comparison between human to human and human to chatbot conversation

2. Jennifer Hill, W. Randolph Ford, Ingrid
G. Farreras

3.  This study analyzed how communication
changes when people communicate with an
intelligent agent as opposed to with another
human.
Keywords
 CMC
 Instant messaging
 Chatbot
 Cleverbot

4.  Obstacles for computers to learn human
language.
 CMC
 Instant messaging vs Text messaging
Objectives
 To determine the linguistic characteristics of
Chatbot.
 Do humans communicate differently when
they know their conversational partner is a
computer?

5.  People would send fewer messages, write
fewer words per message, and exhibit a more
limited vocabulary when communicating with
a chatbot compared to another person.
 Reasons?
 Less experience communicating with chatbot
 Designed to sustain conversation
 No element of common history

6.  Compared 100 random human IM conversations
against 100 random human conversations with
Cleverbot.
 Cleverbot?

7.  Collection of data
 Human to human conversation (students)
 894 IM conversations between 205 unique users
 Human to chatbot conversation (Rollo Carpenter)
 2140 Cleverbot conversations
 100 conversations of each type was selected.
 The two types of conversations contained 7261
messages, with a total of 41,307 words,
collectively.

8.  Seven dependent variables were investigated
in this study.
 Words per conversation
 Messages per conversation
 Average number of words per message
 Type-token ratio
 Profanity
 Shorthand
 emoticons

9.  Messages (fewer words) to chatbot (2-13)
 Messages (more words) to human (2-25)
 No. Of messages to Chatbot (19-248)
 NO. Of messages to human (3-122)
 Human conversing with Cleverbot ( a mean
4.29 words per message
 Cleverbot’s mean 3.57 words per message

10.  Human conversing with another human
averaged 7.95 words per message
 Their responding human partners averaged
8.14 words per message.
 No foul language in 85% of human to human
conversation.
 Only 20% of human-Cleverbot conversations
contained no foul language.

11.  Purpose  comparison of human to human
conversations with human to chatbot
conversations.
 Hypothesis  human would send fewer messages
and write fewer words per message to chatbot
than when communicating with other humans.
 Result  the messages sent to chatbots did contain
fewer words per message, as predicted. People
were inclined to send more than twice as many
messages to chatbots compared to other people,
contrary to expectations and disconfirming the
notion that people feel less confident or
comfortable communicating with chatbot.

12.  Reason  people were not tentative.
 For example: people adapt their language when
conversing with children or non-native speakers.
 To test this hypothesis, words per message were
analyzed of each human-chatbot and human- human
conversations.
 Human conversing with chatbot expressed a
similar(small) number of words per message as
chatbot did.
 Human conversing with human averaged a
similar(higher) number of words per message.

13.  Chatbots are limited.
 In spite of these, many people are willing to
have extensive interactions with chatbots.