The language of over-50s in online dating classified ads

Olga Mudraya, Lancaster University
Paul Rayson, University Centre for Computer Corpus Research on Language, Lancaster University

Abstract

1. Introduction

The number of online dating classified ads have grown rapidly in recent years. As Fiore (2004: 13) describes, they “have shed their stigma as matchmakers for the awkward to claim a prominent role in the social lives of millions of people”, and “the base of users spans generations, breaking the 20- or 30-something age ceiling common in many online social environments”. Researchers have studied online relationships mainly from psychological and sociological perspectives (Van Gelder 1985; Lea & Spears 1995; Scharlott & Christ 1995; Walther 1996; Daneback 2006; Valkenburg & Jochen 2007; etc.), and predominantly among younger people. The methodology employed in these previous studies is usually interview-based.

Academic research into the language of online dating ads is still scarce. In Computer-Mediated Communication research, previous analysis of personal pronoun usage (Sherblom 1990; Witt 2004; Walther 2007) has been reported. More pronoun use is associated with greater immediacy and involvement with a reader or topic. The vocabulary and semantics of online dating ads have not yet been investigated. A number of studies in psychology and evolutionary anthropology have identified important personal trait categories (and words associated with them), such as age, physical attractiveness, resources (current or future earning potential), and commitment to the relationship (Bereczkei & Csanaky 1996; Bereczkei et al. 1997; Greenlees & McGrew 1994; Wiederman 1993), as well as entertainment and social skills (Miller 1998). Robin Dunbar was involved in a series of evolutionary psychology investigations of different categories of words in Lonely Hearts advertisements (Waynforth & Dunbar 1995; Pawłowski & Dunbar 1999a; Pawłowski & Dunbar 1999b; Pawłowski & Dunbar 2001) where it was reported that men and women attached different levels of importance to the following five categories in online ads: attractiveness, resources, commitment, social skills and sexiness.

The purpose of our study is to apply a corpus-based methodology in order to allow a larger scale study of the key words and topics in this type of data. This will also permit a comparison with the results obtained via interview-based methods of data collection in previous research such as Pawłowski and Dunbar (2001). We have collected texts from the online dating classified ads on the Telegraph.co.uk dating website KindredSpirits. In our study we use the Wmatrix web-based corpus processing software tool for linguistic analysis in order to compare the language of men looking for women, men looking for men, women looking for women, and women looking for men. Our focus is the over-50s, as this age group is often overlooked in previous studies.

2. Related work

Pawłowski and Dunbar’s (2001) evolutionary psychology study found the five most important categories of words in Lonely Hearts advertisements were attractiveness, resources, commitment, social skills and sexiness. It is important to note that evolutionary theory provides an explanatory framework for understanding patterns of human mate choice and the preferred characteristics in a potential partner from the point of view of reproduction (Dunbar et al 2007: 91). However our study is not strictly comparable, because we focus on people over 50 years of age, therefore reproduction may not be on their minds, even subconsciously, when they decide to place an online dating advertisement. Pawłowski and Dunbar (2001) sampled only advertisers aged between 18-29, describing themselves as students or graduates and seeking heterosexual relationships. Our study also includes same-sex relationships that were not examined in Pawłowski and Dunbar (2001). Also, their method of investigation was based on subjective manual rating of the appeal of the collected Lonely Hearts advertisements to 200 male and female university students, while we use a corpus processing software tool for linguistic analysis with the aim of producing more objective quantitative results.

According to Pawłowski and Dunbar's (2001: 6-7) study of words in Lonely Hearts advertisements, men and women attached different levels of importance (with the exception of commitment) to five broad categories of traits. Men’s preferences (in order) were: commitment, attractiveness, social skills, resources and sexiness. In contrast, women’s preferences in order were: commitment, social skills, resources, attractiveness and sexiness. Table 1 exemplifies the traits associated with these categories, based on the investigations conducted by Dunbar and his colleagues. The types of words coded for the categories of attractiveness, resources and commitment (to be precise, physical attractiveness, wealth/status and family commitment) were given in Waynforth and Dunbar (1995). The category of social skills (with the focus on traits that “allow individuals to keep a relationship going even during difficult times, at least part of which entails the ability to keep a mate entertained”) was exemplified in Pawłowski and Dunbar (1999b: 55), and was also explained in Dunbar et al (2007: 77) as being considerate, having a sense of humour and being a good listener. The category of sexiness was added in Pawłowski and Dunbar’s (2001) without exemplification.

Overall, in Pawłowski and Dunbar’s (2001: 7) study, females rated resources and social skills higher than males did, while males rated attractiveness and sexiness higher than females did. There was no significant difference between the sexes in respect of commitment, which was “the common overriding preference”, and interestingly, sexiness scored low for both sexes.

3. Methodology

In our study, we use the Wmatrix corpus processing software tool which incorporates the CLAWS and USAS corpus annotation tools to provide part-of-speech and semantic field analysis. The USAS semantic category system was originally based on a semantic taxonomy in McArthur’s Longman Lexicon of Contemporary English (McArthur 1981), but has been significantly revised to enable automatic annotation. The semantic tagset reflects 21 major semantic domains (shown in Table 2), further sub-divided into 232 semantic sub-categories. The core of the semantic tagger is the semantic lexicon knowledge base, in which single words and multiword expressions are mapped to their potential semantic categories.

Through a web interface, Wmatrix allows us to detect key words and key concepts in the corpus, using the keyness statistic applied to the word and semantic tag levels (Rayson 2008). Typically, key words and key semantic tags are not the most frequent words and concepts but the most unusually frequent in a given body of text relative to a reference corpus. For the reference corpus in our study, we used the British National Corpus (BNC) Sampler, which is a broadly representative subcorpus of the BNC, consisting of two million words, 50% written material and 50% spoken material. In addition, sub-corpora of our online data classified ads were compared with each other, rather than a reference corpus to permit direct comparison. The results are displayed with the most significant key items at the top of the list by sorting on the LL (log-likelihood) value which indicates the significance of the difference. To be statistically significant at the 99% level (p < 0.01), the LL value should be over 6.63, or at least 3.84 to ensure 95% confidence of significance (p < 0.05).

We should note that the semantic analysis is not intended to directly discover the intentions of the authors but to help us describe the language that they employ. Undoubtedly, there will be some discrepancy, as in other forms of advertising, where part of the goal of the advert is to sell what the author has to offer to an interested reader.

4. Corpus data

We have selected classified ads from the Telegraph.co.uk online dating service for people over 50 years old (51+, to be precise) in four distinct dating subgroups. These are as follows: men looking for women, women looking for men, men looking for men, and women looking for women. The Telegraph.co.uk online dating service KindredSpirits also includes two other categories of people, men looking for both men and women, and women looking for both men and women, but these are not analysed in our study.

At the time of our study, KindredSpirits had more than 22,600 members with an age range of 18 to 100. Around 21,500 members are in two groups of roughly equal size, men looking for women and women looking for men. As this is a live online dating service, the number of classified ads changes constantly. At the time of writing, there were 517 women looking for women and 722 men looking for men in the UK between the ages 18 and 100.

All KindredSpirits members are required to write a 40 to 1,000 words long narrative, describing themselves and their ideal match. For this study, 80 narratives in each search category were included. This produced a corpus of nearly 45,000 running words, henceforth the 51+ Corpus. Interestingly, narratives of men looking for women and women looking for men were almost twice as long as those of looking for a same-sex relationship (see Table 3).

5. Results

From our analyses, all five of Pawłowski and Dunbar’s (2001) categories appear as statistically significant key semantic domains, although only one of them – social skills (particularly, good sense of humour typically abbreviated to GSOH) – is among the five most statistically significant. Being happy, energetic and enjoying life appear at the top of our list. It is important to note though that there is not an exact match between our semantic category system and Pawłowski and Dunbar’s (2001). For example, some of the words coded for the category of attractiveness by Pawłowski and Dunbar, such as athletic, fit, healthy, nice body, muscular, fall under either Health or Sports category in Wmatrix. This is also the reason why commitment is not prominent in our study as traits associated with this category are classed under a number of different categories in Wmatrix. 

When comparing men’s and women’s preferences across the four dating subgroups, a number of findings have come to light. These are summarised in Table 4. Further details are shown in the following subsections.

5.1 Whole corpus results

When the whole 44,905-word 51+ Corpus was compared with the normative one-million-word BNC Sampler Written, 942 statistically significant (p < 0.01, LL > 6.63) key words and multiword expressions and 100 statistically significant key semantic domains were identified. Figure 1 displays a screenshot of the key word cloud giving the 100 most significant items (more significant items are shown in larger font size).

Table 5 below gives some examples of the words and multiword expressions from the top ten semantic domains.

5.2 Men looking for women

Rather than comparing the whole corpus to a reference dataset, we now turn to results obtained by direct comparison of sub-corpora. Comparing men-looking-for-women ads with women-looking-for-men ads produced the following 14 statistically significant (p < 0.01, LL > 6.63) key words: she, lady, ladies, considerate, her, name, totally, good-looking, called, divorced, restaurants, dog, off, well-educated, and 12 statistically significant key semantic domains. Figure 2 presents a screenshot of the key word cloud.

Table 6 provides examples from the statistically significant key domains in the men-looking-for-women dataset.

5.3 Women looking for men

The comparison between women-looking-for-men and men-looking-for-women ads identified 15 statistically significant (p < 0.01, LL > 6.63) key words: he, would, friends, yes, sea, love, dogs, exploring, n't, afraid, normal, zest, beach, and, man, and 5 statistically significant key semantic domains. Figure 3 presents a screenshot of the key word cloud with more significant items in larger font size.

Table 7 provides examples from the statistically significant key domains in the women-looking-for-men dataset.

Three of Pawłowski and Dunbar’s (2001) categories (resources, sexiness and social skills) do not appear in our significant results. Interestingly, mental / intellectual characteristics emerged as statistically significant in the women-looking-for-men dataset.

5.4 Men looking for men

Men-looking-for-men data was compared with both women-looking-for-women and men-looking-for-women data. Comparing men-looking-for-men data with women-looking-for-women data, 25 statistically significant (p < 0.01, LL > 6.63) key words: guy, I’m, man, younger, men, male, one, straight-acting, relationship, discreet, or, accommodate, gay, each, good sense of humour, guys, prefer, relaxed, older, bloke, regular, affectionate, wants, mature, willing, and 8 statistically significant key semantic domains were identified. Figure 4 presents a screenshot of the key word cloud from Wmatrix.

The comparison of the men-looking-for-men data with men-looking-for-women data produced 38 statistically significant (p < 0.01, LL > 6.63) key words: guy, gay, looking for, male, mature, straight-acting, or, relationship, accommodate, friendship, younger, I’m, each, guys, travel, men, ideas, regular, interests, someone, discreet, shy, as to, casual, clean, dogs, etc., exploring, non-scene, sex, wishes, older, listener, know, who, arts, mate, soul, and 12 statistically significant key semantic domains. Figure 5 presents a screenshot of the key word cloud in this category.

These two sets of key words indicate that words describing social skills (particularly, having a good sense of humour and being a good listener) and words related to resources (in providing accommodation for meetings) appear among the statistically significant key words. This is consistent with Pawłowski and Dunbar’s (2001) study.

Table 8 provides examples from the statistically significant key domains in the men-looking-for-men dataset when compared with women looking for women.

Comparing statistically significant key domains in the men-looking-for-men data with the men-looking-for-women data refines these findings somewhat more. This comparison identified 12 key semantic domains as presented in Table 9. From this, it emerged that both general relationships based on friendship and sexual relationships were most significant for gay men, with general relationships as the second most significant semantic domain and sexual relationship third. Also, this comparison has identified numerous important personality traits, some of which could represent social skills (such as being informal and friendly, broad-minded and tolerant, peaceful and relaxed), rather than just physical characteristics, although age still shows as a significant semantic domain.

5.5 Women looking for women

Women-looking-for-women ads were compared with both men-looking-for-men and women-looking-for-men ads. The comparison between women-looking-for-women ads and men-looking-for-men ads identified 27 statistically significant (p < 0.01, LL > 6.63) key words: I’m, woman, love, her, places, have, I’ve, I, independent, she, am, art, female, lesbian, lively, lady, children, kind, beautiful, garden, got, honesty, in touch, shopping, by, we, food, and 6 statistically significant key semantic domains (Table 10). Figure 6 presents a screenshot of the key word cloud.

The comparison of the women-looking-for-women ads with women-looking-for-men ads has produced 28 statistically significant (p < 0.01, LL > 6.63) key words: I’m, looking for, friendship, I, lesbian, I’ve, her, woman, lively, good sense of humour, going out, lonely, politics, others, get, UK, etc., boring, financially, form, gay, u, she, chat, just, love, a lot, trustworthy, and 5 statistically significant key semantic domains. Figure 7 presents a screenshot of the key word cloud in this category.

These two sets of key words provide support for Pawłowski and Dunbar’s (2001) study because words belonging to four of their categories – attractiveness, sexiness, resources and social skills (particularly, having a good sense of humour) – appear among the statistically significant key words that we have observed. Words describing commitment do not appear in our list, although perhaps words honesty and trustworthy can indirectly indicate commitment.

Table 10 provides examples of the statistically significant key domains in the women-looking-for-women dataset when compared with men looking for men.

On the key semantic domain level, none of Pawłowski and Dunbar’s (2001) categories apply to women looking for women when compared with men looking for men, which may be due to the fact that Dunbar did not investigate same-sex relationships in his study. That said, Pawłowski and Dunbar’s category of social skills may be manifested through our Social Actions, States and Processes category that has emerged as the second most statistically significant semantic domain in the women-looking-for-women data (when compared with men-looking-for-men data). This study has uncovered four other topics, such as history and culture (exemplified by museums), nature / great outdoors (represented by Geographical Terms), medical treatment (including slimming) and email communications.

Comparing statistically significant key domains in the women-looking-for-women data with the women-looking-for-men data identified 5 key semantic domains as presented in Table 11.

The comparison of women looking for women to women looking for men has shown that general relationships based on friendship rather than sexual relationships is a significant category in the language of classified ads for women looking for a same-sex relationship. Also, another of Pawłowski and Dunbar’s categories – social skills (being considerate, in particular) – is statistically significant in this dataset. Interestingly, physical weight (as a possible indicator of attractiveness) is one of the key semantic domains. This comparison shows some similarity with the comparison of men looking for women with women looking for men in that both physical weight and a personality trait of unselfishness emerge as statistically significant.

6. Conclusion

The research described here examined the language used in online dating ads and provided a comparison of four dating subgroups as follows: men looking for women, women looking for men, men looking for men, and women looking for women by identifying their key words and key semantic domains. We employed a corpus-based methodology and a large number of classified ads downloaded from the Telegraph.co.uk online dating service. This study compared the results arrived at using the Wmatrix web-based corpus processing software tool to Pawłowski and Dunbar’s (2001) findings in their evolutionary psychology study. A number of similarities were observed allowing us to confirm some findings from previous work. Here, we summarise these finding across the four subgroup comparisons:

1. Men looking for women. Attractiveness is statistically significant, but only after education and the USAS category Unselfish. Physical weight (as a possible indicator of attractiveness) appears as a key semantic domain. Being considerate (classed under the unselfish category in our study) appears more statistically significant in our dataset than attractiveness. Interestingly, the In-Power category traditionally associated with a male role finds its way into the statistically significant semantic domains in our men-looking-for-women data.

2. Women looking for men. There is some indication that attractiveness (illustrated by expressions set their world on fire and elusive spark in the statistically significant On Fire category) is of importance to women looking for men. Interestingly, mental/intellectual characteristics emerged as statistically significant in this dataset.

3. Men looking for men. On the keyword level, words describing social skills (particularly, having a good sense of humour or GSOH, and being a good listener) and words related to resources (in providing accommodation for meetings) appear among the statistically significant key words in this dataset. On the key semantic domain level, Pawłowski and Dunbar’s category of sexiness is statistically significant in men-looking-for-men subset when compared with women looking for women. However, the comparison with the men-looking-for-women data has established that both general relationship based on friendship and sexual relationship are significant for gay men, with general relationship as the second most significant semantic domain and sexual relationship the third. Our study has found out that gay men give particular importance in the language of classified ads to the age and physical characteristics of their prospective partners, which in Pawłowski and Dunbar’s terms would relate to attractiveness. Discretion is another important feature in online dating gay discourse.

4. Women looking for women. At the keyword level, there is more congruity in this set of results with Pawłowski and Dunbar’s than in any of the other dating subgroups. Words belonging to four of Pawłowski and Dunbar’s categories – attractiveness, sexiness, resources and social skills (particularly, having a good sense of humour) – appear among the statistically significant key words in this dataset. However, on the key semantic domain level, the Social Actions, States and Processes category has emerged as second most statistically significant. Also four important areas, such as history and culture (exemplified by museums), nature / great outdoors (represented by Geographical Terms), medical treatment (including slimming) and email communications are identified. The comparison of women looking for women with women looking for men has shown that a general relationship based on friendship rather than a sexual relationship is portrayed as being significant for women looking for a same-sex relationship.

Our methodology is more detailed and more objective than previous research since it is not based on intuition and interview-based methods. We have also carried out more sub-group comparisons as described above. As further development, it would be interesting to repeat the same procedures for other age groups in order to see if there were age-related differences in the use of language in online dating classified ads.

Sources and resources

British National Corpus. http://www.natcorp.ox.ac.uk/.

Telegraph Dating. http://dating.telegraph.co.uk/s/.

Wmatrix. http://ucrel.lancs.ac.uk/wmatrix/.

References

Bereczkei, Tamas & Andras Csanaky. 1996. “Mate choice, marital success, and reproduction in a modern society”. Ethology and Sociobiology 17: 23-35.

Bereczkei, Tamas, Silvia Voros, Agnes Gal, & Laszlo Bernath. 1997. “Resources, attractiveness, family commitment: Reproductive decisions in human mate choice”. Ethology 103: 681-699.

Buss, David M. 1989. “Sex differences in human mate preferences: Evolutionary hypotheses tested in 37 cultures”. Behavioral and Brain Sciences 12: 1-49.

Daneback, Kristian. 2006. Love and Sexuality on the Internet. Ph. D. thesis. University of Gothenburg Faculty of Social Sciences.

Dunbar, R. I. M., Louise Barrett & John Lycett. 2007. Evolutionary Psychology: A Beginner’s Guide. Oxford: Oneworld.

Fiore, Andrew R. T. 2004. Romantic Regressions: An Analysis of Behavior in Online Dating Systems. Master’s Degree thesis. Massachusetts Institute of Technology.

Greenlees, I. A. & McGrew, W. C. 1994. “Sex and age differences in preferences and tactics of mate attraction: Analysis of published advertisements”. Ethology and Sociobiology 15: 59-72.

Lea, Martin & Russell Spears. 1995. “Love at first byte? Building personal relationships over computer networks”. Under-Studied Relationships: Off The Beaten Track, ed. by Julia T. Wood & Steve Duck, 197-233. Thousand Oaks, Calif.: Sage.

McArthur, Ton. 1981. Longman Lexicon of Contemporary English. Harlow: Longman.

Miller, Geoffrey F. 1998. “How mate choice shaped human nature: A review of sexual selection and human evolution”. Handbook of Human Evolutionary Psychology: Ideas, Issues and Applications, ed. by Charles Crawford & Dennis Krebs, 87-129. New York: Lawrence Erlbaum.

Pawłowski, B. & Dunbar, R. I. M. 1999a. “Impact of market value on human mate choice decisions”. Proceedings of the Royal Society 266B: 281-285.

Pawłowski, B. & Dunbar, R. I. M. 1999b. “Withholding age as putative deception in mate search tactics”. Evolution and Human Behavior 20: 53-69.

Pawłowski, B. & Dunbar, R. I. M. 2001. “Human mate choice strategies”. Economic Models of Animal and Human Behaviour, ed. by J. van Hooff, R. Noë & P. Hammerstein, 187-202. Cambridge University Press.

Rayson, Paul. (2008). “From key words to key semantic domains”. International Journal of Corpus Linguistics 13(4): 519-549.

Scharlott, Bradford W. & William G. Christ. 1995. “Overcoming relationship-initiation barriers: The impact of a computer-dating system on sex role, shyness, and appearance inhibitions”. Computers in Human Behaviour 11(2): 191-204.

Sherblom, John C. 1990. “Organizational involvement expressed through pronoun use in computer mediated communication”. Communication Research Reports 7(1): 45-50.

Valkenburg, Patti M. & Peter Jochen. 2007. “Who visits online dating sites? Exploring some characteristics of online daters”. CyberPsychology & Behavior 10(6): 849-852.

Van Gelder, Lindsy. 1985. “The strange case of the electronic lover”. Ms., October 1985, 94-104, 117, 123-124.

Van Gelder, Lindsy. 1990. “The strange case of the electronic lover”. Talking to Strangers: Mediated Therapeutic Communication, ed. by Gary Gumpert & Sandra L. Fish, 128-142. Norwood: Ablex.

Walther, Joseph B. 1996. “Computer-mediated communication: Impersonal, interpersonal, and hyperpersonal interaction”. Communication Research 23(1): 3-43.

Walther, Joseph B. 2007. “Selective self-presentation in computer-mediated communication: Hyperpersonal dimensions of technology, language, and cognition”. Computers in Human Behavior 23(5): 2538-2557.

Waynforth, David & Dunbar, R. I. M. 1995. “Conditional mate choice strategies in humans: Evidence from ‘Lonely hearts’ advertisements”. Behaviour 132: 755-779.

Wiederman, Michael W. 1993. “Evolved gender differences in mate preferences: Evidence from personal advertisements”. Ethology and Sociobiology 14: 331-352.

Witt, Paul L. 2004. “An initial examination of observed verbal immediacy and participants’ opinions of communication effectiveness in online group interaction”. Journal of Online Behavior 2(1). http://old.behavior.net/JOB/v2n1/witt.html [article no longer available online].