T R A C K I N G H U M A N - C O M P U T E R D I A L O G U E S I N
P R O C E S S C O N T R O L A P P L I C A T I O N S
René van Paassen
Universität Gesamthochschule Kassel, Laboratory for Human-Machine Systems Mönchebergstraße 7, 34109 Kassel, e-mail repa@imat.maschinenbau.uni-kassel.de
Abstract. Communication acts exchanged between a human operator and a machine may be seen as forming part of dialogues, cohérent sets of information exchanged to reach one or more goals in the Man-Machine System. In process control applications, where many click-and-point type interfaces are used, dialogues may easily be interrupted temporarily. This paper describes a dialogue tracking system, that recognises human-computer dialogues and keeps track of their progress. Results from the tracking System may be used for supporting functions in the Man-Machine System.
Keywords. Process control, man/machine interface, computer interface, human/computer dia-logue
1 INTRODUCTION
In order to be able to cope with the complexity of many modem process control applications, the requirements posed on the economical opération, and the safety requirements, an increased atten-tion must be devoted to the design of the Man-Machine Interface (MMI).
In an MMI, two main components may be distin-guished (Johannsen, 1995). One is the présenta-tion system, which is concerned with the format of the information that is presented to the user. The second is the dialogue system, which controls the flow of information between operator and MMI. A dialogue system cannot always be seen as one integral component. Often software modules from différent générations and différent manufacturers have to cooperate. Via parallel Channels, thèse components are then competing for the operator's attention and for room on the terminal screen. When this is the case, a dialogue system cannot provide support for the user in keeping track of the dialogues she or he is maintaining with the various components in the MMI.
These dialogues between the user and the con-trolled process or the support functions consist of logically cohérent séquences of actions and re-actions, called interaction acts. Examples of such interaction acts are the reporting of an alarm,
con-trol actions by the HO, requests for information, etc.
This paper describes a dialogue tracking system, which is able to keep track of the dialogues be-tween various components in the MMI and the hu-man operator. For this function, the tracking Sys-tem interprets the log of the interaction between the Human Operator (HO) and the MMI. The tracking system is in first instance developed for intégration in the software toolkit of the A M I C A1
project (Fabiano et ai, 1994; Averbukh et ai, 1995). This toolkit contains a system for logging the interaction acts that are exchanged between the user and the MMI, and thèse logged data are used as the input data for the tracking; system. The construction of this dialogue tracking system is based on a pragmatic approach. The possible di-alogues between users and the MMI are described in so-called dialogue schemes. These are prototype dialogues, that describe of which interaction acts a typical dialogue consists. Additional information stored in a dialogue scheme describes in how far ac-tual dialogues may deviate from the scheme. Such déviations may take the form of a allowable différ-ences in order of the interaction acts, whether acts may be skipped, or repeated and at what points it is possible to interrupt the dialogue and continue with another dialogue.
^rite/EuRam project 6126 AMICA, Advanced Man-machine Interfaces In process Control Applications, sup-ported by the European Commission
Dialogue Tracking System
Human Operator TES
statu / / i » . \ / / /coriip. status\
\ /|meas. var.;;i^—
presentation a *
Fig. 1. A n information flow view of the dialogue tracking System in relation with other components in the interface. The information flow from the H O to the MMI takes place at the level of actions. A t the hand of information fiows in the présentation System, the tracking System infers which dialogues the H O ïs working on. Information at the baise of the pyramids is raw, unprocessed data, in the middle and at the top information is processed and more abstract.
T h e o u t p u t o f the dialogue t r a c k i n g System is a list of current dialogues t h a t c a n be presented directly to the H O . A t the t o p o f this list the active d i a -logue is presented, this is the dia-logue that the H O is currently w o r k i n g o n . T h e other dialogues i n the list are dialogues t h a t have been interrupted a n d are n o t yet c o m p l e t e d . T h e o u t p u t o f the t r a c k i n g System c a n also b e used t o check the consistency a n d completeness o f the h u m a n ' s actions. W h e n then one o r more essential acts i n a dialogue are forgotten, o r t h e H O starts a dialogue that is prob-a b l y n o t the prob-answer t o urgent problems diprob-agnosed b y the M M I , a w a r n i n g m a y be issued. T h i s w o u l d be s i m i l a r t o the concept o f p i l o t assistants ( W i t -tig a n d O n k e n , 1992; L e Fort et ai, 1993) that are being developed for aeronautical applications.
2 I N F O R M A T I O N F L O W V I E W O F T H E D I A L O G U E T R A C K I N G S Y S T E M It is also possible t o view t h e dialogue t r a c k i n g system at a more abstract level. A schematic rep-résentation o f this view is given i n F i g u r e 1. I n f o r m a t i o n f r o m the p l a n t is t r a n s m i t t e d as r a w d a t a , i.e. measured values, to the m a n m a c h i n e i n -terface ( p a t h a). B y the T e c h n i c a l S y s t e m ( T E S ) , this r a w d a t a is processed into integrated d a t a t h a t provides, at a more abstract level, i n f o r m a -t i o n abou-t -the f u n c -t i o n i n g o f -the plan-t (pa-th b ) . T h e T E S comprises the plant w i t h t r a d i t i o n a l au-t o m a au-t i o n , a n d a supervisory conau-trol sysau-tem, a n d m a y provide analysed d a t a a n d generate a l a r m s . A n alternative for the conventional a l a r m
analy-sis m a y be a System that performs a means-ends based analysis o f the d a t a , w h i c h results i n i n d i -cations o f how various sub-goals a n d goals i n the plant are reached. T h e H O m a y select a t w h i c h level, o r c o m b i h a t i o n o f levels, the i n f o r m a t i o n is presented o n the screen (paths c l , c2, c 3 ) . Information c o m i n g f r o m the H O , i.e. his c o n t r o l actions, i n f o r m a t i o n requests, opération a n d usage of support system, is u s u a l l y passed d i r e c t l y t o the appropriate component, either the p l a n t , o r a supporting function (path d ) .
There is thus a d i s p a r i t y between t h e i n f o r m a -tion c o m i n g f r o m the p l a n t , t h a t is available as b o t h low-level (raw) d a t a a n d as higher-level (processed) d a t a , a n d t h e i n f o r m a t i o n about the h u -m a n ' s intentions a n d actions, that is o n l y available as low-level d a t a .
T h e dialogue t r a c k i n g system a t t e m p t s t o detect a cohérence i n t h e user's actions, a n d t o detect which dialogues are m a i n t a i n e d b y t h e H O a n d the M M I . It provides therefore i n f o r m a t i o n at a more abstract level about t h e tasks a n d goals t h a t the user is w o r k i n g o n ( p a t h e). T h i s i n f o r m a t i o n , together w i t h the i n f o r m a t i o n f r o m the T E S c a n be used b y s u p p o r t i n g functions i n the M M I , a n d fed back t o the user.
3 A P P L I C A T I O N S O F A D I A L O G U E T R A C K I N G S Y S T E M
4 I N T E R A C T I O N L O G G I N G I N T H E A M I C A S Y S T E M
Interpreting t h e user's actions and determining w h a t dialogue (s)he is w o r k i n g o n , does i n itself not constitute a c o n t r i b u t i o n t o the safety, econ-o m y econ-or u s a b i l i t y econ-o f a M a n - M a c h i n e S y s t e m . O n e of the first applications that comes to m i n d is the présentation of the current dialogues i n a list. T h i s allows the user t o keep track of a i l dialogues that have been started. T h i s is especially useful i n sit-uations where the h u m a n operator has to manage a large n u m b e r o f possibly p a r a l l e l tasks. A task analysis for one of the e x a m p l e applications i n the A M I C A project, a cement m i l l , i n d i c a t e d that op-erators w o u l d like t o have a reminder o f pending tasks t h a t have been interrupted when an a l a r m or other urgent s i t u a t i o n h a d t o b e handled (Heuer
et al., 1993). A n example o f how such a support w o u l d f u n c t i o n is given i n T a b l e 1.
O n the basis o f the results f r o m the dialogue track-i n g System track-i t track-is also posstrack-ible to check whether dtrack-ia- dia-logues have been completely h a n d l e d , a n d that the user d i d not forget any essential acts. Such a slip is a n error o f o m i s s i o n f r o m the user. W h e n the d i -alogue describes a control opération o n the plant, p r o v i d i n g a n early w a r n i n g m a y also prevent the plant f r o m getting i n t o an undesirable state. A n a l y s i s o f the plant's state w i t h techniques such as M u l t i l e v e l F l o w M o d e l l i n g ( L i n d , 1990) m a y i n -dicate w h i c h goals o f the plant are not met, a n d thus need the user's attention. These goals m a y be m a t c h e d against the activities of the operator, to détermine whether he or she is devoting atten-t i o n atten-t o atten-the i m p o r atten-t a n atten-t goals a n d atten-thaatten-t atten-the proper stratégies are chosen.
In a multiuser System, a n overview o f the d i a -logues t h a t the different users are w o r k i n g o n can i n a d d i t i o n provide a t o o i t o optimise the coordi-n a t i o coordi-n of the crew.
T h e f o l l o w i n g applications were considered of most use a n d will be the first to be developed:
• C h e c k i n g that the user does not o m i t essential acts f r o m dialogues.
• E n s u r i n g t h a t t i m e - c r i t i c a l dialogues are not left unattended for too long.
T o e x p l a i n the f u n c t i o n i n g o f the dialogue track-i n g System, track-i t track-is necessary t o gtrack-ive a s m a l l descrtrack-ip- descrip-t i o n o f descrip-the way indescrip-teracdescrip-tion acdescrip-ts are logged i n descrip-the A M I C A t o o l k i t . Interaction acts are f o r e x a m -ple notifications o f a l a r m s , i n p u t s b y the user t o be sent to the plant, or to s u p p o r t i n g functions i n the M M S , changes i n the display layout etc. Thèse acts, whether they come f r o m the user o r f r o m the various components i n the M M S , are recorded and logged by the interaction logging f a c i l i t y i n the A M I C A t o o l k i t (Fabiano et al., 1994). A i l logged acts hâve a name, a n d i f applicable i n f o r m a t i o n about t h e topic is also logged. T h e t o p i c is, i n analogy to n o r m a l c o m m u n i c a t i o n , the subject o f the conversation. I n this case t h e t o p i c always refers t o a part of the controlled p l a n t .
N o t a i l the i n f o r m a t i o n i n t h e i n t e r a c t i o n l o g is necessary t o recognise a n i n t e r a c t i o n act. In m a n y cases k n o w i n g t h e n a m e o f t h e a c t suffices t o uniquely identify the act, sometimes the t o p i c is also required.
5 D E S C R I P T I O N O F D I A L O G U E S ; D I A L O G U E S C H E M E S
A t the start o f the A M I C A t o o l k i t , the dialogue tracking system reads a file w i t h descriptions o f t y p i c a l dialogues. Thèse descriptions, the dialogue schemes, contain gênerai i n f o r m a t i o n about the d i -alogue, such as its name, a n associated goal, a n d the description o f the interaction acts i n the d i a -logue, i n the f o r m o f act.atoms, descriptions of a n interaction act that c o n t a i n the necessary i n f o r m a -t i o n -to uniquely dis-tinguish -this ac-t f r o m -the o-ther possible acts. A section o f the file w i t h dialogue schemes is given i n F i g u r e 2.
A d d i t i o n a l l y , the dialogue scheme contains infor-m a t i o n that describes how t h e a c t u a l dialogues m a y difler f r o m the dialogue described i n the d i -alogue scheme. T h e following specifiers m a y b e given t o the act_atoms:
INTERRUPTIBLE
It is possible t h a t the dialogue is interrupted after this act.SKIPPABLE
T h i s act m a y be o m i t t e d .REPEATABLE
T o indicate t h a t a n act m a y be re-peated.T a b l e 1 A h y p o t h e t i c a l scenario for a cement m i l l , and the reactions of the dialogue t r a c k i n g System. T h e user has a synoptic presentation of the plant i n the b a c k g r o u n d , a n d p o p - u p Windows for the electronic diary, for t a k i n g a sample of the m i l l e d product a n d for i n s p e c t i n g the dialogue stack.
A c t i o n o f user or p l a n t Dialogue T r a c k i n g S y s t e m T h e user opens the sample w i n d o w
T h e user issues the c o m m a n d t o take a sample.
T h e user closes the s a m p l e w i n d o w T h e user opens the d i a r y w i n d o w . T h e user enters some text.
T h e T E S reports a n alert for the roller press (a press t h a t crushes the ce-ment clinker before i t is fed t o the m i l l proper).
T h e user closes the d i a r y w i n d o w . T h e user adjusts the roller press speed. T h e T E S reports t h a t the roller press values are O K again.
T h e user opens the d i a r y w i n d o w again. T h e sample results come i n .
T h e user s u b m i t s the d i a r y entry a n d closes the d i a r y w i n d o w .
T h e user checks the dialogue stack a n d sees that "take sample" is s t i l l there. T h e user opens the s a m p l e w i n d o w .
T h i s action m a y belong t o the dialogues "take s a m p l e " a n d "inspect samples". B o t h dialogues are started o n a tenta-tive basis, b u t not yet shown o n the stack.
"take sample" is made t o be the active dialogue. T h e stack remains unchanged.
"diary entry" is made t o be the active dialogue, "take s a m -ple" moves to place t w o o n the stack.
T h e stack remains unchanged.
T h e dialogue t r a c k i n g System starts u p t r a c k i n g o f the d i a -logue "roller press adjustment", b u t since t h e user has n o t yet worked o n i t , i t is not placed o n the stack.
T h e stack remains unchanged.
"roller press adjustment" becomes the active dialogue, "roller press adjustment" is considered complète a n d m a r k e d for éjection.
" d i a r y entry" becomes the active dialogue a g a i n , "roller press adjustment" is removed f r o m the stack.
T h e tracking system takes notice o f this act, b u t t h e stack remains unchanged, because no p a r t i c i p a t i o n o f the user is involved.
"diary entry" is m a r k e d for éjection.
There is no dialogue scheme defined for "inspection o f the dialogue stack", so the dialogue t r a c k i n g system does n o t react.
"take sample" becomes the active dialogue, " d i a r y entry" is removed.
F u r t h e r m o r e i t is possible t o indicate h o w the or-dering o f the interaction acts i n a real dialogue m a y differ f r o m the order i n w h i c h they appear i n the dialogue scheme. F o r this the acts i n the scheme are grouped i n t o blocks, see F i g u r e 3. B l o c k 1 con-tains the acts w h i c h n o r m a l l y indicate the start of the dialogue, the last block usually contains the acts t h a t i n d i c a t e t h e e n d o f the dialogue. T h e strictness w i t h w h i c h the user w i l l follow the block order c a n be specified
(STRICTJDRDER).
A new block starts w i t h a n interaction act that has aNEXTJICTSJBLOCK
specifier, the n u m b e r following this specifier indicates the order strictness o f the acts w i t h i n the block.B l o c k 0 is a special case, i t is reserved for " w i l d -c a r d " a-cts, a-cts that m a y o-c-cur anywhere i n the dialogue. T h e example dialogue i n F i g u r e 2 uses two such w i l d - c a r d acts.
T h e d i v i s i o n o f a dialogue scheme into blocks gives
sufficient freedom t o specify t h e dialogue t h a t re-sults when the user has t o fill o u t t h e questions i n a n u m b e r o f consécutive
Windows,
thus w i t hSTRICTJDRDER
1 . 0 , w h i c h spécifies t h a t the order i n w h i c h the Windows are presented is fixed, a n dNEXT-ACTS-BLOCK
0 . 0 , w h i c h spécifies t h a t the Order i n w h i c h t h e questions i n a w i n d o w are a n -swered is determined by the user.U s i n g the order i n f o r m a t i o n , the dialogues track-ing system c a n résolve conflicts t h a t arise w h e n the same interaction act occurs at m o r e t h a n one place i n a dialogue, or when i t occurs i n m o r e t h a n one dialogue.
T h e example scheme given i n figure 2 describes the m a k i n g o f a n entry i n a n electronic d i a r y ( F a b i a n o
et ai, 1993). Such a dialogue starts w i t h o p e n i n g of a d i a r y w i n d o w o n the computer's screen, then the user fills out a n u m b e r o f items, o f w h i c h some are o p t i o n a l
(SKIPPABLE
1 . 0 ) . T h e order i n w h i c hDIALOGUE " d i a r y entry" {
GOAL "administrativ« work" GOAL.IMPORTANCE 0.3 NORKAL.COHPLETION.TIHE 60.5 STRICT.DRDER 0.1
KUST.BE.COMPLETED 1.0 EJECTION complete.last.block
t »ild-card a c t s , may occur anyvhere i n the diary dialogue t these acts occur vhen a user closes tba diary windo« t u h i l e the procesa of making an entry i s not complete
ACT.ATOK WaOpen { HA.ID diary REPEATABLE 1.0 > ACT.ATON HaClose { HA.ID diary } t f i r s t block, open ACT.ATOK WaOpen < HA.ID diary NEXT.ACTS.BLOCK 1.0 REPEATABLE 0.0 }
t second block, a l l possible acts, unordered, t may be repeated, may be skipped
ACT.ATOK DiaryEnterSystem { NEXT.ACTS.BLQCK 0.0 SKIPPABLE 1.0 REPEATABLE 0.2 > ACT.ATOK DiaryEnterComponent ACT.ATOK Di&ryEnterVariable ACT.ATOK DiaryEnterAction ACT.ATOK DiaryEnterïarious ACT.ATOK DiaryEnterTime
• s e n t i n e l block, submitting an entry and c l o s i n g the
t diary »indos completes the dialogue
ACT.ATOK DiarySubmitEntry { NEXT.ACTS.BLOCK 1.0 SKIPPABLE 0.0 REPEATABLE 1.0 }
ACT.ATOK HaClose {HA.ID diary}
Fig. 2. A section of the datafue with the dialogue schemes, description of a diary entry
WaOpen Block 0, wild-card acts
WaClose
Block 0, wild-card acts
WaOpen Block 1, Start
DiaryEnterSystem Block 2, loose order, acts are repeatable and skippable.
DiaryEnterComponent
Block 2, loose order, acts are repeatable and skippable.
DiaryEnterVariable
Block 2, loose order, acts are repeatable and skippable.
DiaryEntarAcoon
Block 2, loose order, acts are repeatable and skippable.
DiaryEnterVarious
Block 2, loose order, acts are repeatable and skippable.
DiaryEnterTime
Block 2, loose order, acts are repeatable and skippable.
DiarySubmitEntry Block 3, last block, when both acts occurred, dialogue is finished. WaClose
Block 3, last block, when both acts occurred, dialogue is finished.
Fig. 3. Division of the acts into blocks. Block 0 is the block with wild-card acts, that may occur any-where within the dialogue. T h e dialogue Starts with acts from block 1. Block 2 contains the "body" of the dialogue, the last block contains the actions that close off the dialogue. Strict-ness of order within the blocks, eis well as the strietness with which the order of the blocks is followed, can be speeified.
D i a r y E n t r y act is " b l o c k i n g the w a y " . Instead i t assumes that the user has t e m p o r a r i l y closed the w i n d o w , a n d that the W a C l o s e act i n t h e b l o c k w i t h wildcards is the appropriate act. W h e n S u b -m i t D i a r y E n t r y has occurred, a n d t h e n the w o r k i n g area is closed, the t r a c k i n g system assumes that the W a C l o s e act i n the last block has occurred, because that fits better t h a n a mere w i l d - c a r d act.
the acts are performed is o n l y determined b y the user, so the strietness o f ordering i n the second block is set t o 0. W h e n a n entry has been prepared, the user c a n press a b u t t o n t o s u b m i t this entry, and t h e n he c a n close the diary.
T h e t r a c k i n g System considers the d i a r y work finished when at least one entry has been s u b m i t -ted a n d the w i n d o w is closed. T h e w a y to de-t e r m i n e de-the e n d o f de-the dialogue is speeified w i de-t h EJECTIONJIETHOD completeJ.ast_block. I n this case t h e last block contains the act S u b m i t D i -a r y E n t r y , w h i c h is speeified -as not skipp-able, -a n d W a C l o s e , w h i c h signifies the closing of the work-ing area ( = w i n d o w ) o f the diary. T h e orderwork-ing i n this block is strict. W h e n the user temporar-i l y temporar-interrupts the work o n the dtemporar-iary, the tracktemporar-ing system w i l l n o t assume that the W a C l o s e act i n the last block has occurred, because the S u b m i t
-6 S T A G E S I N T R A C K I N G D I A L O G U E S
6.1 The dialogue stach
T h e t r a c k i n g dialogue system m a i n t a i n s a stack o f current dialogues. T h e first place i n this stack is occupied by the dialogue that is believed t o b e t h e active dialogue, i.e. the dialogue t h a t the user is working o n . O t h e r places o n the stack c o n t a i n the dialogues that have been i n t e r r u p t e d .
T h e t r a c k i n g system often w i l l have t o d e a l w i t h more or less ambiguous situations, where i t is n o t clear to w h i c h dialogue a n i n c o m i n g i n t e r a c t i o n act belongs. I n order to express this a m b i g u i t y i n the i n f o r m a t i o n about the dialogue stack, a t r u t h value a c t i v e is m a i n t a i n e d . W h e n a new i n t e r a c t i o n act comes i n , the a c t i v e property is u p d a t e d , i n ac-cordance w i t h the goodness w i t h w h i c h t h e act fits
i n the h i s t o r y o f acts i n the dialogue, a n d i n ac-cordance w i t h the fact whether t h a t act is spécifie for this one dialogue, or whether i t m a y fit i n more dialogues.
6.2 Look-up of an interaction act
A t the i n i t i a l i s a t i o n o f the t r a c k i n g System, when the dialogue schemes are read i n , the tracking Sys-t e m indexes Sys-the acSys-ts i n Sys-the dialogues according Sys-t o their n a m e . E a c h entry i n this i n d e x array con-tains possible e x t r a spécifications (topic, w a J d , etc.), a n d références t o the places i n the dialogue scheme or schemes where the act was specified. T h e t r a c k i n g System interprets a n i n c o m i n g inter-a c t i o n inter-act i n three stinter-ages. I n the first stinter-age inter-a i l the dialogues a n d a i l the places where this act occurs are looked u p i n the i n d e x array, by c o m p a r i n g the act_name o f the i n t e r a c t i o n act, w i t h the act_name specified i n the dialogue scheme a n d , i f that infor-m a t i o n is available, b y c o infor-m p a r i n g t o p i c , a c t o r , r é c i p i e n t a n d wa_id. T h e resuit of this stage is a list o f dialogue schemes were this interaction act m a y fit, a n d for each scheme a list o f one or more places where the act m a y occur.
6.3 Fitting the act in a dialogue scheme
A "goodness o f fit" is calculated for a i l possible places where the act m a y fit i n t o a dialogue. W h e n this dialogue has already been started, a n d thus the previous acts are k n o w n , the r e l a t i o n between the new act a n d the previous acts is considered, a n d one or m o r e of the f o l l o w i n g cases m a y a p p l y :
1. T h e présent act follows the previous the previ-ous act, not o n l y i n the real-life dialogue, but also i n the description i n the dialogue scheme, therefore the fit is o p t i m a l .
2. B o t h acts do not follow each other i n the d i a -logue scheme b u t are i n the saine block of acts. T h e goodness o f fit is ' p u n i s h e d ' according t o the order strietness i n the block. If a low value for order strietness was given, goodness stays h i g h , otherwise goodness w i l l be lowered. 3. W h e n an act was repeated, the goodness w i l l
be corrected b y the value found after the REPEATABLE spécifier.
4. W h e n the previous a n d présent act are n o t f r o m the same block, the goodness is corrected
w i t h the order strietness for the b l o c k s . I t is possible that the correction f r o m I t e m 2 a n d this correction are a c c u m u l a t e d .
5. Lastly, the évaluation o f the goodness o f fit considers the p o s s i b i l i t y t h a t one or m o r e acts were s k i p p e d . T h e goodness is each t i m e cor-rected for the S K I P P A B L E p r o p e r t y of the pos-sibly s k i p p e d acts.
W h e n the dialogue is new, i t is checked whether i t is feasible t h a t the received i n t e r a c t i o n act signais the start of the dialogue. I n p r i n c i p l e o n l y acts i n the first block start the dialogue, b u t i t is possi-ble t o tune the t r a c k i n g behaviour so t h a t is also accepts w i l d - c a r d acts o r acts t h a t are specified elsewhere i n the dialogue scheme as s t a r t i n g acts.
6.4 Ascribing acts to a dialogue
After l o o k i n g u p the acts a n d checking the fit o f thèse acts at each o f the places i n t h e dialogue schemes where they are specified, a décision is made how t o update the i n f o r m a t i o n o n the r u n -n i -n g dialogues. T h r e e cases are possible:
1. T h e l o o k u p o f the act d i d not p r o v i d e any results, i.e. there were n o dialogue schemes specified w i t h this act, o r the a c t was evalu-ated t o have a very l o w goodness o f fit. I n this case no change is m a d e t o the dialogue stack. 2. T h e search for dialogues resulted i n o n l y one dialogue, or the goodness o f fit is o n l y g o o d for one dialogue. I n t h a t case the act is ascribed to that dialogue.
3. M u l t i p l e dialogues proved t o have a c l a i m o n the act, a n d f o r m o r e t h a n one o f t h e m a reasonable goodness o f fit was c a l c u l a t e d . I n this case the act is tentatively described t o
ail thèse dialogues. T h e . r u n t i m e i n f o r m a t i o n for a i l thèse dialogues is u p d a t e d as i f they received the act.
A tentative ascription is m a d e undone a t the next cycle o f the t r a c k i n g p r o g r a m . W h e n a new act cornes i n , a n d i t this act is ascribed t o one d i a -logue, then i t is checked whether this dialogue has a tentative ascription. T h i s tentative d e s c r i p t i o n is then m a d e définitive for this dialogue, a n d taken away f r o m the others. I f the next i n c o m i n g act is not a definite a s c r i p t i o n , b u t a tentative ascrip-t i o n , ascrip-the older ascrip-tenascrip-taascrip-tive descripascrip-tion w i l l b e m a d e définitive for the dialogue that is the most likely candidate t o be active.
7 D I S C U S S I O N
T h e dialogue t r a c k i n g System as described hère is a first, practice oriented solution for interpreting the user's actions. Solutions based o n other p r i n -c i p e s are also i m a g i n a b l e , su-ch as a knowledge-based system w i t h rules describing effect of actions i n view of system goals, or a self-learning system t h a t générâtes its knowledge f r o m actual dialogues a n d questions a n expert when i t does not properly recognise the current dialogue. C o m p a r e d to such Systems the proposed solution is relatively simple a n d hopefully easily m a i n t a i n a b l e a n d adaptable. It is possible to describe schemes i n w h i c h the acts have a strict or a loose order, t o specify acts that m a y b e o m i t t e d o r repeated, a n d acts m a y also occur i n m o r e t h a n one dialogue. T h e possibilities offered for specifying the behaviour o f the track-i n g system s h o u l d be flextrack-ible enough t o allow de-scriptions o f c o m m o n l y o c c u r r i n g dialogue types, such as question a n d answer a n d point a n d click dialogues. T h e dialogue schemes are entered i n a d a t a file w i t h a l e x i c a l l y free f o r m a t , to allow users o f the M M I t o m a k e adjustments t o the schemes i f performance of the t r a c k i n g system is not satis-factory or when o p e r a t i o n a l procédures change. T h e inventory of dialogues produced by the track-i n g system provtrack-ides support for the operator when he h a s t o s w i t c h between m u l t i p l e concurrent tasks. F u r t h e r operator support can be developed w i t h the o u t p u t of the t r a c k i n g system as a basis. T h e a p p l i c a t i o n of a dialogue t r a c k i n g system w i l l keep the operator " i n the l o o p " . T h e system w i l l not automise tasks or dialogues, and the operator w i l l r e m a i n aware of a i l his actions.
C u r r e n t l y , the software o f the dialogue t r a c k i n g system is being tested w i t h a simulated interac-t i o n log. Ininterac-tégrainterac-tion of interac-the sofinterac-tware ininterac-to a démon-strative e x a m p l e is p l a n n e d , this c a n then be used to evaluate the t r a c k i n g system i n a more realistic environment, and interaction w i t h users.
8 A C K N O W L E D G E M E N T S
T h i s work has been carried o u t as part o f the E u r o p e a n C o m m u n i t y research t r a i n i n g project
" V i s u a l i s a t i o n o f Process M e a n s - E n d s Hiérarchies for A d v a n c e d M a n M a c h i n e Interfaces" ( E R B -B R E 2 C T 9 4 3 0 8 5 ) . T h e work has been done i n co-opération w i t h the partners i n the B r i t e / E u R a m II P r o g r a m m e i n Project 6126 ( A M I C A : A d v a n c e d
M a n M a c h i n e Interfaces for Process C o n t r o l A p -plications). T h e c o n s o r t i u m partners are: C I S E (Italy), E N E L (Italy), F L S A u t o m a t i o n A / S (Den-m a r k ) , U n i v e r s i t y o f K a s s e l ( G e r (Den-m a n y ) , M a r c o n i S i m u l a t i o n and T r a i n i n g (Scotland). 9 R E F E R E N C E S A v e r b u k h , E . A . , G . Johannsen, F . M l e t z k o , R . van Paassen, J . R u d e w i g , T . G a v r i l o v a a n d A . V o i n o v (1995). T o o l k i t detailed architec-t u r a l design issue 2 (sofarchitec-tware modules g r o u p 1). A M I C A B r i t e / E u R a m Project 6126 D 5 1 . U n i -versity of K a s s e l , K O N S Y T E C G m b H . Fabiano, A . S . , C . L a n z a a n d J . K w a a n (1994). T o o l k i t architectural design. A M I C A B r i t e / E u R a m Project 6126 D 4 - 2 . C I S E a n d M A R C O N I S i m u l a t i o n . Fabiano, A . S . , C . L a n z a , B . M u l v i h i l l , G . Zardetto, M . A b r a m a n d F . C a l a b r e t t a (1993). A r c h i t e c t u r a l design f o r a p p l i c a t i o n a prototype. A M I C A B r i t e / E u R a m P r o j e c t 6126 D 8 - A . C I S E , E N E L . Heuer, J . , S. Borndorff-Eccarius a n d A v e r b u k h E . A . (1993). Task analysis for a p p l i c a t i o n b . A M I C A B r i t e / E u R a m Project 6126 I R 1 - 0 4 . University o f K a s s e l , L a b o r a t o r y for H u m a n -M a c h i n e Systems.
Johannsen, G . (1995). K n o w l e d g e based design o f human-machine interfaces. Control Eng.
Prac-tice 3(2), 267-273.
L e F o r t , N . , O . A b o u k h a l e d a n d D . R a m a m o n -j i s o a (1993). A co-pilot architecture based o n a multiexpert system a n d a realtime e n v i r o n -ment. I n : IEEE International Conference on
Systems, Man and Cybernetics. L e T o u q u e t , France.
L i n d , M . (1990). Representing goals a n d functions of complex Systems a n i n t r o d u c t i o n t o M u l t i level F l o w M o d e l l i n g . T e c h n i c a l R e p o r t 9 0 D -381. Institute o f A u t o m a t i c C o n t r o l Systems, Technical U n i v e r s i t y of D e n m a r k .
W i t t i g , T . a n d R . O n k e n (1992). K n o w l e d g e based cockpit assistant for controlled airspace opéra-t i o n . I n : 5opéra-th IFAC Symposium on Analysis,
De-sign and Evaluation of Man-Machine Systems.
