chameleons*in*imagined*conversaons - cornell universitycristian/chameleons_in... ·...
TRANSCRIPT
Chameleons in imagined conversa1ons: A new approach to understanding
coordina1on of linguis1c style in dialogs
Cris1an Danescu-‐Niculescu-‐Mizil
Lillian Lee
Department of Computer Science
Cornell University
06-‐23-‐2011 CMCL@ACL
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
When conversing, people non-‐consciously adapt to one another’s communica1ve behaviors. [Giles et al., 1991; Chartrand and Bargh, 1999]
Non-‐conscious coordina-on
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
When conversing, people non-‐consciously adapt to one another’s communica1ve behaviors. [Giles et al., 1991; Chartrand and Bargh, 1999]
Dimension Canonical study
Posture Condon and Ogston, 1967
Non-‐conscious coordina-on
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
When conversing, people non-‐consciously adapt to one another’s communica1ve behaviors. [Giles et al., 1991; Chartrand and Bargh, 1999]
Dimension Canonical study
Posture Condon and Ogston, 1967
Head nodding Hale and Burgoon, 1984
Non-‐conscious coordina-on
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
When conversing, people non-‐consciously adapt to one another’s communica1ve behaviors. [Giles et al., 1991; Chartrand and Bargh, 1999]
Dimension Canonical study
Posture Condon and Ogston, 1967
Head nodding Hale and Burgoon, 1984
Pause length Jaffe and Feldstein, 1970
Backchannels White, 1984
Self-‐disclosure Derlenga et al., 1973
Linguis1c style Niederhoffer and Pennebaker, 2002
Non-‐conscious coordina-on
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
When conversing, people non-‐consciously adapt to one another’s communica1ve behaviors. [Giles et al., 1991; Chartrand and Bargh, 1999]
Dimension Canonical study
Posture Condon and Ogston, 1967
Head nodding Hale and Burgoon, 1984
Pause length Jaffe and Feldstein, 1970
Backchannels White, 1984
Self-‐disclosure Derlenga et al., 1973
Linguis1c style Niederhoffer and Pennebaker, 2002
Communica1ve behaviors are “pa7erned and coordinated, like a dance” [Niederhoffer and Pennebaker, 2002]
Non-‐conscious coordina-on
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
When conversing, people non-‐consciously adapt to one another’s communica1ve behaviors. [Giles et al., 1991; Chartrand and Bargh, 1999]
Dimension Canonical study
Posture Condon and Ogston, 1967
Head nodding Hale and Burgoon, 1984
Pause length Jaffe and Feldstein, 1970
Backchannels White, 1984
Self-‐disclosure Derlenga et al., 1973
Linguis1c style Niederhoffer and Pennebaker, 2002
Non-‐conscious coordina-on
Communica1ve behaviors are “pa7erned and coordinated, like a dance” [Niederhoffer and Pennebaker, 2002]
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
When conversing, people non-‐consciously adapt to one another’s communica1ve behaviors. [Giles et al., 1991; Chartrand and Bargh, 1999]
Dimension Canonical study
Posture Condon and Ogston, 1967
Head nodding Hale and Burgoon, 1984
Pause length Jaffe and Feldstein, 1970
Backchannels White, 1984
Self-‐disclosure Derlenga et al., 1973
Linguis1c style Niederhoffer and Pennebaker, 2002
Non-‐conscious coordina-on
Communica1ve behaviors are “pa7erned and coordinated, like a dance” [Niederhoffer and Pennebaker, 2002]
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Linguis-c style coordina-on
Linguis-c style:
How things are said as opposed to what is said
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Linguis-c style coordina-on
Linguis-c style:
How things are said as opposed to what is said
Example:
Client: “At what 1me does your shop close?”
Shopkeeper: “At five o’clock.” Shopkeeper: “Five o’clock”
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Linguis-c style coordina-on
Linguis-c style:
How things are said as opposed to what is said
Example:
Client: “At what 1me does your shop close?”
Client: “What 1me does your shop close?”
Shopkeeper: “At five o’clock.” Shopkeeper: “Five o’clock”
[Levelt & Kelter, 1982]
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Linguis-c style coordina-on
Linguis-c coordina-on occurs: instantaneously
non-‐consciously
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Linguis-c style coordina-on
Linguis-c style:
for us: func1on word class usage
Coordina-on is not just exact matching:
Client: “At what 1me does your shop close?”
Shopkeeper: “In two hours.”
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Linguis-c style coordina-on
When does it happen?
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Linguis-c style coordina-on
When does it happen?
face-‐to-‐face sedngs [Niederhoffer & Pennebaker, 2002] -‐ Watergate transcripts
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Linguis-c style coordina-on
When does it happen?
face-‐to-‐face sedngs [Niederhoffer & Pennebaker, 2002] -‐ Watergate transcripts
non-‐face-‐to-‐face sedngs [Reifer et al. 2006; Gonzales et al. 2010; N & P, 2002] -‐ phone conversa1ons, online chat
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Linguis-c style coordina-on
When does it happen?
face-‐to-‐face sedngs [Niederhoffer & Pennebaker, 2002] -‐ Watergate transcripts
non-‐face-‐to-‐face sedngs [Reifer et al. 2006; Gonzales et al. 2010; N & P, 2002] -‐ phone conversa1ons, online chat
non-‐real-‐1me sedngs [Danescu-‐Niculescu-‐Mizil et al. 2011]
-‐ Twifer conversa1ons (involves a wide spectrum of rela1onship stages)
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Linguis-c style coordina-on
When does it happen?
face-‐to-‐face sedngs [Niederhoffer & Pennebaker, 2002] -‐ Watergate transcripts
non-‐face-‐to-‐face sedngs [Reifer et al. 2006; Gonzales et al. 2010; N & P, 2002] -‐ phone conversa1ons, online chat
non-‐real-‐1me sedngs [Danescu-‐Niculescu-‐Mizil et al. 2011]
-‐ Twifer conversa1ons (involves a wide spectrum of rela1onship stages)
Across a wide range of conversa-onal seIngs.
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Linguis-c style coordina-on
Why does it happen?
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Linguis-c style coordina-on
Why does it happen?
Ac-vely debated for the last 40 years. [Ireland et al. 2011, Branigan et al., 2010]
disclaimer: we are not solving this debate today
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Linguis-c style coordina-on
Social mo1va1on
to gain the other’s social approval [Giles 2008; Street & Giles, 1982] audience design [Clark, 1996; Borjeld & Brennan, 1997]
Hypotheses:
Why does it happen?
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Linguis-c style coordina-on
Social mo1va1on
to gain the other’s social approval [Giles 2008; Street & Giles, 1982] audience design [Clark, 1996; Borjeld & Brennan, 1997]
Unmediated reflex [Pickering and Garrod, 2004; Chartand and Bargh, 1999] deeply embedded in the language-‐genera1on process
Hypotheses:
Why does it happen?
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Linguis-c style coordina-on
Social mo1va1on
to gain the other’s social approval [Giles 2008; Street & Giles, 1982] audience design [Clark, 1996; Borjeld & Brennan, 1997]
Unmediated reflex [Pickering and Garrod, 2004; Chartand and Bargh, 1999] deeply embedded in the language-‐genera1on process
Hypotheses:
Why does it happen?
Does coordina1on occur when the interlocutors have no social mo1va1ons?
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Linguis-c style coordina-on
Why does it happen?
Does coordina1on occur when the interlocutors have no social mo1va1ons?
Proxy ques1on: Does coordina1on occur when the dialog is not generated by the interlocutors (i.e., by the ones gathering social benefits)?
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Linguis-c style coordina-on
Why does it happen?
Does coordina1on occur when the interlocutors have no social mo1va1ons?
Proxy ques1on: Does coordina1on occur when the dialog is not generated by the interlocutors (i.e., by the ones gathering social benefits)?
Case study: scripted movie dialogs
220,000 imagined conversa1ons between 9000 characters
from 617 movie scripts
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Measuring coordina-on
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Measuring coordina-on
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Measuring coordina-on
VINCENT: Antwan probably didn't expect Marsellus to react like he did, but he had to expect a reac1on.
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Measuring coordina-on
JULES: He just massaged her feet, massaging someone's feet is nothing, I massage my mother's feet.
VINCENT: Antwan probably didn't expect Marsellus to react like he did, but he had to expect a reac1on.
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Measuring coordina-on
JULES: It was a foot massage, a foot massage is nothing, I give my mother a foot massage.
JULES: He just massaged her feet, massaging someone's feet is nothing, I massage my mother's feet.
VINCENT: Antwan probably didn't expect Marsellus to react like he did, but he had to expect a reac1on.
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Measuring coordina-on
JULES: It was a foot massage, a foot massage is nothing, I give my mother a foot massage.
VINCENT: Antwan probably didn't expect Marsellus to react like he did, but he had to expect a reac1on.
Matching on ar-cle presence.
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Measuring coordina-on
JULES: It was a foot massage, a foot massage is nothing, I give my mother a foot massage.
VINCENT: Antwan probably didn't expect Marsellus to react like he did, but he had to expect a reac1on.
Matching on ar-cle presence. But matching ≠ coordina-on !
(e.g., chance, Taran-no likes ar-cles, bad guys like ar-cles)
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Measuring coordina-on
What we want: how much Jules’ inclusion of an ar1cle (say) immediately triggers the usage of ar1cles in Vincent’s reply?
What we don’t want: how similar Jules’ style is to Vincent’s style in general or how consistent Taran1no’s style is.
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Measuring coordina-on
What we want: how much Jules’ inclusion of an ar1cle (say) immediately triggers the usage of ar1cles in Vincent’s reply?
Coordina2on(V to J)(art.) =
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Measuring coordina-on
What we want: how much Jules’ inclusion of an ar1cle (say) immediately triggers the usage of ar1cles in Vincent’s reply?
Trigger
Coordina2on(V to J)(art.) = P(Vart.|V replied to J, Jart)
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Measuring coordina-on
What we want: how much Jules’ inclusion of an ar1cle (say) immediately triggers the usage of ar1cles in Vincent’s reply?
Trigger
Control (for inherent similarity)
Coordina2on(V to J)(art.) = P(Vart.|V replied to J, Jart)
-‐ P(Vart.|V replied to J)
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Measuring coordina-on
Overall coordina-on: average over all pairs of characters (B,A)
“Triggered” probability Control probability
P(Bart.|B replied to A, Aart) P(Bart.|B replied to A)
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Measuring coordina-on
Overall coordina-on: average over all pairs of characters (B,A)
Coordina1on
“Triggered” probability Control probability
P(Bart.|B replied to A, Aart) P(Bart.|B replied to A)
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Movie characters coordinate!
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Movie characters coordinate!
“Triggered” probability Control probability
Sta-s-cally significant (p<0.001) effect of coordina-on on all 9 func-on word categories considered
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Movie characters coordinate!
Does coordina1on occur when the interlocutors have no social mo1va1ons?
Proxy ques1on: Does coordina1on occur when the dialog is not generated by the interlocutors (i.e., by the ones gathering social benefits)?
Yes! Scriptwriters produce coordina1on despite not being the
beneficiaries of the presumed social benefits
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
But what if …
… we are only observing the effect of scriptwriters self-‐coordina-ng?
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
But what if …
… we are only observing the effect of scriptwriters self-‐coordina-ng?
(we know people self-‐coordinate [Reifer et al. 2006])
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Author self-‐coordina-on hypothesis
Coordina1on of Vincent to Jules:
Jules Vincent Jules Vincent Jules Vincent
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Author self-‐coordina-on hypothesis
Coordina1on of Vincent to Jules: Self-‐coordina1on of the scriptwriter:
Jules Vincent Jules Vincent Jules Vincent
Taran1no Taran1no Taran1no Taran1no Taran1no Taran1no
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Author self-‐coordina-on hypothesis
Coordina1on of Vincent to Jules: Self-‐coordina1on of the scriptwriter:
Jules Vincent Jules Vincent Jules Vincent
… would be the same or more than:
Taran1no Taran1no Taran1no Taran1no Taran1no Taran1no
Taran1no Taran1no Taran1no Taran1no Taran1no Taran1no
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Author self-‐coordina-on hypothesis
Coordina1on of Vincent to Jules: Self-‐coordina1on of the scriptwriter:
… would be the same or more than:
Jules Vincent Jules Vincent Jules Vincent
Taran1no Taran1no Taran1no Taran1no
Jules Vincent Jules Vincent Jules Vincent
… would be the same or more than the character self-‐coordina1on of J to J:
Taran1no Taran1no
Taran1no Taran1no Taran1no Taran1no Taran1no Taran1no
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
It is not author self-‐coordina-on !
Coordina-on
Jules Vincent Jules Vincent Jules Vincent
Jules Vincent Jules Vincent Jules Vincent
Coordina1o
n
… it’s actually less than the character self-‐coordina-on
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Movie characters vs. real people
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Movie characters vs. real people
Coordina1o
n
Compare amount of coordina-on for each dimension:
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Movie characters vs. real people
Coordina1o
n
Compare amount of coordina-on for each dimension:
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Movie characters vs. real people
coordina1on
coordina-on self-‐coordina-on
Coordina1o
n
Coordina1o
n
Twi7er Movies
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Movie characters vs. real people
coordina1on
Twi7er Movies
coordina-on self-‐coordina-on
Coordina1o
n
Coordina1o
n
Poten1al applica1on in detec1ng faked conversa1ons.
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Effects of social features
In real life, some people coordinate more than others:
e.g.,
effects of gender [Ireland & Pennebacker, 2010, Bilous and Krauss, 1988] effects of social status [Thakerar et al., 1982; Gregory and Webster, 1996]
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Effects of imagined social features?
In real life, some people coordinate more than others:
e.g.,
effects of gender [Ireland & Pennebacker, 2010, Bilous and Krauss, 1988] effects of social status [Thakerar et al., 1982; Gregory and Webster, 1996]
Do such effects carry on when the social features are imagined?
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Effects of imagined social features?
In real life, some people coordinate more than others:
e.g.,
effects of gender [Ireland & Pennebacker, 2010, Bilous and Krauss, 1988] effects of social status [Thakerar et al., 1982; Gregory and Webster, 1996]
Do such effects carry on when the social features are imagined?
Have preconcep1ons about how certain people coordinate become embedded in this reflex?
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Effects of imagined gender
Proxy for the gender of movie characters:
gender of the respec1ve actors
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Effects of imagined gender
Coordina1o
n
to female
to m
ale
of female
of m
ale
overall
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Coordina1o
n
to female
to m
ale
of female
of m
ale
Effects of imagined gender
overall
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Coordina1o
n
to female
to m
ale
by female
by m
ale
Effects of imagined gender
overall
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
male to m
ale
female to m
ale
male to female
Coordina1n
Effects of imagined gender
fem. to fem.
overall
Coordina1o
n
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
fem. to fem.
male to m
ale
female to m
ale
male to female
Effects of imagined gender
overall
Coordina1o
n
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Discovered coordina1on effects in imagined conversa1ons
Insight into the mechanism behind coordina1on
Preliminary evidence regarding preconcep1ons rela1ng coordina1on to various social features
Large-‐scale, metadata rich corpus of imagined conversa1ons
Contribu-ons
publicly available!
Cris1an Danescu-‐Niculescu-‐Mizil and Lillian Lee Chameleons in imagined conversa1ons
Thank you!
Contribu-ons
JULES: That’s probably a good idea.
VINCENT: I think we oughta leave.