Abstract
News stories have a well-defined generic structure, consisting of components such as headline, lede, and body, with reported speech a prominent feature, especially in hard news stories. Reported speech serves multiple purposes, from providing evidentiality and intertextuality to contributing to the construction of newsworthiness and to the context creation of news. It is also a site of potential bias in who is cited and how, including with respect to the gender of sources. Using a large corpus of English-language news stories for all of 2023 from the main five mainstream news outlets in Canada (over 370,000 articles from news websites), I examine the gender distribution of those quoted, the syntactic variation in the structure of quotes, and the types of reporting verbs. The study provides a comprehensive overview of the extend of gender bias in contemporary Canadian news, at the same time offering insights into the nature of reported speech in modern news and how it endures and evolves, including in news meant for digital-only publication.
Introduction: Reported speech as a feature of news stories
Reported speech is a constant feature of news stories, especially in hard news and current events articles. By reported speech I refer here to phrases, most commonly full sentences, that are reported as either direct speech, that is, written between quotation marks, or as indirect speech, both types as complements of a verbal process such as say, claim, or affirm and traditionally seen as part of the syntactic structure of the clause that contains the reporting verb. Direct speech can also be reported in its own sentence, without a reporting verb, which I refer to as ‘floating quotes’. Example (1) provides instances of each from the data analyzed in this study, with quotes in bold. The first quote is direct speech. The second one, in a sentence immediately following, is a frameless or floating quote, with the quote in a sentence by itself. After a couple of other paragraphs, the article also includes an indirect quote, introduced by said. 1
(1) ‘
‘
[. . .] Storness-Bliss said
Reporters quote somebody else’s words, whether verbatim or rephrased, for various purposes, among them introducing them as characters in the story, distancing the reporter from the content of the quote, or simply placing the reported speech on the record (Gibson and Zillmann, 1993; Nylund, 2003; Sundar, 1998; van Dijk, 1988). Reported speech can also provide evidentiality and intertextuality (Bednarek and Caple, 2012), contribute to the construction of newsworthiness (Bednarek, 2016) and to the context creation of news (Lukin, 2013).
Many news stories are told in the voice of their protagonists, as politicians present or defend a policy decision, witnesses to an incident or crime retell what they saw, athletes recount a game or competition, and experts reflect on and assess the day’s news. This embeddedness of voices is a characteristic of news discourse (Bednarek and Caple, 2012: 90–94; Bell, 1991). Who is quoted and how they are quoted, that is, whose voices we hear in the news, has important consequences for who we think is important in our society. Studies have shown that we see the news’ protagonists as those who are quoted in the news (Manning, 2001), whether in print or as sound bytes in audiovisual format, in broadcast news, online news sites, and in social media sites that post video clips. Mainstream news media have extraordinary power in shaping what we think of as newsworthy and who is worthy of our attention, mainly by who they chose to feature and to quote (Bell, 1991; Caldas-Coulthard, 1994; Goodyear-Grant, 2013; Kassova, 2020b).
Against this backdrop, this paper studies how people are quoted in English-language Canadian mainstream news media in all the stories published in 2023, focusing on the structure of the reporting frame and the form of the reported material. Using data from the Gender Gap Tracker (see Section 3) and analyzing news in digital format (from news websites), I analyze the relative number of quotes by men, women, and other in news stories, focusing on the type of quote (direct, indirect, or floating), and the type of reporting verb. The Gender Gap Tracker was created in a spirit of accountability, providing a visual summary of gender balance across news outlets by extracting quotes and their sources in news stories in Canadian media.
The focus on gender derives from the fact that numerous quantitative and qualitative studies have shown that women are portrayed differently in the media, with women in power (e.g., women in politics or in business) having their gender and their physical characteristics and attire discussed more often (Carlin and Winfrey, 2009; García-Blanco and Wahl-Jorgensen, 2012; Goodall, 2012; Goodyear-Grant, 2013; Power et al., 2019; Trimble, 2018; Trimble et al., 2021; Van der Pas and Aaldering, 2020; Ward and Grower, 2020). Women are not only portrayed differently, but they are also, quantitatively, seen and heard much less often. The Global News Monitoring Project has been studying the gender gap in media across the globe since 1995. It has consistently found that there are fewer women protagonists of news stories and fewer women sources and that (assuming a target of 50% women in news stories), at the rate of progress they observed since 1995, it would take another 67 years to reach parity (Gallagher, 2005; Macharia, 2015, 2020; Ross and Carter, 2011). Numerous other studies, including longitudinal studies (Shor et al., 2015), have found that women are seen and heard less often than men. Non-binary people are very rarely included in studies but, when they are, results show that they are hardly ever present in the news. A series of studies sponsored by the Gates Foundation (Kassova, 2020a, 2020b, 2023) found that the situation was similar in different countries and for specific events, like the Covid-19 pandemic. Explanations for the phenomenon tend to conclude that the persistent underrepresentation of women in the top spheres of power is the ultimate culprit. Studies have also found that bringing awareness to the problem by keeping track of protagonists and sources may increase the representation of women in news stories (British Broadcasting Corporation, 2024; Hawkins-Gaar, 2019; Yong, 2018), thus providing role models and normalizing the presence of women (and men and non-binary people) across all areas of society. These studies typically examine counts, that is, proportion of women and men mentioned in headlines or in news stories; proportion of women and men quoted; or proportions across different sections of the paper (politics, finance, sports, arts, etc.). There are few studies, however, that also study the linguistic structure of the reported speech and whether there are differences not only in how many times women are quoted, but also in how they are quoted (but see Caldas-Coulthard, 1994).
The present study contributes to our understanding of (digital) news today and the role of news to shape and be shaped by the social context in which news media operate. The increasing demand for diversity in news stories is driven by sociological changes in societies such as Canada that see diversity and inclusion as important goals for all aspects of society. At the same time, this push for diversity is shaping the kinds of stories news outlets tell, and the range of voices they include, contributing to the agenda-setting role of news organizations. Although this is a study at a point in time and place (Canada in 2023), longitudinal studies may reveal the extent to which demands for diversity are successful, using this study as a baseline for comparison. Thanks to the wide availability of news in digital format, the analyzed articles were scraped from the websites of news organizations. The study thus also contributes to our understanding of digital news in that it provides an up-to-date analysis of one characteristic of news, reported speech, making use of new tools for its analysis that can be applied to large datasets (see Section 3).
The next section presents an overview of the linguistic literature on reported speech first, followed by a summary of studies in media studies and political science on the influence of whose voices we hear in the news. Section 3 introduces the project that led to a large-scale data collection, the Gender Gap Tracker, and the subset of data collected for this study. The main analysis of the results is presented in Section 4, with a discussion in Section 5, and why who is quoted in the news matters.
Quotations, verbs, and sources in news stories
Reported speech shows extraordinary variation across the world’s languages and is often the site of markers of evidentiality and stance, where the speaker interprets and evaluates the reported speech, indicating affiliation or distance and the source of the speech, for example, as a witness or through hearsay (Chafe and Nichols, 1986). Goddard and Wierzbicka (2019) explore some of that typological diversity across languages, proposing that reported speech is a pivotal human phenomenon: We live ‘among other people’s utterances: they are the stuff of our daily life, our dreams, memories, thoughts, and stories, the fabric of our mental, emotional and social lives’ (Goddard and Wierzbicka, 2019: 197). Our stories have traditionally included reported speech. It was natural then, that news articles, which we often refer to as news stories, also feature reported speech prominently.
Thus, quotes in news media are an important part of the story, a characteristic feature of journalistic discourse. Journalism textbooks teach that quotes, especially direct speech, contribute to make the story more believable and understandable (Gibson and Zillmann, 1993); they ‘lend credit to speakers who use them in their messages’ (Zelizer, 1989: 369). Quotes are so central to news stories that news items have been characterized as ‘talked into being’: ‘news content revolves around the practice of quoting: the (co-)construction, selection, editing and representation of comments, explanations, interpretations, speculations, praise and blame, among others’ (Nylund, 2003: 844). For journalists, obtaining quotes and attributing them to credible sources are essential aspects of journalism discourse, ‘the bread and butter of a news story’ (Sundar, 1998: 56). Quotation practices, and the function of quotes, is different in journalistic discourse than in conversation or in fiction (Waugh, 1995). According to Nylund (2003), quotes serve narrative purposes in the story, among them: confirmation of claims of newsworthiness (novelty, validity, public relevance); criticism and blame (providing conflict and drama, sometimes with a ‘balanced’ perspective, i.e., with conflicting viewpoints); evaluation (e.g., establishing that something is a problem) and emotion (which the reporter does not want to convey themselves); subjective experience and sense of presence and validity (in contrast, with the reporter providing a first-person point of view); or solutions to problems (media finding solutions and sources who will speak to them). These functions can be summarized as engagement and credibility (Zelizer, 1989). Quotes are also deployed as argumentative devices, to either advance a thesis or to provide evidence for it (Smirnova, 2009). Quotes in journalism have to persuade the reader that the reporter’s perspective is the correct one, thus contributing to truthfulness, factuality, believability (van Dijk, 1988), and reliability (Caldas-Coulthard, 1994). Quotes can also be challenged by those being quoted, therefore making the reporter accountable.
The people quoted in the news are referred to as ‘sources’ and are typically politicians or bureaucrats who are either responsible for problems or trying to solve them, experts who identify or elaborate on the problems, and victims who suffer because of the problems (Nylund, 2003). In more positive news stories of the uplifting kind, sources may be people who achieved something (award winners, athletes, lottery winners). Manning (2001), in an influential treatment of news sources, showed the importance of which people and organizations journalists get their information from. He argued that not all sources enjoy the same degree of access and the same ability to communicate their perspectives. Arguably, the issue of who is quoted is central to representation in media: ‘the ability to speak in the news is important for influencing the terms of broader social and political contestation’ (Benson and Wood, 2015: 803). Lazaridou et al. (2017) argue that examining quotes may be the most straightforward and quantitatively feasible way to identify media bias. Compared to selection bias (which news are covered), coverage bias (which aspects of the event are covered) or framing bias (how the event is described), quotes can be studied at large scale and can reveal bias in who is chosen as a source and how they are presented. Lazaridou et al. (2017) found that, in two UK newspapers, reported speech is more frequently by politicians of the governing party and that the two newspapers differ in how faithfully they quote the original speaker. A study by Niculae et al. (2015) uncovered political bias across news outlets by studying how often and how extensively they cite former US President Barack Obama. This study deals with gender bias and who is quoted, by gender, trying to address an important issue of representation.
In addition to the choice of sources, how journalists convey the voice of those sources is also important. A great deal of research in journalism has studied the effect of quotes in perception of newsworthiness and credibility, in addition to readers’ engagement, with particular focus on whether direct or indirect speech increases any of those. Direct quotes are thought to render news stories more lively and trustworthy (Clark and Gerrig, 1990; Gibson and Zillmann, 1993; Short et al., 2002; van Krieken, 2020; Vis et al., 2015), in part because direct speech is supposed to reproduce the exact wording of the source, despite studies that point that the assumption of verbatim reproduction does not always hold (Lehrer, 1989; López Pan, 2010; Short, 1988; Waugh, 1995), and that the surrounding text does much of the work for quote interpretation (Nylund, 2003). Some studies have found that direct quotes do not have a noticeable impact on credibility or engagement and that the conflicting evidence about the importance and credibility of quotes may depend on whether the study is conducted on online news stories or offline (i.e., paper) articles (Sundar, 1998; van Krieken, 2020), because the default reading mode in online news stories ‘might be one of distrust and low credibility’ (van Krieken, 2020: 156). Other factors that influence credibility and engagement, such as trust in the media (Matthews, 2012) or whether the story has a narrative format and its topic (Kelly et al., 2003), may be more important than the type of quote present in the story.
I have, thus far, mentioned quote types and sources as fundamental parts of reported speech. We know something is reported speech because it points to a source, another subjectivity or voice in the text other than the author’s. We also know because there is an introductory, reporting, or quoting verb, a verbal process like say or claim that points to that source. Together, the named source and the reporting verb constitute the quoting frame. And, naturally, we know we are reading reporting speech because there is a quote, content either in direct or indirect speech that is considered important enough to be repeated or summarized. Thompson (1996) argues that reported speech (language reports, in his terminology), include a fourth component, the attitude, that is, the evaluation by the present reporter of the message of the original speaker (see also Bednarek, 2006; Scollon, 1998). Such evaluation can already be contained in the reporting verb, as there is a world of difference between saying something and claiming it, a linguistic manifestation of the heteroglossia inherent in language (Bakhtin, 1986; Martin and White, 2005). Thompson (1996) points out that quoting verbs may also indicate attitude towards the speaker (rather than towards the message), with examples such as brayed, wittering on, or fulminated. Jullian (2011) adds that the evaluation or appraisal inherent in quotes is also an appraisal on the part of the journalist towards the events described and their role in their world, that is, reporting verbs encompass the ideology of the reporter (White, 2000).
Taking into consideration this complexity of reported speech in the news, this paper deals with who the sources are in contemporary Canadian news articles, as well as how they are introduced, together with the structure of the quotes the journalist attributes to them. To do so, it makes use of the Gender Gap Tracker (GGT), which allows large-scale analysis of potential gender bias in news in digital format, including rich language analyses regarding the characteristics of reported speech. I discuss the background of the GGT in the next section.
The Gender Gap Tracker
The Gender Gap Tracker is a collaboration between an Ottawa-based non-profit organization, Informed Opinions, 2 and the Discourse Processing Lab 3 at Simon Fraser University. Data collection for the project started in October 2018 and it was officially launched in 2019, with a public web page that tracks the proportion of men and women quoted in Canadian news media in English. 4 A French-language version, the Radar de parité, 5 was launched in spring 2023. The code to scrape the news articles and the system to extract quotes is publicly available, 6 and the quotation tool is also available as a service through the Australian Text Analytics Platform; 7 albeit without gender analysis; see Bednarek et al. (2024).
The goal of the GGT was to bring awareness to the underrepresentation of women in Canadian media. Informed Opinions had carried out some manual studies (Morris, 2016), but continuous tracking was difficult and thus technology was proposed as a solution. A team of researchers at the Discourse Processing Lab built a Natural Language Processing (NLP) system to identify the people mentioned in an article and the reported speech found therein. After that, a coreference and matching step lines up the names of people with the quotes found in the news article. The system uses a combination of rule-based NLP (to find the beginning and end of segments between quotation marks), syntactic parsing (to determine which clausal complements are complements of verbs of saying), and neural methods (to build coreference chains). We then use external services and a large list of people’s names to assign gender to the speakers of those quotes. The GGT extracts a large amount of information. Importantly for this study, information includes the reporting verb (says, claimed, has stated, etc.) and the type of quotation (direct, indirect, floating, and subtypes of those). With that information and the gender of the speaker, we can provide rich analyses of the relationship between gender and type of quote. I present additional details of this extracted information at the beginning of Section 4, before analyzing the results.
Throughout this article, and on the GGT website, our gender categories are ‘women’, ‘men’, and ‘other’ or ‘unknown’. I would like to acknowledge here that this is an unsatisfactory solution to a complex reality. The world is not simplistically divided into women and men. Our solution is based on names, that is, we only assign a quote to a person if we can assign it to a full name. We do not assign ‘woman’ to a quote that starts with She said that, but instead find the antecedent for she and match it to a full name. This is feasible in news articles, which always quote a person’s full name when they are first introduced. Then, we have two approaches, self-presentation and common association. In the first case, we assign gender based on full name and the self-presentation of that person in public. When a source is not a public figure, then the GGT assigns gender based on the most common association of that first name with a gender, using large databases of genders such as GenderAPI. Because of this reliance on external information, inclusion of non-binary and gender-diverse categories is very poor. In cases when public figures self-present as non-binary (e.g., they use they/them pronouns), we assign them to the ‘other’ category. This is also the case for names that are commonly associated with men, women, or non-binary people (Alex, Amir, Ash).
Evaluation of the GGT shows quite high accuracy in detection of people’s names, quotes, and gender prediction. Most importantly, we found that there is only a small bias in predicting gender. Our main concern is that errors would disproportionately affect one gender or the other, that is, that we would more often assign ‘woman’ to a source that is a man, or vice versa. An error analysis of the top sources showed that we had a slight bias against women, that is, that we assigned the label ‘men’ to women sources slightly more often. The error rates, however, are very small: 0.1% of the cases for men and 0.2% of the cases for women (for full details of the evaluation process, see Asr et al., 2021). Thus, we are confident that the results shown in Section 4 below reflect reality rather accurately.
Technical details and results for the first few years of data are available in two published papers (Asr et al., 2021; Rao and Taboada, 2021). Several blog posts and opinion pieces by our group have drawn attention to the problem of underrepresentation (Rao et al., 2021; Taboada, 2020; Taboada and Asr, 2019; Taboada and Chambers, 2020).
Analysis: Who is quoted, how are they quoted
The GGT database contains about 2 million articles since its inception on October 1, 2018. For this paper, I used all articles published in 2023, a total of 371,724 articles, divided as shown in Table 1 by news organization. The table also includes the numbers of quotes by women and men, and the percentage of those by women. 8
Numbers of articles and quotes included in the study.
The table, thus, addresses the first objective of this section, to answer the question who is quoted. It turns out that it is most often men, even in data as recent as 2023. Women, roughly 50% of the population, are quoted on average 29% of the time. There is some variation across news outlets. At the top of the list is CBC, the Canadian Broadcasting Corporation, a public broadcaster with a commitment to equity, diversity, and inclusion. 9 At the bottom we find The National Post, a right-of-center broadsheet. 10 This shows that, despite the realities of gender representation in the real world, news organizations do have some control over whose voices they feature.
The second objective of the analysis is to study how sources are quoted in the news, and whether this differs according to their gender. To do that, I first briefly summarize the type of information that the GGT extracts and, in the rest of this section, discuss how sources are quoted through the type of reporting verb and the syntactic structure of the quotes.
For each article and each quote within the article, the NLP system in the GGT extracts a great deal of information, including all the people mentioned in the article, all the people quoted, the quotes themselves, and the structure of the quotes (reporting verb, type of quote, and length of the quoted material). For example, for the three quotes in Example (1) above, the system would extract the information presented in Table 2. Acronyms for quote types are explained below; for instance, ‘QCQ’ means ‘quotation mark – content – quotation mark’. Note that token counts include both words and punctuation.
Quote information for quotes in Example (1).
In Section 4.1 I provide quantitative results of the types of reporting verbs found in the data, while in 4.2 I expand on the structure of reported speech. General trends are reported, and potential gender differences are also investigated.
Reporting verbs
A large variety of verbs introduce quotes, in either direct or indirect speech. In the roughly 3 million quotes in our dataset (see Table 1), about 2.5 million had a reporting verb, with the rest being floating quotes. Of those verbs, say is by far the most frequent, certainly when only the lemma is considered (accounting for over 70% of all reporting verbs), but also in its many conjugated forms, with said and says at the top of unlemmatized forms of verbs, as shown in Table 3. The next most used form of quotative form is according to, not strictly speaking a verb. These results seem to align with common trends in news coverage, with similar results found in non-Canadian data (see, e.g., Bednarek et al., 2024: 20), including a general preference for neutral reporting verbs in British (Bednarek, 2006: 141) and US news (Garretson and Ädel, 2008). The unlemmatized side of the table tells us that most of the reporting verbs are in the past. On the lemmatized side, we see common verbs of saying (say, tell, add, note, announce, report, suggest, argue, explain, confirm, warn). Two verbs to note are write, which indicates that many of the verbal reports we see in news come from written sources, perhaps from press releases or digital communication (email, social media posts). The other somewhat surprising verb is think, which seems to be used to introduce a summary of a quote to come. For instance, in (2) the indirect quote introduced by thinks foregrounds the direct quote in the next paragraph. All quotes in the example are in bold.
(2) O’Donnell has high hopes that decriminalization will remove some of the sting of stigma for people who are struggling with addiction. And he thinks
Top conjugated and lemmatized forms of reporting verbs.
After the top 15 verbs in Table 3, the rest of the distribution has a very long tail of 3100 different verb lemmas, with many verbs appearing in a handful of quotes each, such as attach, endorse, equate, excoriate, mourn, or project, indicative of a more formal register and perhaps an effort towards stylistic variation in the reporting verbs. A few of the reporting verbs refer to manner of speaking, such as croak, spit, or mumble, reflecting attitudes towards the speaker (Thompson, 1996). There are very few examples of informal reporting verbs, such as be like or go and, when they appear, they tend to be in the speech of a source, as in Example (3), where the quote introduced by go (in bold) is inside a floating quote which reports what Wainwright said.
(3) ‘I think what I enjoy most about it is just my love for the game of baseball and pitching in general’, Wainwright said. ‘I love to sit on the bench next to our guys, next to our pitchers, and go
As mentioned in Section 2, reporting verbs can carry attitude/evaluation and other connotations, which makes them relevant to an investigation of gender bias. In terms of gender distribution, there are actually few differences in the relative proportion of reporting verbs by men and women. To carry out this analysis, I extracted a smaller sample of quotes (a total of 937,131) which I could be sure were clearly attributed to either a man or a woman (recall that the system also produces ‘unknown’ quotes). Table 4 shows that the top verbs are the same, and in very similar proportions. These results may suggest that the general preference of news reporting for neutral verbs reduces any potential gender bias in the use of non-neutral reporting verbs, perhaps even any kind of bias, with reporters inclined to use very general verbs to avoid the impression that they are interpreting or evaluating the source’s words.
Top 10 reporting verbs, by gender.
Perhaps more interesting is an analysis of verbs that are used by men and not by women, and vice versa, rather than relying on the most frequent verbs which tend to be neutral (Table 4). Caldas-Coulthard (1994) found, like this study, that men are quoted much more often than women, but also that women are more likely to scream, yell, and nag than men. To investigate such potential bias in contemporary Canadian data, I extracted verbs that were only used by either men or women, with samples shown below in (4) and (5):
(4) Verbs that introduce only quotes by men: admonish, articulate, attack, bellow, bemoan, brag, charm, disdain, excoriate, grouse, mutter, object, portray, preach, push, rant, snarl, threaten.
(5) Verbs that introduce only quotes by women: bitch, chalk, curse, diagnose, freak, hypothesize, lack, legislate, purr, recollect, retell, screech, shudder, spew, stammer, strive, wallow, yearn.
There does seem to be some verb usage which corresponds to gender stereotypes (e.g., aggressivity for men, emotionality for women). However, it is difficult to make generalizations, as the numbers are very small. For instance, threatened was only used by men, but only 13 times. The verb curse is only used by women, twice. These low frequencies support the hypothesis that journalistic norms about neutral reporting expressions have important discursive effects on gender bias in reported speech, namely that gender bias may be reflected in the proportion of those quoted, but not necessarily in the verbs used to introduce the quotes.
Quote types
As discussed in section 2, differences in quote type (especially direct/indirect) have been said to influence both credibility and readers’ engagement, although these effects are still debated. In order to accurately parse the different types of quotes, the GGT created a classification system to separate indirect, direct, and floating quotes. Indirect quotes are those introduced as part of the subordinate structure of a sentence with expressions such as They said that, and are different from direct quotes in that the latter are graphically identified with quotation marks. Floating quotes are not part of the syntactic structure of a matrix clause, but appear in a sentence on their own, also identified by quotation marks. In the literature, floating quotes are referred to as reported speech without a framing clause (McGregor, 1994), unintroduced dialogue (Tannen, 1986), ‘defenestrated’ speech (Spronck and Nikitina, 2019), or ‘insubordination’ (McGregor, 2019) because the speech that is reported appears without a matrix or main clause. Following our previous work on the Gender Gap Tracker, I use the term ‘floating quotes’ (Asr et al., 2021).
Indirect quotes were classified according to the order of quoting frame versus content, that is, introductory subject (S) and reporting verb (V) versus content or quote (C). For instance, a quote of type SVC is the prototypical The Minister said that . . . Thus, the possibilities for indirect speech are those in Example (6), with the quote type by acronym at the beginning of each example, with the relevant quote content in bold type.
(6) SVC: The last witness of the day was Brian MacRury, who spent 27 years with the Sudbury police, much of that time as a canine track officer.
McRury led the canine track the morning of Sweeney’s murder. He said
CSV: Coverage of recent presidential elections, the coronavirus pandemic, protests against police killings of Black Americans and other events convinced Janis Fort that the media can’t be believed.
CVS:
VCS: Noting
Direct speech is represented with the same three letters (S, V, C), plus Q to capture quotation marks. Some examples are provided in (7). Floating quotes are a special example of direct speech, where the quote appears in a sentence by itself, as shown in the last example in (7).
(7) QCQSV: Three minutes into the game, sophomore point guard Zakai Zeigler, who brings energy on offense and defense, went down with an injury to his left knee.
QCQVS: The Timoteo Circus is one of the best known of Chile’s 120 circuses.
SVQCQ: Ciotti had announced his party would not vote for either of the two motions of censure – meaning there would not be enough votes to stop the law. Reacting to the vandals, Ciotti tweeted: ‘
QCQ Thomas had made the comments during a committee meeting on Thursday morning as she began asking St-Onge a second round of questions.
‘Minister, I noticed that you answer my questions in French, but other English questions you answer in English, if they’re from your Liberal colleagues’, Thomas said.
Heuristic quotes are a special type of direct speech that spans across multiple sentences. We developed a heuristic method that, when a direct quote was found, would also search back across sentences to find the beginning of the quote, so as not to limit search to within sentence boundaries. Heuristic quotes are of many syntactic types, which is probably why they are the second most frequent type of quote (see Table 5 below). They are typically floating quotes, as shown in the first example in (8), where a direct quote of type QCQSV is followed by a floating quote that spans four sentences. It is the latter that we capture as heuristic. Heuristic may also include the speaker and the reporting verb within the sentence, as we see in the second example in (8), where Rich agrees serves as SV to a quote that spans three sentences.
(8) Heuristic, ‘There’s a lot of meetings that happen’, Singh told Raj. floating ‘
Heuristic, ‘This sounds like a Canadian edition of OpenAI’, I suggest. Rich SVQCQ agrees: ‘
Types of quotes.
The last general type of quote that I will illustrate here are those that are introduced with according to. These also appear in a variety of syntactic patterns, with the prepositional phrase according to introducing the speaker and the quote either before or after the prepositional phrase (Example 9).
(9) according According to a National Institute of Aging study published this to, before month,
according
Turning to the relative frequency of all these types, we see, in Table 5, that SVC (The Prime Minister said that . . .) is the prototypical and, indeed, the most frequent type of quote in news stories. Heuristic quotes appear second, perhaps not unsurprisingly, as they may be composed of many different patterns. What is relevant about the second place of heuristic quotes is that it means that a large amount of reported speech includes multiple sentences. After SVC and heuristic, frequent types include CSV, with a quote followed by subject and verb (Quote, the Prime Minister said) and floating quotes, where there is no reporting verb. Again, similar (but not identical) trends were found in non-Canadian news datasets (see Bednarek et al., 2024: 18), indicating a potential broader reach of such journalistic conventions.
It is interesting to note that, whereas in indirect speech the preferred pattern is SVC (The Prime Minister said that quote), in direct speech it is more common to place the quotation at the beginning, that is, QCQSV (‘Quote’, the Prime Minister said). The second most frequently type of direct speech is floating quotes, with no reporting verb. The third most frequent is SVQCQ, that is, The Prime Minister said ‘quote’. After that, other orders of subject, reporting verb, and quote are vanishingly rare, with only a handful of cases (see Table 5).
The relative frequency of SVC versus QCQSV may indicate that reporters choose direct speech when they want to foreground the quoted content at the beginning of the sentence. In other words, the choice may not be between The Prime Minister said that and The Prime Minister said ‘quote’, but rather between placing the quote after or before the quoting frame. When the quote works better after the quoting frame, then writers choose indirect speech. When the quote is placed before the quoting frame, then reporters choose direct speech.
Turning now to the comparison across genders, and using a smaller subset as above, where the gender of each quote could be clearly identified, we see that there is no difference between the two genders. The summary table (Table 6) shows that the proportions of each type are almost identical for men and women.
Types of quotes, by gender.
To investigate in/direct reported speech, and using a smaller subset of quotes where speakers were clearly identifiable by gender, I grouped the many types into three larger classes. The class of indirect speech includes all the types without a Q: SVC, CSV, CVS, VCS, VSC, SCV, as shown in Example (6) earlier. The second class includes all direct speech, that is, all the types with Q surrounding the content, plus heuristic quotes, which are always direct speech. The third class includes all instances of according to. The relative frequency is shown in Table 7. We see that, overall, reporters seem to prefer indirect speech over other forms of quoting, and indicating that, again, there is no difference across genders in this general classification of quotes.
Relative frequency of direct, indirect, and according to quotes, by gender.
In general, the more frequent use of indirect speech offers an avenue to ‘allow journalists to intertwine their own voices with news sources’ expressions’ (van Krieken and Sanders, 2019: 402). Van Krieken and Sanders, in a historical analysis of Dutch news stories, show that direct speech is becoming more frequent, perhaps as a way to untangle that mix of the reporter’s and the source’s voice (see also Vis et al., 2015). While direct speech may be increasing (although I do not have historical data for English), it is still the case that indirect speech is the more frequent type, especially if we include according to quotes as a form of indirect speech.
This preference for indirect speech over direct speech seems to agree with observations by Waugh (1995), who suggests that indirect speech is the unmarked form of quotation in news stories, and with quantitative corpus analysis results by Semino and Short (2004), who found that news texts contain more indirect speech than other genres such as fiction or biography. It also aligns with very similar findings in a specialized corpus of US print news: It is remarkably close to that found by Garretson and Ädel (2008) in US newspapers during the 2004 presidential election, at 38% vs. 62% across several newspapers. This result suggests that linguistic conventions from older (print) news endure in contemporary digital news. It is also noteworthy that Table 7 shows that there appears to be no gender bias in contemporary Canadian digital news regarding credibility and reader engagement as (potentially) influenced by the use of indirect/direct speech.
Discussion: Why it matters
Bringing together analysis of sources (who is quoted) with analysis of reporting expressions and of quote types (how are they quoted) can provide a richer and more nuanced picture of gender representation in the news. Thus, this study offered insights into diverse rather than uniform usage: There was evidence of strong gender bias in the selection of sources, with women being quoted much less frequently than men. At the same time, there seemed to be no gender bias in the reporting verbs used or in the structure of quotation types. Such nuanced findings are important for initiatives that aim to decrease gender bias in the news, as they can inform us about particular issues that need to be addressed versus aspects that are, in fact, not problematic.
The most striking quantitative result of this data analysis is the lack of balance in gender representation in news stories. Sources are predominantly men, as the Gender Gap Tracker dashboard has shown since we started data collection in October 2018, with average percentages fluctuating between 27 and 32%. The causes, naturally, are not exclusively within news organizations themselves, as they have no control over who is elected to Parliament or appointed as CEO of a company. We saw this clearly during the peak of the Covid-19 crisis, when the percentage of women increased, mostly because the public health officers and ministers of health who were giving daily press briefings tended to be women, as women tend to be overrepresented in such roles (Taboada, 2020). News organizations and reporters, however, do exert control over some of their sources, particularly academics and experts, and some experiments have shown that keeping track of sources helps bring gender balance to news stories (British Broadcasting Corporation, 2024; Yong, 2018).
From a linguistic point of view, the analysis shows a relative lack of variety in reporting verbs, with the single verb say accounting for over 70% of the verbs chosen to introduce quotes. Reporting verbs tend to be in the content or factual class (say, tell), as opposed to evaluative verbs (claim, argue). This helps produce the impression of objectivity or neutrality. Although the structure of quoting frames and quotations shows more variation than the use of reporting verbs, the prototypical SVC structure is the most frequent and indirect speech is more frequent than direct speech, with no gender differences in quote types. One possible conclusion is that, although the most frequent reporting verbs are factual in nature, reporters still exert control over how they rephrase the content of quotes. In addition, the study’s findings indicate that linguistic conventions from print news can still be found in digital news. The fact that there is no gender bias in quote type is a positive finding that could not have been predicted in advance, and shows the advantage of using new tools to identify trends that would be difficult to analyze manually in large datasets.
I have, in this paper, analyzed one year’s worth of data in English. The Gender Gap Tracker and the Radar de parité have such rich data that many other further analyses present themselves. In previous work, we have analyzed the types of sources present in English-language news, by classifying them into politicians, athletes, non-profit leaders, lawyers, experts, or witnesses (Asr et al., 2021), for the years 2018–2020. Further analyses would show whether the same trends continue, and whether they are similar in French-language media. Research by Calsamiglia and Ferrero (2003) examined the differences between academics showing caution in their reporting verbs, while the organizations being quoted were more assertive. Similar analyses could show differences across women and men and across different types of sources. Future research could also analyze the much larger GGT dataset, comprising data going back to October 1, 2018, with continuous updates for the foreseeable future. Examination of the data in French, where the proportion of women quoted in 2023 was similar, at 30%, would highlight potential differences with the English data.
The analyses presented here, and the larger goals of the Gender Gap Tracker, contribute to our understanding of news today, including news available in digital format. The push for more diversity in many of our public institutions leads us, first, to question assumptions about who is important and who needs to be present in news stories. It also pushes advances in corpus and computational projects such as this one, which in turn illuminate aspects of our news discourse.
Footnotes
Acknowledgements
I also want to thank Jillian Anderson, from the Research Computing Group at Simon Fraser University, for help downloading and organizing the data for this study. Special thanks to Monika Bednarek and Teun van Dijk for extremely helpful editorial feedback.
Declaration of conflicting interests
The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The Gender Gap Tracker has received support from multiple sources: Informed Opinions, Simon Fraser University, the Social Sciences and Humanities Research Council of Canada, the Natural Sciences and Engineering Research Council of Canada, and Wage and Gender Equity Canada. It is a true team effort, with research and development led by Fatemeh Torabi Asr, Mohammad Mazraeh, and Prashanth Rao.