perm filename CHAP4[4,KMC]22 blob sn#061023 filedate 1973-08-31 generic text, type T, neo UTF8
00100	LANGUAGE-RECOGNITION PROCESSES FOR UNDERSTANDING DIALOGUES 
00200		IN TELETYPED PSYCHIATRIC INTERVIEWS
00300	
00400		Since the behavior being simulated by this paranoid model  is
00500	the   sequential   language-behavior  of  a  paranoid  patient  in  a
00600	psychiatric interview, the model must have an  ability  to  interpret
00700	and  respond  to  natural  language  input  sufficient to demonstrate
00800	conduct characteristic of the paranoid mode.   By "natural  language"
00900	I  shall  mean  ordinary American English such as is used in everyday
01000	conversations. It  is  still  difficult  to  be  explicit  about  the
01100	processes  which  enable  humans  to interpret and respond to natural
01200	language.  (A mighty maze !  but not without a plan  -  A.     Pope).
01300	Philosophers,  linguists  and psychologists have investigated natural
01400	language with various purposes. Few of the results have  been  useful
01500	to  builders  of  interactive  simulation models.  Attempts have been
01600	made  in  artificial  intelligence   to   write   algorithims   which
01700	"understand"  teletyped  natural language expressions.     (Colby and
01800	Enea,1967; Enea and Colby,1973;  Schank,1973;  Winograd,1973;  Woods,
01900	1970).  Computer  understanding of natural language is actively being
02000	attempted today but it is not  something  to  be  completly  achieved
02100	today  or  even  tomorrow.  The  problem at the moment is not to find
02200	immediately the best way of doing it but to find any way at all.
02300		During the 1960's when machine processing of natural language
02400	was dominated by  syntactic  considerations,  it  became  clear  that
02500	syntactical  information  alone  was  insufficient  to comprehend the
02600	expressions of ordinary conversations. A  current  view  is  that  to
02700	understand  what  is  said  in  linguistic  expressions, knowledge of
02800	syntax and semantics must be combined with beliefs from a  conceptual
02900	structure   capable   of  making  inferences.  How  to  achieve  this
03000	combination  efficiently  with  a  large   data-base   represents   a
03100	monumental task for both theory and implementation.
03200		For performance reasons we did not  attempt  to  construct  a
03300	conventional  linguistic parser to analyze conversational language of
03400	interviews. Parsers to date have had great difficulty  in  performing
03500	well  enough to assign a meaningful interpretation to the expressions
03600	of everyday conversational language in unrestricted English.   Purely
03700	syntactic parsers offer a cancerous proliferation of interpretations.
03800	A conventional parser, lacking neglecting  and  ignoring  mechanisms,
03900	may  simply  halt  when it comes across a word not in its dictionary.
04000	Parsers represent  tight  conjunctions  of  tests  instead  of  loose
04100	disjunctions needed for gleaning some degree of meaning from everyday
04200	language  communication.   It  is   easily   observed   that   people
04300	misunderstand  and  ununderstand  at  times and thus remain partially
04400	opaque to one another, a truth which lies at the core of  human  life
04500	and communication.
04600		How language is understood depends on  how  people  interpret
04700	the meanings of situations they find themselves in.  In  a  dialogue,
04800	language is understood in accordance with a participant's view of the
04900	situation. The participants are interested in both what an  utterance
05000	means  (what  it  refers  to)  and  what  the  utterer  means  (  his
05100	intentions). In a first psychiatric interview the doctor's  intention
05200	is to gather certain kinds of information; the patient's intention is
05300	to give information in order to receive help. Such  an  interview  is
05400	not  small  talk;  a  job is to be done. Our purpose was to develop a
05500	method for recognizing sequences of everyday English  sufficient  for
05600	the  model  to  communicate  linguistically  in a paranoid way in the
05700	circumscribed situation of a psychiatric interview.
05800		We did not try to construct a general-purpose algorithm which
05900	could  understand  anything  said in English by anybody to anybody in
06000	any dialogue situation.   (Does  anyone  believe  it  possible?   The
06100	seductive myth of generalization leads only  to  trivialization).  We
06200	sought  simply to extract, distill or cull some degree of, or partial
06300	idiosyncratic,  idiolectic  meaning  (not  the  "complete"   meaning,
06400	whatever  that means) from the input. We utilized a pattern-directed,
06500	rather than a parsing-directed,  approach  because  of  the  former's
06600	power to ignore irrelevant details.
06700		Natural language is not an agreed-on  universe  of  discourse
06800	such  as arithmetic wherein symbols have a fixed meaning for everyone
06900	who uses them. What we loosely call "natural language" is actually  a
07000	set of history-dependent, selective, and interest-oriented idiolects,
07100	each being unique to the individual  with  a  unique  history.(To  be
07200	unique   does  not  mean  that  no  property  is  shared  with  other
07300	individuals, only that not every property is shared). It is the broad
07400	overlap  of  idiolects  which  allows  the  communication  of  shared
07500	meanings in everyday conversation.
07600		We took as pragmatic measures of "understanding" the
07700	ability  (1)  to  form  a  conceptualization so that questions can be
07800	answered and commands carried out, (2) to determine the intention  of
07900	the  interviewer,  (3)  to  determine the references for pronouns and
08000	other anticipated topics.  This straightforward approach to a complex
08100	problem  has  its  drawbacks,  as  will be shown, but we strove for a
08200	highly individualized idiolect  sufficient  to  demonstrate  paranoid
08300	processes  of an individual in a particular situation rather than for
08400	a general supra-individual or ideal comprehension of English.  If the
08500	language-recognition processes  interfered  with  demonstrating   the
08600	paranoid  processes,  we would consider it defective and insufficient
08700	for our purposes.
08800		The  language-recognition process utilized by the model first
08900	puts the teletyped input in the form of a list  and  then  determines
09000	the  syntactic  type of the input expression - question, statement or
09100	imperative by looking at introductory terms and at  punctuation.  The
09200	expression-type   is   then   scanned  for  conceptualizations,  i.e.
09300	patterns  of  contentives  consisting  of   words   or   word-groups,
09400	stress-forms  of  speech  having  conceptual  meaning relevant to the
09500	model's interests.  The  search  for  conceptualizations  ignores  (as
09600	irrelevant   details)   function  or  closed-class  terms  (articles,
09700	auxiliaries, conjunctions, prepositions, etc.) except as  they  might
09800	represent  a  component  in a contentive word-group. For example, the
09900	word-group (for a living) is defined to mean `work' as  in  "what  do
10000	you  do  for a living?" The conceptualization is classified according
10100	to the rules of Fig. 1 as malevolent, benevolent or neutral. Thus the
10200	language  recognizer  attempts  to judge the intention of the utterer
10300	from the content of the utterance.
10400		(INSERT FIG.1 HERE)
10500		Some  special  problems a dialogue algorithm must handle in a
10600	psychiatric interview  will  now  be  outlined  along  with  a  brief
10700	description of how the model deals with them.
10800	
10900	.F
11000	QUESTIONS
11100	
11200		The principal expression-type used by an interviewer consists
11300	of a question. A question is recognized by its first term being a wh-
11400	or  how  form  and/or the expression ending with a question-mark.  In
11500	teletyped interviews a question may sometimes be put  in  declarative
11600	form followed by a question mark as in:
11700	.V
11800		(1) PT.- I LIKE TO GAMBLE ON THE HORSES.             	
11900		(2) DR.- YOU GAMBLE?
12000	.END
12100	Although  a  question-word  or  auxiliary verb is missing in (2), the
12200	model recognizes that a question is being asked about its gambling 
12300	simply by the question mark.
12400		Particularly  difficult  are  those  `when'  questions  which
12500	require a memory which can assign each event a beginning, an end  and
12600	a  duration.     An  improved  version  of the model should have this
12700	capacity.  Also troublesome are questions such as `how  often',  `how
12800	many', i.e.   a `how' followed by a quantifier. If the model has "how
12900	often" on its expectancy list while a topic is under discussion,  the
13000	appropriate   reply  can  be  made.  Otherwise  the  model  fails  to
13100	understand.
13200		In  constructing  a  simulation  of  symbolic processes it is
13300	arbitrary how much information to represent in the data-base,  Should
13400	the model know what is the capital of Alabama? It is trivial to store
13500	a lot of facts and there always will be boundary conditions.  We took
13600	the  position  that  the  model  should know only what we believed it
13700	reasonable to know relevant to a few hundred topics expectable  in  a
13800	psychiatric  interview. Thus the model performs poorly when subjected
13900	to baiting  `exam'  questions  designed  to  test  its  informational
14000	limitations rather than to seek useful psychiatric information.
14100	
14200	.F
14300	IMPERATIVES
14400	
14500		Typical imperatives in a  psychiatric  interview  consist  of
14600	expressions like:
14700	.V
14800		(3) DR.- TELL ME ABOUT YOURSELF.
14900		(4)  DR.-  LETS  DISCUSS  YOUR  FAMILY. 
15000	.END
15100		Such  imperatives  are   actually   interrogatives   to   the
15200	interviewee  about the topics they refer to.  Since the only physical
15300	action the model can perform is to `talk' , imperatives  are  treated
15400	as  requests  for  information.  They  are  identified  by the common
15500	introductory phrases: "tell me", "lets talk about", etc.
15600	.F
15700	DECLARATIVES
15800	
15900		In  this  category  is  lumped  everything  else. It includes
16000	greetings, farewells, yes-no type answers, existence  assertions  and
16100	the usual predications.
16200	
16300	.F
16400	AMBIGUITIES
16500	
16600		Words  have  more  than  one  sense,  a convenience for human
16700	memories  but  a  struggle  for  language-understanding   algorithms.
16800	Consider the word "bug" in the following expressions:
16900	.V
17000		(5) AM I BUGGING YOU?
17100		(6) AFTER A PERIOD OF HEAVY DRINKING HAVE YOU  FELT  BUGS  ON
17200	            YOUR SKIN?
17300		(7) DO YOU THINK THEY PUT A BUG IN YOUR ROOM? 
17400	.END
17500		In expression (5) the term "bug" means to annoy,  in  (6)  it
17600	refers  to  an  insect  and in (7) it refers to a microphone used for
17700	hidden  surveillence.    The  model  uses  context   to   carry   out
17800	disambiguation.   For example, when the Mafia is under discussion and
17900	the affect-variable of fear is high, the model  interprets  "bug"  to
18000	mean  microphone.    In  constructing this hypothetical individual we
18100	took advantage of the selective nature of idiolects which can have an
18200	arbitrary  restriction  on  word  senses.   One characteristic of the
18300	paranoid mode is that no matter in what sense the interviewer uses  a
18400	word, the patient may idiosyncratically interpret it in some sense of
18500	his own. This property is obviously of great help for an  interactive
18600	simulation with limited language-understanding abilities.
18700	.F
18800	ANAPHORIC REFERENCES
18900		The common anaphoric references consist of the pronouns "it",
19000	"he", "him", "she", "her", "they", "them" as in:
19100	.V
19200		(8) PT.-HORSERACING IS MY HOBBY.
19300		(9) DR.-WHAT DO  YOU  ENJOY  ABOUT  IT?  
19400	.END
19500		When a topic is introduced by  the  patient  as  in  (8),  a
19600	number  of  things  can  be  expected  to be asked about it. Thus the
19700	algorithm has ready an updated expectancy-anaphora list which  allows
19800	it  to  determine  whether the topic introduced by the model is being
19900	responded to or  whether  the  interviewer  is  continuing  with  the
20000	previous topic.
20100		The  algorithm  recognizes  "it"  in  (9)  as  referring   to
20200	"horseracing" because a flag for horseracing was set when horseracing
20300	was introduced in (8), "it" was placed on the expected anaphora list,
20400	and no new topic has been introduced. A more difficult problem arises
20500	when the anaphoric reference points more than one I-O  pair  back  in
20600	the dialogue as in:
20700	.V
20800		(10) PT.-THE MAFIA IS OUT TO GET ME.
20900		(11) DR.- ARE YOU AFRAID OF THEM?
21000		(12) PT.- MAYBE.
21100		(13) DR.- WHY IS THAT? 
21200	.END
21300		The "that" of expression (13) does not refer to (12)  but  to
21400	the  topic  of being afraid which the interviewer introduced in (11).
21500		Another pronominal confusion occurs when the interviewer uses
21600	`we' in two senses as in:
21700	.V
21800		(14) DR.- WE WANT YOU TO STAY IN THE HOSPITAL.
21900		(15) PT.- I WANT TO BE DISCHARGED NOW.
22000		(16) DR.- WE ARE NOT COMMUNICATING. 
22100	.END
22200		In expression (14) the interviewer is using "we" to refer  to
22300	psychiatrists  or the hospital staff while in (16) the term refers to
22400	the interviewer and patient. Identifying the correct  referent  would
22500	require beliefs about the dialogue itself.
22600	
22700	.F
22800	TOPIC SHIFTS
22900	
23000		In the main a psychiatric interviewer is in control of the
23100	interview. When he has gained sufficient information about a topic,
23200	he shifts to a new topic. Naturally the algorithm must detect this
23300	change of topic as in the following:
23400	.V
23500		(17) DR.- HOW DO YOU LIKE THE HOSPITAL?
23600		(18) PT.- ITS NOT HELPING ME TO BE HERE.
23700		(19) DR.- WHAT BROUGHT YOU TO THE HOSPITAL?
23800		(20) PT.- I AM VERY UPSET AND NERVOUS.
23900		(21) DR.- WHAT TENDS TO MAKE YOU NERVOUS?
24000		(23) PT.- JUST BEING AROUND PEOPLE.
24100		(24) DR.- ANYONE IN PARTICULAR?
24200	.END
24300		In (17) and (19) the topic is the hospital. In (21) the topic
24400	changes to causes of the patient's nervous state.
24500		Topics touched upon previously can be  re-introduced  at  any
24600	point  in  the  interview.  The  model  knows  that  a topic has been
24700	discussed previously because a topic-flag is set when a  topic  comes
24800	up.
24900	
25000	.F
25100	META-REFERENCES
25200	
25300		These are references, not about a topic directly, but about
25400	what has been said about the topic as in:
25500	.V
25600		(25) DR.- WHY ARE YOU IN THE HOSPITAL?
25700		(26) PT.- I SHOULDNT BE HERE.
25800		(27) DR.-  WHY DO YOU SAY THAT?
25900	.END
26000	The expression (27 ) is about and meta to expression (26 ). The model
26100	does  not  respond  with  a  reason  why it said something but with a
26200	reason for the content of what it said, i.e. it  interprets  (27)  as
26300	"why shouldnt you be here?"
26400		Sometimes when the patient  makes  a  statement,  the  doctor
26500	replies,  not  with  a  question,  but  with  another statement which
26600	constitutes a rejoinder as in:
26700	.V
26800		(28 ) PT.- I HAVE LOST A LOT OF MONEY GAMBLING.
26900		(29 ) DR.- I GAMBLE QUITE A BIT ALSO.
27000	.END
27100		Here  the  algorithm  interprets  (29  )  as  a  directive to
27200	continue discussing gambling, not as an indication  to  question  the
27300	doctor  about  gambling.  
27400	
27500	.F
27600	ELLIPSES
27700	
27800	
27900		In dialogues one finds many ellipses, expressions from  which
28000	one or more words are omitted as in:
28100	.V
28200		(30 ) PT.- I SHOULDNT BE HERE.
28300		(31) DR.- WHY NOT?
28400	.END
28500		Here the complete construction must be understood as:
28600	.V
28700		(32) DR.- WHY SHOULD YOU NOT BE HERE?
28800	.END
28900	Again   this   is  handled  by  the  expectancy-anaphora  list  which
29000	anticipates a "why not".
29100		The opposite of ellipsis is redundancy which usually provides
29200	no problem since the same thing is being said more than once as in:
29300	.V
29400		(33 ) DR.- LET ME ASK YOU A QUESTION.
29500	.END
29600	The model simply recognizes (33) as a stereotyped pattern.
29700	
29800	.F
29900	SIGNALS
30000	
30100		Some fragmentary expressions serve only as directive  signals
30200	to proceed as in:
30300	.V
30400		(34) PT.- I WENT TO THE TRACK LAST WEEK.
30500		(35) DR.- AND?
30600	.END
30700	The  fragment of (35) requests a continuation of the story introduced
30800	in (34). The common expressions found in interviews are "and",  "so",
30900	"go  on", "go ahead", "really", etc. If an input expression cannot be
31000	recognized at all, the lowest level default condition is to assume it
31100	is  a  signal  and either proceed with the next line in a story under
31200	discussion or if the latter is not the case, begin a new story with a
31300	prompting question or statement.
31400	
31500	.F
31600	IDIOMS
31700	
31800		Since so much of conversational language involves stereotypes
31900	ans  special  cases, the task of recognition is much easier than that
32000	of linguistic analysis.  This is particularly true of idioms.  Either
32100	one knows what an idiom means or one does not. It is usually hopeless
32200	to try to decipher what an  idiom  means  from  an  analysis  of  its
32300	constituent  parts.  If  the  reader  doubts this, let him ponder the
32400	following expressions taken from actual teletyped interviews.
32500	.V
32600		(36) DR.- WHATS EATING YOU?
32700		(37) DR.- YOU SOUND KIND OF PISSED OFF.
32800		(38) DR.- WHAT ARE YOU DRIVING AT?
32900		(39) DR.- ARE YOU PUTTING ME ON?
33000		(40) DR.- WHY ARE THEY AFTER YOU?
33100		(41) DR.- HOW DO YOU GET ALONG WITH THE OTHER PATIENTS?
33200	 	(42) DR.- HOW DO YOU LIKE YOUR WORK?
33300		(43) DR.- HAVE THEY TRIED TO GET EVEN WITH YOU?
33400		(44) DR.- I CANT KEEP UP WITH YOU.
33500	.END
33600		In  people,  the  understanding of idioms is a matter of rote
33700	memory.  In an algorithm, idioms can simply be stored as  such.    As
33800	each    new    idiom    appears    in   teletyped   interviews,   its
33900	recognition-pattern is  added  to  the  data-base  on  the  inductive
34000	grounds that what happens once can happen again.
34100		Another advantage in constructing an idiolect for a model  is
34200	that  it  recognizes  its  own idiomatic expressions which tend to be
34300	used by the interviewer (if he understands them) as in:
34400	.V
34500		(45) PT.- THEY ARE OUT TO GET ME.
34600		(46) DR.- WHAT MAKES YOU THINK THEY ARE OUT TO GET YOU.
34700	.END
34800		The expression (45 ) is really a double idiom in which  "out"
34900	means  `intend'  and  "get" means `harm' in this context. Needless to
35000	say.  an algorithm which tried to pair off the  various  meanings  of
35100	"out"  with  the  various meanings of "get" would have a hard time of
35200	it. But an algorithm which recognizes  what it itself is  capable  of
35300	saying, can easily recognize echoed idioms.
35400	
35500	.F
35600	FUZZ TERMS
35700	
35800		In this category fall a large number of expressions which, as
35900	non-contentives, have little or  no  meaning  and  therefore  can  be
36000	ignored by the algorithm. The lower-case expressions in the following
36100	are examples of fuzz:
36200	.V
36300		(47) DR.- well now perhaps YOU CAN TELL ME something ABOUT 
36400			YOUR FAMILY.
36500		(48) DR.- on the other hand I AM INTERESTED IN YOU.
36600		(49) DR.- hey I ASKED YOU A QUESTION.
36700	.END
36800		The algorithm has "ignoring mechanisms" which allows for  for
36900	an  `anything'  slot  in  its pattern recognition. Fuzz term are thus
37000	easily ignored and no attempt is made to analyze them.
37100	
37200	.F
37300	SUBORDINATE CLAUSES
37400	
37500		A subordinate clause is a complete statement  inside  another
37600	statement.  It  is  most frequently introduced by a relative pronoun,
37700	indicated in the following expressions by lower case:
37800	.V
37900		(50) DR.-  WAS IT THE UNDERWORLD that PUT YOU HERE?
38000		(51) DR.- WHO ARE THE PEOPLE who UPSET YOU?
38100		(52) DR.- HAS ANYTHING HAPPENED which YOU DONT UNDERSTAND?
38200	.END
38300		One of the linguistic weaknesses of  the  model  is  that  it
38400	takes  the  entire  input as a single expression.   When the input is
38500	syntactically complex, such as possessing  subordinate  clauses,  the
38600	algorithm  can become confused. To avoid this, future versions of the
38700	model will  segment  the  input  into  shorter  and  more  manageable
38800	patterns  in  which  an  optimal selection of emphases and neglect of
38900	irrelevant  detail  can  be  achieved  while  avoiding  combinatorial
39000	explosions.
39100	.F
39200	VOCABULARY
39300	
39400		How many words should there be in the algorithm's vocabulary?
39500	It  is  a  rare human speaker of English who can recognize 40% of the
39600	415,000 words in the Oxford  English  Dictionary.   In  his  everyday
39700	conversation  an  educated person uses perhaps 10,000 words and has a
39800	recognition vocabulary of about 50,000  words.  A study  of telephone
39900	conversations  showed  that 96 % of the talk employed only 737 words.
40000	(French, Carter, and Koenig, 1930). Of course if the remaining 4% are
40100	important  but  unrecognized contentives,the result may be ruinous to
40200	he coherence  of a conversation.
40300		In  counting  all  the  words  in  53  teletyped  psychiatric
40400	interviews conducted by psychiatrists, we found  only  721  different
40500	words.    Since  we  are  familiar  with psychiatric vocabularies and
40600	styles of  expression,  we  believed  this  language-algorithm  could
40700	function  adequately  with  a  vocabulary  of  at most a few thousand
40800	contentives. There will always be unrecognized words.  The  algorithm
40900	must  be  able to continue even if it does not have a particular word
41000	in its vocabulary.    This provision represents one  great  advantage
41100	of   pattern-matching   over  conventional  linguistic  parsing.  Our
41200	algorithm can guess while a parser must know with certainty in  order
41300	to proceed.
41400	
41500	.F
41600	MISSPELLINGS AND EXTRA CHARACTERS
41700		There is really no good defense  against  misspellings  in  a
41800	teletyped  interview  except  having a human monitor the conversation
41900	and make the necessary corrections. Spelling correcting programs  are
42000	slow,  inefficient,  and  imperfect.   They experience great problems
42100	when it is the first character in a word which is incorrect.
42200		Extra characters sent over the teletype by the interviewer or
42300	by a bad phone line can be removed  by  a  human  monitor  since  the
42400	output  from  the  interviewer first appears on the monitor's console
42500	and then is typed by her directly to the program.
42600	
42700	.F
42800	META VERBS
42900	
43000		Certain common verbs such as "think", "feel", "believe", etc
43100	can take a clause as their ojects  as in:
43200	.V
43300		(54) DR.- I THINK YOU ARE RIGHT.
43400		(55) DR.- WHY DO YOU FEEL THE GAMBLING IS CROOKED?
43500	.END
43600		The  verb  "believe"  is  peculiar  since it can also take as
43700	object a noun or noun phrase as in:
43800	.V
43900		(56) DR.- I BELIEVE YOU.
44000	.END
44100		In expression (55) the conjunction "that" can follow the word
44200	"feel" signifying a subordinate clause. This is not  the  case  after
44300	"believe"   in   expression   (56).   The  model  makes  the  correct
44400	identification in (56) because nothing follows the "you".
44500	.F
44600	ODD WORDS
44700		From  extensive  experience  with  teletyped  interviews,  we
44800	learned the model must have patterns for "odd" words.  We  term  them
44900	such  since  these  are  words  which  are quite natural in the usual
45000	vis-a-vis interview in which  the  participants  communicate  through
45100	speech  but  which  are  quite  odd  in  the  context  of a teletyped
45200	interview. This should be clear from the following examples in  which
45300	the odd words appear in lower case:
45400	.V
45500		(57) DR.-YOU sound CONFUSED.
45600		(58) DR.- DID YOU hear MY LAST QUESTION?
45700		(59) DR.- WOULD YOU come in AND sit down PLEASE?
45800		(60) DR.- CAN YOU say WHO?
45900		(61) DR.- I WILL see YOU AGAIN TOMORROW.
46000	.END
46100	
46200	
46300	.F
46400	MISUNDERSTANDING
46500	
46600		It is perhaps not fully recognized by  students  of  language
46700	how  often  people  misunderstand one another in conversation and yet
46800	their dialogues proceed as if understanding and being understood  had
46900	taken place.
47000		A classic example is the following man-on-the-street interview.
47100	.V
47200		INTERVIEWER -  WHAT DO YOU THINK OF MARIHUANA?
47300	 	MAN - DIRTIEST TOWN IN MEXICO.
47400		INTERVIEWER - HOW ABOUT LSD?
47500		MAN - I VOTED FOR HIM.
47600		INTERVIEWER - HOW DO YOU FEEL ABOUT THE INDIANAPOLIS 500?
47700		MAN - I THINK THEY SHOULD SHOOT EVERY LAST ONE OF THEM.
47800		INTERVIEWER - AND THE VIET CONG POSITION?
47900		MAN - I'M FOR IT, BUT MY WIFE COMPLAINS ABOUT HER ELBOWS.
48000	.END
48100		Sometimes    a    psychiatric   interviewer   realizes   when
48200	misunderstanding occurs and tries  to  correct  it.  Other  times  he
48300	simply  passes  it  by.  It is characteristic of the paranoid mode to
48400	respond idiosyncratically to particular word-concepts  regardless  of
48500	what the interviewer is saying:
48600	.V
48700		(62) PT.- SOME PEOPLE HERE MAKE ME NERVOUS. 	
48800	        (63) DR.- I BET.
48900	        (64) PT.- GAMBLING HAS BEEN NOTHING BUT TROUBLE FOR ME.
49000	.END
49100	Here one word sense of "bet" (to wager) is confused with the  offered
49200	sense  of expressing agreement. As has been mentioned,  this property
49300	of paranoid conversation eases the task of simulation.
49400	.F
49500	UNUNDERSTANDING
49600	
49700		A dialogue algorithm must be prepared for situations in which
49800	it simply does not understand. It cannot arrive at any interpretation
49900	as to what the interviewer is saying since no pattern can be matched.
50000	It may recognize the topic but not what is being said about it.
50100		The language-recognizer should not be faulted  for  a  simple
50200	lack of facts as in:
50300	.V
50400		(65) DR.- WHO IS THE PRESIDENT OF TURKEY?
50500	.END CONTINUE
50600	when  the  data-base does  not  contain  the  word 
50700	"Turkey". In this default condition it is simplest to reply:
50800	.V
50900		(66) PT.- I DONT KNOW.
51000	.END CONTINUE
51100	and dangerous to reply:
51200	.V
51300		(67) PT.- COULD YOU REPHRASE THE QUESTION?
51400	.END CONTINUE
51500	because of the disastrous loops which can result.
51600		Since  the  main  problem  in  the   default   condition   of
51700	ununderstanding is how to continue, the model employs heuristics such
51800	as  changing  the  level  of  the  dialogue  and  asking  about   the
51900	interviewer's intention as in:
52000	.V
52100		(68) PT.- WHY DO YOU WANT TO KNOW THAT?
52200	.END CONTINUE
52300	or  rigidly  continuing  with  a  previous topic or introducing a new
52400	topic.
52500		These are admittedly desperate measures  intended  to  prompt
52600	the  interviewer  in  directions the algorithm has a better chance of
52700	understanding. Although it is usually the  interviewer  who  controls
52800	the  flow  from  topic to topic, there are times when control must be
52900	assumed by the algorithm.
53000		There  are  many   additional   problems   in   understanding
53100	conversational   language   but   the  above  description  should  be
53200	sufficient to convey  some  of  the  complexities  involved.  Further
53300	examples  will  be  presented  in  the next chapter in describing the
53400	logic of the central processes of the model.