00100		THE CENTRAL PROCESSES OF THE MODEL
00200	
00300	
00400	(THIS CHAPTER REQUIRES MANY FLOW DIAGRAMS- SEE BACK OF MS)
00500	
00600		Only   the  major  processes  will  be  described  in  detail
00700	sufficient  to  illustrate  the  logic  of  the   algorithm.     Many
00800	"housekeeping"  procedures  are  needed  to  run  the  model  but  no
00900	understanding of them is necessary to follow the main flow of  symbol
01000	processing.  In  the  next  paragraph  I  will  give some examples of
01100	"housekeeping" only to illustrate what little interest they have  for
01200	the nonspecialist reader.
01300		The  first  theoretically uninteresting procedure executed is
01400	one of initialization which checks to make  sure  the  data-base  has
01500	been read in and sets a number of variables to their starting values.
01600	Some of these variables serve as flags or  indices  pointing  to  the
01700	topic  under  discussion  or to the last self-topic discussed.  Other
01800	variables are set by the interviewer who can choose to run a weak  or
01900	strong version of the model.  If the weak version is elected, affect-
02000	variables  of  ANGER  and  FEAR  can  be  set  to  "low"  or   "mild"
02100	values, while  MISTRUST   can  be  set  to  "mild"  or  "high".   The
02200	interviewer also has the option of following the internal workings of
02300	the  model  which  can  be displayed in "windows" on a console. After
02400	this initialization the algorithm prints out "Ready" to  indicate  to
02500	the interviewer he may now enter his input.
02600	
02700		After the  input  expression  is  assigned  a  sentence  type
02800	(statement,  question  or  imperative),  it  then serves as the input
02900	argument to the major procedures which deal with (in  order)  special
03000	reactions,  delusional references, self references, flare references,
03100	interviewer-interviewee  relations,  miscellaneous  expressions   and
03200	self-scanning.
03300	
03400	SPECIAL REACTIONS
03500	
03600		This  procedure  produces  appropriate  reactions  to special
03700	types of input expressions.    If the input consists  simply  of  the
03800	letter  "S" {the means by which an interviewer indicates silence over
03900	a teletype), then the algorithm chooses a  linguistic  response  from
04000	the "Silence"list. The linguistic output responses are not generated
04100	word-by-word.  They  consist  of  pre-formatted  English  expressions
04200	stored in the data-base on ordered lists.
04300	
04400		The procedure which selects the next reply from the  relevant
04500	response list also removes that response from the list so it will not
04600	be output twice.   Thus, in this  case,  where  repeated  silence  is
04700	being detected and if there are no more responses  on  the  "Exhaust"
04800	list  {i.e. the "Exhaust" list is itself exhausted} , PARRY would end
04900	the dialogue.
05000	
05100		An "Exhaust" list represents  a  boundary  condition  in  the
05200	model.  That  is,  since the model has a limited number of linguistic
05300	responses for each topic it can discuss,  when  these  responses  are
05400	exhausted   PARRY   must have some way of dealing with a large number
05500	of  conceptually  equivalent  repetitions  on   the   part   of   the
05600	interviewer.   When a response list is exhausted, the model expresses
05700	a wish to change the topic and, as mentioned, when the "Exhaust" list
05800	itself  is exhausted,    PARRY  ends the dialogue. Since this process
05900	is true of all instances in which the response list is exhausted,  it
06000	will  not  be  mentioned again. I trust the reader will remember that
06100	this is what happens each time this boundary condition is reached.
06200	  
06300		The second case handled by this procedure consists  of  input
06400	expressions  in  which  the interviewer states or insinuates that the
06500	model is mentally ill.  This condition is detected by  finding  "you"
06600	and a member of the "Abnormal" list in the input.  The inputs:
06700	
06800	     (1) DR.- YOU NEED TREATMENT.
06900	
07000	     (2) DR.- YOU ARE DELUSIONAL.
07100	
07200	     (3) DR.- DO YOU THINK YOU MIGHT BE PARANOID? 
07300	
07400	would satisfy this condition.
07500		If  the input is a question, as in {3}, ANGER is increased by
07600	a  smaller amount of its current value than if  it  is  a  statement,
07700	Thus  a question is interpreted as an implicit insult compared to the
07800	explicit insult of a direct statement.
07900	
08000	     The linguistic response now chosen is selected from a list of
08100	"alienated" responses,e.g.
08200	
08300	     (4) PT.- I THINK I KNOW WHAT YOU DOCTORS ARE UP TO.  
08400	
08500		If  conditions  for  the procedure handling special reactions
08600	are not found to obtain, the algorithm  next  attempts  to  recognize
08700	references to delusions.
08800	
08900	DELUSIONAL REFERENCES
09000	
09100		The  strong  version of the model contains in its data-base a
09200	delusional network of beliefs about the Mafia.   The  next  procedure
09300	called  scans  the  input  expression looking for a reference to this
09400	delusional network.      As will be  seen,  reactions  to  the  first
09500	reference  differ  from  reactions  to  subsequent  references.   The
09600	conceptual contentives of the delusional net are  classified  in  the
09700	data-base  into  "strong"  and "ambiguous" terms.  Thus "murder" is a
09800	strong term whereas "bug" (as mentioned in chapter 4), is  ambiguous.
09900	If  delusional  terms are detected in the input, a variable is set to
10000	the list of terms found and the  terms  are  then  deleted  from  the
10100	delusional word list for reasons which will become clear later.
10200	
10300		Two situations in the interview must  be  distinguished.  The
10400	first  is  one  in which a delusional topic occurs for the first time
10500	and the second in which some aspect of the delusional  net  is  under
10600	discussion  or  has  been  under discussion and is now being taken up
10700	again.    Since  the  topic  of  the  Mafia  is  fear-eliciting,  any
10800	reference  to  it for the first time raises FEAR by an increment much
10900	greater than if the topic has already been discussed. The concept  of
11000	"Mafia"  is  represented in the data-base by a node in a weighted and
11100	directed conceptual graph.
11200		((DRAWING  OF  GRAPH  HERE)  Horses  →  Horseracing → Bookies
11300	→Gangsters →  Rackets  →  Mafia  →    Gambling → Police  →    Money →
11400	Italians).   The  nodes  in  the graph represent "flare" concepts  to
11500	which the model is particularly  sensitive.   Associated  with  these
11600	nodes  are  small  stories  which the model can narrate about each of
11700	them as a theme. Nodes closer to the Mafia node are  weighted  higher
11800	to  represent  the notion that they are of greater concern since they
11900	bear more directly on  the  delusional  network.  If  a  Mafia  topic
12000	appears  for the first time,  pointers in the directed graph of flare
12100	concepts must be modified accordingly since the Mafia  node  has  the
12200	highest  weight  in  the  graph.    A  topic such as "bookies", while
12300	leading eventually to Mafia beliefs, is of much less importance  than
12400	Mafia-topics.    But  if  "bookies"  comes  up  in the interview, the
12500	algorithm must know  whether  or  not  the  Mafia  has  already  been
12600	discussed.    Also,  if an introductory-topic {see p.0OO} or subtopic
12700	was under discussion when reference to a  Mafia-topic  is  made,  the
12800	algorithm must unset the introductory-topic indicator.
12900	
13000		Since the model strives to tell its story about the Mafia,  a
13100	flag  is  set  to  indicate  that, should the topic be changed by the
13200	interviewer,   PARRY   should return to  the  current  point  in  its
13300	story  under  appropriate  circumstances, e.g.   when the interviewer
13400	asks a non-specific question or requests any information the  patient
13500	wishes to volunteer.
13600	
13700		If the interviewer's input expression contains a reference to
13800	the delusional net, a delusional statement is output.  But which one?
13900	If this is the first time  the  topic  has  come  up,  the  algorithm
14000	outputs  the  first statement of its delusional story.   From then on
14100	the output delusion selected depends on what has been said,  what  is
14200	still  unsaid,  and  what the interviewer has said about the previous
14300	delusional statement.   Thus the most recent delusional statement  is
14400	saved along with expected anaphoric references, anticipating that the
14500	interviewer may subsequently ask a question or make a statement about
14600	it.
14700	
14800		One special case must be noted.  If the values of ANGER, FEAR
14900	and/or MISTRUST are extremely high (above a particular threshold  set
15000	in the algorithm), the program will refuse to discuss Mafia-topics at
15100	all since it is too "upset" to talk about this most sensitive area.
15200	
15300		To make some of these operations more  intelligible,  let  us
15400	consider  interview examples.  Suppose at some point in the interview
15500	the doctor asks a standard first-interview question as follows:
15600	
15700	     (5) DR.- DO YOU EVER HAVE THE FEELING YOU ARE BEING WATCHED? 
15800	
15900	If this is the first reference  to  the  delusional  net,  FEAR  will
16000	increase greatly and the linguistic response will be:
16100	
16200		(6) PT.- YOU KNOW, THEY KNOW ME.   
16300	
16400	In making this response, the model must expect from the interviewer a
16500	number  of  typical  questions  of  the  wh-type as well as rejoinder
16600	statements.  The use of "they" by the interviewer in his response  to
16700	the model's   output  is  assumed to be an anaphoric reference to the
16800	"they"    PARRY  is  talking  about.    Although  it  is  likely  the
16900	interviewer  will  react  to the model's output of {6}, the algorithm
17000	must be prepared for the possibility that he will change  the  topic.
17100	Hence  if  the  interviewer  at  this  point  asks  some non-sequitur
17200	question such as:
17300	
17400	     (7) DR.- HOW LONG HAVE YOU BEEN IN THE HOSPITAL?  
17500	
17600	the  program recognizes that no reference to the delusional topic has
17700	been made and answers the question just as though it had  been  asked
17800	in  any other context.  This ability to deal with input in a flexible
17900	manner is important because of the many contingencies which can occur
18000	in psychiatric dialogues.
18100	
18200		If   the  topic  is  changed  abruptly  in  this  way  by  an
18300	interviewer, the algorithm "remembers" that it has output  its  first
18400	delusional  statement  of  (6).    When the interviewer makes another
18500	neutral delusional reference, the next "line" of the delusional story
18600	will be output, e.g.
18700	
18800	     (8) PT.- THE MAFIA REALLY KNOW ABOUT ME.   
18900	
19000	The ability to answer typical wh- and HOW questions  depends  on  how
19100	much  conceptual  information  is  contained in the delusional belief
19200	being addressed.  For example, suppose   PARRY   replied as in {6}
19300	
19400	     (6) PT.- THEY KNOW ABOUT ME.   
19500	
19600	and the interviewer then asked:
19700	
19800	     (9) DR.- WHERE DO THEY KNOW ABOUT YOU? 
19900	
20000	If the expectancy-anaphoras contain no "where", then a question about
20100	location   cannot  be  answered.   In  this  default  situation,  the
20200	algorithm recognizes the anaphoric "they", "know" and "you". Hence it
20300	knows  at least that the topic has not been changed so it outputs the
20400	next statement in the delusional story;
20500	
20600		(9) PT.- THEY KNOW WHO I AM.       
20700	and again anticipates questions and  rejoinders  pertaining  to  this
20800	statement.
20900	
21000		In constructing the data-base of beliefs, we tried to pack as
21100	much information in each belief as any "reasonable" (like  ourselves)
21200	interviewer might request.  However, one cannot anticipate everything
21300	and  when  some  unanticipated  information  is  requested,   another
21400	relevant reply must be substituted. This heuristic may seem less than
21500	perfect but there is little else to do when the  model  simply  lacks
21600	the pertinent information. By the way, humans do this also.
21700	
21800		When the interviewer shows interest in the delusional  story,
21900	PARRY  continues  to  output  assertions appropriate to the dialogue.
22000	However, when the interviewer expresses doubt or disbelief about  the
22100	delusions,  ANGER  and  FEAR  increase  and  the  interviewer becomes
22200	questioned as in:
22300	
22400	     (10) PT.- YOU DON'T BELIEVE ME, DO YOU?  
22500	
22600	Such an output expression attempts to prompt the dialogue towards the
22700	relation  between  the  interviewer  and  the  model  which  will  be
22800	described later ( see p.000).
22900		If no  delusional  reference  at  all  is  detected  by  this
23000	procedure,  the   algorithm attempts the next function which searches
23100	for certain types of references to the self.
23200	
23300	
23400	SELF REFERENCES
23500	
23600		Since the main concern of a psychiatric interview consists of
23700	the beliefs, feelings, states and actions of the patient,  the  model
23800	must  be able to answer a large number of questions about its "Self".
23900	
24000	
24100		If  the  input  is  recognized  as a question and no topic is
24200	currently under discussion and the question  refers  to  the  "Self",
24300	then  it  is  assumed  temporarily  that it will refer only to a main
24400	self-topic. These  main  or  "introductory"  self-topics  (age,  sex,
24500	marriage, health,  family,  occupation,  hospital stay, etc.) in turn
24600	have  sub-topics  to  varying  depths.   For  example,  suppose   the
24700	interviewer asks:
24800	
24900	     (12) DR.- HOW DO YOU LIKE THE HOSPITAL?  
25000	
25100	Since "hospital" is a main "introductory" topic with several levels
25200	of sub-topics, the algorithm answers the question with
25300	
25400	     (11) PT.- I SHOULDN'T HAVE COME HERE.   
25500	
25600	and  then  anticipates  a  variety  of likely questions such as "What
25700	brought you to the  hospital?",  "How  long  have  you  been  in  the
25800	hospital?",  "How  do  you  get along with the other patients?", etc.
25900	Each of these questions  brings up  further  topics,  some  of  which
26000	represent  a continuation of the main topic "hospital", but others of
26100	which represent a shift to  another  main  introductory  topic,  e.g.
26200	"other  patients".   Since  many  of  the  inputs  of the interviewer
26300	consist of ellipses or fragments, the algorithm assumes them to refer
26400	to the topic or subtopic under discussion.  If some  topic  is  being
26500	discussed,  the algorithm checks first for a new main topic, then for
26600	a follow-up to the last subtopic, then (unless the subtopic is itself
26700	a  main  topic,  as  for example "other patients" in the above) for a
26800	follow-up to the last main topic.  Thus continuity and  coherence  in
26900	the dialogue are maintained.
27000	
27100		If some meaning cannot be extracted from the question but  it
27200	is recognized at least that a question is being asked, a procedure is
27300	called  which  attempts  to  handle  certain   common   miscellaneous
27400	questions  which  are difficult to categorize.      These include the
27500	space-time orientation questions ("What day is this?")  and  everyday
27600	information   ("Who  is  president?)  asked  by  psychiatrists  in  a
27700	mental-status  examination  to  test  a   patient's   awareness   and
27800	orientation.     Some  quantitative "how" questions ("how many", "how
27900	often", "how long") are   recognized here.  Since  any  adjective  or
28000	adverb can follow a "how", one of the limitations of the model is its
28100	inability to handle all of them satisfactorily because  the  relevant
28200	information  is lacking in the data-base.  If absolutely no clues are
28300	recognized in the question, the  algorithm  is  forced  to  output  a
28400	noncomittal reply such as:
28500	
28600	     (12) PT.- WELL, I DON'T KNOW.    
28700	
28800		This function also checks for statements about the self which
28900	are taken to be insulting or complimentary. Naturally the presence of
29000	a negator in the input reverses the meaning.  Thus
29100	
29200	     (13) DR.- YOU DON'T SEEM VERY ALERT.  
29300	
29400	is classified as an insult whereas
29500	
29600	     (14) DR.- YOU ARE RIGHT.   
29700	
29800	is considered complimentary and benevolent.
29900	
30000		Among the introductory self-topics are those which constitute
30100	sensitive  areas,  e.g. sex, religion and family.  If the interviewer
30200	refers to one of these areas, the value of  ANGER  increases  sharply
30300	and  a  response  is  selected  from  one of the lists categorized as
30400	"hostile", "defensive", "personal" or  "guarded",  depending  on  the
30500	level  of  MISTRUST  at  the moment.  For example, if the interviewer
30600	asks a question about   PARRY'S   sex life, it first replies with:
30700	
30800	   (13) PT.- MY SEX LIFE IS MY OWN BUSINESS.   
30900	
31000	If the interviewer persists or even later tries to ask about sex, the
31100	model will respond with a hostile reply, such as:
31200	
31300	     (14) PT.- DO YOU KNOW WHAT YOU ARE DOING?   
31400	
31500		The  particular  sensitive areas in the model are part of the
31600	initial  conditions  specific  for  this  hypothetical  patient.   Of
31700	course,  these  topics  are  commonly  found to be sensitive areas in
31800	human patients.
31900	
32000		The model operates  sequentially  trying  one  major  process
32100	after  another.  If it has come this far, (that is, having tested for
32200	special reactions, delusional references and self references  without
32300	recognizing  anything in the input pertinent to these procedures), it
32400	proceeds to the next process which handles flare references.
32500	
32600	FLARE REFERENCES
32700	
32800		The data-base contains a directed graph of concepts  involved
32900	in  the  model's "stories".     PARRY has small stories to tell about
33000	horseracing, gambling, bookies, etc.  The  major  concepts  of  these
33100	stories are termed "flare" concepts since they activate stories which
33200	are differentially weighted in the graph.
33300	
33400	
33500		In the strong version of the model, the  concept  "Mafia"  is
33600	given  the  highest  weight, while  in  the  weak version the concept
33700	"Rackets" is most heavily weighted.   In both versions  "Horses"  has
33800	the  lowest  weight. The weights are assigned to the concepts and not
33900	individual words or word-groups denoting the concepts.
34000	
34100		The  graph  is  directed  in  the  sense  that  reference  to
34200	horseracing elicits the first line of a story about horseracing. When
34300	a story is ended, a prompt is given to the interviewer to discuss the
34400	next story in the graph which involves "bookies".   The model strives
34500	to tell its  stories  under  appropriate  conditions  and  leads  the
34600	interviewer  along  paths of increasing delusional relevance.    Much
34700	depends on whether the interviewer follows these leads "benevolently"
34800	and reacts to the prompts.
34900	
35000		The  first  step in this procedure is to scan the input for a
35100	flare concept having the highest weight.   Thus if a flare concept is
35200	already under discussion, a weaker new flare will be disregarded.  If
35300	the flare concept is one in a story which has already been  partially
35400	told,  then  a prompt is offered regarding the next story-node in the
35500	graph.
35600	
35700		If a question is asked about the events of a story, the model
35800	tries to answer it.  Also the  model  is  sensitive  to  whether  the
35900	interviewer  is  showing interest in the story or whether he tries to
36000	change  the  subject  or  expresses  a  negative  attitude,  such  as
36100	disbelief.
36200	
36300		If the interviewer indicates a positive attitude towards  the
36400	story,  then  benevolence  is  recognized and the variables of ANGER,
36500	FEAR and MISTRUST  decrease  slightly  after  each  I-O  pair.  ANGER
36600	decreases  more rapidly than FEAR while MISTRUST, being a more stable
36700	variable once it has risen, decreases least.
36800	
36900		If no flare concepts are recognized in the input,  the  model
37000	next  tries  to  detect  if a reference is being made to the relation
37100	between the interviewer and the model.  In an  interview  interaction
37200	there  exist  two  situations, the one being talked about and the one
37300	the participants are in at the moment. Sometimes the latter situation
37400	becomes the former, that is, the one talked about.
37500	
37600	INTERVIEWER-INTERVIEWEE RELATIONS
37700		As described in Chapter 4, the algorithm  must  be  ready  to
37800	handle input referring to the relation between interviewer and model.
37900	The simplest cases are exemplified by expressions such as:
38000		(15) DR.- I UNDERSTAND YOU.  
38100		(16) DR.- YOU DO NOT TRUST ME.  
38200	Those phrases in an expression which can appear between "I" and "you"
38300	or between "you" and "me" we classified as representing a positive or
38400	negative  attitude  on  the  part of the interviewer. Thus expression
38500	(15) is taken  to  be  positive  whereas  (16)  is  negative because,
38600	although it contains a positive verb, the verb is negated.
38700		If a positive attitude is expressd by the  interviewer,  FEAR
38800	and  ANGER  decrease.  FEAR  and  ANGER  increase  depending  on  the
38900	conceptualizations of the input. These attitudes of  the  interviewer,
39000	as interpreted by the model, are reflected in the values of the affect
39100	variables.
39200		Associated in the  data  base  with  each  type  of  attitude
39300	expression expected are lists of appropriate output expressions. Thus
39400	in reply to:
39500		(16) DR.- I UNDERSTAND YOU. 
39600	the model would reply:
39700		(17) PT.- I'M GLAD YOU DO. 
39800	or
39900		(18) PT.- I APPRECIATE YOUR TRYING TO UNDERSTAND.  
40000	or  some  equivalent  expression  depending  on  values of the affect
40100	variables.  When  ANGER  and  FEAR  are   high,   positive   attitude
40200	expressions  are  interpreted  as insincerity and hence evoke hostile
40300	replies.
40400		The   remainder  of  input  expression  types  thus  far  not
40500	discussed are handled by a procedure for miscellaneous expressions.
40600	
40700	MISCELLANEOUS EXPRESSIONS
40800	
40900		This procedure deals with all those  interviewer  expressions
41000	from  which no clear conceptualization can be formed.  The only thing
41100	which can be determined is perhaps the sentence-type  of  the  input.
41200	Presented  with  one  of these expressions, if FEAR is extremely high
41300	PARRY signs off without a farewell expression and cannot be contacted
41400	through  further  natural  language  input.   If FEAR is high but not
41500	extreme, and the input is recognized as a question, the model chooses
41600	a  reply  from a list which brings up the attitude of the interviewer
41700	as in:
41800		(19) PT.- WHY DO YOU WANT TO KNOW?  
41900	or
42000		(20) PT.- YOU PRY TOO MUCH.  
42100	If  the  input is recognized as a statement, a reply is chosen from a
42200	list which indicates some degree of anxiety:
42300		(21) PT.- WHO ARE YOU REALLY?  
42400		(22) PT.- YOU ARE MAKING ME NERVOUS.  
42500	If ANGER is high and the input is a question, a reply is chosen  from
42600	a list designed to express hostility as in:
42700		(23) PT.- DO YOU KNOW WHAT YOU ARE DOING?  
42800		(24) PT.- PERHAPS YOU ARE JUST POSING AS A DOCTOR.  
42900		Sometimes  in  these  default  conditions the flag set in the
43000	procedure for delusional references allows the model to  continue  by
43100	giving  the next line in its delusional story.  If the story is under
43200	discussion, continuity is maintained.  But if it is  not,  the  model
43300	appears  to  ignore  the  input and jumps back to one of its previous
43400	preoccupations.   In this instance the observed property of  rigidity
43500	is a function of linguistic non-comprehension and not of the paranoid
43600	processes per se.   Increasing  the  model's  ability  to  comprehend
43700	conversational language would remedy this deficiency.
43800		If  a story flag has not been set by a previous discussion in
43900	the interview and ANGER and FEAR are not high, the algorithm tries to
44000	see  if the input is some type of general prompt from the interviewer
44100	such as:
44200		(25) DR.- GO ON.  
44300	or
44400		(26) DR.- TELL ME MORE.  
44500	If so, PARRY     continues  with  its  current story  or attempts  to
44600	initiate another story. 
44700		If  none  of  these  conditions hold, the procedure ANSWER is
44800	called.  This  procedure  handles  a  group  of  common  special-case
44900	miscellaneous questions such as:
45000		(27) DR.- HOW DO YOU DO?  
45100	and miscellaneous statements such as:
45200		(28) DR.- HI.  
45300		(29) DR.- GOOD EVENING.  
45400	
45500	SELF SCANNING
45600		The final major procedure in the  algorithm  scans  what  the
45700	model  has  chosen  to  output.  That is, it treats its own output as
45800	input.  If this expression contains a flare or delusional  reference,
45900	the appropriate flags are set and FEAR is raised slightly, but not as
46000	much as if this expession had come from the interviewer. In this  way
46100	the  model  "frightens  itself"  by  what it says about a frightening
46200	topic.
46300	
46400				SUMMARY
46500		To  recapitulate  the  operations  of  the  model,  it  first
46600	attempts a  linguistic  recognition  of  the  input  by  looking  for
46700	patterns  which  are  meaningful  for  it.  The internal and external
46800	reactions of the model depend on whether the meaning is classified as
46900	malevolent,  benevolent,  or  neutral.  Internal reactions consist of
47000	adjusting the values of affect variables of anger, fear and mistrust.
47100	The model also keeps track of the topic under discussion and by means
47200	of anaphora-expectancy functions, anticpates what might be said.  The
47300	external  reactions  of  natural  language expressions depends on the
47400	nature of the input, the topic under discussion and the values of the
47500	affect variables.
47600		The systemicity of the model is obvious. We now come  to  its
47700	testability.  How  can  we  compare  the  model  to  its subject, its
47800	naturally-occurring counterpart, so that we can judge its  degree  of
47900	correspondence to facts of observation?