Emotion and Ageing in Discourse: Do Older People Express More Positive Emotions?

3.1 Spoken British National Corpus 2014

We used the Spoken British National Corpus 2014 (Spoken BNC2014), which contains face-to-face conversations of British English speakers between 2012 and 2016.^[1] This corpus is annotated with rich and standardized metadata, enabling detailed investigation of age groups, gender, dialects, socio-economic status, and so on.

Since we aim to investigate the association between emotion and ageing, we are particularly interested in the metadata on age groups. Spoken BNC2014 has 10 age groups, ranging from children (0–10) to older adults (90–99), as shown in Table 1. Since too many age groups would make the research results difficult to interpret, we classified the age groups into five broad categories. We tried to recategorize the age groups by staying as close as to the standard set by the World Health Organization (WHO). As adolescence is a critical transitional phase between childhood and adulthood, we start by defining this period. The WHO defines adolescents as people aged between 10 and 19 years old, which best suits the second age group in our corpus. As a result, we labelled the second age group “adolescence”. The age group 0–10 is thus labelled children. For adult groups, we evenly distributed those aged between 19 and 99 into three branches: younger adults, middle adults, and older adults. Our five age groups include childhood (0–10), adolescence (11–18), younger adult (19–29 and 30–39), middle adult (40–49, 50–59, and 60–69), and the older adult (70–79, 80–89, and 90–99). The corpus covers 662 speakers, with a total of 11,337,627 words.

Before conducting further analyses, we need to separate the discourse materials for different age groups, as the original conversation files mix multiple speakers in different age groups. Let us first consider the age group 0–10 years as an example. The first step is to extract a list of speaker IDs from the metadata sheet bnc2014spoken-speakerdata.tsv.csv by limiting the age range. By examining the format of each utterance in the text files, we find that each utterance starts with angle brackets indicating the speaker id and ends with the marker </u>, as shown in the example below:

<u who= “S0115^[2]” n= “1”>I used to uh like to listen to classical music uh</u>

Therefore, the second step is to compile a txt sheet with regular expressions for retrieving discourse materials in each age group, with each speaker occupying a single line. After obtaining the speaker IDs in each age group, we filled the speaker IDs into the following regular expression, with each speaker occupying a single line:

<u n= “\S+” who= “S0140”>[\s\S]*?\</u>^[3]

Having prepared the speaker ID lists, we extracted the utterances of each age group using PowerGREP. By repeating this process, the discourse materials of each age group were obtained. We also separated male and female discourse for each age group; therefore, we have 20 files in total (10 groups in total, each group with a discourse file for male and one for female speakers): 0–10, 11–18, 19–29, 30–39, 40–49, 50–59, 60–69, 70–79, 80–89, 0–99.

3.2 Sentiment Analysis of Different Age Groups’ Discourse

There are three major approaches to sentiment analysis: machine learning, lexicon-based, and hybrid approaches, and three levels of sentiment classification: document level, sentence level, and aspect level (D’Andrea et al. 2015). Machine learning sentiment analysis assesses the sentiment of a text with classification algorithms while lexicon-based approaches use sentiment lexicon as the main standard for sentiment identification. Both machine learning and lexicon-based approaches have their strength and weaknesses. The strength of machine learning method lies in its ability to generate new trained data or models; however, it has difficulty in dealing with semantic knowledge not learnt from the data. In comparison, the lexicon-based sentiment analysis is easier for users as sentiment lexicons are readily available, but the research results can be limited by the lexicon involved.

Considering the characteristics of our data and the advantages and applicability of each method, we conducted sentence-level, lexicon-based sentiment analysis using the sentimentr package with the NRC lexicon (Rinker 2018). We first divided the eight emotions into positive and negative groups and then calculated the ratio between positive and negative frequencies in each age group. We then analysed the semantic frequencies of eight broad emotions (anticipation, joy, surprise, trust, anger, disgust, sadness, and fear) and their relationship with ageing. The results of sentiment analysis are accompanied by qualitative explanations of speech data using examples from different age groups.

To determine the developmental curves for the eight basic emotions, we first detected emotion words based on sentiment analysis and then computed the percentage of each basic emotion against the total frequencies of emotions in each age group. For instance, the frequencies of eight basic emotions for male speakers aged 0–10 are shown in Table 2.

The ratio of anger to total emotions is calculated as:

3.3 Structural Topic-Modelling Thematic Analysis of Different Age Groups’ Concerns

To explain what makes speakers in different age groups more positive or negative, we submitted our texts to the structural topic modelling (stm) R package developed by Roberts et al. (2014, 2016 to identify the key thematic terms in different age groups. Unlike traditional keyword analysis, stm extracts words with the highest probability. The word clusters in each age group constitute the topics of concern in each age group. Function words such as prepositions and articles are not theme-related but easily skew our research results, so we have removed all the function words while doing stm. We extracted the top three topics of each age group to identify people’s primary concerns.

4.1 Results of Sentiment Analysis

As can be seen from Table 1, the number of speakers varied erratically across age groups. There are as many as 250 speakers in the 19–29 age group, while the number of speakers in the age groups 0–10 and 90–99 is only seven and four, respectively. Since contributors to the discourse materials in Spoken BNC2014 are instructed to record their conversations using smartphones, the uneven age group distribution of speakers in Spoken BNC2014 is relatively reasonable considering the smartphone ownership by different age groups in the UK. According to Statista, smartphone penetration is highest among those aged 16–24 and 25–34 – as high as 84 % and 88 %, respectively. However, only 26 % of people aged 65 years or older owned and used smartphones in 2014; thus, it is difficult for researchers to collect as many conversation recordings of people aged above 65 years as of those aged between 16 and 24 and 25–34.

Although the uneven distribution of conversation recordings across different age ranges is thus understandable, we still have to minimize its influence on our sentiment analysis and structural topic modelling. Therefore, instead of reporting the positive and negative sentiment frequencies of different age groups, we calculated the ratio between positive and negative frequencies in the respective age groups, as shown in Figure 1.

Figure 1:

Positive/negative words ratio for people of different age groups.

As illustrated in Figure 1, the older generation is more positive than the younger generation, as the ratio between positive and negative frequencies for both male and female speakers increases with age. Male speakers are slightly more optimistic than female speakers at a young age, but this changes as they approach their 60s: female speakers in their 60s start to become more optimistic than their male counterparts. The higher positive/negative ratio in old age confirmed the findings of previous psychological lab studies indicating that older adults are less likely to express negative emotions in discourse, but it also shows that people’s positive emotions do not move straight up but fluctuate across various age spans. There are several V-shaped curves in Figure 1: two for female speakers, at 19–29 and 40–49, and three for male speakers, at 19–29, 40–49, and 60–79. Similar to previous studies (Blanchflower and Oswald 2008; Consedine and Magai 2006), we find a pronounced dip in people’s 40s, but how can we explain the dip in people’s 20s?

Male and female speakers’ first low ebb in emotions comes at 19–29, when they experience major life transitions. They may lack support in coping with schoolwork, independent-living skills, and relationships. Although people in their twenties are considered energetic, depression is not uncommon as they try to find ways to deal with changes and transitions, which can often trigger depression and irritability. Conversation (1) and conversation (2) are extracted from the Spoken BNC2014 conversations of people aged 19–29 on whether one should stay in college and the pressure of student loans. Two speakers in conversation (1) wonder whether college will be helpful in entering the work field. Speakers in conversation (2), though determined to continue their college studies, are burdened with student loans and taxes.

1. S0529: gonna stay in college then? cos I thought you might want to do that

   S0530: I don’t know cos er

   S0529: well don’t feel like you’re pressured to cos

   S0530: no cos I’d rather live with people I like than stay in the college

   S0529: yeah

   S0530: like it’s not a big deal to me, staying in the college or not

   S0529: no

2. S0632: most of it goes in erm goes in taxes and I like I was paying a lot more

   tax and paying back my student loan a lot more at, as well so a lot of that money would go out of my account before I’d seen it whereas now I’m on I’m on such a small amount that I don’t pay back my student loan

   S0631: mm

   S0632: at all and I’m paying tax but they taxed me a stupid amount when I arrived and now they can’t charge me any more because I’ve basically I’ve paid all the tax that I should in a whole year in one go

   S0631: yeah put on emergency tax so you would get a rebate until it balanced out

   S0632: yeah and now it’s like my wages just go in and I just get it

Student debt worries speakers in conversation (2) because a college degree has become a near-necessity in the job market, but tuition fees are quite high, especially the Cameron government massively raised student fees starting in 2012.

The second valley comes at 40–49, with female speakers feeling more depressed than male speakers. There are many reasons for depression in one’s 40s: the responsibility of caring for children and ageing parents, financial and work stress, and relationship issues. Women are less likely to be emotionally stable as menopause and perimenopause cause changes in their hormones, which leads to fluctuations in their mood. Conversation (3) shows two women talking about menopause:

1. S0517: cos some of the treatment for erm cancer puts you into early menopause

   S0558: yeah

   S0388: doesn’t it?

   S0558: well she said she went through menopause when she was forty-two

   S0517: gosh that’s early

   S0558: so she must be forty-seven

   S0517: gosh

Born between 1966 and 1975, speakers in their 40s in the data are trying to save for retirement, taking care of ageing parents and young children, and struggling for work–life balance.

Overcoming menopause, the emotion curve for female speakers in the data is gradually becoming more positive, while male speakers experience another plunge in their emotions from their 60s until their late 70s. When approaching their 60s, people start to lose their family and friends, and their bodies can also decline sharply. However, women and men differ in the ways they express their emotions. Compared with men, women are more open to share their emotions, which leads to greater happiness. In conversation (4), the two speakers are talking about death:

1. S0281: have you got any preference? You’re going to be buried aren’t you? next to your father

   S0355: yeah but in a box of ashes if they’d allow me

   S0281: oh really?

   S0355: to put those in the grave yeah, there’s in the headstone for my name

There is also a sharp rise in positivity among female speakers aged 90 to 99, but a decline among male speakers in the same group. This phenomenon can be explained by women’s stronger willingness to seek help and treatment when experiencing emotional difficulties, thus allowing them to recover more quickly and be happier than men.

With a general understanding of people’s emotion curves, we may wonder how people would express their eight basic emotions: anticipation, joy, surprise, trust, anger, disgust, sadness, and fear? In Section 4.2, we explore the eight basic emotions across different age ranges.

4.2 Eight Basic Emotions Among Different Age Groups

Previous studies only focus on one single emotion such as happiness (Kim et al. 2023), here we take anticipation, surprise, trust, anger, disgust, sadness, and fear into consideration. The developmental paths of eight basic emotions are illustrated in Figures 2–9. As can be seen in Figures 2–5, emotion frequencies for anticipation, joy, surprise, and trust are higher for older adults than for younger generations. The emotion curves for anticipation and joy are generally the same, with pronounced dips in men and women in their 20s and 40s and a low ebb for men in their 60s and 70s. Anticipation is concerned with pleasure or anxiety when considering or waiting for something. Conversations (5) to (7) show the use of verbs such as eager, expecting, or adore to express one’s anticipation. Healthy anticipation can help people get through hard times and energize their lives. Conversations (8) and (9) show people’s expression of joy through words like ardent and bless.

1. S0084: I was just writing I was too eager to write it

   S0083: so you only heard it recently?

   S0084: yeah

2. S0013: she was fine really wasn’t you know we didn’t been know she was expecting really

   S0012: well I thought she was cos she had a big

3. S0327: you get tourists going in there and of course tourists adore it because It’s exactly

   S0445: yeah

4. S0284: what what do you do for a job? you know what do you think about dwarf throwing?

   S0282: yes but was a very ardent feminist in my younger days

5. S0255: so yeah she said she and she got a bit upset about that and

   S0315: oh bless her

Figure 2:

Anticipation frequencies for people of different age groups.

Figure 3:

Joy frequencies for people of different age groups.

Figure 4:

Surprise frequencies for people of different age groups.

Figure 5:

Trust frequencies for people of different age groups.

Figure 6:

Anger frequencies for people of different age groups.

Figure 7:

Disgust frequencies for people of different age groups.

Figure 8:

Sadness frequencies for people of different age groups.

Figure 9:

Fear frequencies for people of different age groups.

Regarding surprise, frequencies are relatively high for children aged 0 to 10, as they are enthusiastic and eager to learn all the time. However, the surprise frequencies quickly fail as people age, more quickly for men than for women. Then, the frequencies gradually regained and fluctuated over time. An example is shown in excerpt 10.

1. S0653: hang on hang on hang on a backup in case you don’t like?

   S0654: in case someone doesn’t want enchanted

   S0653: in case someone doesn’t want enchanted we could watch The Slipper and the Rose

Unlike anticipation, surprise, and joy, trust frequencies are low for children under 10 years old and then increase as they become teenagers. The low trust frequencies in the early years (babies and children) can be partly explained by the ‘trust and mistrust’ stage in Erikson’s (1963) social-emotional development model, which begins at birth. During this stage, the infant is uncertain about the consistency and stability of their caregivers. Moreover, nurturing trust with parents takes time. Similar low ebbs in trust frequencies occur in people in their 20s (male and female speakers), 40s (male speakers), 60s and 70s (male speakers), and 70s and 80s (female speakers). Excerpt 11 is an example of a trust expression.

1. S0257: at one stage I’m really sorry this like you know, I assured you that this had been done because I was being told it was look been looked after I said, and it and it wasn’t done but I sorted it out now I’ve done this done that done the other I said

   S0255: right

When it comes to negative emotions, we find that as people age, they express fewer negative emotions in their discourse compared with their younger counterparts. People are less likely to reveal anger (as shown in conversation 12), disgust (as shown in Figure 7), and fear (as shown in conversation 14). The only exception is sadness – the discourse of older people shows more sadness than that of the younger generation. Chronic diseases and loss of loved ones during old age can give rise to sadness and loneliness during old age.

1. S0439: you just killed a bug you bitch

   S0441: I’m an actual bitch

2. S0278: it’s abuse isn’t it?

   S0013: you know? yes it is

3. S0094: don’t be afraid to say if you don’t later

   S0021: what

4. S0008: and then I thought well, and then I started getting like that the last tooth started aching a bit

   S0012: mm

   S0008: but not bad enough to just an ache not a pain

   S0012: yeah

4.3 Structural Topic Modelling of Different Age Groups

After examining the general positive and negative emotion curves of male and female speakers, we turn to structural topic modelling, which sketches the top three thematic topics across age groups. We divided the 10 age groups into five categories: childhood (0–10), adolescence (11–18), younger adult (19–29 and 30–39), middle adult (40–49, 50–59, and 60–69), and older adult (70–79, 80–89, and 90–99). The factors behind people’s positive and negative emotions are presented in the following subsections. As people are going through major events and life challenges in different age groups, they may experience changes in their mood. This section tries to unravel the major life challenges that will influence the expression of their emotions.