* bugs *

x include "bind" in resource interfaces: depends x include clarifyDependentIssue, or generalize clarifyAction in selection_rules

quit does not work after clarification q x?? (control) godis bug: when plan has been reloaded without producing reraising fb, agenda is empty; this leads to leaving the turn to the user even if there is a "findout" on top of the plan that would have been selected
other bug: y/n answers not fed back so strange they are also denied $S> Hello How are you feeling today? $U> [score 0.0-1.0]: $S> Are you feeling nauseous? $U> yes headache [score 0.0-1.0]:0.8 $S> Okay. head aches. $U> no [score 0.0-1.0]: $S> not head aches. So, Are you feeling nauseous?
bug III: issue reraised since "okay" is relevant -> no backtracking of "issues"
- raised issues should not be reraised! ! same thing happens if the answer needs grounding & is denied S> Hello How are you feeling today? $U> okay $S> So, How are you feeling today? $U>
reraise still produced when taking tests, since no question was raised while testing... x hack: empty output form
- real solution: keep track of previous goal issue
x? cannot quit or request top after initial altq?
- why the final thingies on irrelevantFollowup???

answer(disease...) not integrated here:

[greet]. [ask(X^info(X)), answer(disease(malaria))].

problem with forget(bestq) etc: if plan for Q is interrupted and answer to Q is given in subdialogue, bestq stays on PRIv/BEL
- could be solved by deleting inconsistent info in dev_query
there are two irrelevantFollowup rules X several symptoms at once -> assumes self-correction [answer(symptom(fever)), answer(symptom(headache))]
* OTHER *

x make slides
- choice between IOD, AOD; motivations
  - we learned that IOD may require devices
- add plans
- add sample dialogues
x things we would like to add to plan execution rules:
- precedence constraints
  - see below
- subplans
  - new constructs
    - resolve(Q)
    - do(A)
x things we have added:
- assume, assume_issue
* todo *
get GUI to work
not confirmed vs fail
- fail is not resolving
- need to decide when to give up a plan
- what does fail mean?
  - fail for current info vs. # terminal fail: notexist(X,PX)??? does not work for y/n
  - assume terminal fail this if fail + best_... empty
x "not bestq(_)" is not deleted; perhaps just fail(...)???
- should be notexist(X,bestq(X))

same qs in interview and diagnosis?

! no, since interview presupposes diagnosis

x diagnosis input form for disease q

x only fever test!

w domain needs to handle rejected(X^symptom(X)) without breaking

just dont answer with disease here!

! we could easily have gotten info directly from device, but didn't want to clutter the IS

--- nice extras ---

fix all error reports
rename medical -> diseaseDB
swedish lexicon, to show mutilinguiality by resource switching on-line
- also reinclude domain stuff needed for lang. change ! shows advantage of discoupling lexicon from plan
device_nurse
report should be q-raising? which q? whether to proceed? ! ?action(A)
- action assumed by plan?
- what if "no"?
generalize IOD to work with with device & change to that?
replace device with database? ? user request test actions
domain stuff
- asking for which medications have been taken
- asking for countries traveled to
  - inference to "travel to tropical climate" or "... desert region"
--- half-done ---

x? how treat negative confirmation?
- = disconfirmed???
x? dependent issue accommodation & clarification
- example?
  
  "I have malaria"
  "Do you want general information or recommendations for treatment?"
x? accommodation
- check that it really works with bind; that "bind" is in all accommodation rules where "findout" is
x? additional issues
- provide recommendations about treatment
  - now called "info"
  - add to lexicon, remove from device
  x provide general info x modify device
- get to work with infosharing
x? multiple simultaneous issues? ! interference if try to ask about info for some other disease
- not really a good example of information sharing; not same question ("what disease do I have" vs "what disease do I want info about") ? ask if have to pay extra for test
- also add request(not_take_test)?
- and add request(take_test)
plan(pay_extra, [ findout(X^test(X))

x? issue clarification
- example: info vs. treatment given disease
  - input form disease -> info_disease
? positive_diagnosis as action?
- findout(X^disease(X)) should trigger plan
- rule: if findout(Q) on plan and
trindikit: surpress initalization reports ? user request test actions
? prevention q

? territories q

w GODIS: w findPlan rule: loadplan not triggered if no "findout" or "raise" in plan.
- logical operators for if_then -> can shorten rules for info, treatment x fix reraising fb so generated only when issue is changed; peek in tmp? (but will be vulnurable to grounding subdialogs, but perhaps OK to say "returning to" after subdialog?) x if not q top on tmp/sys/issues, generate reraise
x? now, confirmed_... is reraised
- since on issues from start
- should be raised at end of X^disease(X)! x or empty output form (ugly)
- masterplan?
x? info does not work w rename "treatment"
- get info-share to work ? push info at start
x treatment q
- add to lexicon, remove from device
x? check feedback and sequencing moves
- incl denying
- reraise:fb only in tree variant
x revising by reaccommodation (works?)

g�r ett riktigt beslutstr�d & komplettera dom�ninfo

? why isn't "so" generated in multiissue example?

w bug: if a treatment was discussed during the examination dialogue, it is still the one assumed at the end

x fix godis bug

? sem_sort test? needed?

x check lexicon so all works ? malaise

--- done ---

x alternative version, not push Q's at start?

"I want a diagnosis"
"I want to be tested for malaria", binds X^disease(X)
- change lexicon input form for diseases
! V2 changes in
- domain, l.57
- lexicon, l.356
PROBLEM: does not ask about anything but diseasehistory
- change in domain
confirm_by_tests should also bind X^disease(X)

x should confirmed_by_interview really bind questions?

yes, accommodation should work there too (but can avoid in demo)
no, because X^disease(X) must be answered first
- need to encode ordering constraints
  - precedes( X^disease(X), confirmed... )
    - precedes( Q1, Q ) prevents dependent accommodation of Q unless Q1 already resolved
    - add condition to acc. rule: not( $domain :: precedes( Q1, Q ) and plan( Q1, _ ) and not( resolved( Q1 ) )

x negative test results do not work

subplans in godis? exec?

when loading a Qplan, raise Q

or inference; but what if non-com info needed? assume?

x bestq returns too many qs

x how get out of confirmed_by_interview by failing? it does not resolve!

x disconfirmed not same as not confirmed!!!

ok, so the system cannot disconfirm

x? m�jliga resultat fr�n dev_query:

X^disease(X) vet ej - fail(X^disease(X)) pos - disease(a) neg - notexist(X,disease(X))

confirmed_by_interview vet ej - fail(confirmed_by_interview) pos - confirmed_by_interview neg - not(confirmed_by_interview)

samma f�r confirmed_by_tests

x? how make sure that subsequent tests are not performed?

masterplan?

x dev_query info(Disease)

D device_nurse

get to work with actions ! confirm(test) should trigger??? but never requested, only dev_do
- take_test(T) dev-action -> assert test_result(T)
assume_issue(X^test_result(X))
- add to lexicon
x koppla in nya device & testa

x how capture this:
- when asking for treatment/info, this can be about any disease
- but when a disease has been diagnosed, it is probably about that one # inference disease(X) -> info_disease(X) # hyp_disease isa disease, so answer to disease is answer to hyp_disease x info_disease isa disease, so disease is answer to info_disease
! why necessay to push all issues on agenda at start???

x lexicon! x top x "maybe" and "i think so" -> rej x outputfor for positive_diagnosis x output_form for confirmed_by_tests x output_form for confirmed_by_tests

x? what if test show no disease? answer(not(positive_diagnosis)) ->

x best_labtest

x? what if user cannot answer questions x rejected(Q) added to sh.com *DH device rensar bort alla rejected Q

x restart using "top"

? make sure info(_) and characteristics(_) get deleted so not same answer given second time (perhaps reinstate urule solution?)

? "maybe" to y/n
- regard as reject?
? precedense constraints on accommmodation?

x how make tests?
- tell user take test, ask for result
- provide random result
  call device, where user enters result
x plan for info; share disease/1 with other plan

x anv�nd forget(bestq(_)) f�r att rense priv/bel; ta bort godis�ndring

x koppla in OAA-databas ? ta hand om fail(Q,Reason)

x mutiple-answer questions

x correction using negation

x fever -> symptom(fever), etc.
- "how are you feelding today" = X^symptom(X); "fine"= symptom(none)
tv� l�sningar x1 anv�nd backend f�r att hitta n�sta fr�ga att st�lla ! utg�r fr�n att backend �r mer eller mindre intelligent ! man kan t�nka sig att l�gga till beslutstr�d till backend
- remove old bestqs in priv/bel
- how change plan when no bestq, to do requests
- don't generate reraise fb if q topmost on ISSUES
x2 implementera beslutstr�d i dialogplan ! var hittar vi "r�tt" tr�d? hitta p� sj�lva?
accommodationl�snginar x 1a accommodation, backend/tr�dplan: i slutet av planen finns bind(Q1), bind(Q2) osv d�r Q1, Q2, ... �r alla fr�gor som n�gonsin kan vara relevant f�r X^disease(X)
# kan bli problem med reraising & retraction: ev m�ste vi �ndra ordningen p� accommodationreglerna s� att GoDiS f�rs�ker ackommodera fr�n share/com f�re den provar med bind i planen ! g�r inget; retract �r oberoende om varifr�n acc g�rs!
1b accommodation, backend: fr�ga device om vilka fr�gos som ska bindas ! bind set of questions, retrieved from expert system ! binding alternatives x all q's
- all except already known
- all except bestq
2 (tr�dplan): accommodation kikar in i if_then
- sv�rt; ska planen g�ras om till tr�d?
? ej spara tillst�nd

x AOD or IOD?
- it's AOD since tests are performed
- but IOD since database search
- consider as interface to device containing expert system + test-taker?
MORE COMPLEX
how deal with actions?
- requests by system! exec_instruct
- add tests
  1. request action
  2. device-do action
new move: warn(A): "I'm going to do A" * GoDiS changes: X = copied to src & CVSd *
x modified definition of "depends" in resource_interfaces

x clarifyDependetIssue in AOD selection, or generalize clarifyAction

x exec_assume_issue

x exec_forget- commented out clear actions

x irrelevantFollowup 031023
- but try again with old rule, may not have done []
X integrateUsrAccNegICM: % SL031022 add( /shared/com, rejected(Q) ) ]).

X added exec_assume

X change integrateUsrAnswer so it is possible to give multiple answers to same Q in one turn
- abandoned solution, removes last not first, should be own rule forall_do( in($/private/nim, DP_M2) and DP_M2/snd = answer(A2) and $domain::relevant(A2,Q) and $domain::combine(Q,A2,P2) and $domain::incompatible(P,P2), del( /private/nim, DP_M2 ) )
X definition of incompatible should be partly domain-specific

exec_consultDB removes incompatible propositions

"maybe" as answer to y/n questions X resource_interfaces fixed
- integration rule
questions that are rejected or not resolved when raised first time should be stored for later reraising, but not reraised immedialtely
only if they are contradictory, assume selfcorrection & delete first! ? need to add self-correction move? ocm:corr
* trindikit changes *

x interpret_empty

? vilka predikat ska egentligen exporteras fr�n modulerna? Inga alls?
reports from modules and about initialization should be surpressed in "quiet" mode