3.1 Motivations

Motivations for using online dating sites are considerably diverse, and they are necessarily qualitatively different from one individual to another. Nonetheless, it is clear that online dating and computer-mediated matchmaking have moved from targeting niche audiences to become part of mainstream social practice (Ellison, Gibbs, and Heino, Reference Ellison, Heino and Gibbs2006: 416). This is certainly due—at least in part—to the fact that information and communication technologies more generally have become central, even mundane, components of everyday social and professional interaction (Herring, Reference Herring2004).

The proliferation of information and communication technologies is not, however, the sole reason for the rapid growth of online dating. Online dating sites respond to a number of social needs of their users. For instance, they provide an alternative to meeting singles in bars and clubs, or at work and school (Whitty and Carr, Reference Whitty and Carr2006: 125). Online dating can also provide an environment less susceptible to personal embarrassment and social awkwardness (Hardey, Reference Hardey2004; Yurchisin, Watchcravesringkan, and McCabe, Reference Yurchisin, Watchravesringkan and McCabe2005). In addition, online dating sites enable communication between members to take place relatively quickly, if not synchronously, which can help to foster emotions relatively quickly (Ben-Ze'ev, Reference Ben-Ze'ev2004: 7). Finally, online dating sites provide members with an expanded network of potential social contacts that is not limited to their personal social networks in offline contexts (Hardey, Reference Hardey2004; Ellison et al., Reference Ellison, Heino and Gibbs2006; Whitty and Carr, Reference Whitty and Carr2006).

3.2 Self-presentation and impression-formation

Self-presentation and impression-formation in online environments has received much attention in CMC scholarship (e.g., Bargh, McKenna, and Fitzsimons, Reference Bargh, McKenna and Fitzsimons2002; McLaughlin, Osbourne, and Ellison, Reference McLaughlin, Osbourne, Ellison and Jones1997; Parks and Floyd, Reference Parks and Floyd1996; Roberts and Parks, Reference Roberts and Parks1999; Utz, Reference Utz2000). According to this research, the lack of non-verbal information makes linguistic information all the more important for managing impressions of others in online contexts. This has had important implications for researchers interested in online dating.

Ellison et al. (Reference Ellison, Heino and Gibbs2006), for instance, found that (English-speaking American) online dating site members took advantage of the online environment in order to present their ‘ideal self’ (see Goffman, Reference Goffman1959). In anticipation of offline, face-to-face interaction, however, members attempted ‘to balance their desire for self-promotion with their need for accurate self-presentation’ (p. 430). At the same time, Ellison et al.'s informants reported developing various techniques for evaluating the credibility of other members’ profiles in recognition of the potential for inaccurate self-presentation in online dating contexts. In turn, then, these same techniques enabled them to develop ‘various strategies to demonstrate the credibility of their [own] identity claims, recursively applying the same techniques they employed to uncover representational ruses in others’ (pp. 430–431). Some of these techniques were discursive, for example, providing narrative accounts of personality traits ‘rather than including a “laundry list” of attributes’ (p. 429).

3.3 Discourse and language variation

A few recent studies of online dating sites have explored discursive and linguistic features of dating site members’ profiles. Groom and Pennebaker (Reference Groom and Pennebaker2005), for example, compared women's and men's personal narratives along a rather gender-stereotyped dimension. They report men focusing more on physical traits of themselves and desired partners, money and other possessions, and professional and personal success. Women, on the other hand, centered their ads around personality traits and their desire to construct more intimate interpersonal relationships. van Compernolle (Reference van Compernolle and Koop2008a) found that French-speaking women and men from Quebec engaged in different types of communication styles. Citing Tannen's (Reference Tannen1990) concepts of report and rapport communication styles, this author concludes that women shape their personal narrative in order to build rapport with their reader, while men tend to report their status and goals. Important age differences were also found, in that younger advertisers tend to treat the online personal as a form of interpersonal communication by, for example, including greetings more often than do older advertisers. In addition, van Compernolle found that older advertisers justify their participation in online dating more often than do younger ones.

Linguistic variation has been explored in two previous studies of online personal ads from Quebec. van Compernolle (Reference van Compernolle2008b) investigated the variable use of ne in verbal negation. He reports ne deletion (i.e., absence or omission of ne) at a rate of 22.4%, a surprisingly high frequency given the fact that the personal narrative is a rather static electronic text that can be revised and edited. A VARBRUL analysis revealed age to be a significant independent variable, in that younger advertisers (18–25 years) retained ne less frequently than did advertisers aged 26–35 years and 36–45 years. In addition, those advertisers who used a second-person address form in the ads deleted ne more often than did those who employed only first- and third-person reference. van Compernolle (Reference van Compernolle2008b) speculates that ‘informal linguistic forms are used as a means of creating intimacy between the advertiser and his or her interlocutor,’ arguing that the ‘use of standard written forms could actually seem sociopragmatically inappropriate, since it is unlikely that (hyper)formal linguistic variants (e.g., ne presence) would be used in face-to-face or live voice communication, at least within certain demographic groups’ (‘Conclusion,’ para. 5).

van Compernolle (Reference van Compernolle2008c) analyzed second-person pronoun use and address strategies in the same corpus of online personals from Quebec. His analysis reveals that most online dating site members address their reader with familiar tu; however, young men tend to use vous-plural to address all potential readers. Despite this difference, van Compernolle concludes that second-person address, especially tu, personalizes the narrative. He speculates that this is motivated by a desire to avoid undue formality and instead to create intimacy between an advertiser and his or her reader in anticipation of an offline, face-to-face meeting (2073–2074).

4.1 Overview of Netclub.com

Netclub.com is a multinational, multilingual online dating site owned by Trader Classified Media serving over 4,500,000 people in North America and Europe (Netclub, 2008a). Members of the Quebec site, which is based in Montreal, are typically between the ages of 25 and 44, and live in urban areas, such as Montreal and Quebec City (Netclub, 2008b). The site encourages long-term, serious relationships, as opposed to brief sexual affaires. According to the membership agreement, inappropriate behavior (e.g., looking for sexual partners) may be sanctioned by the site administrators.

Netclub.com provides its members with a number of services. An online profile is first created as a way of advertising members seeking a relationship. Such information as age, gender, occupation, favorite hobbies, and so forth are entered into a searchable database. In addition, members are encouraged to provide more details by writing a personal narrative up to 2,000 characters (~300 words) in length. Members are also allowed to upload up to five digital photographs to their profile (pending content approval from site administrators).

Two levels of membership are available on Netclub.com. A VIP membership is available for a nominal fee, depending on the length of subscription (e.g., one month, six months, or one year). VIP members have full access to all of the site's communication tools (e.g., intranetwork email and real-time chat), and they can contact any member they wish. Alternatively, a basic (i.e., free) membership is available. Non-paying members are entitled to create a profile and search for potential partners, but they can only contact VIP members who have full access to the site's resources. In other words, no two people with a basic, free membership can contact each other.

4.2 Data collection

Data were collected from the Netclub.com Quebec site during late Fall 2006 over a period of one week. In order to access members’ profiles, the author created two basic membership accounts (i.e., by providing a valid email address): one as a man seeking a women and the other as a women seeking a man. The author never published a profile,Footnote ² nor did he contact any members. All data were collected from members’ public profiles (i.e., profiles that were accessible to anyone having provided a valid email address). As such, no data were obtained through deceptive means, such as contacting a member and pretending to be interested in arranging a date.

Profiles were found using the site's searchable database. For each gender, profiles of members in two age groups were sought: 18–25 years and 36–45 years.Footnote ³ These age groups were chosen as the specific focus of this study because they represent two distinct generations who have had different histories of access to computer-mediated communication technologies. The 18–25 year-olds, for instance, are the so-called ‘digital natives’ (Prensky, Reference Prensky2001) of the ‘Net Generation’ (Tapscott, Reference Tapscott1999), who have nearly always had access to computers and Internet technology. By contrast, many of those members in the 36–45 year-old age group could be considered ‘digital immigrants’ (Prensky, Reference Prensky2001), because they have not always lived in a world where access to information and communication technologies was widespread.

Since the focus of this study is on the variable use of diacritics in dating site members’ writing, only those ads that included a written narrative were included. Those members’ profiles that did not have a personal narrative were simply discarded. Profiles were searched until 50 personal narratives from each demographic category (i.e., age and gender) were obtained. In total, then, 200 narratives—equally distributed across social categories—were retained for analysis. The narratives were copied and saved as text files for further review. No personally identifiable data (e.g., photographs, pseudonyms, date of birth) were copied in order to protect the privacy of Netclub.com members. Each narrative was simply coded for gender and age group.

4.3 Circumscribing the variable context

In accordance with the principle of accountability, each potential variable context (i.e., a letter that should—in the formal, standard, written language—or did have an accent or diacritic) was identified using a concordancing software package. In total, over 5,000 potential variable contexts were extracted from the corpus. Each token, then, was reviewed in order to determine whether or not it represented a true variable context. The following exclusions were made:

• – Capitalized letters (e.g., Écris-moi vs. Ecris-moi ‘write to me’), where accent/diacritic use is optional even in the standard written language

• – Misplaced or incorrectly inserted accents/diacritics (e.g., plûtot vs. plutôt ‘rather’, ce né pas vs. ce n’est pas ‘it is not’)

• – Environments in which another orthographic form replaced an accent or diacritic (e.g., j'ai parler vs. j'ai parlé ‘I spoke’, sa vs. ça ‘that’, c t vs. c’était ‘it was’)

• – Proper nouns (e.g., Montréal)

After exclusions, a total of 2,799 tokens were retained for analysis. Each token thus represents a context in which an accent or diacritic would be expected (or required) in the standard written language. Table 1 displays information about the present corpus, including the number of personals, words, and variable contexts found in each of the four data sets.

Table 1.

Corpus information.

5.1 Overall distribution of diacritics

The first level of analysis aims to determine the rate of accent and diacritic use in the online personal ads and to uncover any patterns that emerge from the data. Each token of a variable context was coded for VARBRUL analysis using the latest instantiation of the variable rule program—Goldvarb X (Sankoff, Tagliamonte, and Smith, Reference Sankoff, Tagliamonte and Smith2005). Three independent variables were considered: age, gender, and diacritic type. Table 2 provides observed frequencies (n), relative frequencies (%), and the VARBRUL probability score for accent/diacritic presence in the corpus. A VARBRUL probability score of .50 or higher indicates that accent/diacritic presence is favored in a given context.

Table 2.

VARBRUL analysis of diacritic variation.^α

^α Total n = 2217/2799 (79.2%); Corrected mean (Input) = .826; Total chi-square = 47.45; Chi-square/cell = 2.97; Log likelihood = –1248.919; p = 0.000

Perhaps unsurprisingly, age was found to be a statistically significant factor. Older advertisers are more likely to use accents and diacritics than are younger ones. This finding supports the well known and often-stated tenet of variationist sociolinguistics that younger speakers typically produce innovative or non-standard variants more frequently than do older speakers. Despite this difference, however, older advertisers demonstrate productive variation in the use of accents and diacritics, in that the expected accent/diacritic was absent from nearly 15% of all variable contexts. Gender, too, was selected as a statistically significant independent variable. As a group, women are favored to use accents and diacritics when compared to men. In terms of Labov's (1990) general principles of language change, this result suggests either stable variation (i.e., no change in progress) (Principle I; p. 210) or the emergence of accent/diacritic use as a ‘prestige variant’ (Principle Ia; p. 213). Finally, diacritic type was found to contribute to the variation as well. The acute accent is the most likely one to be present, with a relative frequency of 89.1%. In contrast, the circumflex and the grave accent occur in approximately two thirds of variable contexts, and the cedilla is present in less than one third of variable contexts.

A comparison of ranges (i.e., the difference between the lowest and highest probability score within a given factor group) shows that diacritic type (range = .54) is the most influential of the three independent variables. Gender (range = .18) and age (range = .17) are approximately equally weighted in terms of contribution to the variation. In order to unpack these findings somewhat, a series of partitioned VARBRUL analyses was performed. A partition analysis helps to provide a more nuanced interpretation of results of a VARBRUL analysis by treating each factor group (i.e., independent variable) as an independent data set (Paolillo, Reference Paolillo2002: 89–93).

5.2 Interactions between age and gender

The results of the first level of analysis show both age and gender to be contributing factors in the variable use of accents and diacritics among Quebecker online dating site members. However, a VARBRUL analysis considering only main effects ‘may fail to capture a large part of the variation’ (Sigley, Reference Sigley2003: 250). Testing for interaction effects via cross-tabulation and recoding can help to elucidate some of the underlying sources of variation not immediately perceptible in a main effects analysis.Footnote ⁴

In the present data, the test statistics (see fn. α of Table 2) indicate that the first level of analysis may not represent the best fit to the theoretical model.Footnote ⁵ For instance, the total chi-square is relatively high for a statistical model with only five degrees of freedom, which is indicative of a number of higher-than-ideal error scores for certain cells. In addition, the scattergram produced by Goldvarb X showed some spread among the data points, suggesting that the original coding was not the best fit to the model. Table 3 displays the results of a partition analysis treating each age group as an independent data set. The test statistics (fn. α and β of Table 3) indicate a better fit to the model when compared to the first level of analysis.

Table 3.

Partitioned VARBRUL analysis of age groups.

^α Total n = 859/1206 (71.2%); Corrected mean (Input) = .743; Total chi-square = 4.52; Chi-square/cell = 0.57; Log likelihood = –636.548; p = 0.007

^β Total n = 1358/1593 (85.2%); Corrected mean (Input) = .877; Total chi-square = 25.84; Chi-square/cell = 3.23; Log likelihood = –603.151; p = 0.000

In both age groups, women retain accents and diacritics more often than do men, thereby corroborating the main effect results presented in Table 2. However, the data clearly show that gender is more influential among advertisers in the 36–45 year-old age group (range = .26) than among those in the 18–25 year-old age group (range = .10). Put another way, while gender is significant in both data sets, the difference between women and men is more pronounced among the older advertisers.

Another way to interpret the data is to explore women and men as separate data sets. These results are provided in Table 4. For both women and men, age was selected as a significant factor in accent/diacritic use. It is clear, however, that age has a much greater influence among women (range = .30) than among men (range = .11). Incidentally, the women/men partition analysis shown in Table 4 produced the best fit to the model (fn. α and β of Table 4), suggesting that this is perhaps the most appropriate way to interpret the data.

Table 4.

Partitioned VARBRUL analysis of genders.

^α Total n = 1078/1270 (84.9%); Corrected mean (Input) = .900; Total chi-square = 4.7; Chi-square/cell = .59; Log likelihood = –438.153; p = 0.000

^β Total n = 1139/1529 (74.5%); Corrected mean (Input) = .765; Total chi-square = 15.63; Chi = square/cell = 1.95; Log likelihood = –795.632; p = 0.000

5.3 Differences among the diacritics

The first level of analysis (Table 2) revealed diacritic type to be the most influential factor in determining the presence versus absence of an accent or diacritic. Such was also true in each of the partition analyses presented above (Table 3 and Table 4). The acute accent context—regardless of its presence or absence—is the most widely distributed, and it is also the most frequently realized one (presence = 89.1%). The circumflex and the grave accent are rather similarly distributed, occurring in 68.8 and 67.4% of variable contexts, respectively. The cedilla, however, is a relatively low frequency diacritic, accounting for only 2.5% of all variable contexts (n = 69), with a mere 31.2% rate of realization.

One reason for the acute accent being realized at significantly higher rates relative to the other diacritics is that its presence or absence often distinguishes between two words by marking a phonological difference (e.g., parl-e [ə] or Ø ‘speak’ vs. parl-é [ɛ] or [e] ‘spoken’). Notwithstanding, it was deleted at a rate of 10.9% (n = 176). This is only marginally higher than the rate of acute accent deletion in asynchronous discussion fora (7.3%) reported by van Compernolle and Williams (in press). Given the similarity between discussion fora and online dating ads in terms of synchronicity (i.e., asynchronous message transmission) and the ability to revise and edit one's text, it is not surprising that the acute accent should occur at relatively high rates in both contexts. van Compernolle and Williams also report that the acute accent is least likely to be deleted in word-final position, because this is the context in which there exists the greatest potential for ambiguity or misinterpretation. Such is also true in the present data, inasmuch as the acute accent was nearly categorically retained in word-final position. The only context in which it was deleted in this particular environment was in the case of feminine gender agreement wherein an additional orthographic (but unpronounced) e is added to the ending of an adjective or past participle.Footnote ⁶ One such example is provided in (1), an excerpt taken from a young woman's personal narrative.

1. (1) Je suis une fille tres connectee a la vie, aux sensations et aux emotions!!

   ‘I'm a girl really connected to life, to feelings and to emotions!!’

In (1), the absence of the acute accent poses no interpretive-semantic problems for the reader of the ad: first, because the syntactic environment itself leaves little doubt as to whether connectee should be read as an adjective (i.e., ‘connected’) or the third-person singular verb form (i.e., ‘connect’); and, second, because the morphological marker of feminine gender agreement (i.e., orthographic e) provides the reader with an additional clue as to the intended meaning of the word. Incidentally, excerpt (1) provides two examples of grave accent deletion (i.e., tres ‘very’ and a ‘to’), as well as an additional example of acute accent absence in the word emotions.

In contrast to the acute accent, the circumflex and the grave accent were less widely distributed but were subject to rather more variation. While the acute accent can serve to distinguish between two words on the phonological level, the circumflex and the grave accent have no such function. Instead, their presence versus absence, in many cases, does not result in any linguistic ambiguity (e.g., très vs. tres ‘very’, être vs. etre ‘to be’). A limited number of homonyms, however, are distinguished in writing through the use of the circumflex or the grave accent, such as the feminine article or pronoun la ‘it/her’ and the adverb là ‘there’, or the preposition sur ‘on’ and the adjective sûr(e) ‘sure/safe’. In isolation, the absence of the circumflex or grave accent in such words could lead to a misinterpretation, but, given that their syntactic and discursive distributions are decidedly different, it is difficult to imagine the absence of these diacritics causing a communicative problem in context.

The cedilla, finally, was least often used. This is because it is rather limited in its potential distribution inasmuch as it can only be used with the consonant c to signal a phonological change (i.e., c [k] vs. ç [s]). In the present corpus, nearly all tokens of potential cedilla use were the demonstrative pronoun ça/ca ‘that’ (n = 60/69). The remaining tokens included garçon/garcon ‘boy’, façon/facon ‘fashion/way’, and français/francais ‘French’. Thus the results for the cedilla are in fact more indicative of its presence versus absence in ça/ca. As for most tokens of the circumflex and the grave accent, the absence of the cedilla in ça is unlikely to lead to any ambiguity.