00100	    		CHAPTER FOUR
00200		PROBLEMS FOR COMPUTER UNDERSTANDING OF NATURAL LANGUAGE 
00300		 IN TELETYPED PSYCHIATRIC INTERVIEWS
00400	
00500	
00600	     By `natural language` I shall mean everyday American English
00700	such as is used by readers of this book in ordinary conversations.
00800	It is still difficult to be explicit about the processes which
00900	enable hummans to interpret and respond to natural language.
01000	Philosophers, linguists and psychologists have 
01100	investigated natural language with various purposes and few
01200	useful results.  Now attempts are being made in artificial intelligence to write        
01300	algorithims which `understand' natural language expressions.     
01400	     During the 1960's when machine processing of natural language
01500	was dominated by syntactic considerations, it became clear that
01600	syntactical information alone was insufficient to comprehend the expressions of ordinary conversations.
01700	The current view is that to understand what is said in linguistic expressions, syntax and semantics
01800	must be combined with beliefs   from an underlying conceptual
01900	structure having an ability to draw inferences.
02000	How to achieve this combination efficiently with a large data-base represents a monumental task for
02100	both theory and implementation.
02200	     Since the behavior  being simulated by  our paranoid model is the
02300	language-behavior  of a paranoid patient in a psychiatric
02400	interview, the model must have an ability to interpret and respond to
02500	natural language input sufficient to demonstrate  language-behavior 
02600	characteristic of the paranoid mode. How language is understood
02700	depends on the intentions of the producers and interpreters in the dialogue.
02800	And language is understood in accordance with the game being played.
02900	Our purpose was to develop a method for understanding everyday English sufficient
03000	for the model to behave conversationally in a paranoid way in a
03100	circumscribed situation. 
03200	
03300	We did not try to construct a general-purpose algorithm which could understand anything
03400	said in English by anybody to anybody in any dialogue situation. (Is it possible?)
03500		We took as a pragmatic measure of "understanding" the ability of the algorithm
03600	to `get the message' of an expression by grasping the imperative intent
03700	of the interviewer ,i.e.what effect he is trying to bring about in the interpreter relative to the topic.
03800	This straightforward approach to a complex problem has its
03900	drawbacks, as will be shown, but we strove for an idiosyncratic idiolect
04000	sufficient to demonstrate paranoid processes of an individual in a particular situation rather than 
04100	for a general or standard-ideal comprehension of English. If the language-understanding
04200	process got in the way of demonstrating the paranoid processes, we would
04300	consider it defective and insufficient for our purposes.
04400	           (Insert from Machr here)
04500		The main  problems 
04600	a dialogue algorithm must cope with in a psychiatric interview will now be discussed.
04700	                  QUESTIONS
04800	
04900		The principal sentence-type used by an interviewer consists of
05000	a question. The usual wh-  and yes-no questions must be recognized by
05100	the language-analyzer. In teletyped interviews a question may sometimes be
05200	put in declarative form followed by a question mark as in:
05300	        (1) PT.-  I LIKE TO GAMBLE ON THE HORSES.
05400	            DR.- YOU GAMBLE?
05500	
05600	Particularly difficult are `when' questions which require a memory which
05700	can assign each event a beginning, end and a duration. Also troublesome
05800	are questions such as `how often', `how many', i.e. a `how' followed
05900	by a quantifier.
06000		In constructing a simulation of a thought process it is arbitrary
06100	how much information to represent in memory. Should the model
06200	know what is the capital of Alabama? We took the position that the model
06300	should know only what we believed it reasonable to know about a few
06400	hundred topics expectable in a psychiatric interview. Thus the model performs badly when subjected to baiting
06500	`exam' questions designed to test its limitations rather than to seek
06600	useful psychiatric information.
06700		     IMPERATIVES
06800	
06900		Typical imperatives in a psychiatric interview consist of 
07000	expressions like: 
07100		(2) DR.- TELL ME ABOUT YOURSELF.
07200		(3) DR.- LETS DISCUSS YOUR FAMILY.
07300	Such imperatives are equivalent to interrogatives
07400	about the topics they refer to. Since the only physical action the model
07500	can perform is to `talk' , imperatives should be treated as requests
07600	for information.
07700	              DECLARATIVES
07800	
07900		In this category I shall lump everything else. It includes
08000	greetings, farewells, yes-no type answers, existence assertions 
08100	and predications made upon a subject.
08200	Naturally negations must be checked for.
08300	
08400	              AMBIGUITIES
08500	
08600		Words have more than one sense, a convenience for human memories
08700	but a pain for language-analysing  algorithms. Consider the word `bug' in
08800	the following expressions:
08900		(4) AM I BUGGING YOU?
09000		(5) AFTER A PERIOD OF HEAVY DRINKING HAVE YOU FELT BUGS ON YOUR SKIN?
09100		(6) DO YOU THINK THEY PUT A BUG IN YOUR ROOM?
09200	In (4) the term `bug' means to annoy, in (5) it refers to an insect and
09300	in(6) it refers to a microphone used for survellience. Some common words
09400	like `run' have fifty or more common senses. Context must be used to
09500	carry out disambiguation, as described in 00.0. Also we have the advantage
09600	of an idiolect where we can arbitrarily restrict the word senses. One
09700	characteristic of the paranoid mode is that no matter in what sense the
09800	interviewer uses a word, the patients idiosyncratically interprets it
09900	in the sense relevant to his pathological malevolence beliefs.
10000	  		ANAPHORIC REFERENCES
10100	
10200		The common anaphoric references consist of the pronouns `it',
10300	`he', `him', `she', `her', `they', `them' as in:
10400		(7) PT.-HORSERACING IS MY HOBBY.
10500		(8) DR.-WHAT DO YOU ENJOY ABOUT IT?
10600	The algorithm must recognize that the 'it' refers to `horseracing'. More
10700	difficult is a reference more than one I/O pair back in the dialogue as in:
10800		(9) PT.-THE MAFIA IS OUT TO GET ME.
10900		(10) DR.- ARE YOU AFRAID OF THEM?
11000		(11) PT.- MAYBE.
11100		(12) DR.- WHY IS THAT?
11200	The `that' of (12) does not refer to (11) but to the topic of being
11300	afraid which the interviewer introduced in (10). Another pronominal
11400	confusion occurs when the interviewer uses `we' in two senses as in:
11500		(13) DR.- WE WANT YOU TO STAY IN THE HOSPITAL.
11600		(14) PT.- I WANT TO BE DISCHARGED NOW.
11700		(15) DR.- WE ARE NOT COMMUNICATING.
11800	In (13) the interviewer is using `we' to refer to psychiatrists or 
11900	the hospital staff while in (15) the term refers to the interviewer and patient.
12000	
15500	
15600	FRAGMENTS
15700	
15800	     Another major problem for algorithms which attempt to understand
15900	discourse consists of the fact that many of the input expressions
16000	are not well-formed.  All sorts of fragments and ellipses appear
16100	which must somehow be connected to conceptualizations under discussion.
16200	For example, consider the following exchange:
16300	
16400	   
16500	     {10} Dr. - `How do you like the hospital?'
16600	
16700	     {11} Pt. - `I shouldn't be here.'
16800	
16900	     {12} Dr. - `Why not?'
17000	The question {12} is an elliptical expression for the full concept
17100	ualization
17200	
17300	     `Why should you not be in the hospital?'
17400	
17500	     Junk words {`well now'} {`tell me more'} and go ahead signals
17600	must be responded to by continuation of a topic.
17700	
17800	For example:
17900	
18000	     {13} Pt.- `I went to the track last week.'
18100	
18200	     {14} Dr. - `Really?'
18300	
18400	Such expressions as {14} stand in a meta-relation to the topic and
18500	serve to keep the conversation going.
18600	
18700	REJOINDERS
18800	
18900	     Sometimes the input expression from the interviewer is a rejoinder
19000	, a reply to a reply by the patient.  For instance:
19100	
19200	     {15} Dr. - `Who are you afraid of?'
19300	
19400	     {16} Pt. - `The Mafia is out to get me.'
19500	
19600	     {17} Dr. - `I would be afraid of them also.'
19700	
19800	in which {17} is a rejoinder.  Such expressions are not requests for
19900	information but provide information for the patient's model of the
20000	interviewer.
20100	
20200	INTERVIEWER-INTERVIEWEE RELATIONS
20300	
20400	     It is characteristic of psychiatric interviewing that the
20500	participants from time to time do not simply talk about the
20600	patient.  Two situations exist concurrently in an interview,
20700	one being talked about and one the participants are in.  At
20800	times the second situation becomes the first.  When the partici
20900	pants discuss one another and their relation, the dialogue
21000	expressions contain intentional verbs which in English fit the
21100	pattern `I X you' or `you X me'.  The comprehension process must
21200	distinguish clearly between subject and object in the case of some
21300	of these verbs.  For example in
21400	
21500	     {18} `I like you'
21600	
21700	the speaker 'I' experiences the liking but in
21800	
21900	     {19} `Do I please you?'
22000	
22100	the `you' experiences the pleasure as a consequence of something
22200	`I' does.