Laughing with Letters:

A Corpus Investigation of the Use of Written Laughter on Twitter

Isabel McKay
April, 2015
University of Michigan Linguistics Department
Senior Thesis

Advisor: Deborah Keller-Cohen

Second Reader: Robin Queen

1
1: Introduction:
“Their morning greeting to a friend in a distant city is usually “g.m.,” and the farewell for the
evening, “g.n.,” … The salutation may be accompanied by an inquiry by one as to the health of
the other, which would be expressed thus: ‘Hw r u ts mng?’ And the answer would be: ‘I’m pty
wl; hw r u?’ or ‘I’m nt flg vy wl; fraid I’ve gt t mlaria.’”

The passage above is taken from a New York Times article first published on November 30th,

1890 ("Friends They Never Meet: Acquainteneces Made by the Telegraph Key. Confidences Exchanged

between Men who have Never Seen Each Other - Their Peculiar Conversational Abbreviations.," 1890).

Its topic: off-the-clock communications between telegraph operators. The article describes how the

men and women who worked the wires often became friends with operators in faraway cities; people

who they never met. It provides examples of abbreviations like those above, but it also records several

mythologized stories about long inter-operator feuds, describing both how operators were able to

recognize one another by their typing styles and how they whined to one another over the wire.1 A

large portion of the piece concerns the ways in which these telegraph operators used language. To

modern eyes the snippets of “telegraph speak” described in this article look almost exactly like the

“txtspeak,” language we use today on cell phones and online. In fact, the entire article bears an

impressive resemblance to a Huffington Post exposé on the logistics of e-romance or a BuzzFeed piece

on how kids these days can communicate using only emoji. Though numerous parallels can be drawn

between the way operators seem to have used the telegraph, the world’s first short-message service,

and the way we use text and instant messaging today, the most compelling section of this article comes

towards its conclusion.

1
“Gol hang ts everlasting grind. I wish I ws rich”

2
 “Operators laugh over a wire, or rather, they convey the fact that they are amused. They do this
by telegraphing “ha, ha.” Vary great amusement is indicated by sending “ha” slowly and
repeating it several times, and a smile is expressed by sending “ha” once or perhaps twice.
Transmitting it slowly and repeating it tells the perpetrator of the joke at the end of the wire that
the listener is leaning back in his chair and laughing long and heartily.”

The passage above is fascinating not only because it describe systematic native speaker

intuitions about the use of laughter over the telegraph, but the system described in the New York Times

article is almost identical to the system described in similar articles written today about our present-day

use of written laughter. An article posted to BuzzFeed in February of this year gives the following

definitions between forms of haha:

haha: “I’m acknowledging that you’ve said something you perceive to be funny,
though I don’t find it particularly funny myself,”

hahaha: “That was funny! I legitimately laughed, or at least smiled, and I am slightly
happier now than I was before you said that.”

HAHAHA(etc.): “I am starting to panic that I may never stop laughing!!!!!!!!!!!!!”

(Heaney, 2014a)

To modern readers it is likely not at all surprising that telegraph operators would have invented

some means of expressing laughter over the wire, or even that their system might have been similar to

our own. After all, what are our modern forms, lol (laughing out loud), lmao (laughing my ass off),

mwahaha, *giggle*, and 😂 if not a version of ᛫᛫᛫᛫ ᛫‒ ᛫᛫᛫᛫ ᛫‒ (haha)? We write our laughter out every

day and so it seems only natural that telegraph operators should do the same.

But why isn’t it surprising? It’s not as if haha can be found all through the novels, the essays,

the newspaper articles, or even the personal letters that were written between the 1890s and the birth

of the internet. It wasn’t as though the first users of modern-day Short-Message Services (SMS) decided

to type out their laughter in order to harken back to the days of the telegraph. Rather, each medium’s

convention of laughing with letters was adopted independently, because laughter is something we need

in SMS communication today just as much as telegraph operators needed it in the 19th century. Even

3
today, with lol having found its way into most dictionaries, written laughter is more or less found only in

short, personal communications. There is something about communication that makes us feel the need

to use laughter and there is something about the way the telegraph and the text message force us to

communicate in particular that has made us feel the need to type out that laughter.

This thesis represents a first attempt to understand the conventions and practices surrounding

how and why we write our laughter with letters. In this project, an attempt will also be made to

evaluate the extent to which use of written laughter draws on knowledge of physical laughter. The

study of written laughter is complex, as it is both a phenomenon of short-message communication and

an approximation of a face-to-face conversational tool. I will present evidence concerning the usage

patterns of six forms of written laughter, lol (laughing out loud), lmao (laughing my ass off), haha, hehe,

😄, and 😂 on the social networking site Twitter.

The following sections are included in order to provide an overview of some of the previous

studies of and observations on both face-to-face and written laughter which have informed this

research. First, a literature review of previous research on both face-to-face and written laughter is

provided. Afterwards, as there have been few studies examining written laughter, a section is included

describing the history of each form under consideration and some native intuitions about each.

2: Literature Review
2.1: On the Study of Face-to-face Laughter:

Laughter is a human communicative universal (Edmondson, 1987). Though etiquette rules

surrounding appropriate contexts for laughter may vary, the situations in which laughter arises are more

similar across cultures than almost any other form of nonverbal communication (Glenn, 2003). Laughter

4
also seems to be one of our oldest communicative tools, as parallels of laughter can be found among

several of the great ape lines (Provine, 2001).

Laughter is a deeply-engrained human behavior which can be studied through many different

lenses. There is literature on laughter as physiological process, as a relic of human evolution, as a

behavioral response to certain stimuli, as an indicator of emotional well-being, and as an identifier of

humor. Though these bodies of literature are all important and interesting, this survey is limited to

those studies which have examined the social information conveyed by laughter and its purpose in

communication.

2.1.1: The Nature of Laughter

Early theories of laughter considered laughter to be little more than a side-effect of an internal

psychological state.2 Just as someone who sneezes can be inferred to have a cold, someone who laughs

can be inferred to be experiencing the psychological state (whatever it may be) that accompanies

laughter. Now experts tend see laughter differently. Modern-day conceptualizations treat laughter as a

behavior actively intended to communicate how a laugher would like his or her words to be taken by co-

participants (Glenn, 2003; Holt, 2013). These understandings draw on the fact that laughter heard out

of context is only laughter. It may mean that someone is feeling nervous or that someone finds a joke

2
Until about the last fifty years, laughter was treated in the literature more as a symptom of an
individual’s internal, emotional state than as a conscious communicative tool (Glenn, 2003; Provine,
2001). The assumption that laughter is a reflexive expression of a psychological state has been made
most often in studies which describe laughter as a natural reaction to humor, but other theories have
proposed it as a reaction to various other forms of stimuli. For example, Superiority/Hostility Theory
argued that laughter was a reflection of an individual’s emotions upon victory, a feeling famously
described by Thomas Hobbes as “sudden glory” (Hobbes, 1640). Incongruity Theory, on the other hand,
supported by Immanuel Kant and Schopenhauer, suggested that laughter reveals the shock a person
feels when his or her expectations do not align with observed reality (Glenn, 2003). And Relief Theory,
supported by Sigmund Freud and others, argues that laughter is a reaction to a release (or sometimes
heightening) of some sort of psychological burden (Provine, 2001). Taken together these theories can
fairly well approximate the types of situations during which laughter can occur. However, as shall be
argued, there are some major flaws in their basic assumption: that laughter is symptomatic of a
psychological state.

5
funny, but regardless, the presence of laughter does not necessarily indicate only one kind of emotional

state. Most possible interpretations of laughter out of context have to do more with the surrounding

interaction than with the feelings of the laugher; that a joke was told, that someone is threatening

someone else, that someone just said something embarrassing. Though laughter, like other forms of

nonverbal communication, is not regulated as consciously as language, it is still a socially-targeted

action.

The most important piece of evidence that laughter is a marked act intended to communicate

one’s feelings, is that the presence or absence of others, whether or not those others are laughing, and

the relationships between the laugher and the others in the room, are all important factors in the

production or nonproduction of laughter (Glenn, 2003; Lee & Wagner, 2002; Osborne & Chapman, 1976;

Provine, 2001). People are more likely to laugh in groups than alone, people are more likely to laugh in

a group of friends than in a group of strangers, and people are more likely to laugh when others laugh

with them than they are to laugh alone. Even, “canned” or “laugh-track” laughter can increase laughter

production. Crucially, the variations described above seem to function independently from the

perceived “funniness” of a situation. Most studies exploring the behavioral differences described above

have asked subjects to view sitcoms or standup comedy routines. After the fact subjects were asked to

report how funny they found the material they were asked to view. Generally, the “funniness” rating

for individuals viewing the same content in various social settings was roughly the same, regardless of

how much laughter was produced (Neuendorf & Fennell, 1988).

If laughter were a symptom of an internal psychological state, we might expect to see it

inhibited in certain situations, as coughing can be inhibited in a quiet concert hall, but we would

certainly not expect to see laughter inhibited in private. All of the environments in which laughter is

most freely expressed are environments in which communicative channels to other participants are

6
opened wide. This suggests that laughter, like several other forms of nonverbal communication, is

produced for the benefit of others.

There are, of course, situations in which individuals do produce laughter in private. Most

theories place this sort of laughter production in the same category as talking to oneself or to the

characters in a television show.

Rather than looking strictly to jokes and humor to understand the nature of laughter, most

theories now suggest that an examination of a social interaction as a whole must take place. As feelings

and desires are more or less unreachable for researchers, it is instead the context of laughter in

conversation which must be examined. We must look at why an individual chose to place laughter when

he or she did in conversation (Glenn & Holt, 2013).

2.1.2: The Meaning of Laughter

Conversation analysis (CA) is one of the most effective ways of looking at the effect various

conversational actions have on the flow of an interaction. By examining actions and the precise

locations where they occur in conversation, CA asks the question why that now? For this reason a CA

approach to laughter is very enlightening. It asks, why laughter here?

An assumption of conversation analysis is that one of the major purposes of talk is to encourage

intersubjective understanding between participants. This means that talk is used not only to

communicate factual information but to communicate participants’ opinions on those stances. This

communication of opinion and feelings can even go as far as to express one’s opinion on the ongoing

exchange. Under this assumption it is not only what a person says that communicates information to

other participants, but how and when that person says it. By replying to another participant’s statement

with a factually-relevant utterance an individual may display intersubjective understanding of the factual

content of a previous utterance. However, by adding to that factually-relevant utterance, discourse

markers, tonal inflections, or body movements, other sorts of intersubjective understanding can be

7
achieved as well. The speaker’s level of comfort with the information being communicated may be

understood by co-participants, the speaker’s opinions on word choice in a previous utterance, the

previous speaker’s conversationally-expressed opinions, or the overall shape a conversation can be

communicated through these small conversational actions. This suggests that by looking at utterances

and interactional behaviors in terms of other surrounding, adjacent, utterances, an understanding of

intersubjective meaning can be achieved (Peräkylä, 2007; Sacks, Schegloff, & Jefferson, 1974).

The groundwork for conversation analysis of laughter was laid by Jefferson, Sacks, and Schegloff

in their 1977 paper, Preliminary Notes on the Sequential Organization of Laughter. Since this paper,

several other similar projects have made important observations about laughter in conversation.

The first of these observations seems obvious, namely, laughter is indexical; it refers to

something specific going on in the conversation (Glenn, 2003, 2010; Holt, 2011, 2013; Jefferson, 1984;

Jefferson, Sacks, & Schegloff, 1977; Provine, 2001). Jefferson, Sacks, and Schegloff therefore suggest

that laughter may be a token of understanding, which is a sort of conversational object which takes its

meaning or force from its referent. The conversational object that laughter takes as its referent

(generally known as a “laughable”) may be either come before the laughter (as in a laugh response to a

joke) or after the laughter (as in a funny story being introduced by laughter).

Second, most papers on laughter in conversation distinguish between conversational laughter,

which comes in short bursts and does not drastically interrupt the flow of interaction, and “laughing

together,” which is raucous, extended, and involves most, if not all, of the participants (Glenn, 1989;

Jefferson et al., 1977). “Laughing together” is often considered a separate activity in and of itself and is

far less referential than conversational laughter.

Third, laughter is invited. We tend to think that it is rude to laugh at one’s own jokes but,

especially in small-group interactions, that is exactly what happens (Glenn, 1989, 1991, 2010; Glenn &

Holt, 2013; Holt, 2013; Jefferson, 1979, 1984; Provine, 2001). In the majority of instances the person

8
speaking or creating a laughable is the first person to laugh. After the speaker’s laughter has begun,

others join in and “accept” the invitation. The terms “invited” and “offered” laughter are used, because

it seems, at least outwardly, that the speaker’s laughter gives other participants permission to laugh

themselves.

Fourth, the genders of the individuals offering and accepting laughter seem to be important

factors in determining whether or not an invitation to laugh will be accepted (Jefferson, 2004; Lampert

& Ervin-Tripp, 2006; Provine, 2001; Rees & Monrouxe, 2010). Though several researchers have

observed this difference, informed discussion on the topic has been restricted by the generally-accepted

understanding that gender is performed, and thus any discussion of male and female behavior ought to

be approached with care. Jefferson, however, observes that as long as these differences are discussed

in terms of a “male identity projection” or a “female identity projection” (instead of male and female

behavior) studying the laughter behavior of these groups can be legitimate. These studies more or less

come to the conclusion that people projecting a male identity are more likely to laugh when not invited

and are more likely not to laugh when invited, especially when the other participant is a female. People

projecting a female identity, on the other hand, are more likely to laugh when invited, especially when

the co-participant is male, and are less likely laugh when not invited (Glenn, 2003; Jefferson, 2004;

Lampert & Ervin-Tripp, 2006; Provine, 2001).

The invitation/acceptance pattern of laughter is of particular note when attempting to

determine the function of laughter in interaction. A laughter invitation asks a co-participant to interpret

some laughable in a certain way, and a laughter acceptance indicates that the other participant has

agreed on the interpretation (Glenn, 1989, 1991, 2003, 2010; Jefferson, 1979).

The leading understanding of laughter today is that is an indicator of a participant’s desire to

view the ongoing interaction through a playful or non-serious lens. This lens is known as a ‘ludic frame,’

(or more colloquially as a “play” frame) meaning that the social events taking place in and around

9
instances of laughter are to be interpreted as disconnected from reality; as a “pretend” version of a

serious exchange (Glenn, 2003; Glenn & Knapp, 1987; Holt, 2013).

Play frames were first discussed by Bateson in the 1950s, though he did so using slightly

different language than that we use today (Bateson, 1955). Bateson, in observing otters’ play, realized

that the actions involved were similar enough to those involved in fighting that the otters must have

been indicating to one another through some sort of social signaling that the action taking place was

play and not aggression.

Erving Goffman famously extended the ideas put forth by Bateson, developing the terminology

of “framing” most commonly used today (Goffman, 1974). Goffman argued that by making certain

signals animals like otters as well as humans are able to set up an interpretive frame of interaction. The

same actions taken or words spoken in two different interactive frames may have wildly different

interpretations. For example when a dog wags its tail, indicating a playful frame, and growls, other dogs

will react by initiating a playful tussle, but if a dog were to give the same growl without signaling that a

playful frame was in effect, other dogs might react aggressively or even violently. Humans, according to

Goffman, signal frame in a similar way and for similar purposes, though Goffman proposes that our

system of framing is more complex than a dog’s or an otter’s allowing for us to frame a wider variety of

social contexts.

It is important to note that many behaviors intended to introduce a certain frame must be

offered and accepted. This allows all participants to reassure themselves that everyone else is “keyed-

into” the same frame (Glenn & Holt, 2013). For instance, if one dog growls playfully and another ignores

her, a playful frame has been offered, but not accepted, and therefore a playful frame has not been

ratified between the two participants in the interaction. Similarly if a one person laughs and the other

does not (a situation often associated with hurt feelings) a playful frame has been offered and not

accepted.

10
 This definition of laughter, as a nonverbal cue implying willingness to enter into a playful frame,

satisfactorily explains many of the puzzling observations researchers have made about laughter in

interaction. First of all it explains why laughter is almost exclusively produced in social settings and is

produced more often with a group of friends than a group of strangers. People do not need to signal

play to themselves and people do not often wish to play with people they do not know well (Glenn,

2003; Provine, 2001). People may laugh when they are nervous or embarrassed in order to indicate that

they are not taking an activity seriously and that therefore their failures need not be taken seriously

either. Often a group of people will laugh when bullying a non-laughing victim; they will play with

someone who does not want to be played with. People laugh when they find something funny so as to

indicate both that they understand the playfulness of the laughable, and to welcome others to play with

them.

Laughter in face-to-face conversation therefore provides illocutionary force, just like many other

forms of nonverbal communication. By laughing the laughter indicates a desire for his or her words to

be interpreted as play.

When we communicate over the internet, we need to communicate that play is underway just

as we do face-to-face. If laughter is truly a marked, communicative act, it would not be surprising were

we to find that the same work is being performed through some other means online. In searching for

the conversational mechanism that indicates the offer and acceptance of play, the obvious first step is

to look to written laughter. In the following section I will lay out some of the findings from the limited

previous research which has focused on written laughter.

2.2: On Laughing with Letters:

The majority of the information which has previously been collected about written laughter was

collected in studies focused on understanding other aspects of online communication. Studies which

11
have provided relevant information about written laughter tend to fall into two categories: studies of

emoticons and studies attempting to identify user’s “latent attributes” (gender, race, age etc.) based on

language use.

2.2.1: Studies of Emoticons

The development of emoticons is, historically, very recent. Most accounts date the first

emoticons back to 1982, when they were supposedly created by Scott Fahlman on a Carnegie Mellon

message board (Krohn, 2004). It is unsurprising, therefore, that research on the topic has only just

begun to mature.

An emoticon is traditionally defined as a series of ASCII characters (letters, numbers,

punctuation, etc.) arranged in such a way that they seem to create either sideways or head-on images of

facial expressions (Dresner & Herring, 2010). A few authors who have written on emoticons have also

included certain emotive netspeak abbreviations (such as lol or omg) in this category as well (Krohn,

2004), but theirs is a nonstandard and largely metaphorical use of the word.

The standardization of Unicode has expanded the inventory of possible emoticons in most

online environments to include single character images of facial expressions. Unicode also makes other

single-character images available, such as trees, beer mugs, and small animals. Unicode images as a

whole are known by the Japanese name for emoticons: emoji. Non-facial emoji are typically used either

as decorative additions to a text message or else in making complete statements such as “let’s grab

drinks” (Bennett, 2014). The distinction between traditional emoticons (which generally, though not

always, represent facial expressions) and emoji, especially non-facial emoji, is an important one to

make, but as this is a subject which is neither well-studied nor particularly relevant to the topics under

discussion, I will not delve into it here. In this paper I will use the two words interchangeably.

Etymologically, the word emoticon is a blend of the words “emotional” and “icon” (Dresner &

Herring, 2010). The majority of work on emoticons has assumed, perhaps partly because of this name,

12
that emoticons are, essentially, depictions of a user’s emotions; happy, sad, uncomfortable, etc. (Derks,

Bos, & Grumbkow, 2008; Provine, Spencer, & Mandell, 2007; Walther & D'Addario, 2001). While it is

true that often emoticons are used as stand-alone expressive acts, they are more often used to attach

intent to a textual utterance, such as ‘flirting,’ ‘joking,’ or ‘disapproving’. These intensions have been

called emotions by some (Walther & D'Addario, 2001), but they are much more accurately described in

terms of Speech Act Theory, as illocutionary force indicators (Dresner & Herring, 2010).

In their particularly insightful 2010 paper, Dresner and Herring discussed the illocutionary force

of emoticons. They propose, in accordance with Social Information Processing theory of computer-

mediated communication (CMC), that though emoticons may not map to facial expressions, they may

accomplish similar communicative functions to those accomplished by smiles or winks (Walther, 2006).

That is to say that while the same information may not be conveyed by a smiley-face that is conveyed by

a smile, they convey the same type of information, namely they provide information about a speaker’s

intended meaning. Dresner and Herring also note that this type of information can be conveyed, to a

certain extent, in more traditional textual styles either though textual statements of intent (ex: “just

kidding”) or through traditional punctuation (ex: “!” or “?”). They therefore claim that emoticons are a

sort of internet shorthand which likely makes use of our pre-existing knowledge of nonverbal facial cues.

Though lol, haha, and other similar forms of online laughter do not qualify as graphical

representations, and are not often treated as emoticons, they have been referenced as pragmatic

particles which communicate illocutionary force (Curzan & Mejia, 2012; McWhorter, 2013). In other

words, lol carries social information about speaker intent in much the same way as an emoticon might.

One of the goals of this study was to establish the degree of similarity between different forms

approximating the same face-to-face behavior, by examining abbreviated, onomatopoeic, and emoticon

forms of laughter.

13
2.2.2: Studies of “Latent Attributes”

Though small studies and theoretical proposals such as those put forth by Dresner and Herring

are evocative, the literature on emoticons is, as yet, not very extensive. The literature on other forms of

laughter is even sparer.

Many studies of online communication focus on identifying what are referred to as the “latent

attributes” of users (gender, race, age etc.) (Weller, Bruns, Burgess, Mahrt, & Puschmann, 2013). Such

studies attempt to model the density and types of words and emoticons used by individuals of different

groups in such a way that these groups can be identified by language use alone (Bamman, Eisenstein, &

Schnoebelen, 2014; Lyddy, Farina, Hanney, Farrell, & O'Neil, 2014; Xy, Yi, & Xu, 2007). This information

is particularly valuable to advertising companies that might wish to target a class of individuals in an

anonymous setting. These quantitative studies, largely conducted on Twitter, have discovered that

females use far more emoticons and “emotive” words, including various spelled-out laughter forms,

than do males (Bamman et al., 2014). The work in this vein has been very successful in its practical

goals, achieving greater accuracy in the identification of user gender than humans surveying the same

data. However, in only a few of these studies have results been used to make theoretical claims about

gender performance or gendered language use online (Bamman et al. (2014) being an exception).

Most word-frequency studies summarize trends in their results by placing word forms into

categories, such as “emoticons,” “numbers,” or “hashtags.” Though many of these studies mention

forms of written laughter, none include analyses of “written laughter” as a separate category of analysis.

Rather, these forms were typically separated into three categories; emoticon, abbreviation, and

onomatopoeic (or slight variations on these). This means that lol and lmao are often lumped-in with

forms like bff (best friends forever) or omg (oh my god); haha and hehe grouped with AAAAA! and ouch;

😂 and 😄 were treated as part of the same category as 😢 or 🌴. Though these forms are placed into

14
these categories for a reason, as generalizations can be drawn across these categories, this is the first

work to have examined forms approximating laughter across these boundaries.

3: Laughing with Letters: Native Intuitions

The following two sections provide some information about the six laugh forms in question, lol,

lmao, haha, hehe, 😂 and 😄 and about online laughter in general. Most of these laugh forms, which

were chosen both because they are common and because they fit into several different categories of

“emotive” netspeak expressions, are relatively new contributions to the English language. Where a

project looking at older, more established, or more formal lexical items might turn to a dictionary or

other “official” source for information on the history and meaning of a word, the sources available here

tend to be informal and largely crowd-sourced. The major source used in this section is Urban

Dictionary, an online, crowd-sourced dictionary of slang. While definitions on Urban Dictionary are

crowd-sourced in that any internet user may provide definitions for words, several studies have used

the site as a source for definitions of slang terms (Smith, 2011). Other sources of information include

blog posts, humor articles, and other online media content. While the information taken from these

sites is not academic in nature, it can provide native-speaker intuitions about how these forms are used.

An individual may not think technically about an particular word, but so long as he or she is able to use it

properly that individual has some sort of knowledge about the word’s proper use.

The first section below presents form-dependent information, providing definitions for each of

the laugh forms under observation taken from various sources. In the second section these form-

dependent definitions are drawn together to outline some general intuitions about written laughter.

15
3.1: History and Practices for Six Written Laugh Forms:
3.1.1: Lol
The first documented usage of the initialism, lol, meaning “laughing out loud,” was published in

a 1989 FidoNet newsletter, though by that point the abbreviation had likely been in use in various online

environments for some time3 (Brandon, 2008; Edel, 1989; Hiscott, 2014; Pearson). As more people

began using the internet, this term became both more active and more widespread. It has now even

made the transition to spoken communication, where it is generally pronounced like the word “loll,”

rather than as an initialism (McWhorter, 2013). Lol has since been the basis of numerous lexical

creations, including “lolzfest,” “loller-skates,” and the infamous “lolcats”4 (Morgan, 2011).

The Oxford English Dictionary provides two definitions for lol. The first defines it as an

interjection “used to draw attention to a joke or humorous statement, or to express amusement.” The

second defines it as a noun which is “an instance of the written interjection ‘LOL’.” ("LOL, int. and n.2,"

2015).

The users who have posted about lol on Urban Dictionary over the years have made some other

relevant observations about the meaning of lol. The quotations in 1 have been drawn from several

users’ definitions of the word.

1) Urban Dictionary observations on / definitions of lol:

a. “Now, [lol] is overused to the point where nobody laughs out loud when they say it.
In fact, they probably don’t even give a shit about what you just wrote. More
accurately, the acronym “lol” should be redefined as ‘Lack of laughter.’”
(no_one_2000, 2005)

b. “lol – originally meant “laughing out loud”, but now is the most common expression
in any text conversation, just used instead of HAHA or any giggle or something like

3
This abbreviation was also commonly used in handwritten letter-writing. In this medium, however, the
abbreviation stood for “lots of love” or, according to some, “lots of luck.” (Pearson)
4
“Lolcats” being humorous photos of cats in costumes or silly poses subtitled using intentionally off-kilter spelling
and grammar

16
 that. Also used all the time when there’s nothing else to say… LOL key should be
added to a standard keyboard.” (dude, 2005)

c. “Lol is most commonly used as a silence breaker, a reply to a joke that is SUPPOSED
to be funny but really isn’t, or an answer to an uncomfortable or random statement
that one couldn’t think of a better response to.” (EXPLIZIT, 2005)

The use of lol is extremely commonplace in internet communication, so commonplace, in fact,

that even by 2005 it had already lost a good deal of its power. John McWhorter has argued that lol has

begun to be used as a more general “marker of accommodation,” rather than as an actual indication of

laughter (McWhorter, 2013). The perceived lack of connection between physical laughter or even

humor and the use of lol is clearly an important part of its present-day meaning, and will be discussed at

greater length below.

3.1.2: Lmao
Though the initialism lmao, standing for “laughing my ass off” is immediately identifiable by

most users of the internet, it is neither as widespread nor as commonly used as lol. This initialism has

therefore received less attention from dictionaries and in the media. Lmao and lol had similar origins in

early online chat groups. A similar, though not identical abbreviation, lmto (“laughing my tush off”) can

be found described in the same 1989 article that represents the first written evidence for lol (Edel, 1989;

Hiscott, 2014). Mike Vuolo dates the first recorded instance of lmao to a 1990 online Dungeons and

Dragons game (Vuolo, 2013).

Unlike lol, lmao has not yet won a place in most formal dictionaries. But several Urban

Dictionary definitions are given in 2, below.

2) Urban Dictionary observations on / definitions of lmao:

a. “Used when the situation is considered funnier than a mere lol. I’ve also heard this
said aloud in conversation pronounced: La mayo.” (TheDanishSugarfairy, 2008)

b. “We use [lmao] when something’s funnier and deserves more than a capitalized “lol”
Sometimes people say it in person too, but that’s only when they’re trying to be
funny” (kirshteeen, 2009)

17
 c. “A chatroom acronym used exclusively by morons, meant to stand for “laughing my
ass off”. Generally typed in response to deeply unfunny remarks, and also used to
feign nonchalance when taunting someone.” (Alan., 2006)

The word lmao seems to carry an implication that a comment was funnier than might have been

implied had the user typed lol instead, but definitions like 2c still suggest a disconnect between typing

laughter and actually laughing physically.

3.1.3: Haha
Unlike the two forms discussed above, both of which clearly trace their origins to online

communication, haha is a word that has been around for thousands of years. Versions of haha can be

found in various Indo-European languages dating back to Ancient Greece (Provine, 2001). The first

record of it being used in English can be found in Ælfric’s Grammar of Old English, written around 995

(Baron, 2011; Menzer, 2004).

All that the Merriam-Webster has to say about ha-ha, is that it is an interjection “used to

express amusement or derision” ("ha-ha," 2015). This form is found in a large number of dictionaries,

but always listed as two words (ha ha) or as a hyphenated construction (ha-ha). Some Urban Dictionary

definitions can be found in 3.

3) Urban Dictionary Observations on / Definitions of haha:

a. “Short way to let a person know over text that they are laughing/thought something
you said was funny. It however doesn’t really reflect how much they are actually
laughing/how much they actually thought something you said was funny.”
(Entity1037, 2015)

b. “Used to express laughter anywhere you can’t say it, for example IM, instant
messaging or via email. The minor difference between haha and hehe is that haha is
often used when laughing at someone, while hehe is used while laughing with
someone. It might be unconscious to most people, but it’s true.” (Løkken, 2006)

c. “A form of expressing laughter when oral expression is not available, like on the
internet… Haha is almost never used to express actual laughter occurring. For those
circumstances, rofl [rolling on the floor laughing], lmao, or a whole sentence about
how the comment actually made one laugh is used.” (rikochet, 2008)

18
 In addition to the intuitions evident in the quotations above some users in more recent

definitions stressed the superiority of haha to lol. These claims tend to identify lol as juvenile. Others

were careful to include specific and different definitions for various orthographic variations of haha as

well.5

3.1.4: Hehe
Hehe is closely related etymologically to haha, as both are onomatopoeic. In fact, most formal

definitions include hehe as a spelling variant of haha ("ha-ha," 2015). The users who write definitions on

Urban Dictionary, however, seem to see strong distinctions between the two forms.

4) Urban Dictionary Observations on / Definitions of hehe:

a. “muffled laughter, suggesting a sneaky aspect to that being laughed at, differs from
lol in this way, which is a full on belly laugh.” (monsieur_d, 2005)

b. “hehe, different from lol or haha. Hehe usually has some type of innuendo. It is a
subtle way to flirt via texting or instant messaging.” (guy12345, 2009)

c. “A somewhat irritating giggle. You may find “hehe” pop up in various conversations
in texts and ims. Many girls say this because it’s a step cuter from the original
“haha”.” (paperstars, 2009)

In general, hehe seems to carry both mischievous and diminutive or feminine implications.

These implications separate hehe from more common laugh forms like lol or haha.

3.1.5: 😂 “Face with Tears of Joy”: 1F602

😂, also known as “Face with Tears of Joy,” or by its Unicode notation, 1F602, is the most

commonly used emoji on Twitter ("Face with Tears of Joy," 2015; Rothenberg, 2015). This form was

5
ricochet, the author of 3c, for example, gives the following additional definitions:
“Hah – The person thinks a comment is mildly funny.
Haha – The person does not think the comment is funny, but acknowledges your attempt at humor.
Hahaha – the person thinks the comment is funnier than average, or is just more enthusiastic than those who
normally say haha
Hahahaha – The person is just being excessive. Usually used to strengthen a friendship because an inside joke was
just mentioned.”
Also see (Heaney, 2014a)

19
originally created to provide users with a means of expressing overwhelming joy. Specific information

about the use of emoji is less freely available than is information about the spelled-out forms of laughter

described above. This is partly because many websites like Urban Dictionary have not fully integrated

Unicode into their system of operations (therefore not allowing users to created definitions for more

unusual Unicode characters like emoji), and partly because the traditional dictionary model has not yet

been extended to include non-word characters. Still, some blogs and magazines have published “How

to Decode Emoji” articles attempting to describe the ways in which these emoticons are used and what

they mean. Some examples of definitions for this form can be found in 5 below.

5) The Meaning of 😂 (Various Sources):

a. “LMAO* LMAO* LMAO* LMAO* LMAO* (*May not actually be laughing my ass
off.)” (Heaney, 2014b)

b. “ Translation: ‘I’m not as happy as I was when I sent the sobbing emoji, but I’m still
happy.’ … Alternate uses: When you didn’t actually think something was funny and
decided to use the minimal amount of effort to avoid hurting someone’s
feelings.”(Toole, 2014)

c. “LEGIT LAUGHING OUT LOUD” (Sebastian, 2014)

Like many emoji, the use of 😂 in practice seems to differ considerably from its original

conception. Rather than standing for “joy” this emoticon seems to be used as a form of written

laughter. As with any Unicode character, the Unicode notation, 1F602, is expressed slightly differently

on different operating systems and programs. On Twitter the 1F602 is expressed thus: .

3.1.6: 😄 “Smiling face with Open Mouth and Smiling Eyes”: 1F604

😄, also known as “Smiling Face with Open Mouth and Smiling Eyes,” or by its Unicode notation,

1F604, is the first emoji to appear on most emoji keyboards. It is often considered to be a neutrally

happy emoji. Though most meanings do not directly associate it with laughter, 😄 is often described as

occurring in the same environments in which one might expect to find laughter.

20
 6) The meaning of 😄 (Various Sources):

a. “I’m not really into emoji, but I know you want me to be. This is the first one
available.” (Heaney, 2014b)

b. “The gigolo of emoji, this guy gets around our text chains, popping up to convey joy
or sarcasm or to test your limits. Nothing comes across as too heavy when
punctuated with 😄.” (Moss, 2014)

c. “Hearty-u-really made me smile… This is as genuinely pleased as i can be on a

particular day! So if am not dishing out many of those, well then that says a lot.”

This emoji has been included in this project because it falls somewhere between the traditional

smiling face emoji and the ‘Face with Tears of Joy’ emoji discussed above. Like other emoji, 😄 is

expressed differently on different systems. On Twitter it looks like: .

3.2: General Observations on Written Laughter

The generalizations drawn in this section are based on observations made by native-speakers of

an online “dialect.” They may not prove completely accurate, but the observations made by native

speakers can serve to point the way to substantive research questions.

First of all, several sources report that the different ways of writing laughter are not completely

equivalent. Some are described as implying a certain sort of humor (hehe for example, implies flirty

humor or innuendo). Others are described as being used preferentially in certain social environments

(3b claims haha is used for ‘laughing at’ rather than ‘laughing with’), or used primarily by certain people

(2c and 4c both make claims of this sort). While different written laugh forms are similar in some

respects, they also have some individuality.

Second, while most forms of written laughter are meant to imply that the user who typed the

laugh form is physically laughing, many users report that they and others use them when this is not the

case. This observation was made several times above for multiple forms (1a, 3a, 3c, and 5a). Appendix

1 also provides some examples of this observation being made in multimedia humor.

21
 Third, while most forms of written laughter seem meant to imply that the user who typed the

laugh form thought something in the previous tweet was funny, they are often used in response to

remarks the user does not find humorous. This observation was made above for several forms (1a, 1c,

2c, 3a, 5b), and has also been made by several linguists (Curzan & Mejia, 2012; McWhorter, 2013).

Of these generalizations, the last two seem to make any association between spoken laughter

and written laughter unlikely. If written laughter is neither universally associated with physical laughter,

nor universally associated with humor, it seems as though the two cannot be accomplishing the same

things.

The literature cited above, however, has made it fairly clear that laughter is not strictly

associated with humor and that the purpose of laughter is communicative. Laughter is the means by

which one expresses his or her desire to carry out a nonserious or playful interaction. When using the

internet, expressing this desire out loud, through physical laughter, would be pointless. For all intents

and purposes, an individual carrying on a conversation with others over the internet is sitting alone. Any

physical laughter he or she produces would be akin to physically talking to co-participants far away and

unable to hear. If one wants to communicate the desire for nonseriousness to co-participants who are

not co-present the desire must be communicated through other means. It is thoroughly possible that

the “other means” used on Twitter and other short message systems, from the telegraph to the text

message, is written laughter.

4: Research Questions and Data

4.1: Research Questions
Is it appropriate to refer to lol, lmao, hehe, haha, 😂, and 😄 as written incarnations of laughter?

These forms were originally meant to represent physical laughter, but might they instead be a substitute

for physical laughter? This question has two prongs. The first is meaning-based. It asks whether or not

22
physical laughter and written laughter have the same meaning in a conversation. If face-to-face

laughter indicates a desire to interpret a conversation as non-serious or playful, can written laughter do

the same? The second prong is based in conversation analysis. This question asks whether the same

rules govern the use of written laughter in conversation as govern the use of physical laughter in

conversation. Do we see offer/acceptance patterns in written laughter as we do in face-to-face

laughter? Are those patterns influenced by the gender of the conversational participants as well? Both

prongs are essential to any argument that written laughter is truly a substitute for physical laughter in

the written medium. This thesis, however, will focus on the second: on the placement of laughter in

written conversation.

In order to establish whether or not we draw on our knowledge of laughter in interaction in

order to use it in the written medium, it is first necessary to understand the ways in which laughter is

used conversationally online. The analyses described below are all attempts to answer the question of

how written laughter is used in conversation. As one of the goals of this project is to establish whether

or not we draw on knowledge of spoken laughter to produce written laughter, these analyses are

largely, though not universally, geared towards determining whether or not observations which have

been made about the use of laughter in face-to-face conversation can also be made about the use of

written laughter online.

4.2: Data
This project uses data drawn from the social networking site Twitter. Twitter is what is known

as a “microblogging” site. This means that users create accounts which they use to post information

about whatever they choose (personal life events, cat videos, news, recipes, etc.) completely publicly.

Posts, limited in length to 140 characters, are known as “tweets,” and may include various sorts of

metadata tags. Tweets may include what are called “mentions,” which use the syntax, @username, to

23
call another user’s particular attention to a tweet. They may also use “hashtags,” which use the syntax,

#hashtag. Hashtags allow other users searching for tweets on a particular topic to find those tweets

quickly, but they are also used rhetorically to provide meta-commentary on the underlying subject or

meaning of a tweet (Yang, Sun, Zhang, & Mei, 2012).6 Users may also include hyperlinks, images, videos,

and various other forms of multimedia in tweets. All forms of text, however, count towards the 140

character length restriction. A diagram showing the layout and appearance of a tweet can be found in

Figure 1.

Figure 1: Layout of a Tweet

Data were collected using Twitter’s public streaming API (Application Programming Interface)

("Public API," 2013). An API is a “door” built into a website by its creators, which allows developers to

use the site for their own purposes. Twitter’s streaming API was constructed to allow developers to

take and store limited amounts of data and access them in more flexible ways than are made available

on the official Twitter website. Most major social networking sites provide APIs, but Twitter’s is

especially good for linguistics research because of the public nature of posting on the site. Other APIs,

such as Facebook’s, may allow researchers access to information about a user’s friendship network, but

their privacy settings are such that the text of posts or messages sent between users cannot be

6
For example, in: @meagooon: “@Cynthiaaaxo LMAO I'm smiling thinking about it. Can't wait to love yew all day
tomorrow in zeeee lib #studiousbitches”, the hashtag #studiousbitches is likely not intended to help other users
interested in “studious bitches” to find this tweet. Instead it provides a slightly humorous commentary on the
tweet it is a part of.

24
retrieved (Weller et al., 2013). However, since posts made on Twitter are automatically public, Twitter’s

API can return information on language use not made accessible by other APIs.

There are several avenues, provided by Twitter and data-mining companies with access to

Twitter’s databanks, through which Twitter data can be accessed, but Twitter’s free streaming API is the

most effective for small research projects such as this (Weller et al., 2013). This API was used, for

example in Bamman, Eisenstein and Schnoebelen’s 2014 study of gendered language on Twitter

(Bamman et al., 2014). This API has an effective “search” function, which allows a researcher to open a

stream of all tweets containing a certain word, phrase, or set of words and phrases. From the time this

stream is opened to the time it is closed these tweets are returned in real time, as they are posted to

the site. In addition to returning the text of the tweets themselves, the API returns a wealth of

metadata about the users involved in a tweet exchange. Most importantly for this project the API

returns information about the tweeter’s volunteered name, username, and location, a Twitter-provided

timestamp, and information about whether or not the tweet is a retweet, whether or not the tweet was

a reply, and a certain amount of user information any users mentioned in the body of the tweet ("The

Streaming APIs: Documentation," 2014; Weller et al., 2013).

The free streaming API is intended for small-scale use only, and therefore suffers from many of

the same limitations as most forms of free software. The limitations of the free streaming API are as

follows. Firstly, all APIs are limited by the data provided to the company by users upon subscription. If a

user wants to give a false name or a false location, Twitter allows them to do this. Therefore though the

API can allow one access to who and where a tweeter claims to be, this information cannot be taken as

fact. In many instances users will state that their location is, to name a few examples, “at Hogwarts” or

“lost in the now.” Users of Twitter are asked, but not required, to state their gender, and are not asked

to state their racial background upon registration. Gender and race information is not available through

the streaming API. Secondly, because one of the major ways that companies like Twitter turn a profit is

25
by selling data, they limit the amount of data that can be extracted using the streaming API. The free

streaming API only allows a user to channel 1% of the total information flow on Twitter at any given

time. Higher-level permissions can be purchased, but were not necessary for this project. If the 1% cap

is, at any point, exceeded, Twitter simply returns a random sample of tweets matching the search

criteria ("The Streaming APIs: Documentation," 2014; Weller et al., 2013). As a random sample is all that

was needed for this project, this cap did not present an issue for data collection.

Initial data collection took place in November of 2014. Using the streaming API, and a filter

which removed automatic retweets, 1000 examples of tweets containing lol, and 250 examples of

tweets containing lmao, haha, hehe, 😂 and 😄, were collected in samples extracted five times over the

course of one Saturday. Initially the dataset for lol was larger than the datasets for the other five forms

under observation because lol was to be the sole topic of this paper, with the other forms merely

providing context. Once the analysis began, however, the concept of this paper began to shift and it

became necessary to conduct additional collections.

These collections were necessary for two reasons. Firstly, a few of the analyses in the sections

which follow require large datasets in order to provide statistically powerful results. In order to provide

comparisons between lol and some of the other laugh forms under observation the datasets for haha

and 😂 were expanded to be an equal size. Due to time constraints and the amount of effort required to

hand-tag datasets, the samples for hehe, lmao, and 😄 were left at 250 and were simply left out of some

of the comparisons. The second reason for additional collections was that the initial collection had not

provided a comparison group: a random sample of tweets with which to compare some of the traits of

tweets containing written laughter were emerging. Additionally, some of the analyses required a

comparison group of tweets specifically not containing laughter.

A random sample group of 600 tweets (a number large enough to provide powerful results for

the analyses which needed the sample) was therefore collected in one block on a Monday in January of

26
2015. Also, an additional 750 examples of tweets containing haha were collected in one block on a

Wednesday in January of 2015, and an additional 750 examples of tweets containing 😂 were collected

in one block on a Wednesday in February of 2015. Though this collection of tweets on varying dates was

not ideal, it does not challenge the validity of this study. The difference in collection dates was only a

few months, likely not a long enough time for the use of forms as established as these to undergo major

changes in usage. The goal of this project was broadly to understand the ways in which these laugh

forms are used on Twitter, and these late collections are still able to serve that purpose.

The result of these collections was seven distinct datasets, an unfiltered sample of tweets, and

one dataset of tweets containing each of the six forms in question. After examples for which the desired

form was included in a manual retweet rather than in the new contribution portion of the message were

filtered out, the resultant datasets were as shown in Table 1. Retweets both manual and automatic7

were filtered out of the datasets because they could not be either attributed to the original posters or

located within their original conversational context. As tweeter-identity and conversational contexts

were both relevant variables in the analyses below these tweets could not be used. However, those

tweets containing manual retweets for which the dataset laugh form rested in the part of the tweet

which did not constitute a quotation were used. These will be discussed in greater detail below.8

In addition to the text of each tweet, the user-provided name, user “handle” (address), and

location along with the timestamp of the tweet were automatically added into the dataset. There was

only one instance, in the lol dataset, in which the same tweeter contributed more than once to a

7
Recall, however, that automatic retweets were filtered out on collection.
8
See Section 5.4

27
dataset. The second instance was removed. This collection therefore provides data about broad usage

rather than the repeated practices of a small set of individuals.

In some of the analyses below it became necessary to filter these datasets still further. Most

commonly, the unfiltered dataset was reduced to exclude tweets which contained any form that might

possibly represent written laughter, creating a “no laughter” dataset. Where the laugh forms were

spelled-out with letters the identification of these forms was simple. The forms which are considered

spelled-out laughter throughout this paper are: lol, haha, lmao, lmfao, jk, hehe, hoho, ahah, huhu,

/chuckle, *giggle* and orthographic variations of these. Identification of emoticon forms of laughter

was more difficult. As there is no set list of emoticons which can and cannot be used to represent

laughter, broad identification mechanism which likely over-identified forms was used. Any emoticon

form with an upturned mouth was considered emoticon laughter. Throughout this paper the following

forms, along with ASCII variations on these, will be referred-to as “emoticon laughter”: 😄, 😍, 😂, 😋,

😎, 😏, 😅, 😉, 😊.

Often the laughter datasets were filtered down only to examples of what will be referenced

here as isolate laughter. Tweets containing isolate laughter contain one and only one instance of the

dataset laugh form and no other instances of either emoticon or spelled-out laughter. Instances of

isolate laughter are desirable for two reasons. First, many of the comparisons which follow will make

comparisons between the behavior of individual laugh forms. If more than one laugh form is present

within a tweet these other laugh forms may interfere with data gathered for these comparisons.

Second, many of the comparisons which follow use the position of a laugh particle within a tweet as a

variable. If two laugh forms exist at different locations within a tweet this might similarly interfere with

results. Tweets with isolate laughter constitute over 90% of the total sample.

In addition to the intra-utterance usage of online laughter words, a major concern of this project

was the conversational context in which written laughter could be found. Among other things, this

28
context provides opportunities to seek out the offer/acceptance patterns seen in face-to-face laughter.

That being the case, it was also necessary to collect any conversational context surrounding the use of a

laughter word. Though Twitter’s API has the ability to collect a certain amount of information about the

tweet directly preceding a particular tweet in context, it cannot provide information about subsequent

conversational turns, as tweets are collected in real time. Subsequent context must therefore be

retrieved at a later date. Conversational context for each tweet was collected some time after major

data collection, using the “view conversation” function on Twitter’s website. As the layout and profile

pictures associated with each user was also desired for reasons discussed more extensively below,

conversational context for each tweet was collected as a screenshot. Though many of the tweets found

in these conversational contexts did contain laugh forms, these context tweets were added to the

original datasets only as they pertained to the original data collections: as metadata. Each tweet was

therefore connected to a single screenshot of its conversational context, and each screenshot of a

conversational context was connected to a single tweet.9

These seven datasets were tagged for several variables, and these variables were analyzed in

terms of frequencies. As each method of tagging is specific to an individual analysis, these tagging

methods are described analysis by analysis in the sections which follow.

5: Analyses and Results

The following sections represent the various analyses which were carried out using the datasets

described above. Section 5.1 examines the overall frequency with which various laugh forms occur on

Twitter. Section 5.2 examines the frequency with which various laugh forms co-occur with three of the

forms under observation (lol, haha, and 😂). Section 5.3 examines the influence that the presence of

9
This will be discussed in greater depth in the introduction of Section 5.6 below

29
written laughter within a tweet has on that tweet’s likelihood of being designed for a specific recipient.

Section 5.4 models the placement (initial, medial, final, alone) of laugh particles within a tweet. Section

5.5 concerns the association which was uncovered between tweet-initial laughter and specific recipient-

design, and some possible implications of this association. Section 5.6 contains several analyses which

attempt to unpack this connection by examining instances of tweet-initial and tweet-final laughter in

conversational context. First 5.6.1 examines the possibility that initial-position laughter refers back to

previous or ongoing topics, and second 5.6.2 examines the possibility that initial-position laughter

functions more often as an acceptance of offered laughter. Section 5.7 constitutes several gender

analyses, 5.7.1 being an analysis of whether or not individual laugh forms are tweeted more often by

males or by females, and 5.7.2 examining the possibility that offer and acceptance of written laughter is

influenced by gender.

5.1: Overall Frequency

The unfiltered dataset which was collected represents a random sample of tweets.10 In order to

obtain an initial idea as to the relative frequencies with which different written laugh forms are used,

instances of laughter in this dataset were counted and graphed in Figure 2. Several forms of laughter

which were not found in the unfiltered dataset, but were found in other datasets were also included

here.

Selection of forms considered to be “written laughter” for this analysis was intentionally broad.

As little is known about the nature of written laughter, there was little information on for creating an

informed definition of the term. Therefore this analysis considers any of the forms listed as either

“emoticon” or “spelled-out” laugh forms in Section 4.2 above to be written laughter.

10
It is unclear exactly the extent to which the sample returned by the Twitter API is actually random, as the precise
algorithms used by the API are private, but this functionality is advertised as a random sample and has been used
by numerous studies as such (Bamman et al., 2014; Weller et al., 2013).

30
 Figure 2: Frequencies of Various Forms
4.50%
4.00%
3.50%
3.00%
2.50%
2.00%
1.50%
1.00%
0.50%
0.00%

Notice that all forms of written laughter listed in Figure 2 occur in fewer than 5% of tweets.

Written laughter, while a familiar feature of discourse on Twitter, is not a feature present in a majority

of tweets. 😂 and lol are the two most frequent laugh forms.

5.2: Co-Occurrence
Most of the analyses presented in the sections which follow will require that laugh-form

datasets be reduced only to examples of isolate laughter.11 There were, however, a relatively small

number of tweets collected as part of each filtered sample which could not be used in these

comparisons as they contain more than one instance of written laughter. This section is devoted to

examining the frequencies with which various non-dataset laugh forms co-occur with three of the pre-

specified forms of laughter (lol, haha and 😂).12 Figure 3 displays the frequencies with which the forms

11
Recall the definition of this term. A tweet containing isolate laughter contains one and only one instance of a
pre-specified laugh form and contains no other laugh form as specified by the list, 😄, 😍, 😂, 😋, 😎, 😏, 😅, 😉, 😊,
lol, haha, lmao, lmfao, jk, hehe, hoho, ahah, huhu, /chuckle, *giggle*, and orthographic variations on these.
12
In examples like the following:
1) @jeeessiiccaaa_: “@TheAmbitiousz1 that's too far haha the drivethru😂😂 wow okay lmao”
2) @Blake41Taylor: “@JessicaManuel95 haha so your mad because your so called number one team is
going to get beat? Maybe a shout out? Haha”
3) @yes_its_james: “@pincheehacobb @CesarB_58 I am lol, no one listens to me haha”

31
32
lol, haha, and 😂 co-occur with each of the forms specified for the frequency analysis in 5.1. Notice that

an unfiltered column is also included. These frequencies are the same as those shown in Figure 2 above.

Haha and to a lesser extent lol seem to co-occur with other laugh forms, particularly emoticon

laugh forms, at a higher rate than those forms would be expected to occur in an unfiltered sample. The

presence of either of these two forms increases the likelihood that another laugh form will also be

present. While the presence of 😂 in a tweet seems to reduce the likelihood that another emoticon will

also be present, its presence does powerfully increase the probability that a spelled-out laugh form will

be present.

5.3: Recipient Design

Twitter can be used either as a microblogging platform or as a short message service (SMS)13.

Historically and canonically Twitter has been used as a microblogging platform. This means that a

tweeter posts as one might to a bulletin board. Messages are public and undirected. Any user may read

any tweeter’s content, but readers may pick out relevant content by “following” specific tweeters. This

allows them to receive only those tweeters’ content on their digest page. A consequence of this is that

those who post to the site are constantly conscious that their main audience consists of those

individuals who have chosen to follow their accounts.

Twitter’s SMS functionalities were added shortly after the site’s original publication. A user

desiring to use the site as a short message service will “mention” other users, typically at the beginning

of the tweet, using the @username function (see Figure 1 above). Though these messages are publically

accessible, so long as the tweet begins with the mention, the message will not be broadcasted

automatically to all of the tweeter’s followers. Instead it is sent only to the users mentioned in the

13
Short-message Services are a broad category of internet communication mechanisms. The canonical example of
an SMS is text-messaging, though any instant messaging service where messages are short and delivered (nearly)
instantaneously also fall into this category.

33
tweet. The recipients are then notified in much the same way as they might when receiving a text

message or email. Often, though not always, this targeted mentioning function is used much like a text

or instant messaging service, allowing users to carry out semi-public conversations over the site.

The following examples show tweets composed using the microblog and SMS formats:

7) Nonspecific Recipient: “Microblog” Format:

a. [@mitchieee47]: I hope I remember shit after this bar lmao
b. [@_Gabby915]: Im gonna get so many toys hehe
c. [@dabuekkehabda_x]: this has been me for the last three hours. haha makes me
laugh bc it's so accurate. :)) http://t.co/fvJB2nrEWG

8) Specific Recipient: “SMS” Format:

a. [@ItsOnlyZeek]: @natalietaylor96 haha you're fine that dude followed me today or
yesterday I think
b. [@Zefoh]: @SwitchSP @SPG_AX why would you leave the team to eeko lol
c. [@Lindsay_Hecker]: @dawncflv @WriterAJCarroll @leannrimes @KarlaHoffman
@kpsander @Christina4jc @danielfgalvan and it's a requirement for us to know it. lol!

These two functions provide an interesting avenue of inquiry from a conversation analytic

perspective. The microblogging format allows users to publish tweets which are not designed for a

specific recipient. While there is some level of recipient design, as the tweeter typically knows

something about his or her pool of followers, tweets using this format are directed at a broad and

flexible audience. Tweets published using the SMS format, however, are by definition specifically-

targeted. Users tweeting in SMS format design their tweets to be received by a specific individual.

When evaluating the extent to which written laughter practices draw on our knowledge of

spoken laughter, the distinction between specific and nonspecific recipient design is relevant for several

reasons. Chiefly, it has been shown that the presence or absence of others, as well as the identities of

those others, are major factors in the production or non-production of spoken laughter. More laughter

is produced if others are present, especially if the individual under observation considers his or her

compatriots to be friends. If written laughter is used in a way that is at all analogous to spoken laughter

we would expect to see more laughter in situations in which co-participants are “more present.” A user

34
who publishes tweets in a microblog format knows that his or her tweet will likely be read by someone

in his or her pool of followers, but does not know exactly who will read it or when. The recipient is

abstract. A user who publishes a tweet using the SMS format, however, knows with a reasonable level

of certainty that the mentioned individual (generally someone with whom the tweeter feels comfortable

conversing) will actually receive and read his or her tweet. Specifically-targeted tweets give the tweeter

a greater impression that he or she is actively communicating with others. If written and spoken

laughter do behave similarly, we would therefore expect to see tweets containing written laughter occur

more frequently than tweets without written laughter in contexts of specific recipient design. This is, in

fact, what we see.

Figure 4: Recipient Design - General

Specific Nonspecific

80%

70%

60%

50%

40%

30%

20%

10%

0%
Haha Hehe Lol Lmao 😂 😄 No Laughter

Figure 4 displays the frequency with which tweets containing each laugh form take specifically-

targeted (containing @username mentions) and nonspecifically-targeted (not containing @username

mentions) formats. Of the two formats, tweets containing one of the six laugh forms under observation

seem to be as or more likely to be specifically-targeted as they are to be nonspecifically-targeted. This is

in contrast to tweets which do not contain written laughter, which use nonspecifically-targeted format

35
with a frequency which is almost two times greater than the frequency with which specifically-targeted

format is used.

A chi square test of independence was performed to examine the apparent relation between

the presence of a laugh particle within a tweet and specific recipient design, and a relation was shown.

Tweets containing no form of written laughter were significantly less likely to be designed for a specific

recipient than were tweets containing written laughter x2(1, N=3,765)=130.37, p<.01. This association

between specific recipient design and the presence of written laughter, suggests that written laughter,

just like spoken laughter, is used in environments in which other individuals are more “present.”

Written laughter is less likely to be used by tweeters in relative social isolation.

Another interesting pattern can be seen in Figure 2 as well. Tweets containing emoticon laugh

forms, while more likely to be designed for specific recipients than are tweets containing no laughter

whatsoever, also seem to be less likely to be specifically-targeted than are those tweets containing laugh

forms which are “spelled-out” with letters (i.e. haha, hehe, lol, and lmao). A second chi square test of

independence was performed in order to determine whether or not there was a relationship between

the type of written laughter and the recipient design of the tweets in which the form occurs. A

significant relationship was found, showing that, indeed, emoticon laugh forms are less likely to occur in

specifically-designed tweets than are spelled-out laugh forms x2(1,N=3,154)=101.31, p<.01. This

difference in the recipient-design distributions for spelled-out and emoticon laugh forms implies that

the two types may be used in accordance with different rules or for different purposes.

One final interesting observation which can be made in Figure 2, is that tweets containing haha

are more likely than are tweets containing any other laugh form to be specifically-designed. These haha

tweets are more than 15% more likely to target specific recipients than the form which is the next most-

likely to be used in specifically-targeted tweets (lol). This may suggest that haha is being used in a way

36
which distinguishes its use both from emoticon and from the other forms of spelled-out laughter (lol,

lmao, and hehe). This possibility will be revisited throughout this paper.

The results of this analysis are threefold. First, tweeters are more likely to target a tweet at a

specific individual if it contains written laughter than if it does not. Second, tweeters are more likely to

target a tweet at a specific individual if it contains a spelled-out laugh form than if it contains an

emoticon laugh form. Third, tweeters seem to be more likely to target a tweet at a specific individual if

it contains haha than if it contains any other form of written laughter under observation. Altogether

three subcategories of laugh forms seem to be emerging: emoticon forms, lol-like forms, and haha.

These categories seem to recur in several of the analyses below.

5.4: Laugh Locations within a Tweet

The analysis which follows examines the location of written laugh particles with regard to the

boundaries of a tweet. Essentially this analysis is intended to provide general information about the

distribution of written laughter within the boundaries of a conversational turn, embodied here by the

boundaries of a tweet. Occasionally individuals will tweet several times in a row, but in general the end

of a tweet implies the end of the speakers turn at talk.14

Those tweets containing isolate laughter15 (over 90% of the total sample) were categorized as

containing laughter that was either tweet-initial, tweet-medial, tweet-final, or alone. Hashtags,

hyperlinks, mentions, manually retweeted passages, punctuation and emoticons were ignored when

tagging a laugh form as tweet-initial or tweet-final, so long as these items were not integrated into the

syntax of the message. Examples 9-12 below provide examples of tweets in each of these categories.

14
See Sacks et al., 1974 for more information on conversational turn-taking.
15
Recall the definition of this term, given in section 4.2. Tweets with “isolate laughter” are considered to be those
tweets containing one and only one instance of written laughter where that form is the dataset-specified form.

37
 9) Tweet-Initial Laughter:

[@BJ5995]: @kimmy_dance lol you waiting for her to get out of a store again?

10) Tweet-Medial Laughter:

[@_kgip]: @Dmoss_2up2down hold on lol I’ll text it to you

11) Tweet-Final Laughter:

[@stackzhoe]: I woke up like this party tonight tho lol #Baseboy

#DumandDumber instagram.com/p/vbiRKfPseT/

12) Laughter Alone:

[@Chappells_Show]: “@Afroj3di: I still wanna know why I gotta shut up Joaquin” – lol

The last example above (12) contains what is known as “manual retweet.” Everything between

the quotation marks was originally posted by @Afroj3di, not by @Chappells_Show, the user posting

here. Manual retweets are often used in the same way one might use media content (photos, links

etc.): as a piece of news or as a joke. More historically this was a method by which individuals kept track

of ongoing conversations (Weller et al., 2013). In order to provide a reply to a friend’s tweet, a user

would copy and paste the friend’s tweet at the front end of her own tweet, placing it in quotation

marks, and then add her own contribution at the tail end. This allows the users involved in the

exchange, as well as anyone else wanting to read it, to follow the thread of a conversation. Occasionally

manual retweets are still used in this fashion, and when they are I have treated quoted sections as

previous turns in the conversation, not as components of the tweets in which they occur. In all cases

they were ignored when logging the position of a laugh particle.

The results of this analysis can be found in Figure 5. All forms of written laughter under

examination are most likely to occur at the end of a tweet and least likely to appear alone. Generally,

over in over half of tweets containing any one form of laughter, that laughter occurs tweet-finally.

38
 Figure 5: Percentage of Laugh Forms by
Position within Tweet
Initial Medial Final Alone

90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
Haha Hehe Lol Lmao 😂 😄

The most interesting feature of the analysis in Figure 5 is that the emoticon laugh forms (😄 and

😂) almost never appear in tweet-initial position. While the spelled-out forms are most likely to occur

tweet-finally, they do appear tweet-initially at substantial rates. This observation provides more

evidence for the proposal made above that spelled-out and emoticon forms of laughter may be used

according to different models. Though haha is the most likely of the spelled-out forms to be used

initially, the difference here is negligible. Thus, this analysis shows no real difference between the use of

haha and the use of any other spelled-out laugh form.

It is worth noting that each form listed above occurs non-medially over 80% of the time.

Written laughter, in general, tends to bookend tweets. It has been observed that emoticons tend to

appear only at the ends of phrases, much like phrase-final punctuation (Provine et al., 2007). A cursory

overview of the spelled-out laughter datasets collected for this study suggests that laughter, unlike

emoticons, can be used at the beginnings of phrases as well. As tweets are limited in length, it may be

the case that the fact that written laugh forms bookend tweets is a side-effect of the fact that these

39
laugh forms bookend phrases. While the positioning of written laughter at a sentence level would surely

be a productive route for future inquiry, this is an analysis of written laughter at a tweet level, and so a

sentence-level analysis is left for future work.

5.5: Recipient Design and Laugh Location

The two comparisons above showed that, while there are some similarities between how

spelled-out laugh forms and emoticon laugh forms behave with regard to recipient-design distributions

and placement within a tweet, there are also some differences. Though tweets containing any form of

written laughter are more likely to be specifically targeted than tweets which do not contain laughter,

tweets containing the spelled-out laugh forms lol, haha, hehe, and lmao are even more likely to be

designed for a specific recipient, and are more likely to contain tweet-initial laughter than are tweets

which use the emoticon forms 😂 and 😄. This section is devoted to the following question: Are these

two observations, that spelled-out forms are more likely to occur tweet-initially than are emoticon laugh

forms and that tweets containing spelled-out laugh forms are more likely to be specifically-targeted than

tweets containing emoticon laugh forms, related? A hypothesized connection between the frequency

with which a laugh form appears initially and the frequency with which it occurs in specifically-targeted

environments is appealing chiefly because initial laughter is strongly correlated with specific recipient

design.

Figure 6 shows the recipient design patterns for tweets containing isolate instances of various

tweet-initial laugh forms.16 The column for the 😄 form is empty because the collected dataset

contained no isolate examples of initial 😄.

For all five laugh forms for which information about initial isolate laughter is available, initial

isolate laughter occurs much more often in tweets with specific recipient design. Though the one

16
See example 9 above.

40
emoticon form that occurs initially, the laughter with tears emoticon, does show an association between

tweet-initial 😂 and specific recipient design, it occurs about 15% more often in nonspecifically-targeted

tweets than do the spelled-out initial forms.

Figure 6: Recipient Design - Tweet-Initial Laughter

Specific Nonspecific

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
Haha Hehe Lol Lmao 😂 😄 No Laughter

Figure 7: Recipient Design - Tweet-Final Laughter

Specific Nonspecific

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
Haha Hehe Lol Lmao 😂 😄 No Laughter

41
 Meanwhile, Figure 7 shows the recipient-design breakdowns for those tweets containing tweet-

final isolate laughter.17 The connection between tweet-final laughter and specific recipient design

seems far more tenuous. Though, for all six laugh forms, tweets containing final laughter are more likely

to receive specific targeting than are tweets which do not contain laughter, the behavioral differences

between initial and final written laughter are quite noticeable. In tweet-final position, the spelled-out

laugh forms hehe, lol, and lmao show recipient-design frequencies closely resembling those shown by

the emoticon forms. The one form for which tweet-final laughter is still clearly associated with specific

targeting is haha, and in this its behavior is set apart from the other five written laugh forms.

Spelled-out laugh forms are used more often initially than are emoticon laugh forms. Initial

laughter is used almost exclusively in specifically-targeted tweets. Might the reason that certain laugh

forms have high rates specific-targeting be that these same laugh forms are used more often in the

strongly-specific tweet-initial position? Can we legitimately consider the rate with which a form receives

specific recipient design to be dependent on the rate with which that form is used tweet-initially?

The charts in Figures 8 and 9 below represent comparisons between the rate with which each

form occurs tweet-initially (see 5.4) and the rate with which tweets containing each form receive

specific recipient design (see 5.3). Figure 8 examines the possibility that there is a linear relationship

between these two variables, namely those laugh forms which occur more often initially occur more

often in specifically-targeted environments. Figure 9 instead depicts the suggestion that has been made

several times above, that the various laugh forms under observation are used in accordance with three

separate general models: emoticon laughter, lol-like laughter (lol, lmao, hehe), and haha.

The suggested linear model drawn in Figure 8 serves to account fairly well for four of the forms,

😂, 😄, lol, and hehe, but fails to adequately account for haha and lmao. There does, however, appear

to be a generally positive trend, making this model intriguing. The addition of more data points through

17
See example 11 above

42
the analysis of more laugh forms might show, in the future, that this way of considering the data is more

generally powerful than it appears to be here. However, given the fact that this study has made only six

data points available, the model proposed in Figure 9 is currently more appropriate.

Figure 8: Specific-Targeting Figure 9: Specific-Targeting

and Initialness - Proposed and Initialness - Three
Linear Relationship Proposed Patterns
85% 85%

80% haha 80% haha

75% 75%
Specific-Targeting Frequency

Specific-Targeting Frequency
70% 70%

65% hehe 65% hehe

60% lol 60% lol

lmao lmao

55% 55%

50% 😂 50% 😂
😄 😄
45% 45%

40% 40%
0% 10% 20% 30% 0% 10% 20% 30%
Frequency Form Used Initially Frequency Form Used Initially

It is of note that at no point in this analysis thus far has any real difference in the use of

abbreviated laugh forms (lol, lmao) and onomatopoeic laugh forms (haha, hehe) been evident. Though

haha seems to show some differences from the other three, the usage patterns of hehe are very similar

to those of lol. The three groups which seem to be emerging are, rather, those shown in Figure 9.

43
5.6: Location of Laughter in Conversational Context
The sections above revealed intriguing differences between the recipient design distributions of

tweets containing tweet-initial laughter and tweets containing tweet-final laughter. In this section

tweet-initial and tweet-final laughter are compared with regard to the nature of conversationally

adjacent tweets.

Figure 10: A Dataset Tweet and its Context Tweets

Unlike the sections above, this section, as well as parts of Section 5.7, discuss the relationships

between tweets which are part of the gathered datasets and tweets in their immediate conversational

context. Take for an example a tweet, such as the one in Figure 10, that was collected as a part of the

dataset of tweets containing hehe. If, the user composing this collected dataset tweet did so in order to

respond to something tweeted earlier by another user, this previous tweet was collected. If, sometime

after collection,18 another user posted a response to the collected dataset tweet, this subsequent tweet

was collected as well. These previous and subsequent tweets, which will be referred to collectively as

context tweets below, provide information about the dataset tweet in the same way that the other

18
Recall that because data collection took place in real time conversational context was collected some time later
in order to give any users who might choose to respond to dataset tweets the time to do so.

44
forms of tagged metadata do. Just as we know that the tweet in Figure 10 contains tweet-final hehe, we

know that it has a previous and subsequent tweet, both tweeted by a female,19 neither of which

contains written laughter. When in the sections that follow I refer to a dataset’s previous or subsequent

tweets, I am referring to any previous tweets or any subsequent tweets which were logged as occurring

before or after any of that dataset’s dataset tweets.

Recall that these context tweets were collected using the “view conversation” function on

Twitter’s website.20 Unfortunately, the algorithm which governs this functionality is imperfect. This

means that for some dataset tweets which did respond to previous tweets or elicit subsequent tweets,

these context tweets could not be retrieved. The algorithm seemed to miss subsequent tweets more

often than it did previous tweets. In addition to this, there were rare circumstances in which those

users who composed the context tweets in question had altered their privacy settings so as to make it

impossible for anyone without specific permissions to view their posted content. These two issues with

the retrieval of context tweets, while they do not invalidate the information provided by those context

tweets which were collected, does invalidate any statistic which examines, for example, the frequency

with which tweets in the hehe dataset have previous tweets or subsequent tweets.

Therefore it is vital to remember that when, in the sections that follow, I refer to the frequency

with which the context tweets of a given dataset possess a trait, I am not referring to the frequency with

which dataset tweets follow or precede tweets with that trait. Instead I am referring to the frequency

with which those context tweets that were collected for a given dataset have that trait. For example,

when, in section 5.6.2, I refer to the frequency with which the previous tweets of the haha dataset

contain spelled-out laughter, I am referring to the percentage of the previous tweets collected for the

19
For gender-identification techniques, see Section 5.7
20
See section 4.2

45
haha dataset contained spelled-out laughter, not the percentage of tweets containing haha that were

preceded by tweets containing spelled-out laughter.

Two major analyses were conducted concerning differences in the conversational purpose of

tweet-initial written laughter and tweet-final written laughter. The first, in section 5.6.1, considers the

possibility that tweet-initial laughter refers back to a previous or ongoing topic, while tweet-final

laughter more often refers to tweet-internal topics. The second analysis, outlined in 5.6.2, considers

parallels between the placement of spoken laughter within a conversational turn, and the placement of

written laugh particles within a tweet. It is in this section that the offer/acceptance patterning of face-

to-face laughter is investigated with regard to written laughter.

5.6.1: Tweet-Initial vs. Tweet-Final Laughter: Externally-Referent Pronouns

This analysis considers the possibility that tweet-initial laughter tends to reference an ongoing

topic or laughable first mentioned in a previous conversational turn, while final laughter tends to

reference a laughable produced by the tweeter. Studies of face-to-face laughter have often suggested

that turn-initial laughter tends to have utterance-external referents, as opposed to turn-final laughter

which often references the speaker’s own utterance (Glenn, 2003; Holt, 2011). If this were the case for

written laughter as well, it could go a long way to explaining the strong association between tweet-initial

laughter and specific recipient design, as a previous or ongoing topic must be described in a previous

turn, something which can only exist in a specifically-targeted exchange. It would also still allow for the

relatively high, though less-pronounced frequency of specific recipient design among instances of tweet-

final laughter. Some examples of exchanges for which initial laughter seems to reference an ongoing

topic can be found in Appendix 2.

The most obvious way to approach this question would be to examine the frequency with which

dataset tweets containing tweet-initial laughter have preceding tweets and compare that to the

46
frequency with which dataset tweets containing tweet-final laughs have the same. Unfortunately, for

the reasons described in the introduction of this section, this comparison could not be made.

Instead, another method was used to determine whether tweet-initial and tweet-final laughter

differ in their ability to reference an external topic. This analysis examines the presence of externally-

referent pronouns in the “punctuation sentence” containing the laugh particle. The relevance of this

analysis is based on a single assumption. This assumption is that where written laugh particles occur in

close proximity to externally-referent pronouns, they are more likely to also refer to a tweet-external,

previous or ongoing topic.

For the purposes of this study, an externally-referent pronoun is defined as a third-person

pronoun with a meaning that cannot be inferred from the text of the tweet alone. It refers back to a

topic discussed earlier in a conversation. The tweet in [13], for instance, contains an explicit externally

referent pronoun (it). Notice that a reader reading [13] in isolation does not know what “it” means.

13) Externally-Referent Pronoun (Explicit):

[@Knee_Uhhhhh]: @K_Soto214 it's inactive lol.

This is in contrast to examples like [14] below, in which the pronoun “him” is clearly co-

referenced with “that cute guy in the hat.”

14) Internally-Referent Pronoun:

[@MetalEmpress]: @xHollyGlambertx All I can think about when I see this photo is Adam
pushing that cute guy in the hat against a wall & making out w/ him lol

There is a second category of externally-referent pronouns which has been included in this

analysis. These are the implied externally-referent pronouns. Notice that in [15] there is an implied “it”

or “that” before “not gonna happen.”

15) Externally-Referent Pronoun (Implicit):

[@jduccee]: @JennPortilloo not gonna happen lol

47
 The presence or absence of an externally-referent pronoun was evaluated within the

“punctuation sentence” containing the laugh particle in question. This is a loose definition merely

intended to restrict the portion of a tweet under scrutiny to the general area of the tweet in which the

laughter actually occurs, particularly in cases where tweets contain more than one sentence. The

borders of the punctuation sentence were marked wherever traditional sentence-separating

punctuation, emoticons, hyperlinks, or hashtags were introduced, as well as at the beginnings and ends

of a tweet.

Figure 11: Initial lol refers to an earlier subject without an externally-referent pronoun

Dataset
Tweet

It is important to note that while the presence of an externally-referent pronoun may imply a

topic-continuation, the pronoun need not be present in order for that tweet to be a topic continuation.

Figure 11 below provides an example of an instance in which lol in the dataset tweet is clearly being

used to refer back to the comment made in the previous tweet, but an externally-referent pronoun is

not used. In the datasets collected for this project there are many examples of tweets which seem to

bear some sort of external reference, but leave it unclear whether the laugh particle references the

previous topic or the specific contribution being made by the current tweeter.21 Such tweets were

21
To provide a few examples:
1. [@Blake41Taylor]: @JessicaManuel95 haha so your mad because your so called number one team is
going to get beat?
2. [@whit_law]: @Jro_Chicago haha I’ll be keeping up with the game, tony has it on his phone!
3. [@_azucenaamonic]: @sayhelloirene @559_emm lol no life

48
extremely difficult to tag in a reliable and consistent manner. The choice to tag only those instances of

external references that make use of externally-referent pronouns served to provide a far less

ambiguous set of tweets in which external reference was clear. Unfortunately, this meant that only the

absolute clearest cases of external reference were considered as such for this analysis. The actual rates

of external reference were likely quite a bit higher, both for tweets containing tweet-initial laughter and

for tweets containing tweet-final laughter.

As tagging the presence of externally-referent pronouns must be done by hand, and requires

large datasets in order to provide powerful results, this analysis was conducted using only the three

large-sample datasets: lol, haha, and 😂. These datasets were originally enlarged because they

exemplify both the three initialness/specificity categories drawn up in Figure 9 above (emoticon, lol-like,

and haha) and the three major categories of netspeak items often discussed in other studies (emoticon,

abbreviation, and onomatopoeia). The presence or absence of externally-referent pronouns within the

same punctuation sentence was marked for tweets containing tweet-initial isolate laughter and

containing tweet-final isolate laughter. The results of this comparison are shown in Figure 12.

Percentages are given in terms of tweets with that laughter form in that position, rather than simply in

terms of the number tweets containing that laughter form.

If we assume that tweets containing tweet-initial laughter do, in fact, tend to reference a

previous or ongoing topic then we would expect to see tweets containing tweet-initial written laughter

to contain externally-referent pronouns at a higher rate than do tweets containing tweet-final written

laughter. As far as haha and lol are concerned, this is, in fact, what we see in Figure 12. Contrastively, in

the case of 😂, this trend cannot be shown. However, since the number of examples of tweet-initial 😂

was severely limited, this result may be deceiving.

49
 Figure 12: Percentage of Tweets Containing Three
Laugh Forms in Tweet-Initial and Tweet-Final Position
which Contain Externally-Referent Pronouns
Initial Final

40%
PERCENTAGE W/ PRONOUN

35%
30%
25%
20%
15%
10%
5%
0%
Lol Haha 😂

Though these statistics are drawn from relatively small samples, it may prove worthy of note

that for all three forms the frequency with which externally-referent pronouns co-occur with tweet-final

laughter seems similar (hovering between 13 and 20%). Incidentally, this frequency is the same as the

apparent frequency with which tweet-initial 😂 co-occurs with externally-referent pronouns (though this

frequency was drawn from a small dataset, N=24). If these statistics are representative, this analysis

would suggest not only that tweet-initial and tweet-final 😂 behave similarly to one another, but that

both behave similarly to tweet-final lol and haha. Perhaps emoticon forms are somehow inherently

internally-referent; referring only to the specific contribution being made by an individual tweeter. This

is approximately what has been proposed by those who discuss emoticons as indicators of illocutionary

force (Dresner & Herring, 2010). They argue that emoticons provide information about intended deeper

meaning of the writer’s own contributions to the conversation.

Perhaps spelled-out laugh forms like lol and haha can use initial position to indicate that the

laughable is external. If we then suppose that an emoticon cannot take an external referent, the

50
fronting of an emoticon would not have the same power as the fronting of a spelled-out laugh form.

This would explain the lack of distinction between tweet-initial and tweet-final laughter in Figure 12, but

it might also explain why emoticons appear relatively rarely tweet-initially.

While haha shows the association between tweet-initial position and the presence of externally

referent pronouns, it also shows generally higher rates of co-occurrence with external reference. This

may explain the high rates of specific recipient design seen earlier. Where 😂 is strictly internally-

referent, haha may be somewhat externally-referent. If haha tends to refer to previous or ongoing

topics that might explain why it is so often found in environments that allow for the presence of an

ongoing or previous topic, environments of specific recipient design. This association could, of course,

be drawn in the opposite direction. If a form is more often used in an interpersonal, specifically-

targeted environment, it would make sense for that form to more often reference the surrounding

conversational context as well.

5.6.2: Presence/Absence of Conversationally-Adjacent Laughter

Many of the studies discussed in the literature review of this paper have found that when

laughter is used in face-to-face conversation it is both offered in accepted. One speaker will laugh

towards the end of her utterance, offering laughter, and her conversational partner will begin his

utterance by laughing as well, accepting her offered laughter (Glenn, 1989, 1991, 2003; Jefferson et al.,

1977; Provine, 2001). This section examines the possibility that written laughter may be used in a

similar manner. The comparisons which follow look for associations between presence and/or location

of certain forms of written laughter in a dataset tweet and the presence of laughter in its context

tweets.

Ideally this analysis would include a comparison for which the locations of laugh forms in

dataset tweets and in context were variables. Unfortunately, the data gathered for this project did not

51
provide sufficient sample sizes to conduct such analysis productively and therefore this aspect of the

study is left for future work.

In the analyses which follow, context tweets will be tagged as containing or not containing

spelled-out and emoticon laughter. The distinction made here between emoticon laugh forms and

spelled-out laugh forms was motivated by those of the results presented above which seem to suggest

that emoticon forms are used on Twitter according to a different model than are spelled-out forms.

Most of the analyses in this section concern the presence or absence of spelled-out forms in context

tweets, but a mention of emoticon laughter is made towards the end of this section.

The datasets involved here were filtered down once again to only examples of isolate laughter,

as the location of the laugh particle in a dataset tweet was a major factor in this analysis. Each dataset

tweet’s context tweets were tagged if they contained spelled-out or emoticon laughter, using the lists of

laugh forms given in 4.2.22 This means that the factor being tested in each case was the influence each

dataset laugh form had on the presence or absence of previous laughter of these two types in general,

not the influence these forms had on the specific identity of these forms.

Throughout this section I will refer to the frequency with which the context tweets for a certain

group of dataset tweets (such as those tweets containing tweet-initial haha) contain laughter. For the

reasons given in the introduction of 5.6, these frequencies will be the percentage of the context tweets

collected for that group of dataset tweets which contain laughter. This means that if I am defining the

frequency with which those tweets containing tweet-initial haha have previous tweets containing

laughter, I am reporting the frequency with which the previous tweets of tweet-initial haha contain

laughter, not the frequency with which tweets containing haha both contain tweet-initial haha and are

preceded by a laughing tweet.

22
Spelled-out forms: lol, haha, lmao, lmfao, jk, hehe, hoho, ahaha, huhu, *giggles*, /chuckle and orthographic
variations on these. Emoticon forms: 😄, 😍, 😂, 😋, 😎, 😏, 😅, 😉, 😊, and their ASCII counterparts (any form with
an upturned mouth was included).

52
 This analysis examines data from those of the unfiltered tweets which contain no laughter, as

well as data from five out of the six collected datasets of laugh forms. The one form not discussed in

this section (😄) has been left out because no examples were found in which this form was used initially

in isolate. The location of the dataset laugh form within a tweet is a major variable in the analyses

which are to follow, and so 😄 has been put aside for the time being.

5.6.2.1: General Association between the Presence of Various Written Laugh Forms and the
Presence of Spelled-Out Laughter in Context Tweets

This first analysis examines the relationship between the presence of written laughter within a

dataset tweet and the presence of spelled-out laughter in its context tweets without regard to the

location of the dataset form within the tweet. This is intended to establish first, whether or not the

presence of the dataset forms selected is correlated with the presence of spelled-out laughter in context

tweets in general, and second, whether any of these laugh forms is preferentially used before or after

other tweets containing written laughter. If, for example, one laugh form had been shown to be

associated only with the presence of spelled-out laughter in its previous tweets and not its subsequent

tweets, this laugh form may have served a more responsive purpose.

The prediction of this analysis is that the presence of a written laughter in a dataset tweet is

associated with the presence of spelled-out laughter in its context tweets. This would lead us to predict

that the frequency with which the context tweets of dataset tweets containing each written laugh form

contain spelled-out laughter ought to be higher than the frequency with which the context tweets of

dataset tweets containing no written laughter do the same. This prediction has two halves. The first

half predicts that each form will be associated with elevated frequencies of spelled-out laughter within

its previous tweets, while the second predicts that each form will be associated with elevated

frequencies of spelled-out laughter within its subsequent tweets. Several of the forms in question show

both predicted associations, as can be seen in Figure 13

53
 Figure 13: Frequency with with Context Tweets
Contain Spelled-out Laughter
Frequency of PT Laughter Frequency of ST Laughter

30.00%

25.00%

20.00%

15.00%

10.00%

5.00%

0.00%
Haha Hehe Lol Lmao No Laughter

Notice that dataset tweets containing all four of the spelled-out laugh forms under observation

seem more likely to have context tweets, both previous and subsequent, that contain spelled-out

laughter than are the dataset tweets containing no laughter. This is exactly the result that was

predicted. However, though dataset tweets containing 😂 are slightly more likely to have context

tweets which contain spelled-out laughter than are dataset tweets containing no laughter, this

difference is very small. This seems to be no strong association between the presence of 😂 in a tweet

and the presence of spelled-out laughter in either its previous or subsequent tweet. This provides

further evidence that emoticon and spelled-out forms of written laughter are used differently. This

particular difference will be revisited in section 5.6.2.3 below.

Another observation that can be made based on the information given in Figure 13 is that for

each form, dataset tweets are about as likely to contain spelled-out laughter in previous tweets as they

are to contain spelled-out laughter in subsequent tweets. There is no form which is primarily used

before tweets containing laughter and no one form which is primarily used after tweets containing

laughter.

54
5.6.2.2: Adding Laugh Location as a Factor

This section examines the possibility that the location of a laugh form within a tweet may have

an influence on the frequency with which its previous tweets and subsequent tweets each contain

laughter. This question stems from the observation, made in studies of the offer/acceptance patterns

for face-to-face laughter that offered laughter tends to occur towards the end of an utterance while

accepted laughter tends to occur towards the beginning of an utterance. Examples of the various

combinations of previous and subsequent laughter and locations of laugh particles within tweets can be

found in Appendix 3.

If written laughter follows similar offer/acceptance patterns to face-to-face laughter, we might

expect to see initial laughter more strongly associated with laughter in a previous tweet than is final

laughter, and final laughter more strongly associated with laughter in a subsequent tweet than is initial

laughter. We are essentially looking for two separate predictions here. The first prediction, that tweet-

initial laughter is more strongly associated with laughter in a previous tweet than is final laughter, we

will call the “Initial-Previous Relation.” The second prediction, that tweet-final laughter is more strongly

associated with laughter in a subsequent tweet than is initial laughter, we will call the “Final-Subsequent

Relation”

Figures 14 and 15 show that Initial-Previous and Final-Subsequent Relations are observable at

least for some laugh forms. The Initial-Previous Relation, shown in Figure 12, is observable for hehe,

lmao, lol, and 😂, but, interestingly, not for haha. Meanwhile, the Final-Subsequent Relation, shown in

Figure 13, is observable for haha, hehe, lol, and lmao, though the association only seems strong for hehe

and lmao.

55
 Figure 14: Frequency of Spelled-Out Laughter in
Previous Tweets as a Function of Dataset Laugh
Location
Tweet-Initial DS Form Tweet-Final DS Form

35.00%
30.00%
25.00%
20.00%
15.00%
10.00%
5.00%
0.00%
Haha Hehe Lol Lmao No Laughter

Figure 15: Frequency of Spelled-Out Laughter in

Subsequent Tweets as a Function of Dataset Laugh
Location
Tweet-Initial DS Form Tweet-Final DS Form

40.00%
35.00%
30.00%
25.00%
20.00%
15.00%
10.00%
5.00%
0.00%
Haha Hehe Lol Lmao No Laughter

Interestingly, the Initial-Previous Relation seems more consistent than the Final-Subsequent

one. This may be partially due to the fact that the number of subsequent tweets collected for each

dataset was smaller than the number of previous tweets collected, or it may be because tweet-initial

56
laughter behaves more systematically as a laughter acceptance than tweet-final laughter does as an

invitation. Either way, some forms do show an association between tweet-initial written laughter and

the presence of spelled out laughter in a previous tweet, and some forms do show an association

between tweet-final written laughter and the presence of spelled-out laughter in a subsequent tweet.

5.6.2.3: General Association between the Presence of Various Written Laugh Forms and the
Presence of Emoticon Laughter in Context Tweets

The above two sections have examined the relationship between the presence of various

dataset laugh forms and the frequency with which context tweets contain spelled-out laughter. This

section looks at the relationship between the presence of these dataset laugh forms and the frequency

with which context tweets contain emoticon laughter. Here we go back to a prediction similar to the

one made in 5.6.2.1, that those dataset tweets containing some form of written laughter will be more

likely to have context tweets which contain some form of emoticon laughter than are dataset tweets

containing no laughter whatsoever.

Figure 18: Frequency with which Context Tweets

Contain Emoticon Laughter
Frequency of PT Laughter Frequency of ST Laughter

40.00%

35.00%

30.00%

25.00%

20.00%

15.00%

10.00%

5.00%

0.00%
Haha Hehe Lol Lmao No Laughter

57
 In examining the results of this comparison, shown in Figure 18, it becomes immediately obvious

that dataset tweets containing 😂 seem to be more likely than dataset tweets containing any other form

of written laughter under observation to have context tweets which contain emoticon laughter. This is

the opposite of what we saw in 5.6.2.1, where 😂 was the only form which was not associated with

spelled-out laughter in previous and subsequent tweets.

A second observation which can be made about Figure 18 is that the presence of at least three

of the spelled-out forms, haha, lol, and lmao, does not appear to be associated at all with the presence

of emoticon laughter in context tweets. The one spelled-out laugh form which does show the predicted

association, hehe, shows it only weakly. Yet again a difference between the spelled-out laughter forms

and emoticon forms is evident.

It must be noted that for this comparison tagging of emoticon laughter was, perhaps, overly

flexible. If a large number of emoticon forms which may or may not have actually represented laughter

were tagged as such. If emoticon forms were in fact over-tagged, they may have created enough noise

to have obscured associations which may otherwise have been present, or to have created associations

which otherwise would have been absent.

5.7: Gender Analyses

The following sections represent two analyses concerning the effect of gender on the use of

written laughter. We have seen above that some of the offer/acceptance patterns seen in the use of

face-to-face laughter can be observed in the conversational use of written laughter as well. This section

is motivated by two observations about laughter and gender made fairly often in the literature. First,

many of the studies of netspeak languages have found that most “emotive” expressions such as the

laugh forms under observation in this thesis are used much more often by females than by males.

Second, many studies of face-to-face laughter have found that the extent to which offered laughter is

58
accepted or not accepted is strongly dependent on the gender of the participants. The first part of this

section delves into the first observation, looking specifically at whether the forms under observation

tend to be used more by females or by males in these datasets. The second part of this section concerns

the second observation, that the likelihood that an individual will accept offered laughter is largely

dependent on the gender of the individuals involved. This second part examines the influence of gender

on offer/acceptance patterns of written laughter.

Twitter, unlike many other social networking platforms, neither requires nor asks users to attach

to their account the gender with which they identify. In fact, all information attached to any individual’s

account may be completely fabricated. A user may tweet as Justin Bieber, Mockingjay❤, Batman, or

even God. Most studies examining gendered use of language on Twitter have therefore followed the

accounts of subjects whose gender identity had been previously ascertained.

As data for this study were collected as a random sample, rather than by following specific

individuals whose gender identity was known no real-world information about each tweeter was known.

One previous study (Bamman et al., 2014), did identify the gender of anonymous Twitter users by

running users’ volunteered first names through a computer program intended to identify names as male

or female. If a first name could be found at least 1000 times in the most recent available US Census and

was over 80% associated with individuals of one gender or the other (names like Paula, James, or

Desiree), that users’ data was tagged as either male or female and was used for their project.

In a pilot study I conducted, this method was applied to the lol dataset of this project. The

methods above allowed only about 1/3 of users to be identified as male or female. Even when tagging

was done by hand, allowing nontraditional spellings (ex: ᗰᗩƘᗩᎥℓᗩ) or gendered inventions (ex:

NAWTY❤GAL), only about 55% of the tweeters in the dataset could be identified as male or female

based on username alone. For a small, but not negligible portion of these identifications other

59
information on the tweeter’s homepage suggested that the gender categorization based on username

alone had, in fact, yielded incorrect results.

A second pilot study was therefore conducted to search for a better identification method. I

examined the profile pictures associated with individuals in two friends’ Twitter networks. Both users

were aware of the real-world gender identification of most of the users in their networks so my

identifications could be compared with their expert knowledge. Members of these two twitter

networks were identified as male or female based on the gender of the person or fictional character in

their profile pictures. Groups photos containing only members of one gender were also accepted, and

photographs of celebrities were not included. Not every person in these two Twitter networks could be

identified using this method, but in general a larger percentage of users could be identified using this

method than the previous methods attempted, and these identifications seemed to be more

consistently accurate than previous methods. I could identify the gender of 73.78% of the first

individual’s network and less than 0.05% of those identifications were inaccurate. In the second

individual’s network, tagging was less successful. 58.72% of this individual’s total network could be

gender-identified, though as 24.25% if his twitter network was ungendered businesses and blogs, this

rate is not surprising. Of those pages followed by this second individual that were gendered 77.52%

could be identified. Only 2.90% of the total identifications were explicitly incorrect, though an

additional 8.70% attached genders to non-gendered entities such as blogs and news agencies. Though

this method of identification was much more labor-intensive and was still not completely accurate, it

was a method better suited to this project.

An initial attempt to identify users in the dataset for this project using their profile pictures

revealed that about 80% of users could be identified as male or female if single-person gendered

photographs, or single-character gendered non-photographs were used as an identification method.

Those individuals who could not be identified had profile pictures showing a character of ambiguous

60
gender, a photograph containing individuals of multiple genders, photographs for which gender could

not be established, or pictures which contained neither characters nor photographs of people (such as

an advertisement for an event, or a photo of a sunset).

This second method was the method used for gender identification in this study. For each tweet

in the dataset the gender of the tweeter as well as the genders of any other tweeters found in the

conversational context of the dataset tweet were marked. Particular focus was given to the dataset

tweeter and the user who tweeted directly before and directly after that individual. In the end

identification rates were only slightly lower than the 80% identifiable in the pilot (hovering around 75%,

depending on the laugh form in question.) The precise percentage of each dataset for which gender

could be identified is given in Table 3. In the analyses which follow only that portion of the data for

which gender could be identified was used.

5.7.1: Gender of Tweeter

Most studies of gendered language on twitter have indicated that emoticons as well as “emotive

abbreviations” and laughter words like haha, hehe, lol, and lmao, are much more commonly used by

females than by males (Bamman et al., 2014). This goal of this analysis is to determine whether each of

the forms of written laughter can be shown to occur in tweets composed by males or by females using

the methodology described above. For each laugh form under consideration the percentage of gender-

tagged tweets tweeted by males and by females is compared. These male/female tweeter distributions

61
for tweets containing each laugh form were then compared to the male/female tweeter distributions for

tweets which were a part of the original unfiltered sample.

The results of this analysis, as presented in Figure 19, are somewhat surprising. Previous

research would predict that each of the forms in question ought to be associated with female tweeters.

Though the frequency with which tweets containing each laugh form were tweeted by females was

higher than the frequency by males, in only a few cases (hehe, 😂, and 😄) was this frequency higher

than the frequency with which females were found to tweet in general. This essentially means that

though lol, haha, and lmao, were tweeted more often by females than by males, tweets containing

these three forms were actually more likely than the average tweet to have been contributed by males.

This means that three of the laugh forms in question actually seem to be somewhat associated with

male, rather than female tweeters.

Figure 19: Tweeter Gender

Male Female

90%

80%

70%

60%

50%

40%

30%

20%

10%

0%
Haha Hehe Lol Lmao 😂 😄 Unfiltered

62
 There is, of course, another question which could be asked, namely, is written laughter in

general a marker of female speech? The answer appears to be no. A chi-square test of independence

was performed in order to examine the relation between the presence of written laughter in a tweet

and the likelihood that the tweeter was female. The relation between these variables was not

significant. Tweets containing written laughter are not significantly more likely to be contributed by

females than the average tweet on the site x2(1, N=3325)=0.9821, p>0.1.

It does seem worthy of note that two of the three forms which do appear to be associated with

female tweeters are emoticon forms. A second chi-square test of independence was therefore

preformed in order to examine the relation between the category of written laughter in a tweet

(emoticon or spelled-out) and the likelihood that the tweeter was female. The relation between these

variables was significant. Tweets containing emoticon laughter are significantly more likely to be

contributed by females than are tweets containing spelled-out laughter forms x2(1, N=2897)=48.74,

p<0.01.

5.7.2: The Relationship of Gender to Offer/Acceptance Patterns

As has been discussed repeatedly above, face-to-face laughter generally takes on an

offer/acceptance pattern in conversation. In addition to this, many studies have also found an

association between the gender of conversational participants and their behavior with regard to

laughter acceptance. These studies are discussed in the literature review of this paper, but in general,

males are less likely to accept offered laughter, especially when that laughter is offered by a female, and

are more likely to laugh when laughter is not offered, especially when the previous utterance was

produced by a female. Contrastively females are more likely to accept offered laughter, especially when

it is offered by a male, and less likely to laugh when laughter is not offered, especially when the previous

speaker was male.

63
 The examination of offer/acceptance patterns in Section 5.6.2 above found that the presence of

spelled-out laugh forms in a tweet is associated with higher rates of laughter in previous and subsequent

context tweets.23 The position of laugh forms within the tweet, tweet-initial or tweet-final, was also

found to be a relevant factor in the presence of spelled-out laughter in previous and subsequent tweets.

Tweet-initial laughter was generally associated with spelled-out laughter in a previous tweet, while

tweet-final laughter was generally associated with spelled-out laughter in a subsequent tweet.24 Where

the analyses preformed in Section 5.6.2.2 took the position of a laugh particle within a tweet to be the

primary variable in whether or not context tweets contained laughter, this analysis considers the

possibility that the gender of participants is that primary variable.

Because the analysis in 5.6.2.3 showed that the relationship between the presence of several of

the laugh forms in question in dataset tweets and the presence of emoticon laughter in context tweets

was small to non-existent, this analysis only examines the presence of spelled-out laughter in context

tweets. Otherwise, the same tagging mechanisms were used to tag context tweets as containing

spelled-out laughter as were used above.25 Due to the amount of labor required in order to tag the

gender of the tweeters of context tweets for this analysis, as well as the need for large datasets in order

to obtain powerful statistics, this analysis only considered the forms haha, lol, and 😂.

The graphs in Figures 20 and 21 represent the results of this analysis. Figure 20 examines the

influence that the gender of both the dataset tweeter and the previous tweeter has on the frequency of

spelled-out laughter in previous tweets, while Figure 21 examines the influence that the gender of both

the dataset tweeter and the subsequent tweeter has on the frequency of spelled-out laughter in

23
Here we are once again discussing the frequency with the previous tweets of dataset tweets containing spelled-
out laugh forms contained laughter. See the introduction of Section 5.6.2 for definitions of these terms.
24
See Section 5.6.2.2
25
See Section 5.6.2.2

64
subsequent tweets. N-values are low because frequencies are given in terms of, for example, the

number of previous tweets tweeted by males before dataset tweets also tweeted by males.

Figure 20: Frequencies with which Previous Tweets

Contained Spelled-out Laughter for Various Dataset Tweeter
Gender and Context Tweet Gender Combinations
Male after Male Male after Female Female after Male Female after Female

35.00%

30.00%

25.00%

20.00%

15.00%

10.00%

5.00%

0.00%
Lol Haha 😂

Figure 21: Frequencies with which Subsequent tweets

Contained Spelled-out Laughter for Various Dataset
Tweeter Gender and Context Tweeter Gender
Combinations
Male before Male Male before Female Female before Male Female before Female

50.00%

40.00%

30.00%

20.00%

10.00%

0.00%
Lol Haha 😂

65
 Figure 20 suggests that any influence that gender has on offer/acceptance patterns of written

laughter is unrelated to what is seen in face-to-face laughter. Where in face-to-face communication

males are unlikely to accept the laughter invitations of females, for at least two forms above (lol and 😂),

male tweeters tend to use written laughter more often in response to the laughter of females than that

of males. However, females do seem to show the expected pattern, at least for the forms haha and 😂.

Females more frequently respond to the laughter of males than that of females. The exception to this

trend is lol. Females seem more likely to respond to female laughter with lol than they are to respond to

male laughter this way.

The patterns shown in Figure 21 resembled the patterning observed in Figure 20. Males using

lol or 😂 were more likely to receive laughter acceptances from females. Females using lol or 😂 were

about equally-likely to be answered by laughter from males as from females, while those using haha

were more likely to receive a laughter response from males than from females.

Overall, this analysis revealed that there is no generalizable relationship between the genders of

the dataset tweeter and the context tweeter and the likelihood that context tweets contained spelled-

out laughter. It may be that gender has little influence on laughter acceptance rates because the gender

of Twitter users is less physically obvious than in face-to-face conversation.

6: Discussion
This project had two overlapping goals. The primary was to establish preliminary information

about the use of written laughter on Twitter, the second was to evaluate the extent to which we make

use of our knowledge of physical laughter when writing laughter. In this section the analyses above will

be examined as a whole with regard to the insights they can provide moving towards a better

understanding of written laughter.

66
 There appear to be three major types of written laughter, emoticon laughter, lol-type laughter

(lol, lmao, and hehe), and haha. These are the same three groups which were first discussed in Figure 9

above. A table outlining the general behavior of each of these groups can be found in Table 2. 26

Table 2: Similarities and differences between written laughter groups Spelled-Out Forms
Emoji Lol-like Haha
# Traits
Laughter Laughter
The presence of this laugh form in a tweet increases the
1 likelihood that a second laugh form will also occur in that
tweet (5.2)

2 More likely to co-occur with spelled-out laugh forms (5.2)

Tweets containing this laugh form are more likely to be

3
specifically-targeted than are non-laughing tweets (5.3)

Tweets containing this laugh form are more likely to be

4
specifically-targeted than not (5.3)

5 Form occurs initially in 20-30% of cases (5.4)

Tweets containing this laugh form tweet-initially are more

6 likely to receive specific recipient design than those which
contain this laugh form tweet-finally (5.5)
Tweet-final laughter is associated with specific recipient
7
design (5.5)

Initial-position laughter is associated with the presence of

8
externally-referent pronouns (5.6.1)
Tweets containing these laugh forms are more likely to
9 contain spelled-out laughter in their context tweets than
are tweets containing no laughter (5.6.2.1)
Tweets containing these laugh forms in tweet-initial
10 position are more likely to contain laughter in previous
tweets than tweets containing them finally (5.6.2.2)
Tweets containing these laugh forms in tweet-final position
11 are more likely to contain laughter in subsequent tweets
than tweets containing initially (5.6.2.2)
Tweets containing these laugh forms are more likely to
12 contain emoticon laughter in their context tweets than are
tweets containing no laughter (5.6.2.3)

26
Shaded squares represent situations for which the association is observable

67
 Though lol-type laughter does differ in some respects from haha-type laughter, the most striking

differences can be drawn between emoji laughter and spelled-out laughter. In several cases, lol-type

laughter and haha-type laughter behave similarly, while emoji laughter does not (see rows 2, 4, 5, 8, 9,

11, and 12). In general the spelled-out laugh forms behave in a way which is more reminiscent of

spoken laughter than do the emoji laugh forms. They are more tightly associated with interpersonal

exchanges and the locations in which they occur seem to be more strongly associated with previous and

subsequent spelled laughter as well as with external reference. The lack of association between the

presence of spelled-out laugh forms in a tweet and the presence of emoji in the previous and

subsequent tweets (see Table 2) is strong evidence for the spelled-out/emoji distinction.

The lack of distinction between onomatopoeic laughter (haha, hehe) and abbreviated laughter

(lol, lmao) is also worthy of note. Though haha has been placed in an independent category, hehe,

another onomatopoeic form, behaved very similarly to lol or lmao. Several studies have examined

netspeak abbreviations as if they were all one category. This study suggests that this is likely not the

case. Netspeak abbreviations are often treated as one vague and somewhat mysterious class of lexical

items. Here two of these netspeak abbreviations, lol and lmao, clearly behave a great deal like the other

forms of spelled-out laughter under observations. It is their membership in the category “spelled-out

laughter” which determines their use in conversation, not their membership in the category “netspeak

abbreviations.” Some studies have also attempted to treat forms like lol and lmao as emoticons. This

also does not seem to be an appropriate treatment, as these spelled-out laugh forms do seem to be

used in ways which clearly distinguish them from the emoticon forms.

These spelled-out laughter forms, at the very least, behave similarly enough to spoken laughter

that it is likely that individuals who spell out their laughter are drawing on knowledge of spoken

laughter. There are, however, some significant differences in usage patterns, particularly with regard to

the offer-acceptance patterns of laughter. For the most part written laughter is associated with spelled-

68
out laughter in a previous or subsequent tweet. However, a number of the subtleties of face-to-face

laughter seem not to translate to spelled-out laughter. For example, tweets containing written laughter

contain laughter in previous and subsequent tweets at much lower rates than might be predicted were

they regulated exactly as face-to-face laughter is. In addition, the genders of participants in twitter

exchanges seems to have little to no influence on the rates at which written laughter is offered and

accepted.

These differences might be explained in several ways. First, physical laughter is regulated

almost unconsciously, whereas spelled-out laughter is placed more deliberately (Provine, 2001). The

more unconscious portions of our laughter knowledge may, therefore, be put-aside when writing out

laughter. Another explanation may be the relatively slow pacing of Twitter exchanges. Though often

these conversations happen in real time, “real time” for typists is much slower than for speakers.

Differences in behavior may also be due to the fact that the “laughing together” phenomenon briefly

described in the literature review, cannot take place through text.

7: Future Research
This project is a very preliminary examination of an enormous internet phenomenon. Written

laughter occurs all over the internet, not just on Twitter. It shows up in blogs, “memes”27, text

messaging, and emails, not to mention poetry, movie scripts, spoken language, or graffiti art. This

project has examined the use of only six written laugh forms in only one internet communication

27
I am referring here to the colloquial rather than the academic usage of the word “meme.” In academic circles
“meme” is used to describe ideas or trends which take off through a society (Blackmore, 1999). These are often
the ideas which some describe as “viral.” In the online community, however, this word refers to a specific sort of
viral idea. These tend to be pictures, often screenshots from movies or well-known youtube videos, subtitled with
humorous quips. Over time some pictures become associated with a certain sort of commentary and a certain sort
of language use. Lol-cats are perhaps the most famous sort of “meme.” Examples of “memes” can be found in
Appendix 1.

69
environment. There are therefore a wealth of environments which could be the topics of future

research.

First and foremost, this study was quantitative. Even utilizing this same data, a qualitative study

could be conducted examining the ways in which written laughter is used in order to manipulate the

flow of conversation or the meaning of utterances. A line of research like this might serve to fulfil the

first half of the two pronged question posed in 4.1, it might serve to provide evidence that the

intersubjective meaning of written laughter is similar to the intersubjective meaning of physical laugher.

This is a topic which has been largely ignored in the present project which very much needs to be

examined.

Another promising avenue for research would be to examine the use of written laughter in

different SMS environments, such as text messaging, WhatsApp, Snapchat, and various instant

messaging platforms. It would be interesting to see whether the results found here are repeatable in

other similar environments or if laughter practices are specific to individual SMS platforms.

Similarly it would be interesting to see examine written laughter practices in longer-form

messaging environments, such as email, Reddit, Tumblr, or the comment sections for media posts. To

what degree might written laughter use in these environments mirror its use on Twitter? Moving

forward, we might to look at the use of written laughter in non-messaging online environments like

blogs, “meme” or lol-cat subtitling, and web comics. In this vein, many more traditional artistic genres

such as poetry and script-writing have begun to make use of these online laughter words. Are written

laugh forms used differently in more “formal” genres?

In addition to their online use, many written laugh forms have made a transition to spoken

language as well. In these environments words like lol are able to co-exist with actual, physical laughter.

It might be very enlightening to examine the similarities and difference in the usage of lol or haha in

face-to-face interaction and in online interaction. If lol is being used in face-to-face interaction it must

70
mean something more than simple laughter. Looking at the ways in which each of these laugh words

are used in face-to-face conversation could help to unravel the meaning of forms like haha or lmao,

distinguishing between the meanings which approximate face-to-face laughter from additional

meanings they take on as forms of written laughter.

This study also has revealed some interesting questions relevant not only to written laughter,

but to netspeak forms like emoticons and abbreviations in general. There is a suggestion in these data

that including laugh categories in general analyses for “netspeak abbreviations” may be misleading, as

lol and lmao behaved so similarly to haha and especially hehe. It would be interesting to see whether

the behavior of these two forms differs from the behavior of omg, brb, or wtf. This might be particularly

interesting with regard to the recipient-design distributions of tweets containing these forms.

This study has similarly revealed some fairly strong differences in the behavior of spelled-out

and emoticon laugh forms which deserve more extensive investigation. In particular, the relative lack of

tweet-initial emoticons is interesting. Why are emoticons not used initially? Is what they do to the

meaning and interpretation of a tweet different from what forms like lol or haha do in some more

meaningful way? The intuitions given in section 3.1.6 of this paper about 😄 revealed that this form is

only very weakly related to laughter in the minds of users, and yet its behavior through several

comparisons was very similar to 😂’s. Does that imply that these forms’ identities as emoji are more

important to their use and interpretation than their identities as laugh forms?

Additional phenomena of interest include rules governing the ways in which the orthography of

various laugh forms can be varied in order to achieve different sorts of meaning. What is the difference

between a capitalized laugh form and its lowercase counterpart (lol vs LOL)? What effect do repeated

vowels have (lol vs loooooool)? What about the number of end-to-end repetitions of a laugh form (lol vs

lololol)? Are these effects form-specific or universal? These and other aspects of orthographic variation

such as surrounding punctuation and layout are worthy of future research.

71
 The enormous online linguistic community which has developed on the internet over the last

decade-and-a-half has far outstripped our academic understanding. There is a vast amount of data from

numerous discourse environments available to linguists desiring to conduct their research online. Some

linguists have a tendency to see this information as “just text,” or, more often, as less essential to

understanding the human linguistic system. But just as it is important to study Language from the

perspective of more than just the English language, it is important to study Language from more than

just the perspective of spoken language. This is why sign languages are studied so extensively, and it is

why real-time written language should be studied as well.

Conclusion
Whether it is sent over the telegraph wire or bounced off a satellite, spelled-out laughter is a

fixture of SMS communications of all kinds. This project represents an initial attempt to understand why

that might be. By examining data from the social networking site Twitter, some generalizations about

different patterns of usage were drawn. Though any real understanding of written laughter is still in its

nascent stages, this project did find that emoji forms of laughter behave fairly differently from spelled-

out forms, but that abbreviated forms and onomatopoeic forms behave fairly similarly. Some of the

patterns seen in the use of face-to-face laughter, such as its association with interpersonal exchanges

and its invitation/acceptance patterns, were found to also hold for written laughter, while others, such

as gendered patterning in invitation/acceptance and, in some cases the correlation between the

location of a laugh within a turn and its identity as a laugh invitation or a laugh acceptance, could not be

shown with significance. There is a long way to go in the study of written laughter, but it is my hope

that this project will provide at least a small pool of information from which future projects may draw

ideas.

72
Acknowledgements:
I would first and foremost like to thank my amazing thesis advisor, Debby Keller-Cohen, for all of

the time and energy she’s put into helping me through the various stages of this project. I couldn’t have

done it without her. I would also like to thank my second reader Robin Queen. I would also like to

thank the other students who participated in the University of Michigan Honors Summer Fellows

program over the summer of 2014, along with the program’s organizers. Their help in brainstorming

topics and in getting this project off the ground was invaluable.

73
Appendices
Appendix 1: Observations about Written Laughter from Multimedia
Humor

74
Appendix 2: Tweet-Initial Laughter References Previous/Ongoing Topic

Example 1:

-Lol-0882

Example 2:

-Lol-0883

75
Example 3

-Haha-0604

76
Appendix 3: Examples of Laughter in Previous and Subsequent Tweets
for Various Laugh Locations:

Example 1: Tweet-Initial Laughter with a Previous Tweet containing Spelled-out Laughter (haha 241):

Dataset
Tweet

Example 2: Tweet-Initial Laughter with a Subsequent Tweet containing Spelled-out Laughter (lol 34):

Dataset
Tweet

77
Example 3: Tweet-Final Laughter with a Previous Tweet containing Emoticon Laughter (lol 15)

Dataset
Tweet

Example 4: Tweet-final Laughter with a Subsequent Tweet containing both Spelled-out and Emoticon
Laughter (haha 36):

Dataset
Tweet

78
Works Cited
Alan. (2006). lmao. Retrieved from http://www.urbandictionary.com/define.php?term=lmao&page=9
Bamman, D., Eisenstein, J., & Schnoebelen, T. (2014). Gender Identity and Lexical Variation in Social
Media. Journal of Sociolinguistics, 18(2). doi: 10.1111/josl.12080
Baron, D. (2011). Cyber-laugh: everything old is new again, haha. Retrieved from
https://illinois.edu/blog/view/25/45228
Bateson, G. (1955). A Theory of Play and Fantasy Steps to an Ecology of Mind: Collected Essays in
Anthropology, Psychiatry, Evolution and Epistemology (pp. 183-198). Northvale, New Jersey:
Jason Aronson Inc.
Bennett, J. (2014, July 27). The Emoji Have Won the Battle of Words. The New York Times. Retrieved
from http://www.nytimes.com/2014/07/27/fashion/emoji-have-won-the-battle-of-
words.html?_r=0
Blackmore, S. (1999). The Meme Machine.
Brandon, J. (2008). FWIW -- The Origins of 'Net Shorthand. PCWorld. Retrieved from
http://www.pcworld.com/article/153504/net_shorthand_origins.html
(2012, June 24). That's What They Say [Radio]. Retrieved from http://michiganradio.org/post/ever-
wonder-what-lol-means
Derks, D., Bos, A. E. R., & Grumbkow, J. v. (2008). Emoticons and Online Message Interpretation. Social
Science Computer Review, 26(3), 379-388.
Dresner, E., & Herring, S. C. (2010). Functions of the Nonverbal in CMC: Emoticons and Illocutionary
Force. Communication Theory, 20, 249-268. doi: 10.1111/j.1468-2885.2010.01362.x
dude, l. (2005). lol. Urban Dictionary. from
http://www.urbandictionary.com/define.php?term=Lol&defid=1079038
Edel, L. (1989, May 8, 1989). MO_ICONS_PLEASE. FidoNews: International FidoNet Association
Newsletter. Retrieved from http://www.textfiles.com/fidonet-on-the-
internet/878889/fido0619.txt
Edmondson, M. S. (1987). Notes on Laughter. Anthropological Linguistics, 29, 23-34.
Entity1037. (2015). Haha. Retrieved from
http://www.urbandictionary.com/define.php?term=haha&defid=8075298
EXPLIZIT. (2005). lol. Urban Dictionary. 2015, from
http://www.urbandictionary.com/define.php?term=Lol
Face with Tears of Joy. (2015). iEmoji.com. from http://www.iemoji.com/view/emoji/26/people/face-
with-tears-of-joy
Friends They Never Meet: Acquainteneces Made by the Telegraph Key. Confidences Exchanged
between Men who have Never Seen Each Other - Their Peculiar Conversational Abbreviations.
(1890, November 30, 1980). New York Times.
Glenn, P. J. (1989). Initiating Shared Laughter in Multi-Party Conversations. Western Journal of Speech
Communication, 53(Spring), 127-149.
Glenn, P. J. (1991). Current Speaker Initiation of Two-Party Shared Laughter. Research on Language and
Social Interaction, 25, 139-162.
Glenn, P. J. (2003). Laughter in Interaction. New York: Cambridge University Press.
Glenn, P. J. (2010). Interviewer Laughs: Shared Laughter and Asymmetries in Employment Interviews.
Journal of Pragmatics, 42, 1485-1498.
Glenn, P. J., & Holt, E. (2013). Introduction. In P. J. Glenn & E. Holt (Eds.), Studies of Laughter in
Interaction (pp. 1-22). New York: Bloomsbury.

79
Glenn, P. J., & Knapp, M. L. (1987). The Interactive Framing of Play in Adult Conversations.
Communication Quarterly, 35(1), 48-66.
Goffman, E. (1974). Frame Analysis: An Essay on the Organization of Experience. Boston: Northeastern
University Press.
guy12345. (2009). hehe. Retrieved from
http://www.urbandictionary.com/define.php?term=hehe&defid=3904004
ha-ha. (2015).
Heaney, K. (2014a). The 42 Ways to Type Laughter, Define: Here's what your typed and texted LOL-ing
REALLY means. Retrieved from http://www.buzzfeed.com/katieheaney/the-42-ways-to-type-
laughter-defined#.shwKDjARK
Heaney, K. (2014b). The 58 Yellow Smiley Emoji, Defined. Retrieved from
http://www.buzzfeed.com/katieheaney/the-58-yellow-smiley-emoji-defined#.wgBjRWoMj
Hiscott, R. (2014, 07/17/2014). LOL, WTF? The Origin Stories Of Your Favorite Internet Acronyms.
Huffington Post. Retrieved from http://www.huffingtonpost.com/2014/07/17/internet-
acronyms_n_5585425.html
Hobbes, T. (1640). The Elements of Law, Natural and Politic. London: Unwin Brothers.
Holt, E. (2011). On the Nature of "Laughables": Laughter as a Response to Overdone Figurative Phrases.
Pragmatics, 21(3), 393-410.
Holt, E. (2013). "There's Many a True Word Said in Jest": Seriousness and Nonseriousness in Interaction.
In E. Holt & P. J. Glenn (Eds.), Studies of Laughter in Interaction (pp. 69-89). New York:
Bloomsbury.
Jefferson, G. (1979). A Technique for Inviting Laughter and its Subsequent Acceptance Declination. In G.
Psathas (Ed.), Everyday Language: Studies in Ethnomethodology (pp. 79-96). New York: Irvington
Publishers INC.
Jefferson, G. (1984). On the Organization of Laughter in Talk about Troubles. In J. M. Atkinson & J.
Heritage (Eds.), Structures of Social Action: Studies in Conversation Analysis (pp. 346-369). New
York: Cambridge University Press.
Jefferson, G. (2004). A Note on Laughter in 'Male-Female' Interaction. Discourse Studies, 6(1), 117-133.
Jefferson, G., Sacks, H., & Schegloff, E. (1977). Preliminary Notes on the Sequential Organization of
Laughter. Paper presented at the Informal Seminars of the Ethology-Hormones-Psychosomatics
Research Groups, University of Pennsylvania.
kirshteeen. (2009). lmao. Retrieved from
http://www.urbandictionary.com/define.php?term=lmao&defid=3688532
Krohn, F. B. (2004). A Generational Approach to Using Emoticons as Nonverbal Communication. Journal
of Technical Writing and Communication, 34(4), 321-328.
Lampert, M. D., & Ervin-Tripp, S. M. (2006). Risky Laughter: Teasing and Self-directed Joking among Male
and Female Friends. Journal of Pragmatics, 38, 51-72.
Lee, V., & Wagner, H. (2002). The Effect of Social Presence on the Facial and Verbal Expression of
Emotion and the Interrelationships Among Emotion Components. Journal of Nonverbal
Behavior, 26(1), 3-25.
Løkken, R. (2006). haha. Retrieved from
http://www.urbandictionary.com/define.php?term=haha&defid=2043167
LOL, int. and n.2. (2015). Oxford University Press.
Lyddy, F., Farina, F., Hanney, J., Farrell, L., & O'Neil, N. K. (2014). An Analysis of Language in University
Students' Text Messages. Journal of Computer-Mediated Communication, 19, 546-561. doi:
10.1111/jcc4.12045
McWhorter, J. (2013). Txtng is killing language. JK!!! TED [Video File].

80
Menzer, M. J. (2004). Ælfric's English "Grammar". The Journal of English and Germanic Philology, 103(1),
106-124.
monsieur_d. (2005). hehe. Retrieved from
http://www.urbandictionary.com/define.php?term=hehe&defid=1466854
Morgan, J. (2011, April 8, 2011). Why did LOL infiltrate the language? BBC News Magazine.
Moss, C. (2014). Cosmopolitan Has A Guide On How To 'Decode' Men's Emoji. Retrieved from
http://www.businessinsider.com/cosmopolitan-how-to-decode-his-emoji-2014-10
Neuendorf, K. A., & Fennell, T. (1988). A Social Facilitation View of the Generation of Humor and Mirth
Reactions: Effects of a Laugh Track. Central States Speech Journal, 39(1), 37-48.
no_one_2000. (2005). lol. Urban Dictionary. Retrieved 3/9/2015, from
http://www.urbandictionary.com/define.php?term=Lol&defid=1406597
Osborne, K., & Chapman, A. J. (1976). Suppression of Adult Laughter: An Experimental Approach. Paper
presented at the International Conference on Humour and Laughter, Cardiff, Wales.
paperstars. (2009). hehe. Retrieved from
http://www.urbandictionary.com/define.php?term=hehe&defid=3716398
Pearson, W. The Origin of LOL. 2015, from http://pages.cpsc.ucalgary.ca/~crwth/LOL.html
Peräkylä, A. (2007). Conversation Analysis Blackwell Encyclopedia of Sociology Online.
Provine, R. R. (2001). Laughter: A Scientific Investigation. United States: Penguin Books.
Provine, R. R., Spencer, R. J., & Mandell, D. L. (2007). Emotional Expression Online: Emoticons Punctuate
Website Text Messages. Journal of Language and Social Psychology, 26(3), 299-307. doi:
10.1177/0261927X06303481
. Public API. (2013): Twitter.
Rees, C. E., & Monrouxe, L. V. (2010). "I should be lucky ha ha ha ha": The Construction of Power,
Identity and Gender through Laughter within Medical Workplace Learning Encounters. Journal
of Pragmatics, 42, 3384-3399.
rikochet. (2008). haha. Retrieved from
http://www.urbandictionary.com/define.php?term=haha&defid=3248100
Rothenberg, M. (Producer). (2015). emojitracker. Retrieved from http://www.emojitracker.com/
Sacks, H., Schegloff, E., & Jefferson, G. (1974). A simplest systematics for the organization of turn-taking
for conversation. Language, 50(5), 696-735.
Sebastian. (2014). What your favorite emojis really mean (23 photos). Retrieved from
http://thechive.com/2014/10/14/what-your-favorite-emojis-really-mean-23-photos/
Smith, R. E. (2011). Urban dictionary: youth slanguage and the redefining of definition. English Today,
27(4), 43-48.
The Streaming APIs: Documentation. (2014). Retrieved 11/9/2014, from
https://dev.twitter.com/streaming/overview
TheDanishSugarfairy. (2008). lmao. Urban Dictionary. from
http://www.urbandictionary.com/define.php?term=lmao&defid=3259576
Toole, C. (2014). Emoji Dictionary: The Honest Meanings Behind Your 15 Favorite Emojis. Retrieved
from http://elitedaily.com/humor/emoji-dictonary-true-meanings-definitions/870508/
Vuolo, M. (2013). LMAO: A history. Retrieved from
http://www.slate.com/blogs/lexicon_valley/2013/10/03/lmao_laughing_my_ass_off_has_a_lon
g_lineage_that_dates_back_to_at_least.html
Walther, J. B. (2006). Nonverbal Dynamics in Computer-Mediated Communication or :( and the Net :('s
with You, :) and You :) Alone. In V. Manusov & M. L. Patterson (Eds.), (pp. 461-480): Sage
Publishers, Inc.
Walther, J. B., & D'Addario, K. P. (2001). The Impacts of Emoticons on Message Interpretation in
Computer-Mediated Communication. Social Science Computer Review, 19.

81
Weller, K., Bruns, A., Burgess, J., Mahrt, M., & Puschmann, C. (Eds.). (2013). Twitter and Society (Vol. 89).
New York.
Xy, L., Yi, C., & Xu, Y. (2007). Emotional Expression Online: the Impact of Task, Relationship, and
Personality Perception on Emoticon Usage in Instant Messenger. Paper presented at the Pacific
Asia Conference on Information Systems (PACIS), Auckland, New Zealand.
Yang, L., Sun, T., Zhang, M., & Mei, Q. (2012). We know what @you #tag: does the dual role affect
hashtag adoption? Paper presented at the WWW '12: The 21st international conference on
World Wide Web.

82