New http://nms.sagepub.

com/
Media & Society

Ambient affiliation: A linguistic perspective on Twitter

Michele Zappavigna
New Media Society 2011 13: 788 originally published online 31 May 2011
DOI: 10.1177/1461444810385097
The online version of this article can be found at:
http://nms.sagepub.com/content/13/5/788

Published by:
http://www.sagepublications.com

Additional services and information for New Media & Society can be found at:
Email Alerts: http://nms.sagepub.com/cgi/alerts
Subscriptions: http://nms.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations: http://nms.sagepub.com/content/13/5/788.refs.html

>> Version of Record - Jul 29, 2011

OnlineFirst Version of Record - May 31, 2011
What is This?

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

Article

Ambient affiliation:
A linguistic perspective
on Twitter

new media & society

13(5) 788806
The Author(s) 2011
Reprints and permission:
sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/1461444810385097
nms.sagepub.com

Michele Zappavigna
University of Sydney, Australia

Abstract
This article explores how language is used to build community with the microblogging
service, Twitter (www.twitter.com). Systemic Functional Linguistic (SFL), a theory of
language use in its social context, is employed to analyse the structure and meaning of
tweets (posts to Twitter) in a corpus of 45,000 tweets collected in the 24 hours after
the announcement of Barak Obamas victory in the 2008 US presidential elections.
This analysis examines the evaluative language used to affiliate in tweets. The article
shows how a typographic convention, the hashtag, has extended its meaning potential
to operate as a linguistic marker referencing the target of evaluation in a tweet (e.g.
#Obama). This both renders the language searchable and is used to upscale the call to
affiliate with values expressed in the tweet. We are currently witnessing a cultural shift
in electronic discourse from online conversation to such searchable talk.
Keywords
discourse analysis, social networking sites, systemic functional linguistics, Twitter

Orientation: language and social networking sites

(SNS) such as Twitter
Electronic conversation lends itself to search in a way that face-to-face conversation,
currently, does not. Search engine usage has become a ubiquitous process for locating
information. It is, however, only as we have begun to leave more traces of social

Corresponding author:
Michele Zappavigna, Department of Linguistics, F12, University of Sydney, NSW 2006, Australia
Email: [email protected]

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

789

Zappavigna

interaction online, via microblogging services such as Twitter, that a cultural shift
toward a more interpersonal function for search has emerged. This is a shift from
searching purely for content, to searching what other people are saying online and
forming communities of shared value. In popular terms, it is becoming increasingly
useful to search the hive mind: the stream of online conversation occurring across
semiotic modes (e.g. blogs, online chat and social networking sites). For example, the
kind of discourse-search that Twitter affords has been described as a rival to Google,
with commentators claiming that searching Twitter may soon be one of the most effective ways to gather useful information, since returns capture what users are saying
online in real-time (Rocketboom, 2009).
This cultural shift to interpersonal search has resulted in the emergence of searchable
talk, that is, online discourse where the primary function appears to be affiliation via
findability. This kind of talk expands linguistic meaning potential by using punctuation
to incorporate metadata into language so that online talk can be found. Taking a corpus
of posts to Twitter as a case study, this article aims to suggest how search is beginning to
function as a community-building linguistic activity. It will demonstrate how what are
known as hashtags function as linguistic markers enacting the following social relation:
Search for me and affiliate with my value!.
By enabling users to affiliate online, social networking sites (SNS),1 accessed by millions worldwide, afford a new form of sociality in which language maintains a pivotal
role. While studies of online discourse from a linguistic perspective are relatively established (Baron, 2008; Crystal, 2006; Herring, 1996), whether analysis of linguistic function and structure can serve as evidence for defining communities is an emergent area of
inquiry. This is not to say that there has been little work on language in online communities, but rather that there is yet to be an accumulation of research providing linguistic
models of online, and indeed offline, affiliation. The notion of online community was
popularized in Rheingolds (1993) work on virtual community: virtual communities
are social aggregations that emerge from the Net when enough people carry on those
public discussions long enough, with sufficient human feeling, to form webs of personal
relationships in cyberspace (Rheingold, 1993: 5).
Since the emergence of this definition, often criticized for its vagueness, there has
been a debate surrounding the criteria for establishing the bounds of online communities,
the structure of community and how communities are built or emerge (see for example
Hagel and Armstrong, 1997; Jones, 1997; Burnett, 2000; Wellman, 2001; Herring, 2004,
2008). No stable definition of community has prevailed. However, a linguistic perspective on virtual community might aim to explicitly describe how people use language to
construe social bonds and how they rally around, defer or reject different values construed in language (Knight, 2008). This type of work is emerging as linguists begin to
expand their territory into different modes of communication, such as image, gesture and
music, viewing these modes either as forms of semiosis that are parasitic on language
(Halliday and Matthiessen, 1999) or as themselves having a grammar that can be analysed (Kress and Van Leeuwen, 2006).
This article begins by introducing Twitter and a corpus of tweets (messages posted
to Twitter) containing the keyword Obama collected in the 24 hours after the announcement that Barak Obama had won the 2008 US presidential election. It then introduces
Systemic Functional Linguistics (SFL), the theory of language used in the study. SFL is

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

790

new media & society 13(5)

a social semiotic theory that investigates discourse in context. The Appraisal framework (Martin and White, 2005), a model of evaluative language using the SFL approach,
is then detailed. The aim of the appraisal analysis undertaken in this article is to show
the kinds of interpersonal meanings made in the corpus of tweets and their role in ambient affiliation.

Constrained meaning-making on Twitter

Twitter, developed in 2006, is an example of a microblogging service. These services
allow users to post character-constrained messages via a range of technologies such as
mobile phone, instant messaging clients and the web. Tweets, messages posted to Twitter,
are messages presented to a virtual audience who follow by subscribing to another
users feed and, as the service has evolved, who search. Tweets unfold in time as a twitter stream that is presented in reverse chronological order on a users Twitter page and
also, unless privacy selections are adjusted by the user, as part of the public feed. A tweet,
in reference to its original conception as an SMS-based message, is constrained to no
more than 140 characters in length and may incorporate links to micromedia, small-scale
multimedia, and shortened aliases of longer hyperlinks (Tiny URLs). As such, they are
interesting cases in making meaning within constrained environments.
In contrast to other forms of communication, there is no communal expectation that
anyone respond to a tweet, as the metaphor of twittering continuously like a bird
implies. There is, however, a social need among users to engage with other voices in public and private feeds. Hence we see creative use of punctuation to reference other users
and tag common topics. These expansions in typographic meaning potential are part of a
community-driven movement toward Twitter becoming a form of public conversation. It
is conversation, however, that is multiparty, temporarily fluid and highly intertextual.
As Twitter has evolved, so have the resources for attributing and addressing other
users. Linguistic markers have begun to populate tweets to facilitate heteroglossia
(Bakhtin, 1981), in other words, to bring other voices into tweets by addressing other
users, republishing other tweets, and flagging topics that may be adopted by multiple
users. The first of these conventions is the @ character, used as a deictic marker as in
the following example:
@username I didnt vote Obama. But I wasnt exactly for McCain either. I hate both, to be frank.

The @ character indicates that the username which follows it is addressed in the
tweet and the structure functions like a vocative, that is, as a form of address. The @
character does not have to operate in this initial position in a clause but can also
occupy a medial or final position. In these instances it is more likely to mark a user,
flagging that they are being referred to, but not explicitly inscribing an address. For
example, in the following tweet, the user is not directly addressing username1
although the use of the @ character means that this user, and users who follow him,
are likely to see the message in their feed:
Im joining @username1 in his four commitments to Pres.-elect Obama. Will you? http://
tinyurl.com/579akg.

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

791

Zappavigna

Due to its electronic affordances, this form of punctuation renders the deictic marker
aggregatable and searchable. So, for example, a user may track all @ references to
themselves or to a user that they are following. Another way of bringing external voices
into a tweet is to republish part or all of another tweet intact or modified by retweeting,
using the character combination, RT. In most instances the RT will be followed by the
@ character to source the retweet:
RT @username2 Absolutely bril http://tinyurl.com/2upsz4 Congratulations Obama!

In this way the RT functions as a form of engagement realized by grammatical

projection, in other words, it functions to indicate that the clause following @username2 is a quotation, most often a direct quotation of that users talk.

Hashtags on Twitter
The hashtag2 (#) has a different function to RT and @, functioning instead to mark the
topic of a tweet. However, it is also broadly involved in construing heteroglossia in the
sense that it presupposes a virtual community of interested listeners who are actively
following this keyword or who may use it as a search term. This character usage derives
from Internet Relay Chat (IRC) conventions for naming channels. Hashtags, as they are
used on Twitter, are a form of inline metadata, that is, data about data that is actually
integrated into the linguistic structure of the tweets.
Within a tweet, a tag marked by a # sets up an attributive relationship between the
tweet as a tagged token and the tag as its type. Halliday and Matthiessen (2004: 219)
define attributive relational processes as relationships where an entity has some class
ascribed or attributed to it. In other words, hashtags inscribe a keyword in a tweet as
metadata referencing the topic of the message as assigned by the user. For example, the
following tweet contains the tag #obama indicating the subject matter of the tweet:
I just typed the words Obama presidency. It felt good. #obama.

The tag as type relationship assumes that other users will also adopt this tag and use
it as a keyword for a tweet on this topic. By generating keywords describing their discourse in this way, Twitter users enter into the social realm of collaborative tagging, or
folksonomy (Vander Wal, 2007). Collaborative tagging is a social form of verbal indexing involving a bottom-up approach to the kind of classification previously achieved by
reference librarians. It is, for example, used heavily on photosharing sites such as Flickr.
This type of inline metadata is different to how metadata is usually used because it is
directly visible to the user as part of the text of the tweet. While metadata rendered in markup
languages such as extensible markup language (XML) will typically separate form from
content, social tagging on systems such a Twitter collapses this separation. Thus tweets of
the following kind, where metadata is marked by a hashtag inside a clause, are possible:
Organized the pics from last nights Obama rally #Obama.

The hashtags can also mark functional roles in the linguistic structure. The tweets in
Table 1 are examples of this potential where Classifiers, Things and Processes are

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

792

new media & society 13(5)

Table 1. Examples of the hashtags marking units with different grammatical functions in clauses
ID

Tweet

Function marked

Millions of pathetic #obama supporters woke this morning with

nothing to talk about: http://twurl.nl/yzxt4t Funny stuff (h/t @User)
my daughter is explaining to me why she finds #obama cool. it
sounds something like: tatadatadatitadata. Cant argue.
Alright, lets all sign up. New meme on Election.twitter.com. Time
for Obama. Volunteer to #helpbushpack

Classifier

2
3

Thing
Process

Figure 1. An example of a tweet in the Obama Win Twitter corpus

marked. These are linguistic terms from SFL used to describe the functions of a particular linguistic unit in a clause. They encode both the structure and the meaning of a particular unit. For example, in the following tweet the Process, typically realized by a verb,
(woke) construes an action in the world, the Thing (supporter) encodes the agent that
carries out the action, while the Classifier (obama) describes the type of social categories
to which the Thing belongs: Millions of pathetic #obama (Classfier) supporters (thing)
woke (process) this morning
Tweets and the hashtags which they contain may thus be thought of a two different
orders of experience: a tweet is an instance of language use, while a tag is language about
language, performing what this article will show is an affiliative function.

Data: the Obama Win Twitter corpus

The dataset analysed in this study is a corpus of tweets collected using a Python script
and the Twitter API to scrape all tweets containing the string Obama posted to Twitter
in the 24 hours after the declaration of Barak Obamas victory in the 2008 presidential
elections. The corpus contained 45,290 tweets (813,310 words). The aim was not to construct a representative corpus of the linguistic activity on Twitter, but instead to conduct
a case study in which field variables, that is, the topic of the tweets, was held relatively
constant to afford a rich investigation of meaning-making in a single domain on Twitter.
Figure 1 is an example of a tweet from the corpus. It is presented as it appears on the
web interface to Twitter. The visual appearance would, however, be different depending
on the channel via which the tweet is syndicated and displayed. In this way, a tweet is a
text with multiple expression plane realizations or, in other words, with no single stable
visual or typographic form. Thus tweets are likely to be particularly capricious objects
for multimodal discourse analysis which considers how meanings are made with both
language and paralanguage.

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

793

Zappavigna
Table 2. The 10 most frequent # tags in the Obama Win Twitter corpus
ID

# tag

1
2
3
4
5
6
7
8
9
10

#Obama
#ElectionWrap
#TwitVote
#election08
#web2summit
#3News
#election
#haikus
#puppy
#moc2008

The corpus contained 721 instances of re-tweeting and 7100 instances of @ characters used for attribution and addressivity. There were also 770 tweets employing the
hashtag to refer to the topic that the user wished to ascribe to the tweet. Eighty-five
unique hashtags were used. The twenty most common hashtags in the corpus are presented in Table 2, with the most common tag being #Obama. Two hundred and thirtyfour tweets in the corpus contained this tag. An example is the following: Its still
surreal #Obama.

Theoretical framework: Systemic Functional Linguistics and a

language-based theory of affiliation
Systemic Functional Linguistics (SFL) is a theory of language that is tailored to
answering questions about how meanings function within the particular contexts in
which they are made. SFL is aligned with the functional tradition in linguistics manifest in the Prague School (Jakobson, 1971) and arose out of the linguistic school
known as Firthian Systemics (Firth, 1957). It is a functional theory because it is targeted at answering questions about how meanings operate within the particular contexts in which they are created. As a method for managing the high dimensionality of
language, SFL stratifies language into phonology (systems of sounds/writing), lexicogrammar (systems of wording), discourse semantics (systems of meaning), and context (genre and register). The strata are related to each other in terms of emergent
complexity: as patterns of patterns (Lemke, 1984). For example, discourse semantics
is a higher order patterning of lexicogrammatical patterns, which are in turn patterns
of phonological patterns.
A key functional unit in SFL is the clause. A clause corresponds to the notion of a
simple sentence. For example, I love Obama is a clause. A complex sentence can
contain multiple clauses. For example, I love Obama and I voted for him contains
the additional clause and I voted for him. This is what is referred to as a clause complex because it contains two clauses that enter into a logical relationship. Halliday

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

794

new media & society 13(5)

Figure 2. The institutionalization of affect (adapted from Martin and White, 2005: 45)

(1994: 106) proposes that the clause is central in construing experience, claiming that
it embodies a general principle for modelling experience namely, the principle that
reality is made up of PROCESSES. A process might be thought of as the heart of the
clause. In lay terms, a process is a verb, that is, something that happens such as eating, thinking, saying, and playing, or something that is such as being or
existing.
According to the theory, language enacts three simultaneous functions, referred to as
metafunctions: an ideational function of enacting experience, an interpersonal function
of negotiating relationships, and a textual function of organizing information (Halliday
and Matthiessen, 2004). A linguist using this theory will attempt to consider these three
functions when analysing any instance of linguistic meaning.
This article focuses on relationships set up between ideational and interpersonal
meanings in a corpus of tweets. Evaluation is a domain of interpersonal meaning where
language is used to build power and solidarity by adopting stances and referring to other
texts. In order to analyse evaluative meanings it draws on a theory of appraisal developed
within the SFL paradigm (Martin and White, 2005). This theory considers how the linguistic patterning of a text construes emotional language in three areas: attitude3 (making evaluations), engagement (bringing other voices into the text) and graduation
(scaling up or down evaluations). Figure 2 provides examples of each of these kinds of
evaluation (shown in bold). This figure suggests, following Martin and White (2005: 45),
that the protolinguistic expression of personal reactions via affect that we may see in an
infant, develops as we are socialized into a culture and into cultural institutions. These
feelings become institutionalized as ethics or morality, forming the judgement system,
and as aesthetics or value, forming the appreciation system.
For example, the evaluative lexical item, beautiful in Figure 2 is an example of
appreciation because it makes a value-based assessment. This system is a resource for
expressing attitudes about objects, states and processes, in contrast to the judgement
system, which typically assesses human behaviour and what people say and believe.

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

795

Zappavigna

Recent work in SFL has considered the coupling (Martin, 2000; Zhao, 2010) of
evaluation with other kinds of linguistic meanings in texts as a way of tracking the kinds
of values construed in the process of affiliation; that is, the process by which people
involve themselves in social bonds (Knight, 2008). A coupling is a binding of two meanings across paradigmatic systems of potential and may be involved in larger syndromes
of meaning in a text (Zappavigna et al., 2008). It is related to the simpler notion of collocation, two linguistic items occurring near each other in text. Knight (2008) suggests
that couplings of interpersonal and ideational meanings have an affiliative function. She
develops this idea in the context of conversational humour, which she argues is characterized by the tendency of participants to laugh off couplings that cause wrinkles in
their network of values:
to construe affiliation, conversational participants present couplings that represent what ties
them together as members of particular communities, and variously commune around, reject, or
laugh off these couplings as more or less acceptable bonds between them (Knight, forthcoming).

Culture, when adopting this perspective, becomes a semiotic network of social bonds.
We may investigate these bonds by considering how meanings unfold in texts as they are
instantiated in particular contexts of use, for example, users twittering about Obamas
election win.

Method
The Obama Win Twitter corpus was analysed using the metafunctional approach of
Systemic Functional Linguistics, described in the previous section. In terms of close text
analysis, this involved coding the corpus for evaluative language using the system of
appraisal introduced in the previous section. The schema underlying this analysis was a
system network for evaluative language. Systems networks conform to the SFL modelling strategy that conceives of language as a system of meaning potential, where that
potential is realized as the particular configurations of linguistic choices which can be
identified in texts. This is a perspective that considers language paradigmatically in
terms of what could go instead of what (Halliday and Matthiessen, 2004: 22) rather
than modelling language as a catalogue of structures.
Figure 3 is the system network based on Martin and White (2005) that acted as the
schema for annotating the evaluative language in tweets. In this network, a square bracket
represents a choice between two options (an or relation), while a brace represents
simultaneous choices (an and relation). For example, the attitude system involves
three simultaneous systems: type, explicitness and polarity. Within the type system,
there are three possible choices: affect, judgement or appreciation. The extracts from
tweets in the boxes are example realizations of the features in the network.
The software application, UAM Corpus Tool (ODonnell, 2008), was used to annotate
the data and AntConc (Anthony, 2009) was used for concordancing. The analysis presented in the section that follows draws upon concordance lines, n-grams and close text
analysis. Since it was beyond the scope of manual analysis to analyse the entire corpus
of tweets using the close text analysis involved in appraisal analysis, a random sample of
100 tweets was annotated for appraisal.

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

796

new media & society 13(5)

Figure 3. System network used a schema for appraisal analysis

Twitter and interpersonal meaning

Despite being parodied as a service that allows users to give status updates about their
activities with irritating frequency, Twitter, as some commentators have noted, may be
performing a more interpersonal function: The best Tweets tend to make an observation,
take a stance, or crack a jokenone of which fall under the umbrella of using Twitter to
tell the world what youre doing (McCracken, 2009).
This intuition is supported by the findings of Mischauds (2007) study of the extent to
which Twitter users respond to Twitters question, What are you doing?. This contentbased study found that 65 percent of tweets do not directly reply to this question by
describing activities, although it should be noted that Mischauds criteria that such tweets
do not use ing verbs is not the most rigorous linguistic basis upon which to make this
kind of assertion. Instead topics such a family and friends were the focus of many tweets.
The present study seeks to complement this initial content-based analysis with close
linguistic analysis using the metafunctional lens offered by Systemic Functional
Linguistics. This section will explore how tweets in the Obama Win corpus deploy
resources for making interpersonal meaning.
Consider for example the following tweet in the corpus:
HOLY CRAP. OBAMA WON HE WON!!!! IM SO HAPPY!!!.

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

797

Zappavigna

In this example, the expletive in initial position, the modifier so and the choice of
large-caps and repeated exclamation marks realizes increased graduation of the interpersonal meanings expressed, as annotated below. Obama is coupled with positive judgement about winning the election and in turn with positive affect (appraisal coding in
square brackets, appraisal item in bold):
HOLY CRAP [increased graduation: force]. OBAMA WON [positive judgement] HE WON
[positive judgement]!!!! [increased graduation: force] IM SO [increased graduation: force]
HAPPY [positive affect]!!! [increased graduation: force]

Other evaluatively-charged tweets include the following (appraisal coding in square

brackets, appraisal item in bold):
@username Its not crying [positive affect] over Obama so much as its crying [positive
affect] for what that means. We moved closer to being what we aspire to be [positive
judgement]. #obama
I dont like [negative affect] Obama and I am not happy [negative affect] about his election.
Almost half [increased force] of the US agrees, but I guess theyre not on twitter.
while we breathe we hope [positive affect]!!!!! [increased graduation] love love love [positive
affect] [increased force] obama... he brought hope [positive affect]!
#electionwrap NGO report: McCains coverage was much more [increased
negative [negative judgement] than Obamas

graduation]

The process won was the most frequent evaluative item in the corpus, occurring 2,764
times. Won is an instance of positive judgement, positively appraising the outcome of
Really happy that Obama won
Obama won!

by such a margin! But wtf happend to prop 8 :/

Yay! Lol. And I wanted to tell you something
funny but I forgot.
I still cant believe it! Obama won! I knew it! -hes a modern day heroe, I think and I
hope. Now well just have to wait and see..
its pretty great that obama won, i hope he can deliver on his promises. now
someone decent needs to replace gordon brown.
happy and relieved that obama won the election
i woke up today and Obama had still won. This is awesome
Still stunned Obama won. Its like Christmas.
@username cheer up... OBAMA WON! :)
Still watching msnbc so happily won. Wish Id be in the States to witness this historical
Obama
day.
Thank god Obama won! I dont think the world would have lasted with
another oil-hungry, war-loving, country-invading
Republican

Figure 4. Concordance lines for won in the Obama Win Twitter corpus

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

798

new media & society 13(5)

the election process. Examples are shown in the concordance lines presented in Figure 4.
Instances of positive affect (underlined), positive judgement (bold) and positive
appreciation (bold italics) that are co-realized are also shown. While these concordance
lines are only a random sample of the evaluation surrounding won, they show a general prosody of positive evaluation co-occurring with this item. Negative appraisal was
possible but was generally used in a contrastive relation to Obamas victory.
Some examples of related evaluative 3-grams4 include:
Glad Obama won
Obama has won
won the presidency

Given that the news of Obamas victory was widely known, at least after the first hours
post-announcement, these tweets seems to be performing a function beyond informing
other twits of the news. The kind of evaluative language that sampled above suggests
that the tweets may be forming a more interpersonal social function in which users are
affiliating around values relating to the election result.
In the corpus these values are often construed as couplings of Obama with appraisal.
For example, a common evaluative 4-gram that coupled positive affect and positive
judgement was:
happy that Obama won

This included tweets such as those shown in Figure 5.

Clearly, an intervening variable in the argument being made here about the evaluative
charge and interpersonal-focus of tweets is the historical and political nature of the event
to which these tweets are responding. However, the tendency manifest in this corpus

is really, really, really

is at work til 5 and is
Im happy and sad.
is it weird to feel

happy that Obama won.

happy that obama won
Happy that Obama won,
happy that obama won

Ive talked with a number of

people today who supported
McCain but are still pretty
Im

happy that Obama won.

Very

happy that Obama won!

happy that Obama won!

the election....GO OBAMA!!!!

but sad that Prop 8 passed.
the presidency campaign?
obama changes america,
america changes the world!
hurrah!

its very emotional to witness

history firsthand

Figure 5. Concordance lines for happy that Obama won in the Obama Win Twitter corpus

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

799

Zappavigna

toward presenting evaluative couplings, while perhaps more pronounced due to the political context, seems characteristic of the kind of interpersonal patterning seen throughout
Twitter discourse on a range of topics. However, a large-scale quantitative study, beyond
the scope of this article, would be required to verify this claim.

Hashtags: labelling the ideation we are going to

axiologize around
The affiliative function of the evaluative language seen in the tweets will now be considered
by investigating the role of hashtags in coupling. Hashtagging is an emergent activity.
Hashtags are a typographic convention used to mark the topic of a tweet. We may also think
of them as indicating the target of the appraisal in the tweet. For example, tweets containing
couplings of Obama and positive evaluation sometimes included the tag #Obama:
From fear, hatred and economic collapse to hope, search for common ground and prosperity
again. Good change! #obama.

In these instances the hashtag seems to intensify a call to affiliate with the values in
the tweet by rendering the tweet more searchable. It is more searchable because a user
can find this tweet by searching for Obama even when the body of the tweet does not
contain the string Obama.
Consider the following tweet that couples positive evaluation and Obama without a
hashtag:
Life is Good! And a Good Morning to all in Obama country!

Similar tweets were possible using a hashtag: username: 01:42:23 CET bedtime.
gonite #obama country, land of hope & change!
The hashtag in this example is an instance of inline metadata as the marker operates
within the clause attached to a classifier in the nominal group. This is a classifier as it is
defining the country as an obama type of country. The presence of the hashtag expands
the meaning potential of the tweet when compared with the non-hashtag example by
making the coupling louder in the sense that it is more available for search and more
likely to be automatically followed by those subscribing to this tag.
While Twitter has developed a search facility whereby a user can search via a range
of parameters, this does not override current hashtag practice. Twitter has in fact incorporated a hashtag field into its search form. The utility of the hashtag can be seen when
considering searches, for example, for election 08 which will only return results containing this exact string. A tag such as #Election08 becomes very useful for quickly
retrieving a range of tweets about a topic that do not necessarily reference the topic as a
word in the body of the tweet. In addition, it inscribes a form of intentionality in the
tweet via the tokentype relationship described earlier, because the user has explicitly
indicated this tweet is a token of the category labelled by the tag. Hashtags allow users
to search for potential targets of appraisal and find out, for example, whether Obama is
subject to prosodies of positive or negative evaluation as tweets unfold in time.

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

800

new media & society 13(5)

Interpersonal as well as ideational meanings could occur in a hashtag. For example in

the following tweet positive affect (hope) is used as a tag:
Prediction: Obama as powerful leader starts within month, big speech, call for econmic
legislation #hope #change.

Thus we have a classificatory system that is very different to the principles that would
be employed to categorize a reference collection in library using subject keywords. From
the perspective of information science, this may be perceived as a degradation of classification. However, such a criticism ignores the social function of the classification
which is to not only facilitate efficient relevance and recall, but to make possible what
may be termed ambient affiliation, a concept discussed later in this article.
The inline nature of #tag usage opens up the possibility of play with users creating
tags that are unlikely to be used as search terms and which instead seem to function to
intensify the evaluation made in the tweet. Instances include:
#racialjokeswecanmakenow
#presidentsIhavetheslightestshredofrespectfor
#finallyicansleepatnight
#americastillhasabrainandaconscience

This play is possible because of the close relationship of hashtagging to evaluation. For
example, the final two instances above are part of the following tweet:
Yes! #f*ckin-A #obama #finallyicansleepatnight #byebyebush #americastillhasabrainand
aconscience #whew!

Here the repetition of hashtags functions to scale up the positive evaluation of

Obamas victory via the system of graduation. Without the hashtags this tweet would
lose some of its humourous hyperbole.

Ambient affiliation
Markup is typically hidden by browsers and other display devices; however, hashtags on
Twitter are visible within clauses in tweets and no technology is used to obscure them. In
this way users can choose to mean in an explicitly searchable manner by integrating
metadata into their talk through typographic conventions, such as the hashtag, that
increase the loudness of their discourse by increasing the likelihood that their words
will be found. This, in turn, increases the probability that a users production of texts
over time will be actively followed by others. In other words, it creates the possibility
of ambient affiliation. Here we affiliate with a copresent (Goffman, 1963), impermanent,
community by bonding around evolving topics of interest. This function is directly
inscribed in the web interface to users Twitter accounts as trending topics, a list of

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

801

Zappavigna

Figure 6. Search creating an ambient affiliative network for the #Obama tag

keywords occurring with high frequency in current posts, displayed under the search
box. In increasing loudness in this way, hashtags identify meanings that have become
hyper-charged with an additional semiotic pull that may be likened to a gravitational
field. They act as both a label for the potential discourse community that they establish
and render searchable the coupling that occurs in the tweet.
The expansion of meaning potential seen in the hashtag usage is both a product of the
reduced affordances of the character-constrained mode and part of a multiplication
(Lemke, 1998) of what it means to talk online. It is electronic discourse explicitly
encoded as searchable. Using the #obama tag is, for example, analogous to saying If
you are interested in values about Obama search for me. The social function of the
hashtag is to provide an easy means of grouping tweets, and in turn, creating ad hoc
social groups or sub-communities. Being searchable opens up a new kind of sociality
where microbloggers engage in ambient affiliation. The affiliation is ambient in the sense
that the users may not have interacted directly and likely do not know each other, and
may not interact again. It also could not occur without adequate search functionality.
Users searching to explore online conversations produced on social networking sites in
this way is a new cultural process.
Interpersonally-charged tweets invite with their hashtags an ambient audience to align
with their bonds. The hypercharge of the hashtag involves the tweet in a larger bond
network of values. For example, we might think of the search window for the tag
#Obama as creating a momentary, ambient affiliative network of tweets in which this

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

802

new media & society 13(5)

Figure 7. StreamGraph of the query happy after the Obama election win Produced using
Clarks (2008) system, Twitter StreamGraphs.

tag is potential target of evaluation (Figure 6). Of course in Figure 6 we have factored out
time, perhaps the most important dimension in Twitter discourse. This omission means
that the tweet starts to look like an artefact rather than a text unfolding dynamically over
the twitter stream.
We can, however, gain some insight into what a more time-based representation might
be if we consider Figure 7, a StreamGraph (Byron and Wattenberg, 2008; Clark, 2008;
Havre et al., 2002) showing the unfolding of tweets that contain the string happy. This
graph was produced using Clarks (2008) system Twitter StreamGraphs. StreamGraphs
are an example of a text visualization technique that does not efface logogenesis, the
unfolding of text over time (Zappavigna, forthcoming). While an area graph usually
shows a single data series, StreamGraphs are a form of stacked area graph that represent
multiple data series by stacking one on top of the other. In a stacked area graph the height
of the curve at a given point represents the total frequency of all features at that point and
thus each data series should be read as starting at zero rather than as their accumulative
height. This makes the graphing technique most useful to a linguist interested in the
general trend of a data series, or in other words, the qualitative ebb and flow of the annotated units over the time series. It is also a useful technique for appreciating the relationships between the data series as they unfold by the overall impression of the relative
amount of colour.
Examples of the tweets represented in the StreamGraph include:
Wow, Im so happy, bring it home Obama
Obama:) Im so happy that Im crying! Hahhaa. Yes. He got his daughters a puppy.:) He is an
amazing person and has well deserved this.

The stream along the bottom of the graph shows the coupling of happy (positive affect)
and Obama unfolding over time (represented as intervals along the X-axis). The other

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

803

Zappavigna

streams present in the graph are lexical couplings in temporally related tweets. There is a
fairly consistent relationship between happy and Obama in the time window graphed.
Other evaluative items in the environment of Obama and happy were: proud, sad,
feel, hope, right, won, love, etc.
Given that tweets unfold over time along private and public streams, considering
them from this dynamic perspective is crucial when making claims about the discourse.
The argument that has been made in this article about the construction of ambient community assumes this dynamic perspective: these communities shift as hashtags shift, and
different couplings of ideational and interpersonal meaning are established depending on
what people are talking about at a given time. Theoretically, some topics might remain
fairly constant such as, for example, the microgenre of complaining about something,
usually technological, that is not working for a user, marked by the #fail tag. For example, the following is typical of complaints regarding software problems:
Vista spent 45 minutes installing updates.. only to say after rebooting that the update has failed
and all changes are rolled back. #fail.

Other tags may be more likely to shift with social and political concerns such as #Iran
which at the time of writing was a tag facilitating political protest on Twitter in order to
bypass mainstream media censorship: hopes that people around the world will help civilians rally for independence against unjust religious and political oppression in #Iran.
This functioning of the hashtag as inline metadata is a novel and emergent form of
punctuation usage. The use of these markers appears to be related to a larger movement
that Knox identifies in the evolution of what he refers to as interpersonal punctuation:
The trajectory for interpersonal punctuation begins with boundary marking, moves to
punctuating speech function, and then to punctuating attitude and identity. At the same time, the
prosody of punctuation spans (potentially) longer stretches of text, with the punctuation of
attitude and identity through emoticons now able to spread over entire messages. (Knox, 2009)

For example, commas may be used to mark clause boundaries, question marks to indicate speech function and a smiley emoticon to mark positive affect. To Knoxs trajectory
we would add the punctuation of ambient affiliation via hashtags. This is the beginning
of searchable talk where people mark the ideational targets of appraisal that act as boundaries defining ambient communities. They mark these targets both to indicate a stance
and to make their talk findable.

Conclusion: microblogging and affiliation

This article has presented a case study showing the role of hashtags and evaluative language in affiliation on Twitter. The study involved a corpus-based discourse analysis of
tweets about the victor of the 2008 US Presidential election. It has argued that the outpouring of emotional and evaluative language seen in this corpus is an example of the
way that Twitter seems oriented toward the expression of interpersonal meaning. Criticism
of Twitter as a service facilitating inane and frequent status updates about users activities
seems to have missed the social point of twittering. Twitter offers a medium for expressing personal evaluation to a large body of listeners with which one can affiliate ambiently.

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

804

new media & society 13(5)

This article has argued that by presenting couplings of evaluative and ideational meanings, a tweet invites the follower or searcher to share in the values presented.
The expansion of typographic meaning potential seen in hashtag usage on Twitter is
the beginning of searchable talk. Hashtags are used to mark potential targets of evaluation and to render these as metadata that may be found by other users. Hashtag usage on
Twitter is an example of leveraging one the essential affordances of New Media: the
affordance of the database to render information searchable and to make visible relationships that would not otherwise be recognizable. As Manovich (2002: 228) notes in his
argument that databases have become a key form of cultural logic, regardless of whether
new media objects present themselves as linear narratives, interactive narratives,
databases or something else, underneath on the level of material organization, they are
all databases. Hashtag usage in Microblogging is an example of deploying the semiotic
potential of this material organization.
Searchable talk is online conversation where people actively render their talk more
findable. Talk using typographic strategies such as hashtagging becomes louder and more
bondable. It appears the linguistic corollary of Morvilles (2006) concept of ambient
findability, where information can be found in any location. Now, instead, information
can be found anywhere you have access to online talk and the wealth of online discourse
renders Twitter an interpersonal search engine. In other words, Twitter is the place you go
when you want to find out what people are saying about a topic right now and in order to
involve yourself in communities of shared value that interest you in this given moment.
While this study has focused on Twitter, broader research into how we use
Microblogging to affiliate ambiently is clearly important. A critical issue for any program
of research into Microblogging is to consider it as a semiotic activity, that is, as an activity
of making meaning with language. If we are to understand the relationships enacted by
such meaning-making we require systematic ways of accounting for the role of language
in creating social bonds. Affiliation is about more than connecting; it is about negotiating
meanings within genres of language use. Hence, while studies of who connects with
whom at which level of frequency will be useful, we need to understand the nature of
what is being negotiated with language within particular patterns of social processes. The
approach taken in this article has been informed by two complementary traditions in linguistics: Systemic Functional Linguistics and Corpus Linguistics, both of which offer
useful, replicable strategies for analysing how language works in its social contexts. As
our online talk becomes increasingly searchable and as microblogging changes our patterns of interaction, the data available to analysts for pursuing the question of ambient
affiliation will grow. It is hoped that this article has been a first step into showing how we
can study the inevitable linguistic complexity that arises as people commune online.
Notes
1. For an overview of research into Social Networking Sites see boyd and Ellison (2007).
2. American usage refers to the # character as a pound sign while Australian and UK usage refers
to it as a hash. I will adopt the latter convention.
3. Technical terms of the Appraisal framework are distinguished from their everyday counterparts via small caps font.
4. n-grams are used in corpus linguistics to predict the sequence of items. In this instance, a
3-gram shows clusters of three words.

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

805

Zappavigna
References

Anthony L (2009) AntConc3.2.0. Computer software. Tokyo.

Bakhtin M (1981) The Dialogic Imagination. Austin: University of Texas Press.
Baron N (2008) Always On: Language in an Online and Mobile World. Oxford and New York:
Oxford University Press.
boyd dm and Ellison NB (2007) Social network sites: Definition, history, and scholarship. Journal
of Computer-mediated Communication 13(1). Available at: http://jcmc.indiana.edu/vol13/
issue11/boyd.ellison.html
Burnett G (2000) Information exchange in virtual communities: A typology. Information Research
5(4). Available at: http://jcmc.indiana.edu/vol9/issue2/burnett.html
Byron L and Wattenberg M (2008) Stacked graphs geometry & aesthetics. IEEE Transactions on
Visualization and Computer Graphics 14(6):12451252.
Clark J (2008) Twitter topic stream. Available at: http://www.neoformix.com/Projects/
TwitterStreamGraphs/view.php
Crystal D (2006) Language and the Internet (2nd edn). Cambridge: Cambridge University Press.
Firth JR (1957) Papers in Linguistics 19341951. London: Oxford University Press.
Goffman E (1963) Behavior in Public Places: Notes on the Social Organization of Gatherings.
New York: The Free Press.
Hagel J and Armstrong A (1997) Net gain: Expanding Markets Through Virtual Communities.
Boston, MA: Harvard Business School Press.
Halliday MAK (1994) An Introduction to Functional Grammar (2nd edn). London: Arnold.
Halliday MAK and Matthiessen CMIM (1999) Construing Experience Through Meaning: A
Language-based Approach to Cognition. London: Cassell.
Halliday MAK and Matthiessen CMIM (2004) An Introduction to Functional Grammar (3rd edn).
London: Arnold.
Havre S Hetzler E, Whitney P and Nowell L (2002) ThemeRiver: Visualizing thematic changes
in large document collections. IEEE Transactions on Visualisation and Computer Graphics
8(1): 920.
Herring SC (1996) Computer-Mediated Communication. Linguistic, Social and Cross-Cultural
Perspectives. Amsterdam and Philadelphia: Benjamins.
Herring SC (2004) Computer-mediated discourse analysis: An approach to researching online
communities. In: Barab SA, Kling R and Gray JH (eds) Designing for Virtual Communities
in the Service of Learning. Cambridge and New York: Cambridge University Press, 338376.
Herring SC (2008) Virtual community. In: Given LM (ed.) Encyclopedia of Qualitative Research
Methods. Los Angeles, CA: Sage, 920921.
Jakobson R (1971) Selected writings. Second expanded edition. The Hauge: Mouton.
Jones Q (1997) Virtual-communities, virtual settlements & cyber-archaeology: A theoretical outline. Journal of Computer-mediated Communication 3(3). Available at: http://www.ascusc.
org/jcmc/vol3/issue3/jones.html
Knight N (2008) Still cooland American too!: An SFL analysis of deferred bonds in internet messaging humour. In: Nrgaard N (ed.) Systemic Functional Linguistics in Use, Odense
Working Papers in Language and Communication 29: 481502.
Knight N (forthcoming) Naming culture in convivial conversational humour. In: Mahboob A and
Knight NK (eds) Directions in Appliable Linguistics. London: Continuum.
Knox J (2009) Punctuating the home page: Image as language in an online newspaper. Discourse
and Communication 3(2): 1953.
Kress GR and Van Leeuwen T (2006) Reading Images: The Grammar of Visual Design (2nd edn).
London and New York: Routledge.
Lemke JL (1984) Semiotics and Education. Toronto: Victoria University.

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012

806

new media & society 13(5)

Lemke JL (1998) Multiplying meaning: Visual and verbal semiotics in scientific text. In: Martin
JR and Veel R (eds) Reading Science: Critical and Functional Perspectives on Discourses of
Science. London; New York: Routledge, 87114.
Manovich L (2002) The Language of New Media. Cambridge, MA and London: MIT Press.
Martin JR (2000) Beyond exchange: Appraisal systems in English. In: Hunston S and Thompson G
(eds) Evaluation in Text: Authorial Stance and the Construction of Discourse. Oxford: Oxford
University Press, 142175.
Martin JR and White PRR (2005) The Language of Evaluation: Appraisal in English. New York:
Palgrave Macmillan.
McCracken H (2009) What i know about twitter. Available at: http://technologizer.com/2009/03/15/
what-i-know-about-twitter/
Mischaud E (2007) Twitter: Expressions of the Whole Self. An Investigation into User Appropriation
of a Web-based Communications Platform Masters thesis, Department of Media and
Communications. London: London School of Economics and Political Science.
Morville P (2006) Ambient Findability: What We Find Changes Who We Become. Cambridge:
OReilly.
ODonnell M (2008) UAM Corpus Tool 2.0. Computer Application. Madrid.
Rheingold H (1993) The Virtual Community. Cambridge, Massachusetts: MIT Press.
Rocketboom (2009) The Twitter global mind. Available at: http://www.rocketboom.com/rb_09_
mar_09_twitter_global_mind/
Vander Wal T (2007) Folksonomy coinage and definition. Available at: http://www.vanderwal.net/
folksonomy.html
Wellman B (2001) Physical place and CyberPlace: The rise of personalized networking.
International Journal of Urban and Regional Research 25: 227252.
Zappavigna (forthcoming) Visualizing logogenesis: Preserving the dynamics of meaning. In:
Dreyfus S, Stenglin M and Hood S (eds) Semiotic Margins. London: Continuum.
Zappavigna M, Dwyer P and Martin JR (2008) Syndromes of meaning: Exploring patterned coupling in a NSW Youth Justice Conference. In: Mahboob A, and Knight K (eds) Questioning
Linguistics. Newcastle: Cambridge Scholars Publishing, 103117.
Zhao S (2010) Intersemiotic relations as logogenetic patterns: Towards the restoration of the time
dimension in hypertext description. In: Bednarek M and Martin JR (eds) New Discourse
on Language: Functional Perspectives on Multimodality, Identity, and Affiliation. London:
Continuum, 195218.

Michele Zappavigna is a Research Fellow in Linguistics at the University of Sydney. Her

major research interests are computer-mediated communication and social media.
Currently she is working on a corpus-based study of the language of microblogging
using a 100 million word Twitter corpus. Her book The Discourse of Twitter and Social
Media will be published by Continuum in 2011.Michele alsohas an ongoing interest in
text visualization as a tool to aid discourse analysts.

Downloaded from nms.sagepub.com at ROWAN UNIV on July 9, 2012