Using psycholinguistic inquiry to measure emotional response to music: A feasibility study

Abstract

Linguistic Inquiry and Word Count (LIWC) software, which can analyze elements of language, has been used to measure emotional responses to film clips, yet the viability of LIWC to study emotional responses to music has not been investigated. The purpose of this study was to assess the feasibility of using LIWC software to measure emotional responses to music. Undergraduate education majors (N = 56) listened to two 90-second musical excerpts, one happy and one sad. After each excerpt, participants wrote about induced and perceived emotional responses. A repeated-measures ANOVA used to analyze effects of excerpt, writing prompt, emotion and order revealed a significant main effect for emotion (p = .01). Significant interactions were found between emotion and excerpt (p < .001) and between excerpt, prompt and emotion (p = .005). Participants used significantly more positive emotion words than negative emotion words to describe the happy excerpt, and the reverse was true for the sad excerpt. Both writing prompts elicited the expected differences in emotional response; however, the perceived emotion prompt resulted in greater contrasts between positive and negative emotion words than the induced emotion prompt. Results indicate that it is feasible to use LIWC to measure emotional responses to music.

Keywords

discourse analysis emotion language negative emotions technology valence

One author once wrote that “Emotional responses to music are undoubtedly at the core of why human beings value music so highly” (Hodges, 2010, p. 300). Research demonstrating that emotions are a prime motive for music listening (Juslin & Västfjäll, 2008) supports Hodges’ claim. When responding to music, listeners can successfully discriminate between perceived emotions (emotions that are noticed by the listener) and induced emotions (emotions that are actually felt by the listener; Zentner, Grandjean, & Scherer, 2008).

Various measurement strategies exist for investigating emotional responses to music, including physiological (Hodges, 2010), functional neuroimaging (Koelsch, Siebel, & Fritz, 2010) indirect (Västfjäll, 2010), self-report (Zentner & Eerola, 2010), and a variety of continuous response measures (Schubert, 2010). One potential measurement strategy that has not been fully examined is psycholinguistic analysis, the study of how language relates to underlying psychological processes (Crystal, 2008), including emotional responses. Social psychologists have used a quantitative approach to psycholinguistic analysis in a robust line of research across a variety of fields (Tausczik & Pennebaker, 2010). Through the use of Linguistic Inquiry and Word Count (LIWC) computer software, researchers have studied patterns of word usage in writings and transcribed speech to gain insight into covert psychological processes (Chung & Pennebaker, 2007). The software compares each word in a text document to a comprehensive internal dictionary organized into a variety of linguistic dimensions (e.g., psychological process words, cognitive process words, affect words, pronouns, and non-fluencies) and reports results as percentages of total word count (Pennebaker, Boyd, Jordan, & Blackburn, 2015).

LIWC software has been used in music research to examine psychological processes of songwriters by analyzing the lyrics of songs they wrote (Markowitz & Hancock, 2016) and to study trends in lyrics of songs from specific genres (Yinger & Springer, 2016) or periods of time (DeWall, Pond, Campbell, & Twenge, 2011). A recent study by Springer highlighted the potential for LIWC to be used to measure emotional responses to music. Springer (2014) used LIWC to analyze the presence of positive and negative emotion words in listeners’ free-response writings after listening to musical performances. His results indicated that, after listening to four excerpts of piano music, listeners’ use of positive emotion words and negative emotion words was influenced by the musical excerpt heard. Although evidence is lacking concerning the validity of psycholinguistic analysis as a measure of emotional responses to music, Kahn, Tobin, Massey and Anderson (2007) examined the validity of the LIWC software as a measure of induced emotional responses to film clips. Participants used significantly more positive emotion words to describe the way they felt after viewing a film clip of a comedy routine, and significantly more negative emotion words to describe how they felt after watching a film clip of a funeral.

Aim

The aim of this study was to assess the feasibility of using Linguistic Inquiry and Word Count (LIWC) software to measure emotional responses to music. Specific research questions included:

Will there be differences in proportions of words, as measured by LIWC, used to describe happy vs. sad music?

What are the effects of specific instructional prompts on proportions of words used to describe happy and sad music?

We hypothesized that participants would use (a) more positive than negative emotion words to describe a happy excerpt and (b) more negative than positive emotion words to describe a sad excerpt.

Method

This investigation was a feasibility study using a within-subjects design. Approval to conduct this experiment was obtained from the Institutional Review Board of the authors’ universities. Emotions were defined as relatively intense feelings of a brief duration that usually involve the coordination of various sub-components, including subjective feeling, physiological arousal, expression, action tendency and regulation (Juslin, 2009). Positive emotion was defined as the expression of words suggesting happiness and joy (e.g., love, nice). Negative emotion was defined as the expression of words suggesting sadness, anxiety and anger (e.g., hurt, ugly; Pennebaker, Boyd et al., 2015).

Participants

Participants were 56 undergraduate students, 53 females and 3 males, enrolled in a course called “Teaching Music in Elementary Grades” at a large university in the southeastern United States. They ranged in age from 18 to 28, with an average age of 20.80 years (SD = 1.85), and most were sophomores (n = 20) or juniors (n = 28). Participants were either majoring in elementary education (n = 53) or special education (n = 3).

Stimuli

The musical stimuli included a piece of music that has been described in previous research (Nawrot, 2003) as happy in character (Beethoven’s Symphony No. 6, first movement, Allegro ma non troppo, the annotation of which is often translated as “awakening of cheerful feelings upon arrival in the countryside”) and a piece previously described as sad (Barber’s Adagio for Strings). An excerpt of each musical selection (consisting of the first 90 seconds of each piece) was created using Audacity software, version 2.1.1 (Audacity Team, 2015). Excerpts were presented via stereophonic speakers connected to an iPod.

Instrument

The instrument consisted of two written prompts developed by the researchers about induced and perceived emotional responses. The researchers solicited feedback on these prompts from a panel of three experts, all of whom had served as chair or chair-elect of the National Association for Music Education Affective Response Special Research Interest Group. Their feedback was used to revise the written prompts to the versions used in this study, which were:

(a) How did this musical selection make you feel? (induced emotion prompt)

(b) What feelings or emotions do you think the music was trying to express? (perceived emotion prompt)

Procedure

Students who participated in this study did so as part of an in-class music listening assignment related to induced and perceived emotion. (All students consented to participate and submitted their data for inclusion in the study.) Data were collected from three intact class groupings. After explaining the nature of the study and obtaining informed consent from participants, a member of the research team explained the difference between induced and perceived emotional responses to music using an instructional script. Students were then asked to listen to the two excerpts of music. One of the groups (n = 23) listened to the “happy” excerpt first, and the other two groups (n = 33) listened to the “sad” excerpt first. After listening to each selection, participants responded in writing to two questions about the musical excerpts. Participants were given 2 minutes to respond to these questions. Complete instructions given to participants, along with the instructional script, can be found in the appendix (see supplementary materials). After all data were collected, a research assistant transcribed the content of the handwritten responses into Microsoft Word documents. We then used the LIWC software (Pennebaker, Booth, Boyd, & Francis, 2015) to analyze written responses for positive emotion and negative emotion, both of which were expressed as percentages of total word count.

Results

A series of t-tests for paired samples was used to determine whether there were differences in word count between responses to the happy excerpt and responses to the sad excerpt. Since LIWC expresses positive and negative emotion words as percentages based on total word count, differences in word count between types of excerpts could skew results. Means, standard deviations, and t-test results for analyses of word count are shown in Table 1. There was no significant difference in word count between perceived emotional responses to happy and sad excerpts, t(55) = -0.03, p = .973; however, participants used significantly more words to describe their induced emotional response to the sad excerpt compared to the happy excerpt, t(55) = -2.72, p = .009, d = 0.39.

Table 1.

Means and standard deviations of word count (measured in number of words) and emotion words (shown as a percentage of word count).

Excerpt	Prompt	Word count		Emotion words	Rates of emotion words
		M	SD	Emotion words	M	SD
Happy	Induced	16.48	7.84
				Positive	14.86 %	2.54
				Negative	1.10 %	0.39
	Perceived	14.48	7.42
				Positive	18.81 %	3.18
				Negative	0.18 %	0.15
Sad	Induced	20.38	11.88
				Positive	6.77 %	1.49
				Negative	10.53 %	2.14
	Perceived	14.52	8.69
				Positive	4.09 %	1.87
				Negative	17.97 %	3.28

A factorial ANOVA was conducted to compare the main and interaction effects of three within-subject variables (excerpt, writing prompt, and emotion words) and one between-subjects variable (order) on the number of words elicited. Excerpt included two levels (happy vs. sad music), as did writing prompt (induced vs. perceived emotion), emotion words (positive vs. negative emotion words) and order (happy excerpt first vs. sad excerpt first). Results of the ANOVA are shown in Table 2.

Table 2.

Results of factorial ANOVA.

	df	F	p	ω²
Within-subjects effects
Excerpt	1, 54	1.74	.19
Prompt	1, 54	2.80	.10
Emotion	1, 54	7.13	.01 *	0.03
Excerpt × prompt	1, 54	0.23	.63
Excerpt × emotion	1, 54	29.93	< .001 *	0.20
Excerpt × order	1, 54	1.76	.19
Prompt × emotion	1, 54	0.98	.33
Prompt × order	1, 54	0.05	.83
Emotion × order	1, 54	2.06	.16
Excerpt × prompt × order	1, 54	0.87	.35
Excerpt × emotion × order	1, 54	3.19	.08
Excerpt × prompt × emotion	1, 54	8.39	.005 *	0.03
Excerpt × prompt × emotion × order	1, 54	1.32	.26
Between-subjects effects
Order	1, 54	2.50	.12

Statistically significant (p < .05).

A significant main effect was found for emotion, F(1, 54) = 7.13, p = .01, ω² = .03, indicating a significant difference between the percentage of positive emotion words (M = 11.13, SD = 1.57) and negative emotion words (M = 7.44, SD = 1.11) expressed overall. No significant main effects were found for excerpt, prompt or order (p > .05). A significant two-way interaction was found for excerpt × emotion, F(1, 54) = 29.93, p < .001, ω² = .20. There were more positive emotion words (M = 16.83, SD = 2.54) than negative emotion words (M = 0.64, SD = 0.21) in descriptions of the happy excerpt and more negative emotion words (M = 14.25, SD = 2.25) than positive emotion words (M = 5.43, SD = 1.31) in descriptions of the sad excerpt. A significant three-way interaction was found for excerpt × prompt × emotion, F(1, 54) = 8.39, p = .005, ω² = .03. Profile plots for the three-way interaction between excerpt, prompt and emotion are shown in Figure 1; means and standard deviations are shown in Table 2. The perceived emotion writing prompt resulted in greater contrasts between positive and negative emotion words than the induced emotion writing prompt. No other significant interactions were found (p > .05).

Figure 1.

Profile plot showing three-way interaction (excerpt × prompt × emotion).

Discussion

The purpose of this study was to investigate the feasibility of using LIWC text analysis software as a measure of listeners’ emotional responses to music. As illustrated in Figure 1, results were moderated by the excerpt and type of prompt used. As predicted, participants wrote more positive emotion words in response to the happy excerpt and more negative emotion words in response to the sad excerpt, and a large effect was found for this difference (ω² = .20). This finding supports the feasibility of LIWC as a measure of emotional responses to music. The feasibility of the research methodology used in the present study is further supported by the 100% consent rate and the efficiency of the research protocol, which took no longer than 20 minutes to complete.

Effects differed based on the type of prompt used, and the differences between happy and sad excerpts were most obvious as a reaction to the perceived emotion prompt. Numerous researchers have also indicated differences between induced and perceived emotions (e.g., Wager et al., 2008; Zentner et al., 2008). The explanation for these differences remains unclear; however, listeners may find that identifying perceived emotions being expressed by music is easier than verbalizing their own felt emotions with accuracy. Listeners may have under-reported their own induced emotions due to this difficulty (Zentner & Eerola, 2010). For this reason, researchers should choose their type of instructional prompt based on the specific purposes of their investigation.

It appears that listeners provided the most discrimination between happy and sad excerpts in response to the perceived emotion prompt, which was presented after the induced emotion prompt. It is possible that the order in which the prompts were addressed had an effect on the difference in responses to the two prompts. Future studies could examine the potential for order effects.

Results of this study offer measurement implications for psychologists and music researchers. The data obtained through psycholinguistic analysis provide a different type of self-report—one that lacks constraints of predetermined adjectives, rating scale categories, or continuous-response instrument anchors. Psycholinguistic analysis allows for more unrestricted response mechanisms, which may be more authentic and candid than those obtained from other measures.

It is likely that using a variety of measurement approaches will yield the best results. By using this “method triangulation” (Juslin, 2009, p. 132), researchers can gain a more comprehensive perspective of listeners’ emotional responses by using data from multiple sources, such as static self-report measures, continuous measures, and physiological measures. Although LIWC could allow for greater freedom of responses, it may lack the sensitivity of other measurement tools with regard to determining nuance of emotions, since it only classifies them broadly into “positive” or “negative” emotions. Still, it seems that a form of verbal response (written or spoken) could serve as a complement to other measures of emotional responses to music. For example, when used concurrently with other self-report measures, researchers will be able to analyze both a summative post-performance measure and a more fluid free-response measure, which may offer additional insight into the listening experience.

This study is subject to certain limitations. First, the data were collected from a convenience sample of participant volunteers, almost all of whom were women, from one institution. As such, there was a risk of sampling bias, and it was not possible to test for gender differences. Second, only two stimulus selections were used. Because this was a feasibility study, the use of two excerpts allowed for some degree of parsimony (e.g., brief data collection period and elegant experimental design). Future studies with gender-balanced samples and additional stimuli are needed. Third, participants used an average of 16.5 words to describe musical excerpts, with some participants using very few words. The developers of the LIWC software cautioned that written responses with fewer than 50 words should be interpreted with care (Pennebaker Conglomerates, Inc., 2016), and the validity of the briefest responses should be cautiously interpreted as well. Since the present study compared within-subjects differences in percentages of positive and negative emotion words, the low word count is not likely to have skewed results, provided that the word count for each of the four responses written by each participant was comparable in length. Since participants used significantly more (on average, four more) words to describe the sad excerpt than the happy excerpt when responding to the induced emotion prompt, differences in responses to the induced emotion prompt should be interpreted with some caution.

Finally, it is important to note that LIWC software only provided data describing the percentage of positive and negative emotion words written by participants, which is only one facet of the complex phenomenon of emotional response. It is important for future researchers to compare results from LIWC psycholinguistic analysis with established measures of emotional responses to music, including physiological, functional neuroimaging, indirect, self-report, and continuous measures. Nevertheless, results of this study provide preliminary evidence that LIWC may serve as a useful and effective measure of listeners’ emotional responses to music.

Footnotes

Appendix

Ethical approval

Ethical approval for this project was given by the University of Kentucky Institutional Review Board and the University of South Carolina Institutional Review Board.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

References

Audacity Team. (2015). Audacity: Free audio editor and recorder (version 2.1.1) [computer application]. Retrieved from https://audacityteam.org.

Chung

Pennebaker

(2007). The psychological functions of function words. In Fiedler

(Ed.), Social communication (pp. 343–359). New York: Psychology Press.

Crystal

(2008). A dictionary of linguistics and phonetics (6th ed.). Malden, MA: Blackwell.

DeWall

N. C.

Pond

R. S.

Campbell

W. K.

Twenge

J. M.

(2011). Tuning in to psychological change: Linguistic markers of psychological traits and emotions over time in popular U.S. song lyrics. Psychology of Aesthetics, Creativity, and the Arts, 5, 200–207. doi:10.1037/a0023195

Hodges

D. A.

(2010). Psychophysiological measures. In Juslin

P. N.

Sloboda

J. A.

(Eds.), Handbook of music and emotion: Theory, research, applications (pp. 279–311). Oxford: Oxford University Press.

Juslin

P. N.

(2009). Emotional responses to music. In Hallam

Cross

Thaut

(Eds.), The Oxford handbook of music psychology (pp. 131–140). Oxford: Oxford University Press.

Juslin

P. N.

Västfjäll

(2008). Emotional responses to music: The need to consider underlying mechanisms. Behavioral and Brain Sciences, 31, 559–621. doi:10.1017/S0140525X08005293

Kahn

J. H.

Tobin

R. M.

Massey

A. E.

Anderson

J. A.

(2007). Measuring emotional expression with the Linguistic Inquiry and Word Count. American Journal of Psychology, 120, 263–286. Retrieved from https://www-jstor-org.web.bisu.edu.cn/stable/20445398

Koelsch

Siebel

W. A.

Fritz

(2010). Functional neuroimaging. In Juslin

P. N.

Sloboda

J. A.

(Eds.), Handbook of music and emotion: Theory, research, applications (pp. 313–344). Oxford: Oxford University Press.

10.

Markowitz

D.M.

Hancock

J. T.

(2016). The 27 club: Music lyrics reflect psychological distress. Communication Reports, pp. 1–13. doi:10.1080/08934215.2016.1210663

11.

Nawrot

E. S.

(2003). The perception of emotional expression in music: Evidence from infants, children and adults. Psychology of Music, 31, 75–92. doi:10.1177/0305735603031001325

12.

Pennebaker

J. W.

Booth

R. J.

Boyd

R. L.

Francis

M. E.

(2015). Linguistic Inquiry and Word Count: LIWC2015 [computer software]. Austin, TX: Pennebaker Conglomerates.

13.

Pennebaker

J. W.

Boyd

R. L.

Jordan

Blackburn

(2015). The development and psychometric properties of LIWC2015. Austin, TX: University of Texas. Retrieved from https://repositories.lib.utexas.edu/bitstream/handle/2152/31333/LIWC2015_LanguageManual.pdf?sequence=3

14.

Pennebaker Conglomerates, Inc. (2016). LIWC: How it works. Retrieved from http://liwc.wpengine.com/how-it-works/.

15.

Schubert

(2010). Continuous self-report measures. In Juslin

P. N.

Sloboda

J. A.

(Eds.), Handbook of music and emotion: Theory, research, applications (pp. 223–253). Oxford: Oxford University Press.

16.

Springer

D. G.

(2014). Does slow music always sound sad? An exploratory study of listeners’ affective responses to music using linguistic inquiry. In Forrest

(Ed.), Proceedings of the International Society for Music Education 31st World Conference on Music Education (pp. 313–320). Victoria, Australia: International Society for Music Education.

17.

Tausczik

Y. R.

Pennebaker

J. W.

(2010). The psychological meaning of words: LIWC and computerized text analysis methods. Journal of Language and Social Psychology, 29, 24–54. doi:10.1177/0261927X09351676

18.

Västfjäll

(2010). Indirect perceptual, cognitive, and behavioural measures. In Juslin

P. N.

Sloboda

J. A.

(Eds.), Handbook of music and emotion: Theory, research, applications (pp. 255–278). Oxford: Oxford University Press.

19.

Wager

T. D.

Barrett

L. F.

Bliss-Moreau

Lindquist

K. A.

Duncan

Kober

… Mize

(2008). The neuroimaging of emotion. In Lewis

Haviland-Jones

Barrett

(Eds.), Handbook of emotions (3rd ed., pp. 249–267). New York: Guilford Press.

20.

Yinger

O. S.

Springer

D. G.

(2016). Analyzing recommended songs for older adult populations through linguistic and musical inquiry. Music Therapy Perspectives, 34, 116–125. doi:10.1093/mtp/miu048

21.

Zentner

M. R.

Eerola

(2010). Self-report measures and models. In Juslin

P. N.

Sloboda

J. A.

(Eds.), Handbook of music and emotion: Theory, research, applications (pp. 187–221). Oxford: Oxford University Press.

22.

Zentner

M. R.

Grandjean

Scherer

K. R.

(2008). Emotions evoked by the sound of music: Characterization, classification, and measurement. Emotion, 8, 494–521. doi:10.1037/1528–3542.8.4.484