parts of the FCOM present information about the system; these can be viewed as constituting a function of description. Complementary parts of the FCOM present action-oriented material such as procedures and checklists; these can be viewed as constituting the manual’s function of prescription. 2.3. Phases The tbird component consists of the contexts of use. Viewing the development and use of the FCOM as interactive processes suggests that there are actually two distinct phases, with two corresponding kinds of use: 1. A dialogue between system and its users, specified by the aircraft designer. We calI this the operational phase. 2. A dialogue between the documentation and the users, created by the documentation author. We call this the referential phase. In the operational dialogue, the acts are generally (but not exclusively) domain acts among agents in the flightdeck. The term agent is used in sense that elements of the aircraft (especially its interfaces) and all members of the crew have defined roles, responsibilities and capabilities. Thus the dialogues designed for the operational phase may also be among the users. As a typical example, an airline’s FCOM for the Airbus A320 specifies a procedure for Flight Plan Completion in which key elements are carried out solely by the crew rather than through an action in the interface to the aircraft; in this case, the crew compares (a) flight-plan information from the interface with (b) clearance information received from air-&& control. In the referential dialogue, the acts are generally (but not exclusively) meta acts from the documentation to the users. The design of the understanding dialogue for the FCOM is an especially difficult task because the set of intended users of the manual is not homogeneous. Foreseeable users include not only flightcrew but trainers, future designers and engineers, future authors, and regulators (i.e., for certification). Each of these users has a distinct background, set of goals and way of using the documentation. 3. APPLICATION Having introduced the AFP model generally, we now turn to application of the model to a specific instance of documentation, demonstrating that relevant features of the documentation can be mapped onto the model’s key concepts. 3.1. Example Figure 1 presents an excerpt of the procedure for navigating an Airbus A340 into a holding pattern during the descent phase of a flight. This procedure is part of the FCOM chapter of proceduresf or the Flight Management System (FMS). The interface to the FMS is called the Multipurpose Control and Display Unit (MCDU). To interpret Figure 1 it is important to understand that the flightcrew read information, enter data, snd push buttons to perform commands on the MCDU; some of these actions can modify the current flight plan. 244 (HOLDING PATTERN I Refer to “HOW TO USE” for details. PROCEDURE El F-PLN key . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . DEPRESS . LATERAL revision page . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 SELECT l HOLD prompt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . l HOLDING data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . CHECK / MODIFY l TMPY F-PLN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . CHECK / INSERT Figure I. Excerpt of FMGS Procedure for Holding Pattern Applying the AFP model to the initial actions of the holding- (7) choose-display&J,S , lateral-revision(fix0)) pattern procedure and its documentation, it is important to note first that the procedure itself consists mainly of domain acts carrying out a prescriptive function in the operational phase. In particular, the actions presented in the procedure can be abstracted into acts as follows: Depressing the “F-PLN” key causes the flight-plan page to appear on the MCDU. Among other things, this page presents tbe list of navigation “fixes” that compose the flight plan. This action can be viewed as a set of combined operational-phasea cts: (1) comrnand(U, S, offer-choice(S. U, fixes@))) (2) choose-display(U, S, flight-plan) where U is the user, S is the system, and Fs a list of fixes. Act 1 is a domain act and Act 2 is a meta act. Together, they lead to the domain and meta responses: (3) offer-choice(S, U, fixes(E)) (4) present-display(S, U, flight-plan). Moreover, the documentation of the action can be seen as a referential-phase act that can be linked to Acts 1 and 2: (5) inform(S, U, proc-action@epress-F-PLN key)) in the context of the holding pattern procedure. Selecting the lateral revision page is performed relative to one of the fmes shown on the F-PLN page. This causes the MCDU to present what is essentially a dialogue box for changing the current flight plan at the selected fix. This action, for a fix F, can be viewed as a set of combined operational-phasea cts: (6) select(U. S, fix(F)) where Act 6 is a domain act and Act 7 is a meta act. Together, they lead to domain and meta responses involving a set of action types As.: (8) offer-choice(S, U, actions-types-for(fix(F), action-types(W)) (9) presentJisplay(S, U, lateral-revision(fix(F)) Selecting the hold prompt causes the system to understand that the kind of navigation function to be undertaken relative to the fix will be a hold. Note that this does not cause the hold to happen but simply says that when the modified flight plan is activated there will be a hold associated with this fix. This action can be viewed as an operationalphase domain act: (10) select(U, S, actionJype(hold-at)) that results in the following system acts: (11) S shows data for hold at fix F (12) S offers choice of actions(revise, do or cancel) The succeeding actions in the procedure give rise to further, similarly expressed, acts. Note that this prooedure as presented in the FCOM also contains some clear referential-phase me&acts such as the titling convention for the procedure’s name, the use of the label ‘procedure,” and the icon for the F-PLN key. When modeled formally in the analyses reported below, the meta-referential acts typically require notation that marks parts of the presentation of the procedure. For example, the holding-pattern procedute involves generating tokens to denote the limits of the procedure: 245 mark(S, tO07, procedure): Marked with title and formatting <t007> command&J, S, offer-choice(S, U, fixes@))) (rest of procedure omitted) </t007> Similar tokens mark other me&referential elements such as the denotation of “F-PLN” as a key by surrounding it with a box. Here, then, is a complete AFP representation of an airline’s FCOM for a procedure for completing entry of preflight information on winds through the interface to the FMGS. The specific meaning of the procedure is not particularly interesting or relevant here. The phases and functions are noted in the lefthand columns (for example, RM means referential-meta and OD means operational-domain), interface fields are denoted by a leading question-mark (e.g., ?site), and variables are denoted by lower-case italic characters (e.g., w), and a variable’s class is denoted by an expression following a colon. RM RD OD OM OD RM RD OD OD OM OD mark(S, tO06, procedure): Marked with formatting <t006> if(not(check(U, tropo, ?standard)) then(modify(U, tropo, ?site)) command(U, S, display(S, U, page(history-wind))) inform(U, S, wind(w:flt-lvl(f), windspeed(m mark(S, tOO6ae, xplanation) Marked with indented formatting ct006a> enter&J, wind(w: { cruise wind from afpam)), scratchqad) enter&J, wind(w), field-for(fltJvl(f)) inform(S, U, causes(enter(Uw, ind(w), field-for(fltJvl(f))), effects( (entitle(S, page0listory_wind), wind)))) command(U, S, use(wind(w), page(wind))) </t006a> </t006> This example is representative of the procedures we analand shows how acts in both functions and both phases ate interleaved to produce the documentation. Some prooedures were much longer. Use of AFP along these lines should aid authors of documentation in a number of ways. First, the predicate representation of acts makes possible formal reasoning about the effectiveness of the procedures. If suitable initial and final conditions were formulated, it should be possible to show that a documented procedure does (or does not) connect them. Second, the explicit representation of acts in terms of both functions and phases makes possible connecting them through heuristics or perhaps more formal rules that lead from content to expression. 4. ANALYSIS Using the AFP view, a detailed analysis was conducted, of several FMGS procedures. In addition to determining the practicality of applying the AFP view, the analysis sought to determine typical acts, both domain and meta, and to provide a basis for a rough estimate of the number of acts, relations ard entities contained in the FMGS sections of the FCOM. The analysis found acts and relations that were either dependenot r independent of the domain, plus a large number of domain entities. In addition, an analysis was conducted of a sample of differences observed between FCOMs of various airlines. This analysis sought to characterize these differences in terms of the AFP view, particularly with respect to act (domain or meta) and phase (referential or operational) categories. The analysis found consistent patterns of kinds of changes, as well as patterns of act categories associated with each other in common for single differences. 4.1. Analysis of extended section of FCOM An extended AFP analysis was conducted for a ten-page sample of the FMGS proceduresp resentedi n an airline’s FCOM for the Airbus A320 aircraft. The sections analyzed, which present FMGS procedures for flight-plan entry, were translated into predicate expressions representing acts, entities, and domain and abstract relations. As noted earlier, an act is something performed by an agent, either the user or the system, that has communicative value to the recipient of the act; acts are the “verbs” of the representation. Entities are the “nouns” of the representations. Relations am the “adjectives” of the representation; they qualify or describe entities or relate acts and entities to each other. A domain relation has meaning in terms of the domain of the aircraft; it’s a relation that has meaning that comes from the real world. An abstract relation has meaning outside the domain; it’s a relation like “and’ or “with” that could apply to any domain. We found that both referential- and operational-phase acts were expressed. Note that the acts, relations and entities were not defined with formal semantics; they were treated as defined by the commonsense meanings that arise out of their actual use in the FCOM. In the case of some complex representations of 246 domain entities that were not central to the process of encoding the acts, the analysis substituted bracketed comments in place of predicates. The analysis identified 20 acts in these sections of the airline’s FCOM. Some of these acts are relatively generic, such as command, do, inform, offer-choice, request, select and use. Others are more domain-specific, such as arm, check, clear, display, modify, and power-cycle. Some of the acts, such as entitle and mark, are clearly meta in the referential phase. Overall, the set of acts appears to be relatively stable, in that progressively fewer and fewer new acts were added as the analysis progressed to additional procedures, and most of this growth involved domain-specific acts. This suggests that, at the generic level, the interactions in the FCOM are capable of being abstracted into a finite set of dialogue acts. The analysis identified 53 domain entities. That is, these are unary predicates that represent static domain concepts, such as destination,jlightplan and procedure. As in the case of domaindependent relations, it appears that the set of domain entities will continue to grow as additional procedures are analyzed. Again, to a large extent, this simply reflects the fact that additional procedures relate to new kinds of system functions or flight phases, thus introducing new domain entities such as navaid or departure. The analysis identified ten relations that appear to be relatively domain-independent. That is, these are predicates of one or more arguments that classify or relate other predicates in terms that do not seem to be specifically tied to the FCOM domain. Examples of domain-independent relations include rmd and denotes. As was the case for the domain-independent acts, the rate of growth of the number of domain-independent relations fell as additional procedures were analyzed. This suggests that the set of generic, abstract relations necessary for the AFP view is likely to be stable. The analysis also identified 42 relations that appear to be domain dependent. That is, these are predicates of one or more arguments that classify or relate other predicates in terms of domain-related concepts. Examples of domain-dependent relations include authorized, mode and speed. In contrast to the domain-independent relations, it appears that the set of dornaindependent relations will continue to grow as additional procedures are analyzed. To a large extent, this simply reflects the fact that additional procedures relate to new kinds of system functions or flight phases. As the notions of entity and relation were defined syntactically, based on whether or not the predicate had arguments, the distinction between the two should be regarded with caution. This is not a serious issue for the purposes of the analysis, however, as both domain-dependent relations and domain entities produced basically identical results. Overall, this suggests that, at least for the FMS sections of an FCOM, there are likely to be roughly a dozen domainindependent acts, thirty domain-dependent acts, a dozen domainindependent relations, and several hundred domain-dependent relations and domain entities. 4.2. Analysis of FCOM changes In addition to the extended analysis of part of an individual FCOM, as reported above, the project also examined 88 consecutive differences between A340 FCOMs of two airlines and of Airbus Industrie in terms of the AFP view. The analysis looked as two sections involving standard operating procedures, totaling approximately 60 pages of material. Each observed difference was analyzed using the AFP view, and classified in terms of act (domain or meta) and phase (operational or referential). The results are summarized in Table 1. The choice of sections coded ensured that the analysis would elements across functions (descriptive or prescriptive). cover Table 1. Distribution of acts and phases The data reported in Table 1 suggest, not surprisingly, that most of the changes in FCOMs involve domain acts in the referential phase and meta acts in the operational phase. So, for example, the deletion of a statement such as The pilot’s view from the cockpit of the A340 during approach and landing is particularly good. The cockpit cut off angle is 20 degrees. would be a domain-referential act because it refers explicitly to domain knowledge, and a me&operational act because it does not involve a change in the way that crews are supposed to operate the aircraft. Less frequent are acts that are meta-acts in the referential phase. These would be acts such as changing the presentation style or organization of the FCOM. Even less frequent are acts that are domain acts in the operational phase. This seems logical, as changes to domain-operational acts would be changes to the A34O’s actual procedures. For example, Air France specifies exact wording to be us&l in requests and announces for different cockpit-instrument displays. In fact, the four classes of acts am not independent Just as in the analysis of the individual FMGS procedure discus& above where a single action can constitute more than one kind of a& so too can a single change affect more than one kind of act. Accordingly, the project analyzed the relationships among the classes of acts for the 88 FCOM changes previously classified. The results of this analysis are presented in Table 2. 247 Tally 1 Act/Phase Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . .1a9.. .........ii. . .. . . .. . . . . . . . . . . . . . . .D . . . R. . . .- . M. . . . .O . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . .1 . . 1. . . . . . . . . . . . . . . .i . . . . . . . . . . . . . . . . . . . . . . . .M . . . . R. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . .6 . . . . . . . . . . . . . . . . ji . . . . . . . . . . . . . . . . . . . . . . . . D. . . . R. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . .4 . . . . . . . . . . . . . . . . .! . . . . . . . . . . . . . . . . . . . M. . . . O. . . . -. .M . . . . R. . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . .4 . . . . . . . . . . . . . . . . ii ..................D..o.. ................................. .. . . . . . . . . . . . . . .3 . . . . . . . . . . . . . . . .ii. . . . . . . . . . . . . . .D . . . O. . . . -. .M . . . . O. . . . -. .D . . . R. . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . .3 . . . . . . . . . . . . . . . . iI . . . . . . . . . . . . . . . . . . . . D. . . .O . . . .- . D. . . . R. . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . .3 . . . . . . . . . . . . . . . . ii . . . . . . . . . . . . . . . . . . . D. . . . R. . . .- . M. . . . .R . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 1. . . . . . . . . . . . . . . . .; . . . . . . . . . . . . . . M. . . . O. . . . -. . D. . . .R . . . -. .M . . . . R. . . . . . . . . . . . . . . . . . 1 ; MO Table 2. Distribution of act/phase dunes As could be expected from the data in Table 1, by far the most frequent pairing of act and phase classes was a domainreferential act and a meta-operational act. In fact, this pairing appears to occur more consistently than would be predicted simply by random association of these two frequent categories, indicating that the relationship is systematic. Aside from one instance of a change that was classified MO-DR-MR, there were no instances of act/phase class pairings along the other diagonal of Table 1, even though the relative frequency of acts in these classes might be expected to produce some pairs. In point of fact, the absence of these pairings is entirely logical: a substantive change in domain operations (DO) will have little necessary association with a change in the presentation style (MR) of the FCOM. A relatively frequent kind of FCOM difference was the single meta-act in the referential phase (MR). This classification corresponds to a change in the presentation or organization of the FCOM without an associated change in content. 5. DISCUSSION The sections of FCOM we analyzed primarily presented procedures rather than description of systems. For parts of FCOMs that focus on descriptions, we expect that the mix of kinds of acts would differ. In particular, there number of operational-phasea cts should decline radically, and the number of domain-referential acts should increase correspondingly. This suggests that the trend seen in Table 1 should be even sharper, as the potential for changes in domain-operational acts would be further reduced. The analysis has other limitations. In particular, we concenti more on representation of the semantics of procedures and less on the forms of expression. This focus can be seen in our use of informal rather than predicate expressions in our descriptions of formatting in our examples. Domain and meta acts have different characteristics when used to produce the operational and referential functions of dialogue. Domain acts have a clear and systematic consistency across both functions. That is, an operational domain act will normally have a recognizable counterpart in the set of referential domain acts. In contrast, meta acts of operational dialogue and of referential dialogue are basically different. Meta acts in operationald ialogue can be characterizeds imply on the basis of the interface through which the crew and aimraft communicate. Meta acts in referential dialogue, however, depend on the tools used by authors and on the type of the reader they are addressing. Thus one of our principal aims is to analyze the relationships of information transmitted in the two phases to develop a method that maps domain and, especially, meta acts across the two functions. The coincidence of changes across function and phase classes, as reported in Table 2, is suggestive evidence for the existence of the link we claim exists between the functions. That is, the high incidence of FCOM revisions that involve DR-MO changes means that ways of referring to things are associated with instructions on how to do things. Also, the existence of pairs such as MO-MR and DO-DR suggests links between content and mode of expression. What these links are, in any particular case, may simply reflect the authors’ style practices but may also reflect more fundamental knowledge about systems and documentation that could be captured in heuristic form. To the extent that this is possible, whether for a specific stylebook or more generally, the links between the phases hold out the hope of aiding technical authors in building documentation. Our current work on the AFF’ model involves empirical validation. This includes “concretizing” the model for a useful section of the A34O’s FCOM, using the model to understand reasons why airlines revised the FCOM in particular ways, and obtaining feedback from pilots on whether the model expresses concepts that are of use to them. 6. ACKNOWLEDGMENTS This work was supported by a research contract from Aerospatiale Aeronautique. Airbus Industrie. Air France and Lufthansa provided important assistance. 7. REFERENCES 1. Austin, I. (1962). How to do things with words. Cambridge, MA: Harvard University Press. 2. Carbonell, J. G. (1982). Me&language utterances in purposive discourse, Technical report CMU-CS-82-185, Department of Computer Science, Carnegie-Mellon University. 3. Novick, D. (1988). Control of mixed-initiative discourse through meta-locutionary acts: A computational model. Doctoral dissertation, available as Technical Report CIS- 248 TR-88-18, Department of Computer and Information Science, University of Oregon 4. Novick, D., and Tazi, S. (1998). Plight crew operating manuals as dialogue: The act-function-phase model. Proceedings of HCLAero’98, Montreal, May, 1998, 179- 184. 5. Paris, C., and Hartley, A. (1996). Multilingual document production from support for translating to support for authoring. Information Technology Research Institute Technical Report No. ITRI-96- 17. 6. Power, R., Scott, D., and Evans, R. What you see is what you meant: Direct knowledge editing with natural language feedback Information Technology Research Institute Technical Report No. ITRI-97-03, 1997. 7. Searle, J. (1969). Speech acts. Cambridge: Cambridge University Press. 8. Thimbleby, H. (1996). Creating user manuals for use in collaborative design, CHI 96 Conference Companion, 279- 280. 9. Thimbleby, H., and Ladkin, P. (1995). A proper explanation when you need one. Proceedings of the BCS Conference on Human Computer Interaction, Huddersfield, UK. Cambridge: Cambridge University Press. lO.Traum, D., and Hinkelman, E. (1992). Conversation acts task-oriented spoken ztelligence, S(3) , 575-599. dialogue, Computational 249