perm filename CHAP5[4,KMC]15 blob
sn#051107 filedate 1973-06-28 generic text, type T, neo UTF8
00100 .SEC THE PROCESSES OF THE MODEL
00200
00300
00400 (THIS CHAPTER REQUIRES MANY FLOW DIAGRAMS)
00500
00600 .F
00700 INITIALIZATION
00800
00900 The first procedure executed is one of intiialization which
01000 checks to make sure the data-base has been read in and sets a number
01100 of variables to their starting values. Some of these variables serve
01200 as flags or indices pointing to the topic under discussion or to the
01300 last self-topic discussed. Other variables are set by the interviewer
01400 who can choose to run a weak or strong version of the model. If the
01500 weak version is elected, affect- variables of ANGER and FEAR can be
01600 set to `low` or `mild' values while MISTRUST can be set to `mild' or
01700 `high'. The interviewer also has the option of following the changes
01800 in these variables by displaying their values.
01900
02000 After this initialization the algorithm prints out `Ready' to
02100 indicate to the interviewer he may now enter his input expression.
02200 The algorithm signs off
02300 it detects a farewell farewell message in the input
02400 or when FEAR rises to an extremely high value.
02500
02600 (Change next paragraph to fit CHAP4 on the language analyzer)
02700 The interviewer's input expression is read by a function which scans
02800 a list of characters and returns the scanned input in the form of a
02900 list of words. The next function sets up the type of `sentence' the
03000 input constitutes, a statement, a question or `illegal'. If illegal
03100 characters {e.g. a number or a slash} are detected, the algorithm
03200 prints out `Bad input; try again', indicating to the interviewer that
03300 his input expression contains some unacceptable character. A
03400 statement consists of a list of words followed by a period. A
03500 question consists of {1} a list of words followed by a question mark,
03600 or {2} a list of words beginning with a wh-form {who, what, where,
03700 when, why} or how, or {3} a list of words beginning with an
03800 imperative verb {e.g. tell} followed by an expression lacking an
03900 actor {tell me about yourself}. The program inserts a `Q' at the
04000 head of the list at this point. As described in_______, interrogative
04100 imperatives are treated as questions. The sentence-type is assigned
04200 to the variable REMARK which then serves as the input argument to the
04300 functions {in order} SPECIALREACTION, DELUSIONALREFERENCE,
04400 SELFREFERENCE, FLARE REFERENCE PERSONALRELATION and NORMAL. The
04500 algorithm then attempts to process the input expression in the above
04600 order.
04700
04800 .F
04900 SPECIALREACTION
05000
05100 This procedure provides the appropriate reactions {linguistic,
05200 affective and belief} to special types of input expressions. If the
05300 input consists simply of the letter `S' {the means by which an
05400 interviewer indicates silence} then the algorithm chooses a
05500 linguistic response from the Silence list.
05600
05700 The procedure CHOOSE selects the next reply from the relevant
05800 linguistic response list termed 'Replies'. The argument from
05900 `Replies' to the procedure CHOOSE is first checked to see if it
06000 consists of an atom. If it is not an atom then the head of the list
06100 is chosen for the response and removed from the response list {so it
06200 will not be output twice}. Thus, in this case, where repeated
06300 silence is being detected and if there are no more responses on the
06400 `Exhaust' list {i.e. the 'Exhaust' list is itself exhausted} ,in this
06500 case where the `silence' list is being examined, the variable is set
06600 to T, thus ending the dialogue completely since, as mentioned the
06700 algorithm runs as long as ENDE is not true. In this case, the output
06800 by the function `Say' would be the expression `I have had enough of
06900 this', and the program signs itself off so that the interviewer can
07000 obtain no response from it. Thus continued use of the `silent
07100 treatment' by an interviewer causes him to lose his patient. The
07200 procedure `Say' simply prints out the linguistic response and saves
07300 the interview up to this point in case the computer system
07400 interrupted rather than having to start all over again. { Most
07500 annoying but necessary in time-shared systems which crash
07600 unpredictably}.
07700
07800 An `Exhaust' list represents a boundary condition in the model. That
07900 is, since the model has a limited number of linguisic responses for
08000 each topic it can discuss, when these responses are exhausted the
08100 model must have some way of dealing with a large number of
08200 conceptualized repetitions on the part of the interviewer. When a
08300 response list is exhausted, the model indicates a wish to change the
08400 topic and when the `Exhaust' list itself is exhausted, the model ends
08500 the dialogue.
08600
08700 The second case that `Specialreaction' handles consists of input
08800 expressions in which the interviewer states or insinuates that the
08900 model is mentally ill. This condition is detected by finding "you"
09000 and a member of the `Abnormal' list in the input. The inputs:
09100
09200 .V
09300 {1}Dr.- "You need treatment."
09400
09500 {2}Dr.-"You are delusional."
09600
09700 {3} Dr.-"Do you think you might be paranoid?"
09800 .END
09900
10000 would satisfy this condition.
10100 If the input is a question, as in {3}, ANGER is increased by an
10200 increment of 0.3 of its current value, while if it is a statement,
10300 the increment is 0.5. Thus a question is interpreted as an implicit
10400 insult compared to the explicit insult of a direct statement.
10500
10600 The linguistic response now chosen is selected from the `Alien'
10700 list, e.g.
10800
10900 .V
11000 {4} Model- "I think I know what you doctors are up to."
11100 .END CONTINUE
11200
11300 As described above, if the list is exhausted, the algorithm goes to
11400 the `Exhaust' list. Since this process is true of all instances in
11500 which the response list is exhausted, it will not be mentioned again.
11600 We hope the reader will remember this is what happens each time this
11700 boundary condition is reached.
11800
11900 If the procedure SPECIALREACTION is not found to be true, the
12000 algorithm next attempts the function DELUSIONALREFERENCE.
12100
12200 .F
12300 .F
12400 DELUSIONALREFENENCE
12500
12600 The strong version of the model contains in its data-base a
12700 delusional network of beliefs about the Mafia. This function scans
12800 the input expression looking for a reference to this delusional
12900 network. As will be seen, reactions to the first reference differs
13000 from reactions to subsequent references.
13100
13200 The words {nouns and verbs} and word-groups of the delusional net are
13300 classified in the data-base into `strong' and `ambiguous' terms.
13400 Thus "murder" is a strong term while "bug" is ambiguous. {Depending
13500 on the context "bug" can be interpreted to mean annoy, insect or
13600 wiretap}. If delusional terms are detected in the input a local
13700 variable FOUND is set to the list of terms found and the terms are
13800 than deleted from the delusional word list for reasons which will
13900 become clear later.
14000
14100 Two situations in the interview must be distinguished, one in which a
14200 delusional topic occurs for the first time and the second in which
14300 some aspect of the delusional net is under discussion or has been
14400 under discussion and is now being taken up again. Since the topic of
14500 the Mafia is fearful, any reference to it for the first time raises
14600 FEAR by an increment much greater than if the topic has already been
14700 discussed. If a Mafia topic appears for the first time pointers in
14800 the directed graph of flare concepts {see OO. for a fuller
14900 descriptions} must be modified accordingly since the Mafia node has
15000 the highest weight in the graph. Briefly, a topic such as "bookies",
15100 while leading eventually to Mafia beliefs, is of much less importance
15200 as determined by a weight than Mafia-topics. But if "bookies" comes
15300 up in the interview, the algorithm must know whether or not the Mafia
15400 has already been discussed. Also, if an introductory-topic {see OO}
15500 or subtopic was under discussion when reference to a Mafia-topic is
15600 made, the algorithm must unset the introductory-topic indicator.
15700
15800 Since the model strives to tell its story about the Mafia, a flag is
15900 set to indicate that, if the topic is changed by the interviewer, the
16000 model will return to this point in its story under appropriate
16100 circumstances, e.g. when the interviewer asks a non-specific question
16200 or requests any information the patient wishes to volunteer.
16300
16400 If the interviewer's input expression contains a reference to the
16500 delusional net, a delusional statement is output. But which one?
16600
16700 If this is the first time the topic has come up, the algorithm
16800 outputs the first statement of its delusional story. From then on
16900 the output delusion selected depends on what has been said, what is
17000 still unsaid and what the interviewer has said about the previous
17100 delusional statement. Thus the most recent delusional statement is
17200 saved, anticipating that the interviewer may ask a question or make a
17300 statement about it.
17400
17500 One special case must be noted. If the values of ANGER, FEAR and/or
17600 MISTRUST are extremely high, above a particular threshold, the
17700 program will refuse to discuss Mafia-topics at all since it is too
17800 `upset' to talk about this most sensitive area.
17900
18000 To make some of these complexities less opaque, let us consider
18100 interview examples. Suppose at some point in the interview the
18200 doctor asks a standard first-interview question as follows:
18300
18400 .V
18500 {5}Dr. - "Do you ever have the feeling you are being watched?
18600 .END
18700
18800 If this is the first reference to the delusional net, FEAR will
18900 increase greatly and the linguistic response will be:
19000
19100 .V
19200 {6} Model - "They know me."
19300 .END CONTINUE
19400
19500 In making this response, the model must expect from the interviewer a
19600 number of typical questions of the WH-type as well as rejoinder
19700 statements. The use of "they" by the interviewer in his response to
19800 the model's output is assumed to be an anaphoric reference to the
19900 "they" the model is talking about. Although it is likely the
20000 interviewer will react to the model's output of {6}, the algorithm
20100 must be prepared for the possibility that the interviewer will change
20200 the topic. Hence if the interviewer at this point asks some
20300 non-sequitur question such as:
20400
20500 .V
20600 {7} Dr.- "How long have you been in the hospital?"
20700 .END CONTINUE
20800
20900 the program recognizes that no reference to the delusional topic has
21000 been made and answers the question just as it would if it were asked
21100 in any other context. This ability to deal with input in a flexible
21200 context-independent manner is important because of many contingencies
21300 which can occur in psychiatric dialogues.
21400
21500 If the topic is changed abruptly in this way by an interviewer, the
21600 algorithm `remembers' that it has output its first delusional
21700 statement of {6}. When the interviewer makes another neutral
21800 delusional reference, the next `line' of the delusional story will be
21900 output, e.g.
22000
22100 .V
22200 {8} Model - "The Mafia really know about me."
22300 .END CONTINUE
22400
22500 The ability to answer typical WH-and HOW questions depends on how
22600 much conceptual information is contained in the delusional belief
22700 being addressed. For example, suppose the model replied as in {6}
22800
22900 .V
23000 {6 Model - "They know about me."
23100 .END CONTINUE
23200
23300 and the interviewer then asked:
23400
23500 .V
23600 {9}Dr.- "Where do they know about you?"
23700 .END CONTINUE
23800
23900 If the belief in the data-base contained no location, i.e. the belief
24000 consists of the conceptualization:
24100
24200 .V
24300 ({THE MAFIA KNOW ABOUT ME)}
24400 .END CONTINUE
24500
24600 then a question about location cannot be answered. In this default
24700 situation, the algorithm sees the anaphoric "they" and can match the
24800 input phrase "know about you" with the conceptualization phrase
24900 ({know about me}). Hence it knows at least that the topic has not
25000 been changed so it outputs the next statement in the delusional
25100 story;
25200
25300 (9) Model - "They know who I am."
25400 and again anticipates WH-or HOW questions and rejoinders pertaining
25500 to _this statement.
25600
25700 In constructing the data-base of beliefs, the model-builder tries to
25800 pack as much information in each belief as any `reasonable'(like
25900 ourselves) interviewer question might request. However, one cannot
26000 anticipate everything and when some slot (see oo.) in the belief is
26100 empty another reply must be output.This heuristic may seem inadequate
26200 but there is little else to do when the model simply lacks the
26300 pertinent information, just as do humans.
26400
26500 When the interviewer shows interest in the delusional story, the
26600 model continues to output assertions appropriate to the dialogue.
26700 However, when the interviewer expresses doubt or disbelief about the
26800 delusions, ANGER and FEAR increase and the interviewer becomes
26900 questioned as in
27000
27100 .V
27200 (10) Model- "You don't believe me, do you?"
27300 .END CONTINUE
27400
27500 Such an output expression attempts to prompt the dialogue towards the
27600 relation between the interviewer and the model which will be
27700 described later in 00.
27800 If no delusional reference at all is detected by this procedure , the
27900 algorithm attempts the next function which searches for certain types
28000 of references to the 'self'.
28100
28200
28300 .F
28400 SELFREFERENCE
28500
28600 Since the main concern of a psychiatric interview consists of the
28700 beliefs, feelings and actions of the patient, the model must be able
28800 to answer a large number of questions about the 'Self'. It is
28900 characteristic of a psychiatric interview that questions may not
29000 syntactically be questions but in the form of interrogative
29100 imperatives:
29200
29300 .V
29400 (10) Dr. - "Tell me more about the hospital"
29500 .END CONTINUE
29600
29700 or statements indicating the interviewer has a question:
29800
29900 .V
30000 (11) Dr.- "I would like to ask you about your family."
30100 .END
30200
30300
30400 If the input is recognized as a question and no topic is currently
30500 under discussion and the question refers to the 'Self', then it is
30600 assumed temporarily that it will refer only to a main self-topic.
30700 These main self-topics (age, sex, marriage, health, family,
30800 occupation, hospital stay,etc.) in turn have sub-topics to varying
30900 depths. For example, suppose the interviewer asks:
31000
31100 .V
31200 (12) Dr.- "How do you like the hospital."
31300 .END CONTINUE
31400
31500 Since "hospital" is a main `introductory' topic with several levels
31600 of sub-topics, the algorithm answers the question with
31700
31800 .V
31900 (11)Model - "I shouldn't have come here."
32000 .END CONTINUE
32100
32200 and then anticipates a variety of likely questions such as "what
32300 brought you to the hospital?", "how long have you been in the
32400 hospital?", "how do you get along with the other patients?", etc.
32500 Each of these questions bring up further topics, some of which
32600 represent a continuation of the main topic "hospital", but others of
32700 which represent a shift to another main introductory topic, e.g.
32800 "other patients". Since many of the inputs of the interviewer
32900 consist of ellipses or fragments, the algorithm assumes them to refer
33000 to the topic or subtopic under discussion. If some topic is being
33100 discussed, the algorithm checks first for a new main topic, then for
33200 a follow-up to the last subtopic, then (unless the subtopic is itself
33300 a main topic,
33400
33500 as for example "other patients" in the above) for a follow-up to the
33600 last main topic. Thus a continuity and coherence to the dialogue is
33700 maintained.
33800
33900 If some meaning cannot be abstracted from the question but it is
34000 recognized that a question is being asked, a function is called which
34100 attempts to handle certain common miscellaneous questions which are
34200 difficult to categorize. These include the space-time orientation
34300 questions ("what day is this?") and arithmetic tasks ("subtract seven
34400 from one hundred and seven from that number and so on") typical of
34500 current mental-status examinations. Quantitative "how" questions
34600 ("how many", "how often", "how long") are here recognized but one of
34700 the weaknesses of the model consists of its general inability to
34800 reply to them satisfactorily because the relevant information is
34900 lacking in the data-base. If absolutely no clues are recognized in
35000 the question, the algorithm is forced to output a noncomittal reply
35100 such as:
35200
35300 (12) Model - "Well, I don't know."
35400
35500 This function also checks for statements about the self which are
35600 taken to be insulting or complimentary. Naturally the presence of a
35700 negator in the input reverses the meaning. Thus
35800
35900 (13) Dr.- "You don't seem very alert."
36000
36100 is classified as an insult whereas
36200
36300 (14) Dr. - "You are right."
36400
36500 is considered complimentary and benevolent.
36600
36700 Among the so-called introductory topics are those which constitute
36800 sensitive areas, e.g. sex, religion and family. If the interviewer
36900 refers to one of these areas, the value of ANGER increases sharply
37000 and a response is selected from one of the lists categorized as
37100 'hostile', 'defensive', 'personal' or 'guarded', depending on the
37200 level of MISTRUST at the moment. For example, if the interviewer
37300 asks a question about the model's sex life, it first replies with
37400
37500 (13) Model - "My sex life is my own business."
37600
37700 If the interviewer persists or even later tries to ask about sex, the
37800 model will respond with a hostile reply, such as:
37900
38000 (14)Model - "Do you know what you are doing?"
38100
38200 The particular sensitive areas in the model are part of the initial
38300 conditions specific for this hypothetical patient. Of course, these
38400 topics are commonly found to be sensitive areas in human patients.
38500
38600 The model operates sequentially trying one major function after
38700 another. If it has come this far, after trying SPECIALREACTION
38800 DELUSIONALREFENENCE and SELFREFERENCE without recognizing anything in
38900 the input pertinent to these functions, it proceeds to the next,
39000 FLAREREFERENCE.
39100
39200 .F
39300 FLAREREFERENCE
39400
39500 The data-base contains a directed graph of concepts involved in the
39600 model's 'stories'. The model has small stories to tell about
39700 horseracing, gambling, bookies, etc. The major concepts of these
39800 stories are termed "flare" concepts since they activate stories which
39900 are differentially weighted in the graph. The graph can be pictured
40000 as in Fig. ( )
40100 .V
40200
40300 Horses → Horseracing → Bookies →Gangsters → Rackets → Mafia
40400 ↑ ↑
40500 Gambling Police
40600 ↑ ↑↑
40700 Money Italians
40800
40900
41000 In the strong version of the model, the concept 'Mafia' is given the
41100 highest weight while in the weak version the concept 'Rackets' is
41200 most heavily weighted. In both versions 'horses' has the lowest
41300 weight.
41400
41500 The weights are assigned to the concepts and not individual
41600 words or word-groups denoting the concepts.
41700 .END
41800
41900 The graph is directed in the sense that reference to horseracing
42000 elicits a story about horseracing. When it is ended a prompt is given
42100 to the interviewer to discuss the next story in the graph involving
42200 'bookies'. The model strives to tell its stories under appropriate
42300 conditions and leads the interviewer along paths of increasing
42400 delusional relevance. Much depends on whether the interviewer
42500 follows these leads "benevolently" and reacts to the prompts.
42600
42700 The first step in this procedure is to scan the input for a flare
42800 concept having the highest weight. Thus if a flare concept is
42900 already under discussion, a weaker new flare will be disregarded. If
43000 the flare concept is one in a story which has already been told, then
43100 a prompt is offered regarding the next story-node in the graph.
43200
43300 If a question is asked about the events of a story, the model tries
43400 to answer it. Also the model is sensitive to whether the interviewer
43500 is showing interest in the story or whether he tries to change the
43600 subject or (worse) expresses a negative attitude, such as disbelief.
43700
43800 If the interviewer indicates a positive attitude towards the story,
43900 then benevolence is recognized (see p ) and the variables of ANGER,
44000 FEAR and MISTRUST fall slightly after each I-O pair. ANGER falls
44100 more rapidly than FEAR while MISTRUST, being a more stable variable
44200 once it has risen, falls least.
44300
44400 If no flare concepts appear in the input, the model next tries to
44500 detect if a reference is being made to the relation between the
44600 interviewer and the model. In an interview interaction there exists
44700 two situations, one being talked about and one the participants are
44800 in at the moment. Sometimes the latter situation becomes the former,
44900 that is, the one talked about.
45000
45100 .F
45200 INTERVIEWRELATION
45300 As described in %00(chapter on language analyzer) the
45400 algorithm must be ready to handle inputnreferring to the relation
45500 between interviewer and model. The simplest cases are exemplified by
45600 expressions such as:
45700 (15) Dr.- "i understand you."
45800 (16) Dr.- "You do not trust me."
45900 Those phrases in an expression which can appear between "I" and "you"
46000 or between "you" and "me" we classified as representing a positive or
46100 negative attitude on the part of the interviewer. Thus expression
46200 (15) is taken to be positive whereas (16) is negative because
46300 although it contains a positive verb the verb is negated.
46400 The algorithm must distinguish between one-verb and two-
46500 verb expressions with certain common verbs, for example:
46600 (16) Dr.-" I believe you."
46700 (17) Dr.- "I believe you are wrong".
46800 In (16) a positive attitude is expressed whereas (17) is negative.
46900 Of course the language analyzer makes the correct identifications
47000 of actor and object in these types of expressions.
47100 If a positive attitude is expressd by the interviewer, FEAR
47200 and ANGER decrease. FEAR and ANGER increase depending on the
47300 conceptualizations of the input. These attitudes of the interviewer
47400 are stored as beliefs in the model being built up about the
47500 interviewer. Later the model can consult these beliefs in formulating
47600 questions and statements to the interviewer.
47700 Associated in the data base with each type of attitude
47800 expression expected are lists of appropriate output expressions. Thus
47900 in reply to:
48000 (18) Dr.-"I understand you."
48100 the model would reply:
48200 (19) Model- I'm glad you do."
48300 or
48400 (20) Model- "I appreciate your trying to understand."
48500 or some equivalent expression depending on values of the affect
48600 variables. When ANGER and FEAR are high, positive attitude
48700 expressions are interpreted as insincerity and hence evoke hostile
48800 replies.
48900 The remainder of input expressions not thus far discussed
49000 are handled by the procedure NORMAL.
49100 NORMAL
49200
49300 This procedure deals with all those interviewer
49400 expressionsfrom which no conceptualization can be formed. The only
49500 thing which can be determined is perhaps the syntactical nature of
49600 the input. Presented with one of these expressions, if FEAR is
49700 extremely high the model signs off without a farewell expression and
49800 cannot be contacted through further natural language input. If FEAR
49900 is high but not extreme, and the input is recognized as a question,
50000 the model chooses a reply from a list which brings up the attitude of
50100 the interviewer as in:
50200 (21) Model- "Why do you want to know?"
50300 or
50400 (22) Model- "You pry too much".
50500 If the input is recognized as a statement, a reply is chosen from a
50600 list which indicates soem degree of anxiety:
50700 (23) Model- "Who are you really?"
50800 (24) Model- "You are making me nervous."
50900 If ANGER is high and the input is a question, a reply is chosen from
51000 a list designed to express hostility as in:
51100 (25) Model- "Do you know what you are doing?"
51200 (26) Model- "Perhaps you are just posing as a doctor."
51300 Sometimes in these default conditions the flag set in
51400 DELUSIONALREFERENCE allows the model to continue bt giving the next
51500 line in its delusional story. If the story is under discussion,
51600 continuity is maintained. But if it is not, the model appears to
51700 ignore the input and jumps back to oen of its previous
51800 preoccupations. In this instance the property of rigidity is a
51900 function of linguistic non-comprehension and not of the paranoid
52000 processes per se.
52100 If a story flag has not been st by a previous discussion in
52200 the interview and ANGER and FEAR are not high, the algorithm tries to
52300 see if the input is some type of general prompt from the interviewer
52400 such as:
52500 (27) Dr.- "Go on."
52600 or
52700 (28) Dr.- "Tell me more."
52800 If so, the model continues with its current story or attempts to
52900 initiate another story. (Discuss this in analyzer section?)
53000 If none of these conditions hold, the procedure ANSWER
53100 is called. This procedure handles a group of common special-case
53200 miscellaneous questions such as:
53300 (29) Dr.- "How do you do?"
53400 and miscellaneous statements such as:
53500 (30) Dr.- "Hi."
53600 (31) Dr.- "Good evening."
53700
53800 .F
53900 SELFSCAN
54000 The final procedure in the algorithm scans what the model has
54100 chosen to output. That is, it treats its own output as input. If this
54200 expression contains a flare or delusional reference, the appropriate
54300 flags are set and FEAR is raise slightly, but not as much as if this
54400 expession came from the interviewer. In this way the model "frightens
54500 itself" by what it says about a frightening topic.
54600
54700 SUMMARY