Berkeley Linguistics Society. 2013. 152-166. http://dx.doi.org/10.3765/bls.v39i1.

3877
39 Published by the Linguistic Society of America

Language Change as a Social Process: Diffusion Patterns of

Lexical Innovations in Twitter

REBECCA MAYBAUM
University of Haifa

One of the major concerns of sociolinguists is to better understand and explain the
mechanisms driving language change, in particular the process by which
innovative variants appear and subsequently spread throughout a population.
Questions regarding the diffusion of new variants over time have been explored
from a variety of perspectives (most prominently in socio- and historical
linguistics), and a consistent finding is that the diffusion of innovative variants
through the linguistic system forms an S-shaped curve with respect to time
(Labov 2001).
Similar observations are reported from the field of innovation diffusion
research, an interdisciplinary area of the social sciences concerned with how, why,
and at what rate innovative ideas and technologies spread through social systems.
Studies from innovation diffusion research have shown that the rate of diffusion
of (non-linguistic) innovations—including medical, agricultural, political, and
technological examples—also forms an S-shaped curve with respect to time
(Rogers 1995).
The similarities between findings from language change research and
innovation diffusion research suggest that language change may be explained by
the same mechanisms that govern the social diffusion of non-linguistic
innovations. In this paper I apply the theoretical framework of innovation
diffusion research to an instance of language change. By approaching the
diffusion of linguistic innovations as a social process, I hope to gain insights into
the mechanisms of language change from a new perspective.
Section 1 gives a background of the S-curve model of diffusion from both the

152
 Rebecca Maybaum

sociolinguistic and the innovation diffusion perspectives, highlighting similarities

as well as critical differences between the two models. Section 2 introduces the
dataset and gives an overview of the research design. Section 3 presents the
results of the analysis. Section 4 discusses the implications of the results, in
particular how the innovation diffusion model can be used to uncover nuances in
the innovation-decision process of language users. Finally, section 5 outlines
some conclusions.

1 Background: S-Curves of Diffusion

1.1 Innovation Diffusion Research

Sociologists and anthropologists in the 1920s and 1930s were already

investigating the process of innovation diffusion in various domains on the
premise that the spread of new ideas and products had to do with social
connections and informal communication among people in a society (Katz et al.
1963). A major breakthrough came when Ryan and Gross (1943) reported that the
diffusion of innovative agricultural technology formed an S-shaped curve—
following a “slow-quick-slow” pattern—with respect to time. Since then, studies
in medicine (Coleman et al. 1966), family planning (Rogers and Kincaid 1981),
and other areas have provided more evidence that the S-curve pattern of
innovation diffusion is generalizable to all different types of socially diffused
innovations.
Threshold models (Granovetter and Soong 1988; Valente 1996) have been
proposed to explain why diffusion occurs in an S-curve pattern. These models
propose that there is a “tipping point,” both within individuals and the population
as a whole. In the case of an individual, a very low adoption threshold would
mean that the individual would only need superficial exposure to the innovation in
order to adopt it, whereas a high adoption threshold would require many other
members of the individual’s social network to adopt before arriving at his or her
tipping point. At the level of the population, the threshold refers to the critical
mass of overall adoption among the population before the rate of diffusion begins
to accelerate.
A related concept is the innovation-decision process (Rogers 1995), which
recognizes that an individual’s decision to adopt an innovation is not
instantaneous, but progresses through five stages: 1) knowledge, 2) persuasion, 3)
decision, 4) implementation, and 5) confirmation (Rogers 1995). At any point in
the innovation-decision process, the individual may reevaluate his or her previous
behavior and choose to adopt or reject the innovation.
In spite of the evidence showing that the S-curve is the most typical pattern
for innovation diffusion, Rogers (1995) points out that this pattern is only
predicted for successful innovations, while in fact a great many innovations never

153
 Social Diffusion of Lexical Innovations in Twitter

reach a critical mass of adoption and fail to diffuse widely. Furthermore, even in
the case of successful innovations, some diffusion patterns may not form an S-
shaped curve due to specific conditions related to the social system or to the
innovation itself. The exact shape of the diffusion curve must be empirically
determined for each individual innovation; deviations from the prototypical S-
shaped curve may be interpreted based on the idiosyncratic conditions of the
specific innovation and the specific social system.

1.2 Sociolinguistic Research

Sociolinguists have classically studied the process of language change by

correlating linguistic variation with social factors, such as class (Labov 1964;
1994), gender (Ochs 1992; Bucholtz 1999; Cameron and Kulick 2003), age
(Sankoff and Blondeau 2007), ethnicity (Cukor-Avila and Bailey 2001), and
social network structure (Labov 1972; Milroy and Milroy 1985; Eckert 2000;
Paolillo 2001). The progression of linguistic change over time has typically been
analyzed using either the apparent time construct (comparing language use by
speakers of different ages at a single point in time) or real time data (comparing
language use by matched speakers at different points in time). While these two
approaches have indeed spawned advancements in language change theory, the
time gaps between cohorts are often too coarsely grained to allow in-depth
analysis of the precise diffusion pattern over time.
A parallel stream of language change research has developed that sidesteps
the size limitations of the traditional variationist study by employing computer
modeling to try to understand language change on a large scale. Landsbergen and
Lachlan (2004) use a computer simulation to model the historical semantic shift
of English will and Dutch willen. Nettle (1999) used computer simulations
modeled on Social Impact Theory to test various social and network parameters
that may influence the spread and adoption of linguistic changes. And Ke, Gong,
and Wang (2008) draw from and improve upon Nettle’s model by comparing
simulations using different types of network structures. The cumulative body of
sociolinguistic research, from both the computational modeling and simulation
stream and the empirically based real- and apparent time research tradition,
consistently reports an S-shaped curve of linguistic diffusion over time (Bailey
1973; Labov 1994, 2001; Denison 2002; Tagliamonte and D’Arcy 2009).
In spite of the superficial similarities between the sociological and
sociolinguistic S-curve models of diffusion, however, there is a crucial difference
between the two constructs: one measures diffusion through a linguistic system,
while the other measures diffusion through a social system. This difference is
illustrated in (1), which shows idealized S-curves of diffusion in the socio-
historical linguistic traditions (on the left), as well as in the tradition of
sociological diffusion of innovation (on the right).

154
 Rebecca Maybaum

(1) Diffusion through a linguistic system (left) and social system (right)

In both graphs the x-axis indicates time; the y-axes, however, represent
different measures. On the left, the y-axis represents all occurrences of a particular
linguistic context y—a context in which evidence of variation of some sort has
been identified, and which is the suspected locus of a change in progress. Each
point in the curve represents the percentage of all instances of context y that are
realized as variant z, at each point in time (for instance, the percentage of ne
deletion in negation contexts in Montreal French).
On the right, the y-axis represents the population of potential adopters—that is,
individuals within the community who could conceivably be exposed to the
innovation and might eventually adopt it themselves. Each point in the curve
represents the percentage of the total potential adopter population who have
adopted the innovation at each point in time.
The distinct y-axis labels in both graphs mean that the S-curves of diffusion
discussed in sociolinguistics and innovation diffusion in fact measure two entirely
different concepts. Socio- and historical linguists too often fail to make this
distinction, referring to the increase of a linguistic variant relative to the text data
as social diffusion, when in fact social diffusion is measured by the proportion of
individual language users who adopt the variant.

2 Data and Research Design

2.1 Twitter Data

The study is based on a 19-million-word corpus collected from the online

microblogging service Twitter, consisting of all Twitter posts containing the
following community-specific lexical innovations: tweeps, tweeties, tweeple,
tweethearts, tweople, twerps, tweetheads, twitterbugs, tweebs, and twittertwatters.
These variants, henceforth referred to as Twitter People variants, all share the
approximate meaning “twitter friends.” For example, tweople combines “Twitter”
and “people” (2), tweeps comes from “Twitter peeps” (3), and “Twitter
sweethearts” becomes tweethearts (4):

155
 Social Diffusion of Lexical Innovations in Twitter

(2) How many Tweople got hair cut today?

(3) For all you working tweeps out there...apparently tweeting at work is a
good thing

(4) Time for bed - busy day tomorrow. Goodnight Tweethearts! Thanks for
the fun and tweet dreams:)

One advantage of using Twitter data to investigate language change stems

from the existence of automatically collected digital archives, containing
complete records of past Twitter posts (tweets). In most cases, sociolinguistic
researchers are able to identify a change in progress only in advanced stages of
the diffusion process. Yet using archived collections of historical Twitter data, it
is possible to identify variants of interest at later stages of the change in progress,
and then subsequently trace the diffusion trajectory backward in time to the
earliest appearances of the innovative variants. In this way, the Twitter dataset
collected for the present study provides a rare opportunity to examine the early
stages of the innovation and diffusion process.
The use of Twitter as a data source also avoids some of the practical
limitations encountered by previous research into the diffusion of linguistic
innovations over time. Most of the sociolinguistic studies investigating the spread
of innovative variants are based on empirical, real-world language data collected
from a relatively small number of speakers (e.g. Milroy and Milroy 1985; Eckert
2000), requiring time-consuming methodologies that effectively limit the overall
sample size of the study. Another thread of research has begun using computer
models to simulate large-scale diffusion of linguistic innovations (Nettle 1999; Ke
et al. 2008), utilizing powerful statistical tools to gain insight into aspects of
language change that are undetectable on a small scale. The Twitter data
capitalize on the advantages of both streams of research described above,
combining empirical language data with large quantities of time-stamped data.

2.2 Research Questions

The paper addresses the following research questions: 1) Do the Twitter People
variants follow an S-curve pattern of social diffusion? If so, that would suggest
that language change shares characteristics with other kinds of social processes,
and may be governed by the same mechanisms that shape the social diffusion
patterns of non-linguistic innovations. 2) Are the diffusion patterns consistent
across all Twitter People variants? If not, are these differences correlated with
other factors, such as the relative success/failure of the variant (as measured by
overall frequency)? 3) How many times must a user post a Twitter People variant
to be considered an adopter? And, 4) is the shape of the diffusion curve affected

156
 Rebecca Maybaum

by the criteria used to define adoption? (See below for discussion of adopter
criteria.)

2.3 Corpus Overview

The Twitter People Corpus consists of all Twitter posts that contained any of the
Twitter People variants listed previously, posted from 2006 through 2011. The
data collected includes the full text of the tweets, the user name of the author, a
timestamp of the publication accurate to the second, and a variety of additional
metadata. Table (5) displays the word count, number of tweets, and number of
individual users for each Twitter People variant subcorpus.

(5) Twitter People variants—word count, tweet count, and user count

In total the corpus contains close to 19 million words. By far the most popular
and most widespread variant of the group is tweeps, appearing in more than 800
thousand tweets. At the other extreme is twittertwatters, appearing just 16 times
throughout the entire time period represented by the corpus.

2.4 Time of Adoption

One possible way to represent the rate of diffusion would be to calculate the raw
frequencies of each keyword based on the total number of occurrences in the
corpus. However, the diffusion rate of a socially-diffused innovation is more
often—and more usefully—measured based on the time at which individuals
adopt the innovative term, with no regard to the times of subsequent productions.
For the individual Twitter People variant diffusion graphs, time of adoption is
defined by the Unix timestamp that corresponds to the earliest tweet of each
unique user in each variant subcorpus.

157
 Social Diffusion of Lexical Innovations in Twitter

2.5 Accounting for Twitter Population Growth

Traditional mathematical models of innovation diffusion have generally assumed

a stable overall population size. This assumption, while convenient conceptually
and mathematically, often fails to represent the reality (see Mahajan and Peterson
1978). In the current example, the Twitter population is far from constant; on the
contrary, it has increased—and continued to increase—over time, as more and
more people continue to sign up for new Twitter accounts.
Twitter has not published official records of its population increase over time,
although numerous external observers have calculated the approximate population
increased based on a variety of secondary information. In this investigation I used
Twitter population data calculated by Pelzer (2012). The data were published in
graphical form, but the raw population numbers are not publicly available. I
converted the visual Twitter population data into numerical coordinates to
reproduce the corresponding population data.
Figure (6) shows the total Twitter population increase in monthly increments,
from January 2007 through December 2012 (reanalyzed from Pelzer 2012
visualization). The most dramatic increase in population growth occurred in
February 2009, though there is also a noticeable increase in March 2007. The
population growth continued to increase through the end of 2011.

(6) Twitter Population over Time, 2007–2011

In order to account for this growing potential adopter population in Twitter, I

calculated proportional frequency of adopters over time, rather than using the raw
frequency of individual adoption. Proportional adoption rates were calculated by
dividing the raw frequency of adopters in each Twitter People variant subcorpus
by the number of overall Twitter users, for each point in time along the x-axis.

158
 Rebecca Maybaum

The resulting value represents proportional adoption rates of each variant adjusted
for the simultaneous increase in the total population of potential adopters.

2.6 Adopter Criteria Conditions

The question of how to define adoption in the context of Twitter People variants
in Twitter is one that must be carefully considered, as it may have significant
consequences for the analysis itself and for the interpretations of the results. The
simplest and most straightforward definition would be to consider any single use
of the variant in question as adoption. However, it is possible that some users
adopted a Twitter People variant on a trial basis (as part of the innovation-
decision process) process before subsequently rejecting it. In this case, a single
post containing the variant would not constitute final adoption. Two adopter
criteria conditions, based on number of posts per user (all users vs. multiple-post
users only), are assessed in the diffusion analysis.

3 Results

The results of the analysis are displayed using graphical representations of the
adoption/diffusion patterns of each Twitter People variant, as well as the entire
group of Twitter People terms combined, over the time period represented by the
data.
Although the data collection spanned the time range from March 2006
through January 1, 2012, none of the variants appeared prior to 2007. Because of
this, all of the diffusion graph results are presented with an x-axis time scale of
January 2007 until January 2012. The y-axis range varies according to the overall
frequency of each variant for best visual comparison of the overall trajectories of
the diffusion curves.

3.1 Overall Diffusion Patterns

When looking at the diffusion patterns for the Twitter People variants based on all
users, two common patterns emerge. The first pattern resembles the prototypical
S-curve predicted by both the sociolinguistic and the innovation diffusion
literature. Examples of this pattern are shown in (8). The theoretical S-curve
model measures cumulative frequency of the innovation over time, which would
mean that the y-value can never decrease over time. However, because the
diffusion curves of this study are based on proportional frequency relative to a
simultaneously increasing Twitter population, it is possible for the number of
potential adopters to increase more rapidly over time than the number of
cumulative adopters, as seen in (7).

159
 Social Diffusion of Lexical Innovations in Twitter

(7) S-Curve Diffusion Pattern (All Users)

Tweeple!–!adop2on!over!2me!(all!users)& Twee$es!–!adop2on!over!2me!(all!users)&
0.018%$ 0.030%$
(as!%!of!total!Twi0er!popula2on)&

(as!%!of!total!Twi0er!popula2on)&
0.016%$
0.025%$
0.014%$
0.012%$ 0.020%$

Twee$es!users!
Tweeple!users!

0.010%$
0.015%$
0.008%$
0.006%$ 0.010%$
0.004%$
0.005%$
0.002%$
0.000%$ 0.000%$

Jan,07$
May,07$
Sep,07$
Jan,08$
May,08$
Sep,08$
Jan,09$
May,09$
Sep,09$
Jan,10$
May,10$
Sep,10$
Jan,11$
May,11$
Sep,11$
Jan,12$
Jan-07$
May-07$
Sep-07$
Jan-08$
May-08$
Sep-08$
Jan-09$
May-09$
Sep-09$

May-10$
Sep-10$
Jan-11$
May-11$
Sep-11$
Jan-10$

Jan-12$
Examples of non-S-curve diffusion patterns are shown in (8). These diffusion
curves are stepwise or near-linear in pattern, and are characterized by continuous
increase over time, in some cases interspersed with periods of stable proportional
frequency.

(8) Non–S-Curve Diffusion Pattern (All Users)

Tweeps!–!adop2on!over!2me!(all!users)! Tweethearts!–!adop2on!over!2me!(all!users))
0.08%$ 0.008%$
(as!%!of!total!Twi0er!popula2on)!

(as!%!of!total!Twi0er!popula2on))

0.07%$ 0.007%$
0.06%$ 0.006%$
Tweethearts!users!

0.05%$ 0.005%$
Tweeps!users!

0.04%$ 0.004%$
0.03%$ 0.003%$
0.02%$ 0.002%$
0.01%$ 0.001%$
0.00%$ 0.000%$
Jan007$
May007$
Sep007$
Jan008$
May008$
Sep008$
Jan009$
May009$
Sep009$
Jan010$

Sep010$
Jan011$
May011$
Sep011$
Jan012$
Jan007$
May007$
Sep007$
Jan008$
May008$
Sep008$
Jan009$
May009$
Sep009$
Jan010$

Sep010$
Jan011$
May011$
Sep011$
Jan012$
May010$

May010$

Table (9) shows the distribution of diffusion patterns for all variants.

(9) S-curve Summary of Twitter People Variants (All Users)

160
 Rebecca Maybaum

3.2 Adopter Criteria

As discussed in Section 2, the criteria used to determine whether a Twitter user is,
in fact, an adopter of the Twitter People variants may be adjusted based on the
total number of posts per user. Over 70 percent of users posted only once, while
less than 14 percent posted three or more times.
Figure (10) shows the proportion of adopter type (based on total posts per
user) in each Twitter People variant subcorpus, arranged in order of overall
frequency from left to right. The graph shows a positive correlation between
overall frequency of the variant (indicating relative success of diffusion) and the
percentage of users with multiple posts (2+ and 3+ post users), and a negative
correlation between overall frequency of the variant and the percentage of users
who posted only a single time throughout the period represented in the corpus.

(10) User Population by Adopter Criteria

3.3 Diffusion Patterns—Users with Multiple Posts

When the adopter criteria are limited to include only those users who posted
multiple times (2+ and 3+ users), the resulting diffusion curves are altered. More
of the Twitter People variants exhibited S-shaped diffusion curves under the
multiple-posts-per-user condition than in the all-users condition. Figure (11)
shows the diffusion curves for tweeps and tweethearts for users with three or
more posts each. The same variants that did not produce S-shaped diffusion
curves in the all-user filter (8) now follow the “slow-quick-slow” S-curves under
the multiple-posts-per-user filter.

161
 Social Diffusion of Lexical Innovations in Twitter

(11) S-curve Pattern (3+ Post Users)

Tweeps!–!adop2on!over!2me!(3+!post!users)! Tweethearts!–!adop2on!over!2me!(3+!post!users))
0.02%$ 0.001%$
(as!%!of!total!Twi0er!popula2on)!

(as!%!of!total!Twi0er!popula2on))
0.01%$ 0.001%$
0.01%$ 0.001%$

Tweethearts!users!
0.01%$
Tweeps!users!

0.000%$
0.01%$
0.000%$
0.01%$
0.00%$ 0.000%$
0.00%$ 0.000%$
0.00%$ 0.000%$
Jan*07$
May*07$
Sep*07$
Jan*08$
May*08$
Sep*08$
Jan*09$
May*09$
Sep*09$
Jan*10$

Sep*10$
Jan*11$
May*11$
Sep*11$
Jan*12$

Jan)07$
May)07$
Sep)07$
Jan)08$
May)08$
Sep)08$
Jan)09$
May)09$
Sep)09$
Jan)10$

Sep)10$
Jan)11$
May)11$
Sep)11$
Jan)12$
May*10$

May)10$
Figure (12) summarizes the diffusion patterns for Twitter People variants in
the multiple-posts-per-user filter. (Twittertwatters was excluded because there are
not enough data points for multiple-post users to produce a diffusion curve for
that variant.) The only variant that does not follow an S-curve is tweetheads.

(12) Diffusion Pattern Summary, Multiple-post Users Only

4 Discussion

In this section I discuss the significance of the main findings for the diffusion
analysis, beginning with the S-shaped diffusion patterns for the all-user adopter
criteria exemplified in (7) and (8), and summarized in (9). Although Rogers
(1995) claims that the S-curve pattern occurs only in cases of successful diffusion,
the results show an even distribution of S-curve versus non–S-curve diffusion
patterns across the range of frequencies of the variants. I found no significant
difference between the likelihood of a popular slang term (e.g. tweeps) vs. an
unpopular slang term (e.g. twittertwatters) to diffuse in an S-shaped pattern.
While it is not a trivial finding that five out of ten of the Twitter People
variants follow the S-curve pattern of diffusion—this at least partially confirms
the hypothesis that language change may diffuse socially via the same
mechanisms as non-linguistic innovations—neither is it overwhelmingly

162
 Rebecca Maybaum

conclusive. With the introduction of the varying adopter criteria, however, the
results become much more telling.
The clear majority of all Twitter users in the corpus only authored a single
post employing the Twitter People variant. In other words, most of the Twitter
users tried out the new slang term once, but never fully integrated it into their
permanent lexicon. This raises the question of the degree of perceived
trialability—the ability to try out an innovation on a trial basis without making a
commitment—of Twitter People variants. This allows the individual to judge the
merits and/or consequences of the innovation under real conditions. In this case, a
Twitter user can try out one of the innovative Twitter People variants one time
with a minimum of risk or inconvenience. The attribute of trialability is positively
correlated with rate of adoption (Rogers 1995), meaning the Twitter People
variants (and likely for the same reasons, other innovations within Twitter and on
the Internet as a whole) are predicted to diffuse rapidly.
The relationship found between the number of posts per user and overall
frequency of the variant (10) also supports the interpretation that single-post users
were engaging in a trial period before deciding whether or not to adopt the
innovation. The more popular variants retained more adopters after the trial period
than did the less popular variants, thus the more successful variant subcorpora
have a higher proportion of repeat posters than the less successful variants.
The innovation-diffusion process (Rogers 1995), briefly described in Section
1, conceptualizes the act of adoption as a five-stage process. The first stage is
knowledge, or exposure to the innovation, followed by persuasion, when the
individual forms an initial favorable or unfavorable attitude toward the innovation.
The third stage is the decision stage, and it is here that the trialability of an
innovation comes into play. The first time a Twitter user tries out a Twitter People
variant (or any other innovative element), he or she is engaged in a decision-stage
activity with the purpose of informing the decision to adopt or reject the
innovation. If at this point the individual decides to adopt the innovation, this
stage is followed by implementation (with possible re-invention) and
confirmation. Rejection can occur at any stage in the innovation-diffusion process.
The minimum requirement for an individual to be considered an adopter is the
implementation (post-trial) stage; in the Twitter People data, this can be defined
as a user’s second post using the same Twitter People variant. We can be even
more confident of the adopter classification, however, if the individual has
advanced through the confirmation stage—signaled by a user’s third post.
Following this model, single-post users should not be considered adopters.
A comparison of (8) and (11) illustrates the effect of altering the adopter
criteria to exclude non-adopter single-post trial users from the diffusion data. The
most dramatic transformation occurred for the most successfully diffused variant
in the corpus, tweeps. The summary of diffusion patterns for Twitter People
variants using the multiple-posts-per-user adopter criteria (12) reveals that all but

163
 Social Diffusion of Lexical Innovations in Twitter

one variant—tweetheads—followed the S-curve “slow-quick-slow” pattern of

diffusion. A look inside the tweetheads corpus quickly uncovers the reason for its
exception: the “Twitter People” meaning of tweethead is overshadowed by the
posts and references to public persona @tweethead. As such, it is to be expected
that the tweetheads corpus not follow the typical pattern of innovation diffusion,
since the primary context of tweethead is not as an innovative lexical item but
rather as a personal and/or shared culture referent.

5 Conclusion

The results of the Twitter People diffusion analysis lend support to the view that
language change is a socially driven process, and can be successfully analyzed
using methods and theoretical frameworks from social science disciplines beyond
linguistics. While some of the details of the Twitter People analysis varied from
specific assumptions of the innovation diffusion theoretical framework (for
instance, the failed Twitter People variants were as likely as successful ones to
follow an S-curve pattern of diffusion), the major tendencies found across
innovation diffusion studies held true for the Twitter People variants. The
established concepts of the innovation-decision process and innovation attributes
(in particular the notion of trialability) also provided a cohesive framework and
valuable explanations for interpreting the results.
Applying classic innovation diffusion research methods to the study of
language change gives sociolinguists a powerful tool for verifying and
interpreting the results of both theoretical simulations of large-scale linguistic
diffusion and in-depth empirical research investigating real-world language
change on a smaller scale. Although some quantitative methods were used, this
has remained essentially a qualitative study of diffusion over time. In the future, a
fully quantitative research design may be able to more precisely compare the
diffusion patterns than was possible here. The intersection between innovation
diffusion and language change is a relatively unexplored area, which will benefit
from the analysis of new data sources, as well as further theoretical development.
As a whole, this study represents the successful application of a new (to
linguistics) approach that can add another dimension to the study of the
mechanisms of language change as a social process.

References

Bailey, Charles-James N. 1973. Variation and Linguistic Theory. Washington,

DC: Center for Applied Linguistics.
Bucholtz, Mary. 1999. “Why be Normal?”: Language and Identity Practices in a
Community of Nerd Girls. Language in Society, 28(2):203–223.

164
 Rebecca Maybaum

Cameron, Deborah, and Don Kulick. 2003. Language and Sexuality. Cambridge:
Cambridge University Press.
Coleman, James S., Elihu Katz, and Herbert Menzel. 1966. Medical Innovation: A
Diffusion Study. Indianapolis: Bobbs-Merrill Co.
Cukor-Avila, Patricia, and Guy Bailey. 2001. The Effects of the Race of the
Interviewer on Sociolinguistic Fieldwork. Journal of Sociolinguistics,
5(2):252–270.
Denison, David. 2002. Log(ist)ic and Simplistic S-curves. In R. Hickey, ed.,
Motives for Language Change, 54–70, Cambridge: Cambridge University Press.
Eckert, Penelope. 2000. Linguistic Variation as Social Practice: The Linguistic
Construction of Identity in Belten High. Oxford: Wiley-Blackwell.
Granovetter, Mark, and Roland Soong. 1988. Threshold Models of Diversity:
Chinese Restaurants, Residential Segregation, and the Spiral of Silence. In C. C.
Clogg, ed., Sociological Methodology, 69–104, Washington, DC: American
Sociological Association.
Katz, Elihu, Martin L. Levin, and Herbert Hamilton. 1963. Traditions of Research
on the Diffusion of Innovation. American Sociological Review, 28(2):237–252.
Ke, Jinyun, Tao Gong, and William S.-Y. Wang. 2008. Language Change and
Social Networks. Communications in Computational Physics, 3(4):935–949.
Labov, William. 1964. Phonological Correlates of Social Stratification. American
Anthropologist, New Series, 66(6):164–176.
Labov, William. 1972. Sociolinguistic Patterns. Philadelphia: University of
Pennsylvania Press.
Labov, William. 1994. Principles of Linguistic Change, vol 1: Internal Factors.
Oxford: Wiley-Blackwell.
Labov, William. 2001. Principles of Linguistic Change, vol 2: Social Factors.
Oxford: Blackwell.
Landsbergen, Frank, and Robert F. Lachlan. 2004. A Cultural-Evolutionary
Perspective on Semantic Change: The Role of Non-Social Factors in the Spread
of Innovations. In Proceedings of the Sixth Annual High Desert Linguistics
Society Conference, 6:47–58. Presented at the High Desert Linguistics Society
Conference, Albuquerque, New Mexico.
Mahajan, Vijay, and Robert A. Peterson. 1978. Innovation Diffusion in a
Dynamic Potential Adopter Population. Management Science, 24(15):1589–
1597.
Milroy, James, and Leslie Milroy. 1985. Linguistic Change, Social Network and
Speaker Innovation. Journal of Linguistics, 21(02):339–384.
Nettle, Daniel. 1999. Using Social Impact Theory to Simulate Language Change.
Lingua, 108:95–117.
Ochs, Elinor. 1992. Indexing Gender. In A. Duranti and C. Goodwin, eds.,
Rethinking Context: Language as an Interactive Phenomenon, 335–358,
Cambridge: Cambridge University Press.

165
 Social Diffusion of Lexical Innovations in Twitter

Paolillo, John C. 2001. Language Variation on Internet Relay Chat: A Social

Network Approach. Journal of Sociolinguistics, 5(2):180–213.
Pelzer, Adriaan. 2012, February 20. Twitter Will Break Through the 500 Million
User Mark on Wednesday. Message posted to http://wewillraakyou.com/ .
Rogers, Everett M. 1995. Diffusion of Innovations. New York: Simon and
Schuster.
Rogers, Everett M., and D. Lawrence Kincaid. 1981. Communication Networks:
Toward a New Paradigm for Research. New York: The Free Press.
Ryan, Bryce, and Neal C. Gross. 1943. The Diffusion of Hybrid Seed Corn in
Two Iowa Communities. Rural Sociology, 8(1):15–24.
Sankoff, Gillian, and Hélène Blondeau. 2007. Language Change across the
Lifespan: /r/ in Montreal French. Language, 83(3):560–588.
Tagliamonte, Sali. A., and Alexandra D’Arcy. 2009. Peaks Beyond Phonology:
Adolescence, Incrementation, and Language Change. Language, 85(1):58–108.
Valente, Thomas. W. 1996. Social Network Thresholds in the Diffusion of
Innovations. Social Networks, 18(1):69–89.

Rebecca Maybaum
Department of English Language & Literature
University of Haifa
Mount Carmel
199 Aba-Hushi Avenue
Haifa 31905

[email protected]

166