Love is heterosexual-by-default: Cultural heterosexism in default prototypes of romantic love

Cultural heterosexist ideologies assume heterosexuality to be the default norm. Four studies investigated when concepts of romantic love are heterosexual-by-default ( N = 685). In Studies 1 – 2, participants generated features of romantic love, in general (i.e., the default prototype) or among one of three sexual orientation-speciﬁc couples (lesbian, gay, or heterosexual). Heterosexual-identiﬁed participants’ default prototypes were more similar to heterosexual than same-gender prototypes (Study 1). Lesbian-and gay-identiﬁed participants’ default prototypes were more similar to both heterosexual and gay male than lesbian prototypes, whereas bisexual-identiﬁed participants’ sexual orientation-speciﬁc prototypes were equivalently similar to the default (Study 2). However, heterosexual-identiﬁed participants rated presented features of love similarly acrosssexualorientation-speciﬁc conditions(Study3). Inatimedfeature-veriﬁcationtask (Study 4), participants categorized fewer peripheral features of romantic love as relevant to same-gender than mixed-gender couples. Activating sexual orientation-speciﬁc representations affected subsequent default concepts of romantic love. We discuss implications for heterosexism theories and intervention.

sentence, only to find that at the end it clashes with the image of the man who died in the third sentence.Because such automatic category activation can go unnoticed, people often fail to resolve the riddle.
The present research concerns defaulting, but in regard to the sexual orientation of romantic couples rather than the gender of surgeons.Four studies examined the contexts in which people do and do not conceptualize romantic love as heterosexual-by-default. Herek's (2007) concept of cultural heterosexism suggests that such defaulting may reinforce heterosexist attitudes at the individual and societal level.Such defaulting in category activation can be understood by Kahneman and Miller's (1986) norm theory.Both are introduced below.

Cultural heterosexism
Cultural heterosexism is a socially structured ideology that disadvantages sexual minorities, via assumptions that people are heterosexual-by-default, and that homosexuality, when acknowledged, is 'unnatural,' exceptional, and requiring of particular explanation (Herek, 2007).The concept overlaps considerably with heteronormativity, which was also used in the 1990s to refer to the erasure and othering of non-heterosexual identities that occurs when general categories, but in queer theory (see Seidman, 1991;Warner, 1993).Queer theorists in the 1990s took on the task of rooting out concealed heteronormative ideologies (see Turner, 2000 for a history).Queer theory was deeply indebted to Foucaultian critiques of the psychological disciplines as an operation of ideological power/knowledge (see e.g., Butler, 1990;Halperin, 1997;Sedgwick, 2008), and it had limited impact on social psychology during this period (see Minton, 1997 for a notable exception).Thus 'cultural heterosexism' and 'heteronormativity' have overlapping meanings, but very different intellectual histories.Heteronormativity has since come to take on a wider range of meanings in the interdisciplinary literature (see Marchia & Sommer, 2019).Our preference for 'cultural heterosexism' here reflects more the fact that this paper is psychological in its scope, than any particular difference between these two terms that both aim to name this oppressive ideology.In this context, the psychological term is the more parsimonious one that makes no false promise of an interdisciplinary conversation.We have described how materialist feminism may be used to diffract this seeming impasse between cognitive analysis of concepts and the interdisciplinary demands to queer theory elsewhere.
The cultural heterosexism concept suggests that same-gender couples might still be conceptualized as exceptional, and heterosexual couples as the default, even in the context of equal marriage laws.Social behaviour that is newer, uncommon, lower status, unexpected, or perceived as less-than-ideal can all appear to require particular explanation (Bruckm€ uller, Hegarty, Teigen, B€ ohm, & Luminet, 2017).Heterosexual relationships are historically older, more common, higher status, and reflected in cultural ideal scripts such as gender complementarity (Glick & Fiske, 1996).Consistent with the cultural heterosexism concept, recent media accounts of divorce rates explain samegender couples' divorces against the background norm of opposite-gender couples' divorces almost all of the time (Hegarty, Szczerba, & Skelton,2020).

Category norms
We argue that ideologies such as cultural heterosexism might impact cognition (see also Bailey et al., 2019) both through the graded structure of categories and the ways that categories are activated.Graded structure refers to the notion that category members vary in how prototypical they are (Rosch, 1973).For example, a robin is a more prototypical bird than a penguin.As such, cultural heterosexism may render heterosexual category members more prototypical than LGBQ+ members in long-term memory representations of romantic love.In addition, heterosexual forms of loving may be activated more readily than LGBQ+ forms when thinking about 'romantic love.' Kahneman and Miller's (1986) category norms theory argues that both the graded structure of category knowledge in long-term memory and the activation of accessible exemplars determine what becomes the normative, taken-for-granted assumptions about a category.Category norms are working-memory representations evoked by reference to a specific category (e.g., surgeon).Such references recruit the most accessible exemplars from long-term memory (e.g., male surgeons).The accessibility of such exemplars is influenced by their prototypicality, but also by recently activated knowledge.For example, recent thoughts about an unrelated female doctor might affect how a person interprets the surgeon riddle.Category norms also differ from prototypes because category norms' modal values become implicit in working memory (Kahneman & Miller, 1986).Thus, this theory provides a cognitive understanding of how and why heterosexuality could become the default representation of love in many, but not all, circumstances, and that defaulting might pass unnoticed.
Ideologies may influence category norms by determining their contents.As well as prototypical and recently activated exemplars, category norms are more likely to contain members of more numerous groups and higher-status groups (e.g., men, White people, and heterosexual people; Hegarty & Bruckm€ uller, 2013).For example, participants tend to explain empirical differences between groups by focusing on factors particular to women, ethnic minorities, or LGBQ+ individualsevidencing the normative status of the complimentary higher-status groups.This paper does not aim to weigh up which factors might make heterosexual love the norm.Rather, we examine how cultural heterosexism might be a consequence of category knowledge and category activation processes.Little research has yet tested whether similar social identity defaulting occurs in perceptions of emotion categories (but see Johnson, McKay, & Pollick, 2011).However, evidence shows that people have prototypes of romantic love, which include diverse traits that vary in their centrality (Fehr, 1988;Fehr & Sprecher, 2009;Regan, Kocan, & Whitlock, 1998; see also Rosch & Mervis, 1975).Hence, the activation and implementation of these prototypes may be susceptible to heterosexist category norms.Indeed, many past studies of romantic love conceptualized heterosexuality as the default relationship status without justification (Thorne, Hegarty, & Hepper, 2019).The present studies examine if cultural heterosexism (Herek, 2007) can be extended to concepts of romantic love, by studying prototypes of romantic love through the lens of defaulting-to-heterosexuality.Our research question is whether ordinary people conceptualize mixed-gender love as being more similar to 'love' in general than same-gender love, due to processes of category norm activation.

The present research
Four studies examined when romantic love is conceptualized as heterosexual-by-default.Since the work of Rosch (1977), category prototypes have been studied through feature generation, typicality ratings, and timed feature classification tasks, including the prototypical structure of lay concepts of love (e.g., Fehr, 1988;Regan et al., 1998) and relationships (e.g., Fitness & Fletcher, 1993).The present research uses this prototype approach to examine concepts of romantic love across sexual orientation target groups for the first time.Following classic prototype research we asked participants to recruit features of romantic love (Studies 1-2), rate their prototypicality (Study 3), and judge their inclusion in the category under time pressure (Study 4).We examined participants' default concept of love (e.g., Regan et al., 1998), as well as sexual orientation-specific concepts that referred to heterosexual, lesbian, or gay male couples.Our primary heterosexual-by-default hypothesis predicted that default concepts of love would overlap to a greater extent with heterosexual-specific concepts than with either lesbian-specific or gay-specific concepts.We also examined the boundaries of this effect in terms of participant groups (Study 2) and recent category activation (Study 4).

Study 1: Feature generation among heterosexual participants
Heterosexual participants retrieved features associated with romantic love from memory in one of four conditions (feature generation).In these conditions, the couple was unspecified (i.e., default condition) or was specified by one of three sexual orientations (i.e., heterosexual-specific, lesbian-specific, or gay-specific conditions).All participants were heterosexual-identified.We predicted that their heterosexual-specific concepts would have a more similar graded structure to the default concept than would either their lesbian-or gay-specific concepts.

Participants
Previous studies indicate a sample of 200 is sufficient to identify a prototypical structure (e.g., Fehr, 1988;Regan et al., 1998).Via research websites, social media, and snowball sampling we recruited 583 respondents for a study on perceptions of romantic lovethere was no mention to sexual orientation-specific romantic love in recruitment.Of the 583 respondents, 329 left the survey incomplete.Attrition was significantly higher in the lesbian-and gay-specific conditions (59%) than the default and heterosexual-specific conditions (52%; v 2 (1) = 4.34, p = .04).We excluded participants whose sexual orientation was non-heterosexual (n = 35) or unreported (n = 12).The final sample comprised 207 heterosexual participants (77% female; aged 16-73, M = 23.42,SD = 8.76).1 Details of participants' self-identified ethnic backgrounds are presented in Table 1.

Materials and procedure
The online survey software randomly assigned participants to the Default (n = 63), Heterosexual-Specific (n = 51), Lesbian-Specific (n = 47), or Gay-Specific (n = 46) condition. 2In the default condition, participants read the following prompt adapted from Regan et al. (1998): We would like you to think of the concept of romantic love between heterosexual people, or how heterosexual people experience love.Romantic love has a number of different features or characteristics.Some of these features may be very central (important or essential) to the experience of romantic love.Other features may be less central to romantic lovewe don't automatically think of them when we think of the state of 'being in romantic love'.These features can include a range of emotions, feelings, thoughts, behaviours, actions or events and can range from positive to negative.We would like you to list as many features of romantic love as you can.We remind you to think about love between heterosexual people when making your list of features.
In the sexual orientation-specific conditions, participants were additionally prompted to think about the respective target group (e.g., opening sentence for the gay-specific condition: 'We would like you to think of the concept of romantic love between gay men, or how gay men experience love').Participants were provided with an open-ended text box and engaged with this task for at least 5 min.
Among several metacognitive and demographic items (see Supplementary Materials), participants selected which one of four options best described the type of relationship they were thinking about during the task (heterosexual relationship, homosexual relationship, other relationships, other things).We refer to this variable as 'Relationship Focus.' Participants reported their age, occupation, and self-identified gender and ethnicity.
To test our main hypothesis, we examined frequencies of specific features across conditions (Table 2; see Supplementary Materials for full list).We calculated Spearman's Rho correlations between feature frequencies in each of the four conditions, taking the feature as the unit of analysis (Table 3).All correlations were moderately strong, positive, and significant, showing considerable overlap in the graded structure of prototypes across conditions.Fisher's Exact Z-Test showed that feature frequencies in the default condition correlated more strongly with those in the heterosexual-specific condition than with those in the lesbian-specific (z = 3.67, p < .001) or gay-specific conditions (z = 3.33, p < .001).Feature frequencies in the default condition correlated similarly with those in the lesbian-and gay-specific conditions (z = 0.34, p = .73).As predicted, heterosexualspecific concepts of romantic love resembled the default concept more than the lesbianand gay-specific concepts did.
To describe this difference between concepts further, chi-square tests compared the frequencies of all 41 features mentioned by at least 10% of participants in the heterosexualspecific condition to feature frequencies in the lesbian-and gay-specific conditions.After Bonferroni, frequencies differed for two features (personal sacrifices and references to gender and sexuality).Participants generated significantly fewer references to gender and sexuality (e.g., 'knowing the needs of a female,' 'same as heterosexuals basically') in the heterosexual-specific condition (2%) than the lesbian-specific condition (23%; v 2 (1) = 13.45,p < .001)but not the gay-specific condition (17%; v 2 (1) = 9.67, p = .002).

Relationship focus
Participants' self-reported focus during the task varied by condition (Table 4).As few participants (5.8%) indicated thinking about 'other relationships' or 'other', these responses were excluded.Supporting the heterosexual-as-default hypothesis, all other participants in the default and heterosexual-specific conditions reported thinking about heterosexual couples.However, only a minority of participants in the lesbian-specific, and half of participants in the gay-specific condition, reported thinking about the specified same-gender couple during the task.

Discussion
In Study 1 heterosexual-identified participants conceptualized default and sexualityspecific concepts of love in overlapping ways.Consistent with the heterosexual-bydefault hypothesis, the default concept of romantic love resembled the concept generated by those thinking about heterosexual couples more than it resembled the concept of generated by those thinking about either lesbian women or gay male relationships.Tellingly, participants who were asked to conceptualize same-gender love were more likely both to drop-out, and to report thinking about heterosexual relationships.Together, these patterns suggest that participants found it more difficult to access and report their knowledge of same-gender relationships.
Study 2: Feature generation among lesbian, gay and bisexual participants Results of Study 1 were consistent with our heterosexual-by-default hypothesis, but are open to alternative explanations.Given that participants were heterosexualidentified, the similarity between heterosexual-specific and default concepts could have been caused by in-group projection from heterosexual attributes to the default prototype (Mummendey & Wenzel, 1999).Accordingly, Study 2 replicated Study 1 among lesbian-, gay-, and bisexual-identified people.In-group projection would be supported only if lesbian women and gay men were to conceptualize same-gender romantic love as closer to the default than heterosexual love.However, cultural heterosexism can be internalized by sexual minority individuals, meaning that heterosexual-specific concepts may resemble the default concept even for individuals in these minority groups.Lesbian women and gay men understand their sexual orientation to comprise monosexual (Anderlini-D'Onofrio & Alexander, 2009) samegender love and attraction, whilst bisexual people experience more of a 'dual sexual orientation' (Hemmings & Blumenfeld, 1996).Accordingly, we analysed these two sexual minority participant groups separately.Note.In Studies 1-2, participants had to select one option, whereas in Studies 3-4, participants could select more than one option.

Participants
As in Study 1, we aimed to recruit 200 participants, via UK-based research websites, Facebook groups, the researchers' social networks, and Prolific.ac.Participants were unaware that we were only recruiting LGB participants.As with Study 1, participants were only informed that we were interested in their perceptions of romantic love.Among 587 respondents, 281 left the survey incomplete.We excluded participants with heterosexualidentified (n = 81) or unreported (n = 7) sexual orientation, or who failed to follow instructions (n = 16).The final sample included 99 female, 90 male, 11 non-binary, two unspecified participants (see Table 1 for self-identified ethnic background) aged 18-60 (M = 28.81,SD = 8.96).Participants self-identified as lesbian or gay (n = 98), bisexual (n = 74), and non-heterosexual (n = 30).Most (n = 111) received compensation of GBP £1.

Coding
The coding procedure was identical to Study 1. Two independent judges coded 2,480 items using the 119 codes from Regan et al.

Results
Overall, LGB participants generated a large number of unique features of love that were not generated by Study 1's heterosexual participants (e.g., monogamy, freedom; see Supplementary Materials), indicating distinct features of LGB participants' representations of romantic love that are not shared by the heterosexual sample.Unlike Study 1, the average number of features generated per participant did not differ between conditions (F (3, 198) = .88,p = .45, 2 = .01). 4s in Study 1, we calculated Spearman's Rho correlations between the four conditions' feature frequencies, separately by sexual identity group (Table 3). 5Among lesbian women and gay men, feature frequencies in the default condition correlated significantly more strongly with those in the heterosexual-specific condition than with those in the lesbianspecific condition (z = 1.68, p = .047),but not those in the gay-specific condition (z = .04,p = .48).Also, feature frequencies in the default condition correlated significantly more strongly with those in the gay-specific condition than with those in the lesbian-specific condition (z = 1.72, p = .04).Thus, among self-identified lesbian women and gay men, heterosexual love and gay love overlapped equivalently with the default concept, and both overlapped with it more than the lesbian concept did.Among bisexual participants, correlations between feature frequencies across all conditions did not differ (ps > .56).
As with Study 1, chi-square tests compared frequency of generation of the 31 features mentioned by at least 10% of participants in the heterosexual condition to the frequencies of the same features in the lesbian-specific and gay-specific conditions.No significant differences emerged for any of these features across conditions after Bonferroni correction, for either lesbian/gay male or bisexual participants.Notably, unlike Study 1, the number of participants referencing gender and sexuality did not differ significantly between conditions for either lesbian/gay male (v 2 (3) = 5.34, p = .15)or bisexual participants (v 2 (3) = 3.85, p = .28).

Relationship focus
In contrast to Study 1's heterosexual participants, most sexual minority participants reported focusing on the specified couple type in all sexuality-specific conditions (Table 4).Also, in the default condition, lesbian women and gay men almost all focused on same-gender love most often, whilst a majority of bisexual participants focused on heterosexual love.

Discussion
In contrast to the heterosexual-identified participants in Study 1, LGB participants had no difficulty in calling to mind diverse couples.Lesbians' and gay men's default prototypes of romantic love overlapped with concepts of heterosexual and gay male relationships, more than they overlapped with concepts of lesbian relationships.The findings indicate a pattern of intersecting category norms more subtle than either simple cultural heterosexism or in-group projection predictions.We return to this point in the General Discussion.Bisexual people's default concepts overlapped to a similar extent with all sexual orientation-specific concepts, perhaps reflecting the experience of being attracted to more than one gender in a monosexual world (Meyer & Wilson, 2009).
Studies 3 and 4 examined more closely the roles of long-term knowledge and cognitive activation in explaining the heterosexual default concepts of romantic love among heterosexual people.

Study 3: Feature ratings among heterosexual participants
Study 1 found default and sexual orientation-specific concepts of romantic love to be distinct but overlapping in a feature-generation task.In Study 3 participants rated the importance of features of romantic love in a four-condition design that resembled Studies 1 and 2 (feature rating).Previous research has indicated medium to high correlations between concepts inferred from the feature-generation task used in Studies 1 and 2 and feature ratings elicited in Study 3 (rs ~.30-.58;see Fehr, 1988;Hassebrauck, 1997;Hepper et al., 2012).Unlike feature generation, feature rating is not affected by the accessibility of the feature in long-term memory (Higgins, 1996).As such, a conceptual replication of Study 1's patterns in a feature-rating task would suggest that cultural heterosexism is a consequence of difference in participants' long-term memory representations of romantic couples.A null result would point instead to the possibility that cultural heterosexism arises during category activation, due to the different accessibility of sexual orientationspecific concepts of romantic love.

Participants
Our recruitment target and strategy followed those of Study 1.As with Studies 1 and 2, participants were only informed that the study was about romantic love.Of 437 respondents, 239 left the survey incomplete.Attrition, though lower than Study 1, was again higher in the lesbian-and gay-specific conditions (49%) than the default and heterosexual-specific conditions (38%; v 2 (1) = 4.62, p = .03).We excluded participants whose sexual orientation was non-heterosexual (n = 35) or unreported (n = 9).The drop-out rate left the sample smaller than projected (N = 154; 82% female; aged 17-62, M = 27.61,SD = 9.24).See Table 1 for details of participants' self-identified ethnic backgrounds.

Materials and procedure
The online survey software randomly assigned participants to default (n = 45), heterosexual-specific (n = 39), lesbian-specific (n = 34), or gay-specific (n = 36) conditions.Participants viewed 50 features of romantic love, including the 46 most frequently generated features overall in Study 1 and four features generated more often in the lesbianand gay-specific conditions than in the default and heterosexual-specific conditions of Study 1 (e.g., gender and sexuality).Following opening instructions as in Study 1 (adapted from Regan et al., 1998), participants in the default condition were asked for each feature to 'indicate how central (important or essential) you think it is to the concept of 'romantic love' (1 = not at all important, 7 = very important).Participants in the three sexual orientationspecific conditions were additionally prompted to think about the relevant target group.Finally, to assess Relationship Focus, participants were asked to indicate all the relationship types that they were thinking about (heterosexual, gay, lesbian).Participants also reported their age, occupation, and self-identified gender and ethnicity.

Ratings
After excluding two multivariate outliers (Mahalanobis Distance > critical chi-square [df = 7]), the mean rating for each feature in each condition was calculated.Overall, ratings varied considerably across features, with some rated extremely prototypical, or central (e.g., happiness) and others less prototypical, or peripheral (e.g., shyness; see Supplementary Materials for mean rating of each feature). 6Taking the feature rating as the 6 A MANCOVA analysis was conducted with sexuality condition as the IV, the 50 features of love as the DV, and age, gender (Female vs. Male vs. Other), and occupation (Student vs. Other) as covariates.Gender emerged significant (p = .016)because male participants generally gave slightly higher ratings than female participants.Age also emerged as significant covariate (p = .025);younger participants were found to give higher ratings than older participants.However, even when the covariates variables were controlled for, no significant differences emerged between sexual orientation conditions.We did not include ethnicity as we did not want to take away from people's self-identifications, this variable was simply used to get a better idea of the characteristics of our sample.
unit of analysis, we calculated Spearman's Rho correlations between conditions (Table 5).Two informative findings warrant comment.First, all correlations were very high, indicating near-identical conceptual structure across default and sexual orientationspecific conditions.Fisher's Exact Z-Test showed that none differed significantly.Second, correlations between the frequencies of generation of these 50 features in Study 1 and their respective ratings in Study 3 were moderate, replicating prior findings that the concepts derived from feature-generation and feature-rating tasks are not identical.

Relationship focus
All participants in the default and heterosexual-specific conditions reported thinking about heterosexual couples, whilst few reported thinking about lesbian or gay couples (Table 4).In the lesbian-and gay-specific conditions, most participants reported thinking about both heterosexual and relevant same-gender couples, whilst a few reported thinking about condition-irrelevant same-gender couples.Notably, a higher number of participants indicated thinking a heterosexual relationship in the lesbian-specific condition than in the gay-specific condition, similar to Study 1.

Discussion
Study 3 showed no differences between sexual orientation-specific and default concepts.This finding suggests that the differences between conditions observed in Study 1 were likely a consequence of category activation differences.Distinct but overlapping concepts were elicited by feature-generation and feature-rating tasks across the two experiments as in past research.Jointly, Studies 1 and 3 suggest that heterosexual people can conceptualize love among same-gender couples, but that such concepts are harder to access, leading to a greater overlap between heterosexual and default conceptions of romantic love category features must be accessed.Study 4 examined category activation effects directly.

Study 4: Feature accessibility among heterosexual participants
Study 4 investigated heterosexual participants' activation of romantic love concepts after being prompted (or not) with sexual orientation-specific concepts.Participants performed timed forced-choice tasks in which they verified whether features were relevant or not to default and sexual orientation-specific prototypes of romantic love (Fehr, 1988;Hepper et al., 2012).This speeded featureverification task was more cognitively demanding than the untimed feature-rating task in Study 3, but did not require the recruitment of exemplars as did the feature-generation task in Study 1.
Each participant completed two tasks, in which they verified features to the default concept of love, and to one of three sexual orientation-specific concepts of love (i.e., heterosexual-, gay-or lesbian-specific).We measured the number of features verified by participants to assess the breadth of their concepts in each task.We predicted that participants would verify more features in the heterosexual-specific condition than in the other sexual orientation-specific conditions.We used reaction times to examine if the matching task created a similar or different cognitive load across tasks (Fehr, 1988;Hepper et al., 2012).As in past research, we identified central and peripheral features of romantic love for use as stimuli using a median split of feature ratings from Study 3 (Fehr, 1988;Hassebrauck, 1997).
Crucially, we manipulated the order of the two tasks, to examine possible effects of activating one representation of romantic love on another (see also Herr, Sherman, & Fazio, 1983).Participants completed either the default task or a sexual orientation-specific task first.This design first allowed for comparisons of the default and sexuality-specific concepts activated by the first task.Second, this design allowed a test of the effect of activating the first concept on the second concept.We anticipated that when default concepts of love are activated first, participants would distinguish less between sexual orientation-specific concepts in the second task, as the activated default concept would be similar across conditions.In contrast, when sexual orientation-specific concepts are activated first, we anticipated that there would be differences between the different sexual orientation conditions (as in Study 1), and these differences would affect the default task as they will be interpreting the task in terms of their already activated concept.

Participants
We aimed to recruit as many participants as we could within the academic semester (or year, depending).A sensitivity analysis with .80 power indicated that the obtained sample provides enough power to detect an effect size of .30across the three sexual orientation conditions and an effect size of .27across the two order conditions.The obtained sample size also has sufficient power to detect an interaction effect size of .21.We recruited 137 students and staff at a UK university in exchange for £5 (n = 64) or course research-participation token (n = 73).Participants were only informed that the study was about romantic love.Participants were excluded if they identified as non-heterosexual (n = 15) or engaged in random button-pressing, meaning they gave the same response to every trial (n = 10).The final sample comprised 112 heterosexual-identified participants (85% female, age 18-54, M = 21.54,SD = 5.64).See Table 1 for details of participants' self-identified ethnic backgrounds.

Design
Study 4 had a 3 (sexual orientation-specific task) 9 2 (task order) design.All participants completed the default task and one of three sexual orientation-specific tasks (i.e., heterosexual-, lesbian-, or gay-specific), with the order of these two tasks randomly assigned.Within each task, participants verified central, peripheral, and distractor features.The percentage of features verified, and the mean reaction time to verify those features, were the main dependent variables.

Materials and procedure
The task was presented using E-Prime v2.0.Prior to the main blocks, participants completed a 40-trial practice block.Each task (default and sexual orientation-specific) had 101 trials, presented in two blocks.The trials included 50 features of romantic love from Study 3 (the feature Gender and Sexuality was decomposed into two items; gender and sexuality).These 51 items were categorized as central (n = 25) or peripheral (n = 26).
The remaining 50 items were distractors unrelated to love.Participants were instructed in each trial to indicate as quickly as possible whether or not each feature was associated with the relevant concept of romantic love.Each feature appeared in the centre of the screen and remained until the participant responded using the 'P' and 'Q' keys (counterbalanced across participants).Finally, participants completed metacognitive and demographic items.Participants also reported their age, occupation, and self-identified gender and ethnicity.

Transformation and screening
Latencies below 300 ms and above 3,000 ms, indicating extremely fast or slow responses, were recoded to 300 and 3,000 ms, respectively (Greenwald, Nosek, & Banaji, 2003).Only reaction times for correctly verified responses (i.e., 'yes' responses) were analysed following previous prototype research (e.g., Hepper et al., 2012).A logarithmic transformation was applied to all variables.

Feature verification
Consistent with prior prototype research, participants verified more central features (M = 95.75%,SD = 3.95) than peripheral features (M = 76.01%,SD = 12.31; z = 8.85, p < .001),and more peripheral features than distractor features (M = 2.66%, SD = 3.90; z = 9.19, p < .001).These results confirmed the categorization of features as central or peripheral.We did not analyse responses to the distractor features further.
To test our main hypotheses, we conducted a 3 (Sexual orientation-Specific Condition: Heterosexual, Lesbian, Gay Male) 9 2 (Task Order: Default First, Sexual orientation-Specific First) 9 2 (Feature Type: Central, Peripheral) 9 2 (Task Number: First, Second) mixed-ANOVA.A large Feature Type main effect was observed; participants verified more central (M = 95.75,SD = 3.95) than peripheral features (M = 76.01,SD = 12.31; F (1,106) = 287.03,p < .001,g 2 = .73).Two significant interactions were observed.There was a significant interaction between Feature Type 9 Task Number (F(1,106) = 13.90, p < .001,g 2 = .12)which was qualified by a significant four-way interaction involving all factors (F(2,106) = 4.36, p = .02,g 2 = .08).Accordingly, and reflecting our hypotheses about the impact of category activation in Task 1 on Task 2, we interpreted this interaction by splitting the data by feature type and order condition.In other words, for each of the four intersections of Note.Means that do not share a subscript within columns and order manipulation differ significantly after Bonferroni correction (p < .017).There were no significant differences or interactions with central features (Ms ranged from 93.33 to 96.63, SDs ranged from 3.34 to 7.13).
feature type (central vs. peripheral) and order conditions (i.e., default first vs.sexual orientation-specific first), we conducted a 2 (Task Number) 9 3 (Sexual orientation-Specific Condition) ANOVA on the percentage of features verified (see Table 6 for means).
We found no effects of task number or sexual orientation-specific condition on the percentage of central features verified in either of the order conditions.Participants verified equivalent central features across blocks and across sexual orientation conditions, regardless of whether they completed a sexual orientation-specific task first (Fs < 1.58, ps > .21)or completed the default task first (Fs < 1, ps > .48).Endorsement of the most central concepts of romantic love did not vary across task, order, or sexual orientation condition.
Next, we analysed the verification of peripheral features (Table 6).In the sexual orientation-specific task first conditions, we observed a main effect of sexual orientation condition (F(2,57) = 4.21, p = .02,g 2 = .13).Planned contrasts found that the heterosexual-specific concept was attributed significantly more peripheral features than the gay and lesbian-specific concepts (t(30.44)= 2.82, p = .008).We observed no effect of task number (F(1,57) = 2.53, p = .12,g 2 = .04).These results imply that participants called to mind different sexual orientation-specific concepts in the absence of other prompts, as predicted (and in line with Study 1).Moreover, the lack of any task number by condition interaction implies that concepts remained distinct between conditions even during the second, default task.It seems that in this condition, different sexual orientation-specific concepts were activated in the first task, and shaped the subsequently activated 'default' concept.
We turned to the verification of peripheral features in the default task first conditions (Table 6, upper panel).Participants verified significantly fewer peripheral features in the first (default) task (M = 75.74%,SD = 13.76)than in the second (sexual orientationspecific) task (M = 78.18%,SD = 12.35), (F(1,49) = 7.69, p = .01,g 2 = .14).However, there was no main effect of sexual orientation-specific condition (F(2,49) = .19,p = .85,g 2 = .01),or condition 9 task number interaction (F(2,57) = 1.78, p = .18,g 2 = .07).Participants all completed the default task in this first trial and unsurprisingly did not differ in the number of concepts they recruited.In the second task, all concepts of romantic love were broader, but did not differ across sexual orientation conditions.As a consequence, lesbian-and gay-specific concepts were broader here than when sexual orientationspecific concepts were elicited first.Indeed, participants who conceptualized gay or lesbian romantic love after first conceptualizing the default concept verified 7% more peripheral features than those who conceptualized gay or lesbian romantic love as their first task (Table 6).

Reaction times
We examined reaction times using a 3 (Sexual orientation-Specific Condition: Heterosexual, Lesbian, Gay Male) 9 2 (Task Order: Default First, Sexual orientation-Specific First) 9 2 (Feature Type: Central, Peripheral) 9 2 (Task Number: First, Second) mixed-ANOVA.Participants responded faster to central features (M = 862.66ms, SD = 179.70ms) than to peripheral features (M = 970.36ms, SD = 233.89ms; F (1,106) = 91.28,p < .001,g 2 = .46),consistent with prototype theory.Participants also responded faster as they moved from the first task (M = 959.57ms, SD = 233.32ms) to the second task (M = 873.45ms, SD = 184.72 ms; F(1,106) = 52.71,p < .001,g 2 = .33),consistent with a practice effect.These effects were qualified by two interactions.First, a Feature Type 9 Task Number interaction (F(1,106) = 10.09,p = .002,g 2 = .087)was observed.Participants responded faster to central features than peripheral features in all conditions, but this effect was weaker when the sexual orientation-specific tasks were completed first (d = .41)than when the default tasks were completed first (d = .55;ps < .001).In other words, participants were had more restricted concepts in the first trial, particularly if they completed a sexual orientation-specific task first.Second, a Task Order 9 Task Number interaction (F(1,106) = 12.72, p = .001,g 2 = .11)was observed.The difference in speed of responses was greater in the task completed first (d = .51)than the task completed second (d = .32;ps < .001).Furthermore, this practice effect was larger when the sexual orientation-specific task was completed first than when the default task was first.This suggests a greater effect of category activation when participants moved from a sexual orientation-specific task to the default task than the reverse.7There were no significant main effects or interactions involving sexual orientation-specific conditions (ps > .58).

Discussion
In Study 4, participants conceptualized sexual orientation-specific love as both different, and as similar, depending upon conditions in a timed feature-verification task.Participants who first conceptualized sexual orientation-specific love recruited the broadest concepts in regard to heterosexual-specific love (mirroring Study 1), and these differences persisted when they subsequently conceptualized a default conception of romantic love.However, when the default concept of love was assessed first, it led participants to verify a similar number of features to when they subsequently conceptualized romantic love between any of the three couples.As a consequence, the lesbian-and gay-specific concepts were more inclusive when they were constructed after calling to mind the default concept, compared to when they were recruited first.
Reaction time data suggest no difference in the ease of bringing sexuality-specific concepts to mind, and that all concepts became more accessible across tasks.Moreover, participants sped up across the experiment more when responding to peripheral than central features, and more where sexual orientation-specific concepts were recruited first.Sexual orientation-specific concepts looked most distinct in this experiment in the sexual orientation-specific conditions and in regard to these peripheral features.Calling to mind sexual orientation-specific concepts particularly facilitated the activation of somewhat different default conceptions of romantic love.

General discussion
Four studies showed how heterosexual-identified people can conceptualize love as equivalent irrespective of romantic partners' genders, but sometimes find it cognitively effortful to do so.Results of three different cognitive experiments suggest that although heterosexual-identified participants can fairly match the features of romantic love to lesbian and gay couples (Study 3), the concepts of love that they recruit under more effortful conditions can be sexual orientation-specific (Studies 1 and 4).Unlike LGB participants, heterosexual-identified participants were the only group who showed a unique overlap between a default conception of love and one that was specific to their own sexual orientation group (Studies 1-2).Heterosexual participants were also disproportionately reluctant to remain in studies in which they were asked to conceptualize same-gender love.Finally, heterosexual-identified participants, unlike their LGB counterparts, struggled to hold same-gender loving couples in mind when they did remain in those studies (Studies 1-3).These findings suggest that cultural heterosexism may be instantiated at the cognitive level by differences in the ease of accessing equivalent concepts of love when thinking about diverse romantic couples.
Consistent with Herek's (2007) claims about cultural heterosexism, and Kahneman and Miller's (1986) norm theory, these experiments explain how a common and higherstatus feature such as heterosexuality can be 'conflated with' or 'assumed to be' the norm in social, political, and everyday discourse.They add to recent findings that explanations of differences between couples frame heterosexual couples as the linguistic norm, thus making group inequalities appear more legitimate (Hegarty et al., 2020).Herek also emphasized that cultural heterosexism was long-standing but malleable.By demonstrating variability in the cognitive instantiation of the same concept by task, the present research illustrates that ideological point at the level of category activation, showing how and why 'romantic love' can be conceptualized in limited cultural heterosexist or heteronormative terms, and conceptualized in more inclusive terms.In other words, our findings suggest that cultural heterosexism shapes the accessibility of LGB exemplars (Higgins, 1996).
In developing concepts of love, people may draw upon their experiences in their ingroup.However, the tendency to take heterosexual couples as default should not simply be attributed to in-group projection, motivated by social identity concerns, particularly because lesbian and gay male participants, continued to recruit heterosexual couples as the default (Study 2).In addition, recruiting both in-group-specific concepts and outgroup-specific concepts influenced subsequent default concepts of romantic love among heterosexual participants (Study 3).This flexibility in the categories that people can access cannot, on its own, explain why heterosexuality might be the default norm for romantic love in some contexts.In addition to such proximal cognitive factors, it seems necessary to also invoke distal factors that structure the social environment in ways that position heterosexuality as a default norm, as LGBT psychologists have long argued to be the case.To do this it is additionally necessary to consider social cognition as occurring not in a vacuum but in an ideological context in which certain forms of relationship are positioned as more normative than others.
Of course, lesbian-and gay-specific concepts may also be harder to call to mind for heterosexual participants, but not LGB-identified participants, because these social identity labels overlay real differences in life experience, which also inform concepts.Rich accounts of such life experiences are beyond our cognitive studies, but LGB-identity formation has long been conceptualized as a process of creating a worldview that is epistemologically centred on the LGB identity rather than heterosexual norms (Cass, 1979;Johnson, 2015).The influence of cultural heterosexism on the psychology of sexual minority individuals is most often described as occurring through the internalization of negative cultural attitudes about sexual minorities (e.g., Meyer, 2003).However, cognitive processes can impact members of stigmatized groups without internalization (Biernat & Deaux, 2012).In our studies, LGB individuals reported no difficulty in calling diverse couples to mind, but gay-specific love was closer to the default concept than lesbian-specific love for lesbian and gay participants.LGB individuals' concepts may now be affected by a 'homonormative' ideology resulting from impactful resistance to cultural heterosexism that has centred the identities of more privileged LGB subgroups of people (e.g., white middle-class men; Duggan, 2002).This interpretation is consistent with Petsko and Bodenhausen's (2019) findings that cultural defaulting resulted in some social groups being harder to bring to mind as members of social categories than others.Both media analyses and related studies of category norm representations suggest that gay male couples may be better represented than lesbian couples in understandings of marriage and divorce (Hegarty et al., 2020).
It is important to note the modest sample sizes in some studies here which suggest the need for replication and extension of this work.To our knowledge, these studies are the first to compare the overlap between default and more specific prototypes of an emotion concept, and we recruited in anticipation of observing medium size effects (Cohen, 1992).Further, like many studies that study heterosexism, participants were largely drawn from the UK and USA, where attitudes towards sexual minorities have improved in recent decades (Morrison & Morrison, 2002).Also, the majority of our sample selfidentified as Caucasian and female, which precluded analysis of gender or ethnicity effects.Studies of heterosexism among men in less-liberal national contexts may more appropriately address other levels of analysis than individual cognition (Bartos ß et al., 2014;West, 2018).Although our studies were conducted in a more liberal context, political views can still have an influence on a more individual level.Further research could also benefit from examining political orientation prior to questions on sexual orientation groups.Conservatives may feel more threatened by imagining same-sex behaviours (Mooijman & Stern, 2016) and find taking the perspective of same-gender love particularly difficult.This could explain the high drop-out rates in Studies 1 and 3. We also need to be cautious of the generalizability of our sample in Study 2. Following the advice of Herek et al. (1991), we used a variety of recruitment strategies; however, our results are based on opportunity samples of those LGB individuals who engage with universities and Prolific academic and should not be generalized to all LGB individuals.
Much research on romantic relationships makes normative assumptions about those relationship forms, including the assumption that romantic love occurs first and foremost in heterosexual contexts (Thorne, et al., 2019).We argue that research on prototypes of love, such as the work presented here, allows psychologists to examine empirically the gap between existing conceptualizations of love in the literature and the conceptualizations of lay individuals (Kelley, 1992).Other default characteristics or relationships, such as monogamous relationships, 'love marriages,' or age-discordant relationships could be analysed using a similar approach.In these cases, prototype research can advance understanding by moving such taken-for-granted default features out of the background and examining the process of defaulting itself.
Limitations notwithstanding, the present findings have implications for understanding and addressing the consequences of cultural heterosexism.As cultural heterosexism influences category accessibility, it may well be the case that heterosexuality, like Whiteness, is associated with some overarching national groups in implicit representations (cf.Devos & Banaji, 2005).Our finding that activating a superordinate category impacts subsequent sexual orientation-specific categorization implies that presenting same-gender couples as general examples of romantic love could be beneficial (see Dasgupta & Greenwald, 2001).Further research should also examine the impact of activating non-typical same-gender couples on attitudes via cognitive processes such as the experience of disfluency and surprise (West & Bruckm€ uller, 2013).In sum, as some societies approach legal equality for same-gender couples, our research shows how social cognition can contribute to both the description and explanation of 'hierarchyenhancing' ideologies (Pratto, 1999) that lag behind.By studying how taken-for-granted concepts (such as love) have developed in such minority groups, we could unlock a wider understanding of these concepts and a deeper understanding of 'equality'.
(1998).Undecipherable responses, synonyms of love, and messages to the experimenter were excluded (n = 78).The resulting 2,402 items were assigned to 231 categories: 110 of Regan et al.'s codes and 121 new categories (e.g., meeting families/friends, polyamory, knowing the partner).Inter-rater reliability was very high (Krippendorff's a =.97); in the few unresolved discrepancies, the first judge's code was used.Unique items generated only once were excluded (n = 58), leaving 173 features.As with Study 1, the feature frequencies (i.e., number of participants generating each feature) operationalized each prototype.

Table 2 .
The top 50 generated features and the percentage of participants in each condition generating each featureCodingOpen-ended responses were coded in two stages.First, the lead author classified responses into 2,814 items, breaking complex responses into multiple items.Responses stating that love is 'love' (n = 26) were excluded.Two independent judges, blind to condition, categorized the remaining 2,788 items into the 119 codes described by Regan et al. (1998), and an Other code included items that did not fit within Regan et al.'s scheme.Both judges devised and agreed upon 26 new codes for the Other items, and reliably coded the items into the resulting 145 categories (Krippendorff's a = .82).Most discrepancies were resolved, but when not possible the first judge's codes were used.Unique responses generated by only one participant that could not be classified into one of the 145 categories (n = 151) were excluded from analysis.The number of participants generating each feature at least once per condition was calculated; these feature frequencies operationalized prototypes of romantic love (see Thorne, 2018 for further details).

Table 3 .
Rank correlations [and 95% confidence intervals]between feature generations in each condition, among heterosexual (Study 1), lesbian/gay male, and bisexual participants (Study 2) Note.Correlations that do not share a subscript within rows are significantly different (p < .05,Fisher's Z).

Table 5 .
Rank correlations [and 95% confidence intervals]in ratings of features between sexuality conditions (Study 3) and between ratings and respective feature-generation frequencies from Study 1

Table 6 .
Percentage of peripheral features verified by trial type, trial order and sexuality condition (Study 4)