On Tue, Jan 10, 2012 at 10:12 PM, Gwern Branwen <gwern0(a)gmail.com> wrote:
https://www.technologyreview.com/blog/arxiv/27437/
discussing
http://arxiv.org/abs/1112.3670 "Echoes of power: Language effects and
power differences in social interaction", abstract:
Understanding social interaction within
groups is key to analyzing online communities. Most current work focuses on structural
properties: who talks to whom, and how such interactions form larger network structures.
The interactions themselves, however, generally take place in the form of natural language
--- either spoken or written --- and one could reasonably suppose that signals manifested
in language might also provide information about roles, status, and other aspects of the
group's dynamics. To date, however, finding such domain-independent language-based
signals has been a challenge.
Here, we show that in group discussions power differentials between participants are
subtly revealed by how much one individual immediately echoes the linguistic style of the
person they are responding to. Starting from this observation, we propose an analysis
framework based on linguistic coordination that can be used to shed light on power
relationships and that works consistently across multiple types of power --- including a
more "static" form of power based on status differences, and a more
"situational" form of power in which one individual experiences a type of
dependence on another. Using this framework, we study how conversational behavior can
reveal power relationships in two very different settings: discussions among Wikipedians
and arguments before the U.S. Supreme Court.
From the paper proper:
Status change. Wikipedians can be promoted to
administrator status through a public election, and almost always after extensive prior
involvement in the community. Since we track the communications of editors over time, we
can examine how linguistic coordination behavior changes when a Wikipedian becomes an
“admin”. To our knowledge, our study is the first to analyze the effects of status change
on specific forms of language use.
Users are promoted to admins through a
transparent election process known as requests for adminship4 , or RfAs, where the
community decides who will become admins. Since RfAs are well documented and timestamped,
not only do we have the current status of editors, we can also extract the exact time when
editors underwent role changes from non-admins to admins.
Textual exchanges. Editors on Wikipedia interact on talk pages5 to discuss changes to
article or project pages. We gathered 240,436 conversational exchanges carried out on the
talk pages, where the participants of these (asynchonous) discussions were associated with
rich status and social interaction information: status, timestamp of status change if
there is one, as well as activity level on talk pages, which can serve as a proxy of their
sociability, or how socially inclined they are. In addition, there is a discussion phase
during RfAs, where users “give their opinions, ask questions, and make comments” over an
open nomination. Candidates can reply to existing posts during this time. We also
extracted conversations that occurred in RfA discussions, and obtained a total of 32,000
conversational exchanges. Most of our experiments were carried out on the larger dataset
extracted from talk pages, unless otherwise noted. (The dataset will be distributed
publicly.)
We measure the linguistic style of a person by
their usage of function words that have little lexical meaning, thereby marking style
rather than content. For consistency with prior work, we employed the nine LIWC-derived
categories [36] deemed to be processed by humans in a generally non-conscious fashion
[25]. The nine categories are: articles, auxiliary verbs, conjunctions, high-frequency
adverbs, impersonal pronouns, negations, personal pronouns, prepositions, and quanti-
fiers (451 lexemes total).
Results, starting page 5:
...communication behavior on Wikipedia provides
evidence for hypothesis Ptarget : users coordinate more toward the (higher-powered) admins
than toward the non-admins (Figure 1(a)12 ).
In the other direction, however, when comparing admins and non-admins as speakers, the
data provides evidence that is initially at odds with Pspeaker : as illustrated in Figure
1(b), admins coordinate to other people more than non-admins do (while the hypothesis
predicted that they would coordinate less).13 We now explore some of the subtleties
underlying this result, showing how it arises as a superposition of two effects.
One possible explanations for the inconsistency
of our observations with Pspeaker is the effect of personal characteristics suggested in
Hypothesis B from Section 2. Specifically, admin status was not conferred arbitrarily on a
set of users; rather, admins are those people who sought out this higher status and
succeeded in achieving it. It is thus natural to suppose that, as a group, they may have
distinguishing individual traits that are reflected in their level of language
coordination.
...to investigate whether the effects observed in Figure 1(b) are purely tied to status,
we look at communication differences between these same two populations over time periods
when there was no status difference between them: we compare the set of admins-to-be —
future admins before they were promoted via their RfA — with non-admins. Figure 2(a) shows
that the same differences in language coordination were already present in these two
populations — hence, they are not an effect of status alone, since they were visible
before the former population ever achieved its increase in status.
One way to separate the second issue from the
first is to look at differences in coordination between users who were promoted
(admins-to-be), and those who went through the RfA process but were denied admin status
(failed-to-be). Both admins-to-be and failed-to-be had the ambition to become admins, but
only members of the former group were successful. We investigate coordination differnces
between these two groups during a period when their adminship ambitions are arguably most
salient: during the discussions in each user’s own RfA process. Figure 2(b) shows that
even in the conversations they had on their RfA pages, the admins-to-be were coordinating
more to the others than the failed-to-be, providing evidence for a strong form of
Hypothesis B.
... it is interesting to note that the most dramatic change in coordination is visible in
the second month after the change in status occurred. This suggests a period of
accommodation to the newly gained status, both for the person that undergoes the change
and for those witnessing it.
To study Pspeaker, we create two populations for
comparison: the interactions of each admin before his or her promotion via RfA (i.e., when
they were admins-to-be), and the interactions of each admin after his or her respective
promotion. Figure 3(a) shows how the resulting comparison confirms Pspeaker : admins-to-be
decrease their level of coordination once they gain power.14 Interestingly, the reverse
seems to be true for failed-to-be: after failing in their RfAs — an event that arguably
reinforces their failure to achieve high status in the community — they coordinate more
(p-value 0.05; we omit the figure due to space limitations.)
So, suck-ups tend to pass RfA more often than those who don't suck up
to whom they are talking to. An interesting analysis, altogether.
--
gwern
http://www.gwern.net
Methodology and analysis leaves a lot to be desired and doesn't really
support either their conclusion or your bolder restatement of it.