THE MECHANICS OF HUMAN CONVERSATION

Matthew Probert

GENERAL OVERVIEW:
Conversation is the continual reaction to the receipt of language based
stimuli by a subject. These language stimuli may take the form of: spoken
words and sounds such as normally associated with conversation; written words
and symbols in teletype and computer based communication; body language as
with the movements and symbols created with the body.
Upon receipt of symbols and sounds the subject's perception processes
recognise these as "language", although they may not have been intended as
such and this is an important point. It is important to realise that a subject
when in an interactive situation with another subject is continually receiving
visual, auditory and tactile signs from the other subject. These signs are
continually being processed by the perception process and a guess is made as
to their meaning. This guess is rarely accurate.
The guessing process is comprised of two indistinct phases:
1) A very quick, spontaneous response system very quickly searches a subset memory store for matching words/phrases. Such as the reaction to signs
that are believed to be greetings. A subject receives the sign "hello" and
responds immediately with the appropriate response, in England that might
be a reciprocal "hello" back to the originator.
2) A slow, considered analytic search of the subject's entire memory store
takes place even while the first phase is occuring. This analysis or
"thinking" process helps to improve the accuracy of the guess made as to
the meaning of the received sign. Upon receipt of new signs, this process
may wander to analyse a new sign or may simply give-up and work on
something else or lie dormant.
Before a response can be made or an action taken, the details of the response
or action must be formulated within the mind. This formulation takes place in
an area I call the "response holding area". This area is in turn fed by an
area of dynamic short-term memory that holds a small amount of data related to
the last conversation piece.
To react the subject retrieves from the response holding area the response and
may or may not analyse this proposed response before either putting it into
action, or discounting it and attempting to formulate a new course of action.
Should this process of formulation, analysis and discounting take place
several times in an apparently endless cycle we have materialisation of the
situation known as confusion. This confusion cycle affects the analysis and
retrieval of other data and the subject becomes disorientated and frightened.
The process of analysis, fetching data from the long-term memory, comparing it
with the data held in the short-term memory and comparing the results with
memories of simular situations fetched from the long-term memory requires
effort by the subject. The longer this process continues, the more effort is
required and in turn this causes pain. This may be seen manifesting itself in

situations where two participants are conversing. One participant is providing

the other participant with unexpected symbols, perhaps words that are unknown
to him. After a short while the participant who is having difficulty in
understanding will abort the interaction.
Unexpected stimuli can also cause spontaneous emotional reactions; amusement,
anger, bemusement or fear. Consider the reactions to the receipt of
significant, and yet unexpected news. The process of analysis can be clearly
detected as the recipient remains silent and motionless for a time, trying
desperately to understand the stimuli and to collate all known relevant facts
from the long-term memory.

STIMULI

------>

PERCEPTION ------> REACTION

Fig 1. Basic System of Conversation

Fig 2 illustrates the response process to a perceived stimuli. I stress again

"perceived" stimuli, as misunderstandings are very common during human
interaction.
+------------------+
|
|
|
\|/
PS ----> STM ----> FR ----> ROK ----> REACTION
/|\
| |
|
| |
|
| |
LTM <-------------+ |
|
/|\
|
|
+--------->---------+
Fig 2. Response
Key to Fig 2:
PS
STM
FR
ROK
LTM

Perceived stimuli
Short-term memory
Formulated Response
Response okay analysis process
Long-term memory

Consider a typical scenario containing two imaginary participants; Fred and

John. Both are English men in their mid-twenties, and have been interacting on
a mutualy beneficial level - they are friends - for some time. Both relax at a
public house called the Kings Head, where they frequently meet each other
although such meetings are not planned. It will not be unusual then for Fred
to enter the Kings Head pub on a Thursday evening at around 8 pm and to meet
John standing at the bar. Upon observing his friend, Fred will move towards
John and instigate interaction with a greeting, perhaps he will say "Alright
John, how's it going?" This is a common greeting among this type of person in

the South of England. The expected response is something along the lines of
"fine" or "okay". If however, Fred greets John with a new greeting, perhaps
"Hello", John may still perceive only the greeting and respond with the first
response extracted from the closest part of memory that is the normal response
to the expected greeting of "Alright John, how's it going?", so John may
reply, "yeah fine Fred". If John is not currently engaged in any other
concious mental activity he may take the time to consider the received
stimuli, and respond in a more intelligent manner.
Spontaneous conversation makes use of short-term and close recall memory. It
is an effortless and therefore enjoyable process. Examples include:
Greetings - "Hello", "hello". "How are you?", "I'm fine thankyou"
Customer-assistant interactions in shops such as; "Yes sir what would you
like?", "I should like a pound of sprouts please."
Analytic conversation requires effort. It involves searching the long-term
memory and recalling data into the short-term memory. The requirement on a
degree of effort makes this process less generally enjoyable. Although, many
people do enjoy the stimulation and the process of recall. To them it becomes
a challenge. A process of proving their ability and therefore of receiving
gratification and reassurance. A notable example of this is debating where two
individuals may be locked in conversational combat that is as aggressive and
fullfilling as any physical combat.
The active process of searching long-term memory varies between individuals.
Those which are proficient at it are thought of as quick witted and sharp. The
successful ad-liber is an example. Stand-up comedians if they are to be
successful must develope the skill of recalling and developing responses very
quickly if they are to be successful in their vocation.
During the spring and summer months of 1994 I conducted an experiment into
conversation. Participants were invited to contact a computer bulletin board
service. The participants generally were familiar with the phenomena of
bulletin boards, which are a computer system from which messages to other
callers may be left, and computer files exchanged. It is also common for
conversations to take place on-line between callers or a caller and the
operator of the host computer. This operator being known as a "systems
operator". The bulletin board in question was advertissed as a forum for
discussion into artificial intelligence. Participants were also informed that
the bulletin board would provide facilities for conversing with computer
personalities as well as the human operator. However, the particpants were not
informed when they would be conversing with which. On some occassions they
would instigate a conversation and it would be carried out with the human
operator, and at other times a computer program would respond to them.
The computer programs used for the experiment were programmed to simulate the
type of spontaneous conversation that would be expected to occur between two
parties who could predict what the other would say. When an unexpected stimuli
was received by the computer program, it would respond either by changing the
subject, or with a humourous indication of its confusion. While many
particpants realised after varying times that they were conversing with a
machine, more signifcant was the number of particpants who frequently mistook
the human operator for a machine.
The computer programs were frequently caught out by unexpected questions that
they had not been programmed to respond to, and as such they responded in an
incoherrent manner. This was detected by the more experienced human callers
quite quickly, although callers who had never conversed with a machine before

were still unaware that they were not talking to a human. More often the speed
at which the computer typed, and the regularity of its typing speed (it's body
language) gave the human caller an indication as to the mechanical nature than
the responses. Therefore, when speaking to a human operator with a similarly
fast and uniform typing rate the particpants mis-perceived the body language
to be that of a computer. From this we can see that conversation is not
restricted to word symbols. Inflexion in spoken conversation, typing speed in
teletype conversation and body movements in close contact conversation all
assist and hinder the perception process in putting meaning to the received
stimuli.
THE ANALYSIS PROCESS:
Having discussed the general picture of conversation I should like to turn
attention to a more detailed look at the analysis process that occurs. The
process of analysis of language is called "parsing" and the mechanisms used
are called "parsers". The two indistinct processes already mentioned may now
be examined in more detail. The first, the "spontaneous reaction" is very
quick analysis carried out by a mechanism called a "slot-and-frame parser".

+----------------+
| INPUT PHRASE |
+-------|--------+
\|/
+----------+----------+
+----------->-------------+ COMPARE INPUT PHRASE|
|
| WITH STORED PHRASES |
|
+-----------|---------+
|
\|/
|
+--+
+--+---+
+---+
|
|NO+---------<------+MATCH?|------>-------+YES|
|
+|-+
+------+
+-|-+
/|\
\|/
\|/
|
|
+------+-------+
|
|
|
RESPOND |
|
|
+--------------+
|
|
++--+
+------+-------+
|YES|-----+ ANY MORE |
+---+
|STORED PHRASES|
+------|-------+
\|/
+----------+
+-++
|
|
|NO+----->-----+ FAILURE |
+--+
|
|
+----------+
Fig 3. Flow diagram of a slot-and-frame parser

A slot-and-frame parser is simply an implementation of two tables of data: one

of known or expected phrases, a second with corresponding responses.

Phrase
Response
+--------------------|------------------------+
|HELLO
| HELLO
|
+--------------------+------------------------+
|HOW ARE YOU?
| VERY WELL THANK YOU
|
+--------------------+------------------------+
|HOW'S THE WIFE?
| STILL BREATHING
|
+--------------------+------------------------+
|DO YOU WANT A BEER? | IS THE POPE A CATHOLIC?|
+--------------------+------------------------+
|DO YOU TAKE SUGAR? | NO THANKS
|
+--------------------+------------------------+
Fig 4. An example slot-and-frame phrase and response table
(based upon observations of Southern England, male pub culture)
The second process, the analytical consideration requires the stimuli phrase
to be broken down into components of sufficiently small size that data
associated with them can be recalled from the long-term memory. For example.
Consider an unexpected phrase:
"Get that fat cat off my grass!"
Since this is not an every-day stimuli, there is no corresponding response
programmed in the short-term memory, and analysis passes to the analytical
process. The analytical process scans the phrase to break it down into
components such as:
Question/Order (The nature of the phrase)
Object

(What the phrase refers to)

Subject

(What the phrase says about the object)

Qualifiers

Colour

)
)
) Descriptions qualifying both the object and subject
)
)

Number

(All, one, two etc)

Location

(Where is the object)

Action

(Verbs)

Person

(You, me, us, them, they, him, her)

Tense

(Past, present, future or indeterminable)

Size

In this phrase the components are:

The
The
The
The
The
The

question is
order is
object is
qualifiers are
colours are
sizes are

DO (implied)
CAT
MY
FAT

The
The
The
The
The
The

subject is
person is
number is
action is
location is
tense is

GRASS
YOU (implied)
GET OFF
GRASS

A recall then takes place of the object - "CAT" - this being a common object
it is located in fairly close recall memory. Having collected the data about
"CAT" the same occurs for the subject - "GRASS" - and the qualifications;
"FAT" and "MY". Finally the order is interpreted and the action taken. Little
wonder then that often unexpected stimuli of this nature will produce
responses like;
"What about the cat?"
Where the recall process in retrieving data about the object has taken so much
effort that it has resulted in the loss of the original stimuli from the short
term memory, as in instances of "What were we talking about?"
If, upon breaking down the stimuli phrase we find that the "object" is
missing, such as in the phrase "They're at it again", we may require
qualification of the statement. Before requesting qualification, however, a
search will be made of the short-term memory in an attempt to qualify the
statement ourselves. Should we find that an onject exists from a previous
stimuli, then we may substitute this previous object into the new stimuli
phrase. For example;
"Those damned cats are always fighting"
"Oh yes?"
"I don't believe it! They're at it again"
In the last phrase we will substitute "they" with "those damned cats" from the
previously received stimuli phrase that is still alive in the short-term
memory, and we may also, upon finding a lack on a required action substitute
the action from the previous phrase so that we perceive the stimuli as:
"I don't believe it! Those damned cats are fighting again"
If we are unable to or we do not exert sufficient effort to qualify the
stimuli phrase ourself, we may request qualification from the originator. As
in this imaginary intercourse:
Mrs A.
Mr B.
Mrs A.
Mr B.
Mrs A.

"They're at it again"
"Who?"
"The cats"
"Doing what again?"
"Fighting!"

(c) Copyright 1994

Matthew Probert