Understanding Diverse Interpretations

of Animated GIFs

Abstract

Animated GIFs are increasingly popular in text-based

communication. Like other forms of nonverbal

and analyzed the variance in sentiment based on the

emotions participants used to describe GIFs. We find

diverse interpretations of GIFs, and that duration of

GIFs has a significant impact on interpretation. Positive

GIFs also have more variance in interpretation than

negative GIFs. Overall, we show that there is potential

for miscommunication in animated GIFs, and animated

GIFs may be a more nuanced form of nonverbal

communication than emoticons and emoji.

Author Keywords

Permission to make digital or hard copies of part or all of this work for

personal or classroom use is granted without fee provided that copies are

not made or distributed for profit or commercial advantage and that

copies bear this notice and the full citation on the first page. Copyrights

for third-party components of this work must be honored. For all other

uses, contact the Owner/Author.

Copyright is held by the owner/author(s).

CHI'17 Extended Abstracts, May 06-11, 2017, Denver, CO, USA

ACM 978-1-4503-4656-6/17/05.

Introduction

Animated GIFs have become pervasive online. These

silent, short, usually low-resolution video clips are more

engaging than any other kind of media on Tumblr [1],

and many other social network sites and instant

messaging tools, such as Facebook, iMessage, and

Slack, have incorporated animated GIFs as part of their

feelings and thoughts in ways beyond words and even

photographs” [4].

Prior work has shown that many kinds of nonverbal

communication, such as emoticons and emoji, are

interpreted in inconsistent ways [7,8]; therefore it is

likely that animated GIFs are also susceptible to a wide

range of interpretations. Varied interpretations could be

particularly problematic since animated GIFs are often

used to express emotions [2]. To explore this area

further, we asked the following research questions:

towards examining this space. Our findings reveal that

different people do interpret the same animated GIFs

differently. We also find that the duration of the GIF

contributes to degree of variance in interpretation.

These findings point to future directions for research

around using GIFs in communication, as well as for the

design of communication tools that incorporate

animated GIFs.

Related Work

Due to the limited scholarship on animated GIFs, we

emoticons and emoji. We conclude with a discussion of

existing research on animated GIFs.

Nonverbal Communication in CMC

Ambiguity in the interpretation of mediated

communication is a staple of CMC scholarship. Early

research in CMC suggested that CMC might be

inherently impersonal due to the lack of nonverbal

cues. Kiesler argued that one of the characteristics of

CMC was the scarcity of social context information

because CMC conveyed fewer contextual and nonverbal

cues abundantly available in face-to-face

In the development of SIDE theory, Lea and Spears

CMC information that people use to form impressions of

each other when communicating [6]. In the absence of

interpersonal cues, communicators form impressions

about each other from whatever limited cues are

available. Impression formation is more socially

categorical, rather than personal, impression of others.

Lea et al. also found that when communicators were

geographically separated, paralinguistic cues were

perceived positively if the group relationship was more

salient and negatively when individual identities were

more salient.

negate the effects of CMC [11]. Building on this theory,

Hyperpersonal model argues that CMC message

receivers have idealized perceptions of senders, not

only due to their over-reliance on minimal cues, but

also because the senders are able to selectively present

themselves [12]. This loop of perception intensification

made CMC exceed the level of affection of interpersonal

communication–it became “hyperpersonal.”

While only a brief summary, this research shows the

ability of nonverbal cues to shape interpretations in

CMC and provides a foundation to understanding the

and novel forms of interaction.

Emoticon and Emoji Interpretation

Emoticons and emoji are common nonverbal cues in

today’s CMC, and recent research has shown their ability

to shape perception and their varying interpretations.

Researchers found emoticons were able to shift the

interpretation of messages [13]. The interpretations of

emoticons were consistent within cultures, but varying

cross-culturally [8]. Interpretations of emoji were less

consistent both within-platform and cross-platform [7,10].

However, little is known about interpretations of animated

GIFs.

Despite animated GIFs’ high level of engagement,

interpretations of animated GIFs may not be consistent.

The GIFGIF project

vote on the one GIF that best represents a given emotion

between two animated GIFs. From the results on GIFGIF’s

website, we can see a high number of votes for multiple

emotions for a single GIF, implying various interpretations

of GIFs.

Methods

We created an online survey to solicit people’s

interpretation of a sample of animated GIFs. We

We collected a dataset of animated GIFs in the Emotion

randomly collected 100 animated GIFs along with their

meta-information from each of 26 subcategories in the

Emotion category using the Giphy API (e.g., awkward,

excited, unimpressed). Figure 1 is an example GIF from

the “excited” emotion category.

To consider what characteristics of GIFs might impact

interpretation, we grouped the GIFs in terms of two

having lengths above the median length of the sample,

and short GIFs are defined as having lengths equal to

or below the median (26 seconds). We decided to use

the median instead of average because the GIF lengths

followed a long-tail distribution and the average length

was skewed by very long GIFs (M = 38.75, SD =

38.28, min = 2, max = 540). We randomly chose 10

GIFs for each of the four groups, resulting in 40 GIFs in

total for our survey.

Survey Design

The survey began by asking participants for their

provide an emotion they associated with the GIF, and

provide any additional information about the GIF.

Participants

We recruited participants from social media (seeded

from the authors’ social networks), as well as online

communities such as Reddit and Tumblr. We

encouraged participants to share the call for

participation, resulting in a snowball sample. There

were no participation restrictions beyond a requirement

participants send or receive animated GIFs multiple

times a day, 21% multiple times a week, 23% multiple

times a month, 17% less than once a month, and 7%

have never sent or received any animated GIFs. We

collected 1606 interpretations across all 40 GIFs, with

an average of 40 interpretations per GIF (median = 40,

max = 47, min = 34).

Data Analysis

To analyze the difference in interpretations, we

computationally analyzed the variance of the sentiment

of reported emotions using the sentiment analysis tool

weighted composite sentiment score of the sentence,

To examine the variance in sentiment on the VADER

compound metric between the text and no-text groups,

and between the long and short groups. Bartlett’s test

for homogeneity of variance was used to test for

significant differences in variance between our

categories.

We found differences in variance of interpretations

between groups. A summary of the descriptive

statistics can be seen in Table 1. A summary of the

short group (SD = 0.275). This means when one sends

his or her friend a long GIF, they are more likely to

interpret it differently than the sender, as opposed to if

the sender sent a short one. Some participants also

described long GIFs as difficult to understand. For one

of the longest GIFs in the sample (Figure 3), different

participants reported: “I'm not really sure on the

emotion associated with this gif” and “hard to read,

very confusing.”

We did not find a significant difference in the variance

of interpretation between GIFs with embedded text (SD

had a higher variance than GIFs rated negative. We

positive average compound metric and GIFs that had a

negative average compound metric, and we observed

that positive GIFs (SD = 0.145) had a higher variance

in compound than negative GIFs (SD = 0.121). This

means that one’s friend is more likely to have a

different interpretation of a GIF than the sender if that

GIF is positive rather than negative in sentiment.

Overall, we found that people have different

interpretations of animated GIFs and have identified

significant effect from embedded text on the variance

of interpretation. While both text and length provide

GIFs with more context and more information, one

explanation could be that additional visual information

in GIFs results in ambiguous or multiple emotions while

textual information only serves to reinforce the existing

visual content.

Our findings also suggest that positive GIFs tend to

have more diverse interpretations than negative GIFs;

however, we are not clear what caused this result. We

can speculate two possible reasons behind this: First,

This is an interesting question that could be examined

These findings suggest that there is potential for

miscommunication when using animated GIFs, and it is

affected by duration and sentiment polarity. These

multiple factors could mean that GIFs are a more

nuanced form of nonverbal communication than

emoticons and emoji. These factors, therefore, need to

be taken into consideration when using animated GIFs

in communication and when designing communication

tools that incorporate animated GIFs. For example,

communicators can intentionally choose simple GIFs to

nuanced phenomenon, and thus there are some

limitations to our work that suggest avenues for future

studies. We considered emotions on the one-

dimensional sentiment scale VADER provided: negative

to positive. However, emotions are more complex and

subtle than a single dimension.

In future work, we plan to analyze emotions on other

dimensions (e.g. anger, happiness, sadness) to capture

the complexity of emotions. We will also look at how

reported emotions diverge from the category provided

by Giphy for each GIF on various dimensions, which

directions for further inquiry, we plan to conduct

representative samples, as well as to examine more in-

depth data by conducting interviews.

References

1. Saeideh Bakhshi, David A. Shamma, Lyndon

Kennedy, Yale Song, Paloma de Juan, and Joseph

“Jofish” Kaye. 2016. Fast, Cheap, and Good: Why

Animated GIFs Engage Us. In Proceedings of the

2016 CHI Conference on Human Factors in

Computing Systems (CHI ’16), 575–586.

3. C. J. Hutto and Eric Gilbert. 2014. VADER: A

Parsimonious Rule-Based Model for Sentiment

Analysis of Social Media Text. In Eighth International

AAAI Conference on Weblogs and Social Media.

Retrieved December 12, 2016 from

http://www.aaai.org/ocs/index.php/ICWSM/ICWSM1

4/paper/view/8109

4. Mike Isaac. 2015. For Mobile Messaging, GIFs Prove

to Be Worth at Least a Thousand Words. The New

York Times. Retrieved January 9, 2017 from

http://www.nytimes.com/2015/08/04/technology/gif

s-go-beyond-emoji-to-express-thoughts-without-

066X.39.10.1123

communication. Journal of Organizational Computing

2, 3–4: 321–341.

https://doi.org/10.1080/10919399209540190

7. Hannah Jean Miller, Jacob Thebault-Spieker, Shuo

Chang, Isaac Johnson, Loren Terveen, and Brent

Hecht. 2016. “Blissfully Happy” or “Ready toFight”:

Varying Interpretations of Emoji. In Tenth

International AAAI Conference on Web and Social

Media. Retrieved September 16, 2016 from

http://www.aaai.org/ocs/index.php/ICWSM/ICWSM1

http://www.aaai.org/ocs/index.php/ICWSM/ICWSM1

3/paper/view/6132

9. Lee Sproull and Sara Kiesler. 1986. Reducing Social

Context Cues: Electronic Mail in Organizational

Communication. Management Science 32, 11:

1492–1512.

https://doi.org/10.1287/mnsc.32.11.1492

10. Garreth W. Tigwell and David R. Flatla. 2016.

Oh That’s What You Meant!: Reducing Emoji

Misunderstanding. In Proceedings of the 18th

International Conference on Human-Computer

Interaction with Mobile Devices and Services Adjunct

(MobileHCI ’16), 859–866.

https://doi.org/10.1145/2957265.2961844

11. JOSEPH B. WALTHER. 1992. Interpersonal

Effects in Computer-Mediated Interaction: A

23, 1: 3–43.

https://doi.org/10.1177/009365096023001001

13. Joseph B. Walther and Kyle P. D’Addario. 2001.

The Impacts of Emoticons on Message Interpretation

in Computer-Mediated Communication. Social

Science Computer Review 19, 3: 324–347.

https://doi.org/10.1177/089443930101900307