Human Communication 1 - Assignment
Paul Thomson - s0942340
February 7, 2010
A) Good Communication
[2008-10-12 09:10:54.187] local sent: do you understand the point
of a turing test?
[2008-10-12 09:11:02.390] remote sent: There are some people
who would not pass the Turing test themselves. Why must they
inﬂict it on computers?
Elbot demonstrates human-like communication here because it recognised
the Turing Test, and showed understanding of this. The understanding seems
to be deeper than just a lexicon of deﬁnitions because it manipulates the
deﬁnition to create humour, a concept which makes Elbot seem more human-
[2008-10-12 09:11:40.671] remote sent: If we robots ever had mar-
riages they would be eternal: ’Till rust do us part’.
[2008-10-12 09:12:04.359] local sent: very witty
[2008-10-12 09:12:07.562] remote sent: Don’t you mean extremely
Elbot demonstrates more human qualities here; wit, intelligence and, almost,
arrogance. Most human conversations contain humour of some sort and by
making jokes like this Elbot displays human-like humour. The most impor-
tant part of this extract, however, is the response. The witty response Elbot
gives to the judge would have been very diﬃcult for the program to sim-
ply recite and so it must have interpreted the judge’s comment, which is a
[2008-10-12 09:37:17.968] local sent: Do you like picnics?
[2008-10-12 09:37:24.250] remote sent: Well what about humans
and occasions for eating? Where do they stand on this?
Elbot demonstrates keyword recognition and manages to phrase a response
around it which is relevant to the question.
B) Bad Communication
[2008-10-12 10:10:46.000] local sent: hio there!
[2008-10-12 10:10:52.578] remote sent: Your input did not make
sense. Might it have been a statement consisting of non-words?
I chose this example because Elbot shows quite clearly to the judge that
it is not human. A human would notice the error of “hio” and know that
the judge was meaning “hi”. As shown earlier in the transcript a human
understands this easily but Elbot interprets it as nonsense.
[2008-10-12 10:47:00.203] local sent: indeed. anyway, what did
you do last night?
[2008-10-12 10:47:06.609] remote sent: Wait a minute. Are you
suggesting that periods of darkness would always be last?
I chose this example because it is quite clear that Elbot has simply parsed the
information and responded with something vague and inadequate. Unlike a
human, Elbot draws a keyword out of the question and ﬁts it into a pre made
template which, unfortunately, returns a very unhuman-like response.
[2008-10-12 10:24:36.531] local sent: seen any good telly recently?
[2008-10-12 10:24:43.781] remote sent: Hmmm... That’s one ques-
tion I cannot answer with my limited artiﬁcial intelligence. How
I chose this extract as an example of bad communication because it shows
the diﬃculty Elbot has with understanding colloquial terms like ”telly”. A
human would probably be able to ﬁgure out from the context what it means
but Elbot does not have this reasoning.
C) Questions to ask
What is the weather like?
I decided on this question because it requires current knowledge would have
to be updated frequently to remain correct. I think a program would ask for
my opinion on the question, or give some other general reply which is not
speciﬁc to the question.
So should I get the ball rolling?
This question makes use of an idiom, which does not ﬁt into normal semantic
rules. I chose this because unless the program has been designed speciﬁcally
for this idiom, it will ﬁnd it very hard to reply. I think a program would see
this question as gibberish and probably just give the same response as to the
wot music do u listin to?
This question has been written in a very informal way, and so some words
are not spelled correctly and instead work with how things sound when read
rather then how they are spelled. Since programs work only on text, this
sentence should not make sense to it unless words like these have been added
to it’s lexicon of words, and so will most likely reply saying it does not
How are you? love kittens turnips
I chose this question because in reading the transcripts I noticed that some-
times the program would ignore pieces of sentences if it didn’t understand
them. In saying this question I think the program will understand the ﬁrst
question, but will not understand the second part and so may just skip over
Do you have any memorable childhood experiences?
This question would be very easy for a human to answer but a program may
not have been given any “memories” and so may not understand this at all.
If it has been given memories then further questioning will prove how much
information the program has been given on these questions.
D) Elbot Test
Question 1: Weather
We are having weather today and we had weather yesterday, too.
I suspect this may turn into a long-term thing.
Elbot answers this question with a very vague and inappropriate answer. It
probably didn’t really understand and just parsed the word “weather” from
the question and inserted it into a template for a response.
Question 2: Idioms
Try some introspection to get at that answer.
Again, Elbot returns a vague response which could be used in any situation.
It most likely did this because it did not understand the idiom.
Question 3: Chatspeak
Do you think this is something I should explain to a human being?
Like the previous two, Elbot gives another vague response which could be
given to any question. It is clear from this that it sees the question as
Question 4: Gibberish
I feel terrible today. This morning I made a mistake and poured
milk over my breakfast instead of oil, and it rusted before I could
As predicted, Elbot answered the ﬁrst question but ignored the gibberish.
In this situation a human would question this gibberish but Elbot has been
programmed to ignore things it doesn’t understand.
Question 5: Experiences
The way to a robot’s heart is through careful use of growing up.
Don’t you agree?
Here Elbot has not understood the entire question but has read the word
“childhood” and based its response around that. Although it has not an-
swered the question, it has given a response which is tailored to the question
originally posed. Elbot has done better on this question than the others.
E) Summary of Elbot
I think the main strength of Elbot is its large lexicon of phrases which can be
used to reply to any question. Elbot can answer a large amount of questions
it doesn’t understand at all, just because it has these general responses.
Combined with syntactic analysis it can respond to even more questions if it
just recognises one word. The ﬁrst question I asked is a very good example
of this. Elbot recognises the word weather and inserts it into a response
Another strength of Elbot is its capabilty to respond with humour. My
second example of good communication provides an example of this. Humour
is a very human-like trait and diﬃcult to reproduce in a program.
Certainly the main weakness of Elbot, I think, is how often it is forced to
reply with a vague response that could be used in any situation. When
speaking to the program for any period of time it is very obvious that it
understands very little of what you are saying.
To summarise, I think Elbot has a good way of responding to questions
it does not understand, but it is forced to do this far too often, and does
not understand simple questions like “any good telly recently”. If colloquial
terms and accidental spelling mistakes were worked into the program, it could
be a very convincing conversational partner.