Abstract
Brian Wilson of The Beach Boys is considered one of the most prolific artists of the 20th century. At the same time, he is also known for his persisting struggles with mental health. Considering that most of his 60+-year career consists of published songs and lyrics, he makes an interesting case study to understand the dynamics of language use in relation to psychological experiences over the course of an individual’s life and career. We identified song lyrics for which he is credited as (co-)writer and analyzed their lyrics (N = 250) using the Linguistic Inquiry and Word Count (LIWC) program. Considering features previously associated with mental health (I-talk, Emotional Tone, Analytic language and Cognitive Processes words) over time, we found that I-talk and Cognitive Processes showed an initial increase, followed by a plateau (in 1967), while Emotional Tone showed a cubic relationship with time, being sensitive to changes in Brian Wilson’s life. Analytic language was highly variable and did not show any patterns. Our data suggest that even carefully crafted language, such as song lyrics, can carry information about the writer’s mental health and that considering language use over the course of an individual’s lifespan can provide crucial insights into psychological processes.
It is almost impossible to imagine a world without music and songs, just like it is near impossible to imagine a world without language. Music is deeply rooted in human culture and songs and their lyrics are often used to form and express identities. Song lyrics are a form of artistic expression that uses language and can evoke deep thoughts and feelings, enrich music with narrative detail, and impact the interpretation and perceived meaningfulness of a song. While language in a variety of forms has been the subject of psychological inquiry (with the goal of extracting information about a person from words they produced, see Dehghani & Boyd, 2022), understanding psychological aspects of song lyrics, and the writers behind them, has only recently been of interest to Psychology.
Early research has shown that lyrics in popular songs may change depending on the wider social and economic conditions of society at the time of their creation. For example, Pettijohn and Sacco (2009) connected lyrical trends to external events by analyzing the lyrics of every song that hit #1 on the Billboard charts from 1955 to 2003 for emotional content and social processes. Results showed that songs have more meaningful themes and tend to be more romantic and comforting in times of financial and social hardship. Songs that were popular during these times also focused on the future and emphasized intergroup dependence. Another study (DeWall et al., 2011), found that language in U.S. pop-song lyrics changed over time, with songs becoming less positive and increasingly self-focused, rather than other-focused, which they interpret as song lyrics reflecting cultural changes.
Several famous artists have inspired generations and left a permanent mark on music history by creating new genres, using unique technical approaches to music, or creating especially memorable songs. One such prolific artist is Brian Wilson, founding member of The Beach Boys, an influential pop-rock band. The band consisting of Brian, his brothers Dennis and Carl, his cousin Mike Love, and family friend Al Jardine, rapidly rose to fame with easy-going songs discussing surfing, romance, and youth, also known as “California sound” in the 1960s, then faltered in popularity in the early 1970s, and later emerged as a successful “classic hits” band in the late 1970s. Throughout his career, which continues until the present day, Brian Wilson has been described as “genius” for his musical aptitude and novel and virtuous ways of musical composition, which resulted in highlights such as The Beach Boy’s album “Pet Sounds”, considered one of the best albums ever made according to the Rolling Stone Magazine Best Albums of all Times list, and the legendary abandoned project “Smile”. However, Brian is also known for his complicated mental health, with documented episodes of depressive symptoms, auditory hallucinations, substance abuse, and hospitalization in psychiatric facilities stating in the late 1960s, and prompting the involvement of Eugene Landy, an abusive therapist that dominated Wilson’s life from 1975 to 1976 and again from 1982 to 1991. Brian Wilson’s life and music continues to capture the interest of the public.
The work of some musical icons has been studied in prior research, considering the collective body of work of the individual bands or artists. For example, Petrie and colleagues (2008) analyzed the song lyrics of The Beatles, specifically focusing on emotional tone, cognitive processing, and emotional disclosure, amongst others. They found that The Beatles’ lyrics became more negative and less personal as time progressed. Bob Dylan’s lyrics (N = 241) were analyzed by Czechowski and colleagues (2016) using a mixed-methods approach, showing that quantitative and qualitative measures can inform each other. Kalichman and Smyth (2021) analyzed Bruce Springsteen’s lyrics and found that authenticity in his lyrics changed with each album, not showing any relationship with time, but fewer cognitive processing words were used over time. In 2024, the same authors also analyzed the lyrics of Tom Petty (N = 200), finding dynamic changes in language features indicating analytical thinking (Kalichman & Smyth, 2024). Finally, representing the only female artist, Dolly Parton’s lyrics were investigated by Eaton and colleagues (2022), showing that her lyrics remained positive throughout her career, contrary to other popular music’s lyrics around the same time.
All of these studies used LIWC, the Linguistic Inquiry and Word Count Program (2022 version: Boyd et al., 2022) to extract psychologically relevant information from the song lyrics. LIWC is currently one of the most widely used and extensively validated word-count based language analysis programs. It is used in psychology and beyond to investigate word use in both spoken and written language (Boyd & Schwartz, 2021; Pennebaker et al., 2003). It sorts words from a given input text into linguistic categories from the internal dictionary, and then calculates the percentage of words in the text that belong in each category. Categories are carefully developed to reflect psychologically meaningful features (Tausczik & Pennebaker, 2010). While the aforementioned studies employed LIWC to study song lyrics, none of them has attempted to include a psychologically meaningful aspect of the personal life of the artist, for example their mental health. Some especially remarkable artists, such as Brian Wilson, also experienced struggles with mental health, and due to their lives within the public eye, there is information available regarding both their artistic output and their mental health.
Prior research on language markers of mental health has identified I-talk (as a marker of self-focus; for example, Berry-Blunt et al., 2021; Rude et al., 2004; Tackman et al., 2019; Zimmermann et al., 2017), Emotional Tone (as marker of sentiment; for example, Cohn et al., 2004; Lyons et al., 2018; Seraj et al., 2021), Cognitive Processing words (as markers of meaning making concerns; for example, Konopasky et al., 2020; Pennebaker & Francis, 1996), and Analytic language (as marker of an involved experiential vs. detached rationale processing; for example, Jordan et al., 2019; Monzani et al., 2021) as central linguistic variables. Specifically, greater I-talk, or first-person singular pronoun use (I, me, my) has been reliably associated with increased negative emotionality and depressive symptoms (Edwards & Holtzman, 2017; Tackman et al., 2019). Lower Emotional Tone, indicating a strong negative sentiment, has also been associated with depressive symptoms as well as other conditions characterized by mental distress (e.g., Borderline Personality Disorder, Generalized Anxiety Disorder; Hartnagel et al., 2024; Lyons et al., 2018). Analytical language was recently linked to the need for cognition, or motivation to engage in thinking (Markowitz, 2023). Lower Analytical language scores reflect lower motivation to engage in thinking. Last, All-or-None Thinking, a variable reflecting thinking in extremes using words such as “always” and “never” has been suggested to relate to anxiety, depression, and suicidal ideation, tracking severity of these disorders more closely than Emotional Tone (Al-Mosaiwi & Johnstone, 2018). Specifically, more All-or-None Thinking words reflect greater symptom severity, but it is unclear if this is a symptom of depressive episodes or a trait-like marker of disposition or vulnerability.
Most prior work showing associations between LIWC features and mental health did not track depressive language over time but relied on single timepoint comparisons between depressed and non-depressed individuals. Depression is a dynamic condition; depending on the person, depressive episodes may last from weeks to years and individuals may experience intervals between depressive episodes ranging from months to multiple years, while some may never experience remission. Hartnagel and colleagues (2024) investigated emotion words in diary entries of participants with depression undergoing sleep deprivation therapy. Over the course of 26 days, they found that variation in emotion word use in diary entries mirrored variation in momentary affective states. One study, O’Dea et al. (2021), followed bloggers over the course of 36 weeks, recording both their blog posts and depression and anxiety symptoms every other week. They found that within-person analyses failed to show correlations of LIWC-features and depression and anxiety symptoms and suggested that these features may not be sensitive enough to show symptom change over time. However, their study was limited to 36 months in length and offered a fine-grained (2 weeks) temporal resolution. Considering that language change might be slow, investigating it over the course of multiple decades, such as within a person’s career, could provide additional insights. Another study, found that individuals with higher depressive symptoms used, among others, more first-person singular pronouns and negative emotion words, when texting with close contacts, but not distant contacts (Meyerhoff et al., 2023). Thus, it is currently unclear whether language use dynamically changes with a person’s depression levels, especially over a longer timespan, or whether language features reflect more trait-like behaviors and are insensitive to changes in mental health.
How language use varies within a single person’s body of work, for example over the course of their career, or in relation to their psychological health, has not been considered extensively by prior work. One case study looked at language changes in explorer Henry Hellyer, examining changes in first-person singular pronoun use and negative emotion words leading to his apparent suicide (Baddeley et al., 2011). The study of language used in song lyrics of the same artist over the course of their career can give a unique perspective on the writer and provide insights into how dynamic such features are. Due to his creative output and the information available to the public about his life and mental health, Brian Wilson makes an interesting case to understand variations in language use reflected in his writing over the course of his life.
The aim of the current study was to characterize the change in linguistic features in songs written and co-written by Brian Wilson over the course of his career (data available for 1961−2005) as a case study. Specifically, we were interested in the trajectories over time of features that have previously been associated with mental health, including first-person singular personal pronouns (“I-talk”), Emotional Tone (an overall sentiment index composed of positive and negative emotion language), Analytic language, and Cognitive Processes words, extracted with the LIWC Program. While we do not have access to fine-grained psychometric measures, such as systematically assessed depressive symptoms or anxiety scores throughout Wilson’s life, we expected variation in features to temporally, at the scale of years, coincide with major mental health life events for Wilson. We expected to see increases in I-talk and Cognitive Processes words (indexing more self-centered, ruminative thought) during periods with worse mental health or depressive episodes, as well as an increase in All-or-None Thinking, indexing more absolutist thinking. Furthermore, we predicted a decrease in Emotional Tone (indexing verbal expressions of negative emotionality) and Analytic language use (indexing a more informal, narrative, and experiential as compared with formal, logical, and analytical thinking style).
Method
All data used in this project are public information from various sources. Our datasheets are available on the accompanying OSF project page (https://osf.io/ksxvh/). We collected two kinds of data, song lyrics with metadata and major life events providing biographical context of Brian Wilson’s life.
Song lyric data
For the song lyrics data, we used the appendix of Philip Lambert’s “Inside the Music of Brian Wilson” book to extract song titles, years recorded and released, songwriting credits, and others. We then cross-checked each song’s writing credits with the Broadcast Music, Inc. (BMI) website (https://repertoire.bmi.com), the performance rights organization that handles Wilson’s music licenses, to verify Wilson’s involvement for each song. This resulted in a dataset of 405 songs. Of those, 49 songs were excluded because we could not verify Brian’s involvement via the BMI website. An additional 106 songs were excluded because while we could verify information from both sources, we could not find recordings and/or lyrics (e.g., songs had no lyrics), the lyrics consisted of a single word or phrase (N = 2) or were not original writing (lyrics from “Here comes Santa Claus,” N = 1). For 34 songs, conflicting information about the involvement of co-authors was available from both sources. We decided to keep these 34 songs in the final dataset, as Brian Wilson was still listed as a (co-) author in both sources. The final dataset for subsequent analyses contained 250 songs written or co-written by Brian Wilson. Though the initial recording of some songs was sometimes documented down to the day, only year information was retained to harmonize the dataset and match cases in granularity.
Lyrics were collected using websites AZLyrics (https://www.azlyrics.com/) and Genius Lyrics (https://genius.com/) in line with prior studies (e.g., Kalichman & Smyth, 2021). If available, demo version lyrics were used, as we believe that earlier versions would better reflect Wilson’s state of mind at the time of writing. If demo lyrics were unavailable, the first official release version was used. Lyrics were copied into text files and manually compared with audio recordings on Spotify (https://open.spotify.com/) and YouTube (https://www.youtube.com/). Filled pauses to match rhythmic patterns, such as “ooh” and “aah,” were deleted from text files, as they were not of interest for the current analysis and often varied between different sources and recordings.
Biographical information
For the biographical information, a timeline of significant life events for Brian Wilson was compiled. We used two written biographies as sources: “God Only Knows” by David Leaf (2022) and the autobiography “I am Brian Wilson,” co-written by Wilson and Ben Greenman (2016). In addition, we used three documentaries on Wilson’s life: “Long Promised Road” (2021), “I Just Wasn’t Made For These Times” (1995), and “Brian Wilson–Songwriter 1962-1969” (2010), as well as Brian Wilson’s official website (https://www.brianwilson.com). The extracted events were grouped by year to match song lyric metadata and because sources often did not specify the month or day on which events took place. Events broadly fell into four categories: (1) mental health, including hospitalizations, (2) personal events, including platonic and romantic relationships, (3) band relationships (The Beach Boys), and (4) music & career, which included events such as album releases or band finances. Events were assigned a valence rating of “positive” (e.g., Brian’s marriage to his recently deceased wife Melinda in 1995), “negative” (e.g., onset of auditory hallucinations in 1962), and “neutral” if a valence could not be unambiguously assigned or the event was of ambivalent valence (e.g., increased competition between The Beach Boys and The Beatles in 1964). The full list of events and categorizations is available in the study repository on OSF. While the extraction of life events was done as systematically as possible, we acknowledge that there is a subjective component involved in the collection and coding of these events, as well as differences in the amount of information available about his life during different times, including very little after the 1990s.
Data processing
Text files of the final 250 song dataset were run through the LIWC Program 2022, version 1.4.0. LIWC has been used in the analysis of song lyrics before (e.g., Kalichman & Smyth, 2021, 2024; Petrie et al., 2008; Pettijohn & Sacco, 2009). It uses an internal dictionary to categorize and summarize words and outputs scores per category that represent the percentage of words within that category from the total number of words in the document. Language categories considered sensitive to mental health based on previous research included first-person singular pronouns (“i”), emotional tone (“Tone”), words reflecting cognition (“cogproc”), words reflecting analytical language (“Analytic”), and word reflecting absolutist thinking (“allnone”). We focused on the linguistic dimension of “I-talk”, which is the relative frequency of first-person singular personal pronouns (I, me, my, myself) used, and the summary variable Emotional Tone, which is the degree of positive emotional tone, derived from the difference between positive tone words (e.g., good, new, love) and negative tone words (e.g., bad, wrong, hate) and reflecting sentiment (not emotion) on a scale from 0 (solely negative, no positive tone) to 100 (solely positive, no negative tone). In other words, a higher score on Emotional Tone indicates a more positive emotional tone. In addition, we considered the psychological processes category Cognitive Processes, which includes words related to cognitive processing (but, not, if, or, know) and can indicate the degree to which meaning has been made, and Analytical Thinking, which is a metric of logical, formal thinking (e.g., using more articles and prepositions). Lower analytical thinking scores indicate more personal, narrative, and experiential thinking patterns as well as a lower interest/motivation to engage in thinking. Last, we considered All or None Thinking, a variable indicating absolutist thinking, including words such as “all, none, never, always”.
Analytic strategy
All analyses were conducted in the statistical software R and Rstudio, using the packages ‘earth’ and ‘tidyverse’, and analysis scripts are available in the study repository on OSF. We were interested in the changes in LIWC variables of interest (I-talk, Emotional tone, Analytic, Cognitive Processes, and All-or-None) over the course of Wilson’s career. We separately plotted these four language feature trajectories over time, using the year they were first recorded, and applied a locally estimated scatterplot smooth (LOESS) function. LOESS is a non-parametric smoothing function that applies linear regression locally within the data, to show local trends. The relationships between time (earliest year recorded) and linguistic category did not appear linear (see Figures in Appendix on OSF). To parsimoniously characterize the data using a single mathematical function, we systematically compared three non-linear functions with the linear relationship between time and linguistic category, including a (1) logarithmic, (2) quadratic, and (3) cubic function. We then statistically compared model fit via analysis of variance (ANOVA), so that more complex models were only preferred if they also led to a significantly improved model fit as measured via change in R2 (additional variance explained). In addition, we used Multivariate Adaptive Regression Splines (MARS) to understand if and when changes occurred in the association of time and the respective LIWC variable. The MARS algorithm allows combining several linear regressions with different slopes. It assesses all potential changing points (“knots”) within the association, then prunes to the optimal number of knots by monitoring the change in R2, relying on Generalized Cross-Validation.
Results
A narrative timeline of Brian Wilson’s life and mental health events
Brian Wilson co-founded The Beach Boys in 1961, when they recorded their first Top 20 Billboard song “Surfin’” in his childhood home. The Beach Boys quickly became increasingly popular and commercially successful and signed a contract with Capitol Records in 1962. With increasing success in the years 1962−1965, both within and outside of the US, pressure on the band mounted to produce successful records, play live concerts on tours and festivals, and maintain their “surf culture” image. Brian Wilson, who started to experience auditory hallucinations (1963) and anxiety attacks (1964), withdrew more and more from the performance operations of The Beach Boys. He shifted into producing songs with studio musicians while the band toured, using novel and unusual ways of approaching music. In 1966, his creative endeavors resulted in the mature-sounding album “pet sounds,” a project Brian was proud of but which only sparked a lukewarm response at the time. Increasingly under pressure, Brian began working on Smile (1966), a project he later abandoned (1967) due to disagreements with collaborators, increased external pressure, and continued mental health struggles, resulting in a complete withdrawal (1967) and hospitalization (1968). During this time, the popularity of The Beach Boys faltered and the band openly admitted to struggling financially (1969).
Brian’s involvement in The Beach Boys declined and his mental health continued to deteriorate over the next few years, including substance abuse issues and reclusiveness (1973−1975), until 1975, when Brian’s family hired Eugene Landy, a psychotherapist that later turned abusive toward Brian. In 1976, after Landy was released from his duties, Brian briefly rejoined the band more publicly, but Mike Love ultimately took charge of the group (1977) and Brian had minimal involvement in the band thereafter. After separating from his first wife (1978), Brian was checked into a mental hospital again (1979) and continued to experience substance abuse issues, until Eugene Landy was rehired in 1982. Landy took complete control of Brian’s life, dictating who he had contact with, and Brian often was strongly medicated while under Landy’s supervision. In 1983, Brian’s brother Dennis drowned. The control of Brian through Landy continued until 1988, when abuse allegations against Landy surfaced, his medical license was revoked (1989), and ultimately, a restraining order was placed against him by the Wilson family (1992). In 1995, Brian remarried, and eventually published a solo album in 1997, followed by solo-tours in 1999 and 2000. His brother Carl died of lung cancer in 1998. While less acute, Brian continued to struggle with anxiety until 2005, possibly longer. When considering Brian Wilson’s mental health, we see a drastic deterioration with the increase in pressure following The Beach Boys’ commercial success in the 1960s, resulting in his hospitalization in 1968. A decrease in productivity is visible, and temporally coincides with significant sustained mental health challenges in Brian’s life, including depression (1973), substance abuse and reclusiveness (1973−1975, 1980), suicidal ideation (1975), and a stay in a mental hospital (1979). By the 1990s, Brian’s mental health improved, after a restraining order was placed on Landy in 1992, and Brian met (1986) and married (1995) his second wife Melinda. A visual representation of Brian Wilson’s life events can be found in Figure 1.
Timeline of Brian Wilson’s Significant Life Events Established via the Procedure Described in Methodology. Green Reflects Events Evaluated as Positive, Red Reflects Events Evaluated as Negative, and Black Reflects Events Evaluated as Neutral or Mixed.
One aspect to consider about this dataset is that the number of songs varies over time, with periods of abundance and scarcity. There is a concentration of songs in the 1960s, followed by a period of few songs available in the early 1970s. A spurt of productivity occurs in the late 1970s, followed by a decline in productivity until the mid-1980s, followed by another period of productivity until 1990, and from then on, sporadic data for most years, except for a spike in productivity in 1997. When considering our results, this natural variation in creative output (or published creative output) should be kept in mind.
Empirical analysis of the association between time and linguistic features
For I-talk, the linear model explained 0.5% of variance (F(1, 248) = 1.24, p = .27) in I-talk (b = 0.03, confidence interval [CI] = [−0.03, 0,09]), compared with 0.8% of variance in both the logarithmic (F(2, 247) = 0.97, p = .38) and quadratic models (F(2, 247) = 0.97, p = .38), and 1.0% of variance in the cubic model (F(3, 46) = 0.80, p = .49). An ANOVA comparison of the models shows no statistically significant difference in model fit between these four models.
A MARS model identified one knot in 1967, and explained 2.0% of variance in I-talk. The knot in 1967 showed that I-talk increased (7.99) until 1967, followed by a period of slight decrease (b = −0.44). This is visualized in Figure 2, top left panel. Based on R2, we would favor the MARS model, showing 1967 as a significant turning point in I-talk.
Selected Models. MARS for I-talk (Left Top) and Cognitive Processing Words (Left Bottom), Cubic Association for Emotional Tone (Right top) in Brian Wilson’s Songs, MARS for All-or-None Thinking (Right Bottom). Periods of Positive (=Green), Negative (=Red), and Neutral (=Gray) Valence in Wilson’s Life Are Marked, see Figure 1 and Narrative Timeline Results for Details.
For Cognitive Processes, the linear model explained 1.68% of the variance (F(1, 248) = 4.24, p = .04) in cognitive processes words (b = 0.06, CI = [2.72, 0.12]), compared with 1.68% for the logarithmic (F(2, 247) = 2.11, p = .12) and quadratic models (F(2, 247) = 2.11, p = .12), and 2.04% of variance for the cubic model (F(3, 246) = 1.71, p = .17). An ANOVA comparing model fit showed no significant differences between either model.
A MARS model identified one knot in 1967 and explained 2.13% of the variance in Cognitive Processes words. The knot in 1967 showed that Cognitive Processes words increased (9.10) until 1967, followed by a mild decrease (b = −0.47). This is visualized in Figure 2, left bottom panel. Based on R2, we would favor the MARS model, showing 1967 as a significant turning point in Cognitive Processes words, similar to I-talk, which also identified 1967 as a turning point.
For Emotional Tone, the linear model explained 0.60% of variance (F(1, 248) = 1.48, p = .22) in emotional tone (b = 0.22, CI = [−0.14, 0.5]), compared with 1.76% in the logarithmic model (F(2, 247) = 2.22, p = .11), 1.75% in the quadratic model (F(2, 247) = 2.20, p = .11), and 4.96% in the cubic model (F(3, 246) = 4.28, p = .006). An ANOVA comparison of the models showed no difference between the first three models, and a significant increase in model fit for the cubic model (p < .001). The cubic model is visualized in Figure 2, right top panel. A MARS model identified one knot in 1969 and explained 2.98% of the variance in Emotional Tone. The knot in 1969 indicated a period of strong increase in Emotional Tone until 1969 (63.6), followed by a decline (b = −2.24) for years following 1969. Based on R2, the cubic model explained the most amount of variance. It shows that there was an initial increase in Emotional Tone (meaning song lyrics became more positive), followed by a period of decline and/or stagnation, followed by a second increase as time progressed.
For Analytic, the linear model explained 0.38% of the variance in Analytic, compared with 0.72% of variance for the logarithmic and quadratic models, and 1.4% of variance in the cubic one. An ANOVA comparison of the models showed no significant difference between either model. A MARS model converged on no knots, explaining none of the variance in Analytic words. Based on the absence of any significant differences in variance explained, we conclude that there is no single (and therefore no clear) relationship between Analytic words and time.
For All-or-None Thinking, the linear model explained 3.20% of variance (F(1, 248) = 8.19, p = .005) in all-or-none thinking (b = 0.04, CI = [0.01, 0.07]), compared with 3.91 % in the logarithmic model (F(2, 247) = 5.02, p = .007), 3.91% in the quadratic model (F(2, 247) = 5.02, p = .007), and 3.92% in the cubic model (F(3, 246) = 3.35, p = .02). An ANOVA comparison of the models showed no difference between the three models. A MARS model identified one knot in 1974 and explained 3.82% of the variance in All-or-None Thinking. The knot in 1974 indicated a period of increase in all-or-none thinking after 1974 (b = 0.07), following stagnation in the years before 1974. Based on R2, the cubic, logarithmic, and quadratic models explained near similar amounts of variance as the MARS model. These models show that there was an increase in All-or-None Thinking (meaning song lyrics became more absolutist in thinking, containing more all-or-none words), following the initial period of stagnation or very small changes.
Discussion
We set out to investigate changes in linguistic features extracted from Brian Wilson’s written and co-written song lyrics over time, specifically in relation to his mental health. Our hypotheses were that we would observe an increase in I-talk and Cognitive Processes words during periods of mental distress and decreases in Emotional Tone (song lyrics becoming more negative) and Analytic language during those times.
Research findings
I-talk and cognitive processes
In line with our hypotheses for I-talk and Cognitive Processes words, we found an increase in both during Brian’s early years, but did not observe further discernible differences in later years. Interestingly, MARS models for I-talk and Cognitive Processes showed a knot in 1967, with increases until 1967, followed by periods of mild decline, which we interpret as indexing a plateauing effect. Elevated I-talk and Cognitive Processes words have previously been associated with depression and mental distress (e.g., Konopasky et al., 2020; Tackman et al., 2019), and our findings are in line with the development of mental health challenges in Brian Wilson’s life in the late 1960s. After the initial increase, however, we see a plateau of both I-talk and Cognitive Processes words following 1967, which could indicate that either Brian’s mental health continued to be poor, or that the language he used in his lyrics is less sensitive to changes in depressive symptoms but rather, depressive disposition (see Rude et al., 2004). While the present data cannot tease these two explanations apart, they raise interesting questions regarding what aspects of depression I-talk and Cognitive Processes words specifically reflect. In addition, it seems that, from our data, changes in I-talk and Cognitive Processes words go hand-in-hand, behaving in similar ways within our dataset and thus, possibly reflect a shared psychological process.
One could argue that variability in I-talk over time, as an approximation for mental health fluctuations, confounds potential developmental changes in I-talk with changes due to mental health. Developmental changes in I-talk have been investigated in Pennebaker and Stone (2003), showing that I-talk, as well as negative affect words, decrease across the lifespan (also see Kern et al., 2014). Such a decrease in our data is mild at best and rather seems like a stagnation at a fixed level across time. Furthermore, changes in Pennebaker and Stone (2003) were based on between-individual differences, not within-person variation of I-talk. An absence of change in I-talk related to mental health, at least from 1967 onwards in our data, would be in line with O’Dea et al. (2021), who did not observe a relationship between depressive symptoms and I-talk in blogposts of the same authors across 36 months. Thus, our findings highlight the need of the field to understand changes in language use within a person’s life.
Analytic
Our data do not support the hypothesis that Analytic language varies with changes in Brian Wilson’s mental health. No knots were identified using MARS in the Analytic word data and none of the non-linear functions explained significantly more variance than the linear relationship between Analytic words and time. Analytic word use indicates the level of personal, narrative thinking versus more formal, logical thinking. With a mean Analytic word score of 34.43, a standard deviation of 30.91, and a range of 98, meaning almost the entire 0 to 100 scale, it seems clear that there is a lot of variation in Analytic word use across Wilson’s body of work, suggesting that perhaps no single relationship could be identified by our models, or no (simple) relationship between time, an approximation for mental health, and Analytic word use is present. What stands out is the very large variability in Analytic word use, one that seems to be specific to Wilson’s writing. In fact, analytic word use, formerly known as Categorical-Dynamic Index (CDI) was identified in prior work (e.g., Boyd & Pennebaker, 2015) as a differentiator in author attribution work. Perhaps, the large variation within this category is due to the thematic diversity of Wilson’s songs, spanning from or due to variations in the degree of engagement in the songwriting process.
All-or-none thinking
Our data do not support the idea of increased All-or-None Thinking with increased depressive symptoms. In addition, All-or-None Thinking seems to capture underlying psychological processes different from both I-talk and Cognitive Processing words. We saw near stagnation of All-or-None Thinking in early years, followed by an increase in All-or-None Thinking after 1974. In the years after 1974, Brian lived a life more removed from the public eye, marked by continued mental health struggles, negative life events, and being under the influence of Eugene Landy. However, he also went through more positive phases in life starting in 1985, a timepoint not reflected in all-or-none thinking. Thus, from our data, it seems that All-or-None Thinking does not reflect depressive episodes (cf. Al-Mosaiwi & Johnstone, 2018) but could potentially indicate more vulnerability to, or accumulated life experience with, depression and anxiety.
Emotional tone
Our data appear to support the hypothesis that Emotional Tone varies with Brian Wilson’s mental health. However, contrary to our hypothesis, Emotion Tone initially became more positive during time periods of worsening mental health (early years), then, in line with our hypothesis, decreased during continued mental health treatment, and increased in later years during improved mental health periods. Emotional Tone in songs written by Brian Wilson showed the most variation over time with the cubic model explaining 4.9% of the variance in emotional tone scores. The model showed an initial increase in positive emotional tone, followed by a decrease (or plateau), followed by another increase. The cubic model does not identify when change happens, but it demonstrates that there is a non-linear relationship between Emotional Tone and time, showing the dynamic nature of this feature. Emotional Tone reflecting external life circumstances would be in line with findings from Pettijohn and Sacco (2009), who found that song themes varied with the social and economical aspects of the times songs were published in.
One last possible explanation for the increase in Emotion Tone in the 1960s, despite Brian Wilson’s deteriorating mental health, could be the band persona of The Beach Boys. In an exploratory analysis, we investigated the idea that the increase in Emotional Tone in Brian’s early song lyrics may have been primarily affected by The Beach Boys developing easy-going, “California” band persona, discussing positive, experiential themes like surfing and romance. To test this, we repeated our analyses of Emotional Tone and time on a subset of songs written specifically for the Beach Boys (n = 171). What stands out is that the bulk of songs written for The Beach Boys were written between 1960 and 1970, with a handful of additional songs in the mid-1970s, and few songs after that. Thus, from this indication, it seems that especially Brian’s later work is independent of The Beach Boys.
Statistically, a linear model had an R2 = .037 (F(1, 169) = 6.42, p = .01) with Emotional Tone (b = 1.14, CI = [0.25, 2.04]), while the logarithmic model (F(2, 168) = 3.92, p = .02), and quadratic (F(2, 168) = 3.92, p = .02) showed an R2 = .045. The cubic term did not converge. An ANOVA showed no significant difference between the three models. A MARS analysis for Emotional Tone indicated a single knot in 1964, with an R2 = .036. Emotional Tone sharply increased until 1964 to 62.01, then decreased at −10.43 units per year. Compared with all songs written by Brian Wilson, The Beach Boys songs showed a similar initial increase in Emotional Tone, but the decline in Emotional Tone for the Beach Boys songs is more pronounced (b = −10.43) and earlier (1964) than for all songs (b = −2.24, 1969). Such differences might reflect differences between Brian Wilson and the Beach Boys or be caused by the longer time interval covered within the full Brian Wilson dataset. While the number of The Beach Boys songs with Brian Wilson’s involvement declines after 1976 and ends in the late 1980s, Brian Wilson’s songs written outside of The Beach Boys are represented in our dataset until 2005. In addition, we see Emotional Tone in the full dataset best described with a cubic function. For example, around the time of Brian Wilson’s marriage to his second wife Melinda in 1995, Emotional Tone in his lyrics becomes more positive again, tracing some more positive life events. Thus, we believe that the initial increase in Emotional Tone might be fueled by the forming Beach Boys band persona, but later changes more clearly reflect Brian’s life circumstances.
Implications and outlook
Research limitations
Our research comes with important constraints on generalizability. First, we want to caution the reader that our analyses here are limited to The Beach Boys songs that involved Brian Wilson, not all The Beach Boys’ songs, so we abstain from making inferences about the changes in the musical character of The Beach Boys. Brian Wilson’s career and The Beach Boys are so enmeshed that it is not possible for us, or anyone, to cleanly tease apart effects of changes in The Beach Boys dynamics and changes in Brian’s mental health, as both temporally overlap and are likely entangled. Brian in the 1960s, was considered the de-facto leader of the Beach Boys and considered the band “his” project, but later, he and The Beach Boys developed in different directions, and ultimately, the band continued without his involvement, while Wilson continued with other projects. Despite this, our results overall point to the idea that Brian Wilson’s personal situation shines through in his song lyrics, regardless of whether he wrote songs alone or in collaboration and regardless of whether he wrote them for The Beach Boys or not. However, our data cannot conclusively address the degree to which such coloration is visible, or directly link the linguistic markers in his lyrics to Brian’s mental health due to the lack of objective measures for the latter.
A potential confound is that Brian Wilson did not write all songs by himself, but instead, many songs are co-authored by other Beach Boys members. Petrie et al. (2008) studied The Beatles lyrics and found differences between John Lennon’s and Paul McCartney’s solo-writing, as well as differences in their collaborative songs. However, in our data, while several songs were co-written with single authors, especially Mike Love (n = 80) and Al Jardine (n = 15) not all sources attributed the songs to the same set of co-authors, cautioning any comparisons based on authorship. We did create a subset of solo-songs written only by Brian Wilson (n = 49). This subset is small and none of the MARS models showed any knots, while the R2 for linear models for all four variables was below 1% of variance explained, likely simply due to a lack of statistical power fitting such models with so few data points. Thus, we refrain from interpreting these models here due to the small number of datapoints.
Last, one caveat is that the data we composed, albeit reflecting the whole body of Wilson’s available work, is sparse during certain periods of his life (as described) and temporally coarse (measured in years, not months or days). Perhaps, such sparsity of creative output alone reflects aspects of mental health. However, these periods might be especially relevant, and thus, the lack of lyrics from them might limit the insights we can gain from Wilson’s language use during those times. It is unclear if these periods reflect a creative drought, recording delays, or struggles with self-confidence, resulting in abandoned projects. Relatedly, while we carefully extracted recording dates for songs, lyrics may have been composed at a different time, been significantly changed over time, or dates might simply be inaccurate. Thus, some caution should be applied when interpreting exact dates. This limitation further applies to the conclusions based on LIWC features, which were extracted from this dataset.
Future directions
Brian Wilson often experimented with sounds in general, including wordless vocals, unconventional instruments, and objects used as instruments, and produced several purely instrumental/acoustic tracks, which means that language analysis as done here, might only capture a fraction of his musical expression. Future research could attempt to combine music and lyric analyses, as for example done in North et al. (2021). While we think that his lyrics provide an interesting source of insight, Brian’s exceptionality as a musician cannot be quantified by lyrics alone.
Moving beyond Brian Wilson’s life, this case study of a psychobiography raises important questions on what aspects of mental health I-talk, cognitive processes words, and all-or-none thinking reflect, states or traits (see Rude et al., 2004; Tackman et al., 2019). From the present data, it seems that I-talk and cognitive processes variables are interconnected, and might be capturing a facet that is different from Emotional Tone, Analytic language, and All-or-None Thinking. From our data, Emotional Tone seems to track depressive symptom severity (state), while I-talk, Cognitive Processes, and All-or None Thinking might be capturing more dispositional aspects of, or vulnerability for, depression (trait). On a bigger picture, our work shows that even carefully crafted language output, such as song lyrics, and even when released under a band known for a positive, easy-going band persona, can give rise to the psychological states of songwriters. Our work reflects an application of psychological text analysis to naturalistic text data, and further highlights the value of such analyses,
Conclusion
We set out to investigate changes in linguistic features of Brian Wilson’s song lyrics over time, considering temporally coinciding events relevant to his mental health. Contrary to our predictions, we found that I-talk and Cognitive Processes words increased until 1967, after which they plateaued, being insensitive to changes in Brian’s mental health, and thus, suggesting that these variables represent mental health disposition, not symptoms. In addition, we found that Emotional Tone was sensitive to life circumstances, increasing with a forming band persona (1960s), decreasing during periods of mental health challenges (1970s−1980s), and increasing again with more positive outlooks (1990s). Together, our results highlight the need to understand language use over the course of an individual’s lifespan, and the ability of mental health to shine through in song lyrics.
Footnotes
Author contributions
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
