Marriage for all (‘Ehe fuer alle’)?! A corpus-assisted discourse analysis of the marriage equality debate in Germany

ABSTRACT This study is situated within corpus-assisted discourse analysis (e.g. Baker, P., & McEnery, T. (2015). Corpora and discourse studies: Integrating discourse and corpora. London: Palgrave Macmillan.) and provides a critical discussion of key topics and stances in the marriage equality debate in Germany. The ways in which the German print media covered the debate are explored through two corpora (created using Nexis) which include relevant texts from three German newspapers (TAZ, Welt, Welt am Sonntag) and two magazines (Der Spiegel, Bunte) from two key periods (July 2000–August 2001 and May 2016–July 2017, respectively); in 2001, Germany introduced ‘civil unions’ (‘eingetragene Lebenspartnerschaften’) but it was only in 2017 that marriage was opened to same-sex couples. Qualitative and quantitative analyses using AntConc and ProtAnt suggest that while there are parallels to marriage equality debates in other countries, there are also language-/culture-specific features. For example, there is a stronger focus on children’s welfare in the contemporary data, since marriage equality meant joint adoption rights for same-sex couples. Through an exploration of keywords/collocations and close reading of prototypical texts, it is shown how linguistic choices contribute to framing the arguments for/against marriage equality and how these stances are connected to culture-specific and cross-cultural norms.


Introduction
This article provides a corpus-assisted critical discourse analysis of the marriage equality debate in Germany, as covered by the German print media. The quantitative and qualitative analyses are based on two corpora consisting of texts from three national newspapers (TAZ, Welt, Welt am Sonntag) and two magazines (Der Spiegel, Bunte) published during two key periods (July 2000-August 2001and May 2016-July 2017. In 2001, Germany introduced 'civil unions' (referred to as 'eingetragene Lebenspartnerschaften', registered partnerships) but it was only in 2017 that marriage was opened to same-sex couples. Focussing on the periods located at the start and end of this timeline, respectively, allows not only for an exploration of salient topics and discourses but also for a detection of longitudinal shifts. As far as I am aware, this is the first study on the German marriage equality debate which is situated within linguistics (but see Clayton, 2014;and Raab, 2011, for qualitative analyses drawing on critical theory, with a strong focus on changes in legislation and the history of political activism related to this issue). Apart from providing first insights into the discursive construction of the debate in Germany, this article discusses the results in light of previous findings on marriage equality debates in other countries. Therefore, the next section will offer a brief discussion of the most relevant studies, followed by a brief summary of the history of marriage equality in Germany. I will then move on to methodological issues and the compilation of the corpus before presenting and discussing the results.

Previous research on marriage equality debates
There are several studies dealing with discourses of same-sex relationships, covering a wide range of discourse types and cultural contexts. In recent years, many analyses have focussed on marriage equality debates in particular, driven by the fact that (proposed) changes in legislation in several countries sparked a fairly wide public discussion. For example, there are studies on discourses of marriage equality in the UK focussing on discussions in Parliament (Bachmann, 2011;Findlay, 2017;Love & Baker, 2015), the radio show Moral Maze (Van der Bom, Coffey-Glover, Jones, Mills, & Paterson, 2015), and the UK press Turner et al., 2018). One of the key observations (made by Love & Baker, 2011;Van der Bom et al., 2015; as well as Turner et al., 2018) is that there has been a shift in strategies used by those opposing same-sex marriage, from explicit to implicit homophobic discourse, attributable to their desire to make their case while not coming across as 'bigoted and prejudiced against LGBT people' (Turner et al., 2018, p. 3).
For the debate in the US, there are studies taking into account data from the General Social Survey (Baunach, 2011) as well as news magazines and journals (Crehan & Rickenbacker, 2006), aiming at uncovering key topics and longitudinal shifts in attitudes towards same-sex marriage. Crehan and Rickenbacker (2006) identified five major themes related to the marriage equality debate in the US: 1. religion/morality, 2. procreation/welfare of children, 3. family values, 4. (equal/constitutional) rights and 5. judicial activism. For example, they observe that there is a strong focus on the rejection of homosexuality on moral and religious grounds in the early data (1996)(1997)(1998)(1999)(2000)(2001)(2002) as opposed to the later data (2003)(2004)(2005)(2006), in which arguments regarding procreation/welfare of children are a lot more prominent. Furthermore, in the context of family values, they notice that samesex marriages are construed as the beginning of a 'slippery slope', which might have negative consequences for marriage (as a social institution) and society in general (also see Bachmann, 2011;and Van der Bom et al., 2015).
Even though there are far fewer studies on marriage equality debates in other countries, they are important since they demonstrate what can be gained from taking not only a longitudinal but also a cross-cultural or culture-specific perspective. For example, Vigo (2015) makes use of a diachronic corpus of Italian as well as British newspapers, suggesting that the analysis of (changes in) lexical choices when referring to same-sex partnerships is a fruitful way to uncover (shifts in) social attitudes. Vigo and Milani (2017) explore the tension between Catholicism and secularism in Italian politics, suggesting that the specific cultural context has an impact on how the debate is framed. 1 Overall, it can be said that the key topics identified by Crehan and Rickenbacker (2006) are found (albeit to varying degrees) in all cultural/linguistic contexts that have been studied so far. However, there also seem to be cultural differences regarding the prominence of each topic and longitudinal shifts in the topics and key arguments. Therefore, it will be interesting to see which key themes are found in the German data (and to which degree), how the topics and arguments shift over time and how all this can be interpreted with reference to political and cultural developments. Therefore, the next section will provide an overview of legislative changes in Germany that led to the introduction of same-sex marriages in 2017, before moving on to methodological issues.

Marriage equality in Germany
In early 1994, the European Parliament issued a resolution on 'Equal rights for homosexuals and lesbians in the European Community', which called for the introduction of a legal framework that would grant homosexual couples the full rights and benefits of marriage. 2 However, it was only in July 2000 that a reading of the draft of the Lebenspartnerschaftsgesetz (Act on Registered Life Partnerships), brought forward by Volker Beck from The Greens took place in the Bundestag, the German federal parliament. In November 2000, it was approved under the coalition of The Greens and the Social Democratic Party. When it became law in February 2001, it was referred to as Gesetz zur Beendigung der Diskriminierung gleichgeschlechtlicher Gemeinschaften: Lebenspartnerschaften (Ending Discrimination Against Same-Sex Communities: Life Partnerships), even though it actually did not put same-sex partnerships on par with married couples in legal terms. When the law entered into force in August 2001, entering into a registered life partnership meant having the same rights regarding, e.g. alimony, as married couples. However, the right to joint adoption, pension rights for widow(er)s and tax benefits were only granted to married couples. Still, having legally recognised same-sex unions at all put Germany amongst the more progressive countries within the European Union (at this point, only the Netherlands had already opened marriage to same-sex couple, while other members such as the UK, Ireland, Austria, and Italy had not introduced civil unions/registered partnerships at all).
Over the next 16 years, the law was revised several times, e.g. in January 2005, when stepchild adoption became possible and there were changes to alimony and divorce regulations (Gesetz zur Überarbeitung des Lebenspartnerschaftsrechts, Registered Life Partnership Law Revision Act). The next revisions mostly occurred as a result of rulings by the Bundesverfassungsgericht (Federal Constitutional Court of Germany, henceforth FCC), giving registered partners the same rights as married couples with regard to step child adoption (October 2009), inheritance tax (August 2010), successive adoption (a legal process during which a person is allowed to adopt a child which has already been adopted by their partner, February 2013), and joint tax filing benefits (June 2013).
With joint adoption being the last major right not granted to registered partners, a first attempt was made to officially open marriage to same-sex couples, in the form of a bill put forward by nine (of the sixteen) federal states in association with The Greens in June 2015, about a month after the Ehe fuer alle (marriage for all) campaign had been launched by the non-profit 'Aktionsbündnis gegen Homophobie e.V'. (Coalition against Homophobia). However, in December 2015, a vote on the bill was blocked by the Social Democratic Party (SPD), the Christian Democratic Union of Germany (CDU), and the Christian Social Union in Bavaria (CSU). It was thus seen as a surprising development when Chancellor Angela Merkel (affiliated with the CDU) signalled she was open to a conscience vote on this issue in the Bundestag, triggering a snap vote on the same-sex marriage bill in June 2017 that led to the bill being signed into law in July, with the law entering into force on 1 October 2017. At this point, several countries in the European Union had already opened marriage to same-sex couples (including Belgium, Denmark, France, Iceland, Ireland, Luxembourg, Norway, Portugal, Spain, Sweden, and the UK).
Overall, then, two major changes in legislation can be identified: the introduction of registered partnerships in 2001, and the opening of marriage to same-sex couples in 2017, which are mirrored in the time-frames chosen for the corpora. Before moving on to a description of the data, though, methodological issues will be outlined.

Corpus-assisted critical discourse analysis
The current study is situated within corpus-assisted (critical) discourse analysis (e.g. Baker & McEnery, 2015), i.e. it makes use of a combination of (quantitative) corpus-linguistic (CL) and (qualitative) CDA tools in order to uncover salient topics, stances, and ideologies in newspaper discourse on the German marriage equality debate. As was noted by Baker et al. (2008, p. 274), the term 'corpus-assisted' may imply a hierarchical relationship between CDA and CL but they go on to argue that both can (and should) 'contribute […] equally and distinctly to a methodological synergy'. In order to outline how this can look in practice they propose possible stages in corpus-assisted critical discourse analysis, acknowledging that 'both [CDA and CL] approaches can be used as entry points, creating a virtuous research cycle' (Baker et al., 2008, p. 295). As a result, most studies adopting this framework combine quantitative with qualitative analyses and incorporate both corpus-based as well as corpus-driven elements at various stages (e.g. Bachmann, 2011; for more background on the CL terminology, see McEnery & Gabrielatos, 2006). In corpus-driven analyses, statistical tests are used in order to identify words that (co)occur with a high frequency, while corpus-based analyses allow for an exploration of terms that have been found to be ideologically salient. The underlying assumption is that the methods complement each other and that this will allow, e.g. for a triangulation of the findings, ultimately leading to a better understanding of the phenomenon under analysis. Like with all mixed-methods approaches, the researcher is required to carefully evaluate their options at every stage of the project and to make their choices as well as the underlying rationale transparent. This is why a summary of the different stages of this study, including a brief discussion of the methodological decisions involved, will be provided before moving on to the presentation of the data and the results.
In the case of the current study, the topic ('press discourse on the marriage-equality debate in Germany') was chosen because of its perceived ideological salience. A purely corpus-driven approach would not have led to the identification of this very topic because only a relatively small percentage of newspaper and magazine articles mentions it.
The time-frames were chosen on the basis of the political developments in Germany and the search terms were identified through reading sample texts and (CDA) studies on the marriage-equality debate in other countries. Nevertheless, using corpus-driven analyses in the next step was useful in order to identify statistically salient keywords and collocations, which in turn led to a closer examination of the discourse context (e.g. through the analysis of concordance lines). Previous CDA research on the topic also helped inform the identification of existing topoi, discourses, and strategies. Even though it could be argued that this process should be driven by statistically significant patterns found in the corpus, it is important to acknowledge that keywords themselves (without an exploration of the context) will 'not reveal discourses' (Baker, 2004, p. 347). Furthermore, ideological salience does not always map directly onto formal (statistical) salience in the corpus sense. For example, single word forms may not be frequent enough to be included in keyword lists but taken together with semantically similar terms they may form a 'key semantic field' that cannot be detected using corpus-driven analyses alone (see Bachmann, 2011, for a more comprehensive discussion of this issue).
Finally, when it came to the CDA analysis of prevalent (shifts in) discourse strategies, a decision was made to use ProtAnt, a tool that allows for a corpus-driven identification of the most prototypical texts in a corpus (Anthony & Baker, 2015. This was done in order to downsize the sample to allow for a close manual analysis of the selected texts and to avoid 'cherry picking' by not basing the selection on largely subjective criteria.

Data
In order to analyse the print media coverage of the marriage-equality debate in Germany, Nexis UK was used to compile two corpora of German national newspapers and magazines. Corpus I includes texts spanning from the first reading of the 'Act on Registered Life Partnerships' to the first registered partnerships after the law had entered into force, covering a period of 13 months overall (July 2000 to August 2001). Corpus II covers the 13-monthperiod leading up to the bill on same-sex marriage being signed into law (June 2016-July 2017). While the focus of the current study is on the contemporary data (corpus II), corpus I has been included as a baseline for comparison. This is based on the assumption that such a comparison can be used to identify longitudinal trends and to highlight key issues in the contemporary debate. For example, changes in the prominence of keywords and their associated collocates may indicate shifts in how the marriage equality debate is represented in the German print media (e.g. with regard to the arguments being put forward for and against marriage equality, respectively). In order to be able to compare the two corpora in a meaningful way, a decision was made to keep the time frame constant (13 months for both corpora) and to include only the print versions of national German newspapers and magazines that are available through Nexis UK for both time periods. An overview of the publications, including the title, type, frequency of publication, and circulation can be found in Table 1. 3 Using the Power Search tool on Nexis UK, a limited set of terms (such as eingetragene Lebenspartnerschaft, registered life partnership and Homo-Ehe, homo-marriage) was used in an exploratory search, and close reading of the initial 50 hits for both timeperiods led to the identification of additional relevant search terms (such as Schwulenehe, gay marriage). 4 Filters were used to eliminate duplicates and to exclude websites and documents with fewer than 500 words. The latter filter was applied in order to include only articles discussing the issue in some detail (rather than just reporting on legislative changes as part of a very short news segment). Other irrelevant hits (mostly texts focussing on same-sex marriage in other countries but also articles in which the issue was mentioned only in passing) were discarded manually. 5 The remaining results were then exported as.txt files for further analysis.

Results and discussion
As can be seen in Table 2, corpus I contains 119 texts (111,461 words) while corpus II contains 76 texts (94,148 words), yielding a total of 195 texts and 205,609 words (approx. 1050 words per text). While the overall number of both texts and words is smaller in corpus II, the texts in corpus II are, on average, slightly longer (1238 words per text, as opposed to 936 words per text in corpus Ikeep in mind, however, that all articles with fewer than 500 words were excluded).
Within both corpora, there are clear peaks in the press coverage, closely related to key developments. For corpus I, these occur around the first reading of the 'Act on Registered Life Partnerships' in the Bundestag (July 2000, n = 15), the approval of the act (November 2000, n = 14) and the law entering into force (July/August 2001, n = 39). For corpus II, the peaks are less clearly linked to the stages involved in implementing a new law. Instead, there is one peak as a result of then-EU commissioner Guenther Oettinger's remarks about 'gay marriage' becoming 'compulsory' under Angela Merkel's liberal government, which triggered a more general discussion about same-sex marriage (Pflicht-Homoehe, November 2016, n = 9). Unsurprisingly, the second peak occurs towards the end of the selected time-frame after the snap vote had been announced (June/July 2017, n = 37).
The publication to cover the issue most consistently throughout both periods is the broadsheet TAZ, with a minimum of 1 article/month with the exception of three months across both corpora (March 2001, February and April 2017). 6

Keyword analysis
In a nextcorpus-drivenstep, keywords, i.e. words 'whose frequency is unusually high in comparison with some norm' (Scott, 1996, p. 53) were identified separately for each corpus. As outlined by Bachmann (2011, p. 82), this can be done by comparing two corpora that contain roughly the same number of tokens (resulting in 'two lists of words which are key in each corpus taking the other as the norm'). Alternatively, a larger reference corpus (such as the BNC for British English) can be used as a baseline for comparison with the corpus/corpora at hand. While the first method can be useful for the detection of longitudinal shifts (see, e.g. Fairclough, 2000, on changes in the ideology of the Labour party), this study adopts the second method. Most importantly, comparing corpus I and II to a reference corpus makes it possible to focus on keywords that are shared, which is useful for identifying similarities, especially with regard to key themes that are important throughout both time-frames. This seems suitable for this first, exploratory study of the coverage of this issue in the German print media and it would not be possible to do this via an approach that pitches both corpora against each other. Note that pronounced shifts should also be evident when using the reference corpus method, indicated by a difference in keyness, for example. 7 Furthermore, differences between the two corpora will be focussed on at a more fine-grained level in the (collocation) analyses of particular keywords.
The German PAROLE corpus was selected as a reference corpus. It was created in 2003 by Wolfang Teubert and contains about 23 million words from four domains (books, newspapers, periodicals, and miscellaneous). It was chosen becauseunlike the corpora hosted at the Institut fuer Deutsche Sprache (Institute for German Language)it is not only freely available but it can also be downloaded (and hence be processed using AntConc). It is available via the Oxford Text Archive. 8 For the identification of keywords in corpus I and II, a PAROLE subcorpus containing all data from newspapers and periodicals was created (containing approx. 16.49 million words). In a next step, a keyword analysis was done for corpus I and II (including only words that were used in at least 5% of the texts, with a cut-off point of 500 results). The results were imported into MS Excel in order to identify the degree of overlap in keywords between corpus I and II, which was found to be extremely high, with 17 of the first 25 hits (i.e. 68%) and 205 out of the 500 results (i.e. 41% overall). Furthermore, keywords were ranked according to effect size, based on Log Ratio. In contrast to measures of statistical significance (such as log-likelihood), which indicate that there is a statistically significant difference between corpora for the identified keywords, Log Ratio indicates how big this difference is (Hardie, 2014).
Since even a fairly superficial analysis of the full lists was not deemed feasible, a decision was made to focus on the top 25 shared keywords (including only the top 25 nouns/ proper names and adjectives for each corpus). An overview of the results can be found in Table 3. 9 First of all, it can be observed that the vast majority of keywords refer to relationships and their labelling, followed by ones referring to people and their identities. In a way this is not surprising, since the majority of these were used as search terms when compiling the corpora. However, within these categories, shifts in keyness can be observedfor example, Homoehe is more prominent in corpus I, which may indicate either that the concept is referred to more indirectly in corpus II and/or that another label is used more frequently instead.
In order to follow up on this point, an analysis of different terms referring to same-sex partnerships was done in order to find out more about their absolute and relative frequency of use, their distribution across and within the corpora, and their strongest collocates (presented in the next section below).
Furthermore, we also see a shift in keyness in the 'Legislation (and implications)' category, with both Ehegattensplitting (joint tax filing benefits granted only to married couples) and Adoptionsrecht (adoption rights) being more prominent in corpus II. The latter example hints at the fact that (joint) adoption rights might be discussed more frequently in the contemporary data (this will be taken up in the section on the ProtAnt analysis below).
From Homo-Ehe (homo-marriage) to Ehe fuer alle (marriage for all) As can be seen in Table 4, there is a clear prevalence of the term Homo-Ehe (homo-marriage, n = 229) in corpus I while the official term eingetragene Lebenspartnerschaft (registered life-partnership) is often referred to simply as Lebenspartnerschaft (life-partnership, n = 43; see table for relative frequencies per million words). The fact that Ehe (marriage) rather than Lebenspartnerschaft (partnership)is used in the compound Homo-Ehe (homo-marriage) stresses the perceived similarity between legally recognised same-sex unions and the traditional institution of heterosexual marriage and construes the former as a new subtype of the latter (by premodifying it with Homo-). Legally, however, the two institutions were kept entirely separate andas was outlined in more detail above registered life-partnerships were not equal to marriages in many respects. In fact, the term Homo-Ehe was often flagged up as potentially misleading, as evident in this statement by then-chairman of the SPD Peter Struck:  (2011). Boldface refers to items that have a higher keyness ranking in corpus I, while underlined items have a higher keyness ranking in corpus II (with a difference in the Log Ratio score of at least 1, indicating at least 'a doubling in size of the difference between the two corpora, for the keyword under consideration' [Hardie, 2014] It is not surprising, then, that in about 19% of all cases (n = 43), the term is used with either single or double inverted commas ('Homo-Ehe') that stress that this term is a colloquialism (apart from being a convenient shorthand for the quite lengthy eingetragene Lebenspartnerschaft). This interpretation is backed up by the analysis below, which shows that (so) genannt* (so-called) strongly collocates with Homo-Ehe.
In four cases Homo-Ehe occurred as part of even more complex compoundsthere were two hits for Homo-Ehepaar (homo-couple) and one hit for Homo-Ehen-Problematik (homo-marriage problem) and Homo-Ehen-Propaganda (homo-marriage propaganda), respectively, with both hinting at the fact that this was seen as a controversial topic.
In corpus II, both Homo-Ehe and Lebenspartnerschaft occur a lot less frequently (n = 64 and 60, respectively). Instead, the term Ehe fuer alle (marriage for all) is used in the vast majority of cases (note that the use of this term is not attested in corpus I at all). In about 10% of all cases (n = 14), Ehe fuer alle is put in either single or double inverted commas. They are used here in order to signal that this is the name of an official campaign, initiated by the non-profit Aktionsbündnis gegen Homophobie e.V. (Coalition against Homophobia). 10 It is the German translation of the name for the French law that opened marriage to same-sex couples in 2013called mariage pour tous (marriage for all). Even though Ehe fuer alle had been used sporadically before (e.g. as a hashtag on Twitter), it really gained momentum after the campaign Ehe fuer alle was launched in May 2015.
In order to go beyond mere keyness/frequency of occurrence of the different terms in the two corpora, a collocation analysis was done for Homo-Ehe, Lebenspartnerschaft and (for corpus II) Ehe fuer alle. Collocations allow for a statistically sound way of exploring strong lexical associations providing insights into the 'company' a word keeps (cf., Firth, 1957), which in turn can be used as a starting point for an exploration of prototypical contexts and associations. In order to establish the degree of overlap of lexical collocates for each term between corpora, each corpus was analysed separately. 11 Based on the top results (with 50 being the cut-off point), collocates were compared and categorised as either 'shared' or 'unique'. As can be seen in Table 5, the only shared collocates are fuer and gegen (for/in favour of and against, respectively), suggesting that Homo-Ehe is framed as a polarising issue (tying in with the findings by Turner et al., 2018, on press coverage of the introduction of samesex marriage in Britain). Further evidence for the construal of the topic as a controversial one comes from other collocates, particularly in corpus I (e.g. Konflikt, conflict). This is much less pronounced in corpus IIthis might be due to the fact that Homo-Ehe refers to the already established (and hence less frequently challenged) part of legislation. This is also mirrored in the collocate eingefuehrt (introduced), which is used in corpus II in contexts where some background on the timeline is provided. The fact that so genannt* (so-called) appears in corpus I but not corpus II shows that the term is no longer explicitly marked as a colloquialism. Most of the other collocates (e.g. Bundestag, Gesetz, law) clearly frame the debate as a legal issue. Note, however, that Gesellschaft (society) opens up another domain. On closer examination of the concordance lines in corpus I, it is evident thatalong the lines of the 'slippery slope argument'the legalisation of same-sex marriage is associated with a potential threat to society as a whole: (2)  For Lebenspartnerschaft*, the breakdown (in Table 6) does not include the numbers for eingetragene*/gleichgeschlechtliche*/homosexuelle*~, since these are provided in Table 4. At first glance, it seems to be an ideologically less loaded and less polarising term than Homo-Ehe. Many collocates refer to the legal domain (e.g. Gesetz, law) and overall, the collocates are largely neutral or even carry positive connotations (fuer, for/in favour of in corpus I and Moeglichkeit, possibility in corpus II). Amongst the shared collocates, Einfuehrung (introduction) is used to refer to the possibility of the bill becoming law in corpus I while in corpus II it refers to this past change in legislation. An exploration of the concordance lines for the collocate Ehe (marriage) reveals that in 6 out of the 9 cases, the term is used in corpus I in the context of making registered partnerships more similar to marriage in legal terms. In all instances, the newspapers give a voice to those opposed to this possibility: Note that what is opposed here is not complete equalityrather, even the possibility of granting registered partnerships most of the rights associated with marriage is strongly rejected.
By the time the Ehe fuer alle initiative was started, registered partnerships had been granted most of the rights previously only enjoyed by married heterosexual couples (such as pension benefits and income tax relief) with the exception of joint adoption rights. When exploring the concordance lines for the pairing of Lebenspartnerschaft with Ehe in corpus II, then, it becomes apparent that there is a very pervasive 'equal but separate' discourse (in 6 out of the 12 examples). In these cases, it is argued that same rights should be granted but separate labels should be kept, as can be seen in this interview with CSU-member Alexander Hoffmann: (4) Eine deutliche Mehrheit sagt: Wir wollen die Gleichberechtigung von gleichgeschlechtlichen Paaren […]. Wenn ich aber frage, ob die Lebenspartnerschaft und die Ehe auch gleich bezeichnet werden sollten, dann bekomme ich sehr wohl unverblümt die Rückmeldung: Ja, aber das sind doch zwei unterschiedliche Dinge! (The vast majority [of people] says: we want equality for same-sex couples […]. But if I ask if life-partnerships and marriage should be called the same, I get the response: but they are two different things!) (Die Welt, 28 June 2017) The fact that people allegedly perceive differences between legally recognised homosexual and heterosexual unions is taken up in this next example (5), in which Catholic bishop Heiner Koch claims that demanding the use of the label 'marriage' is an attempt to make existing differences between the two types of relationship invisible. The underlying argument is that using the same term will lead to people perceiving both types of relationships as the same, even though they are not. In a way, the same argument is taken up by advocates for marriage equality, suggesting that different terms will (mis)lead people into thinking that there are differences (even though there really are not), which in turn might encourage discrimination, as made clear in this example from an interview with Soeren Landmann, chairman of the 'Coalition against Homophobia': (6) Solange die Institution der Lebenspartnerschaft und Zivilehe voneinander getrennt werden, gibt es eine Basis für Diskriminierung. (As long as the institutions of marriage and life-partnerships are kept separate, there is a basis for discrimination). (TAZ, 6 June 2016) Overall, this implies a clear shift in focus (from corpus I to corpus II): whereas the debate had previously been about which rights should be granted to same-sex unions, the more recent discussions take equal rights as a given (for the most part) and the question becomes whether same-sex unions should be 'marriages in all but name' or marriage should be opened to same-sex couples. The fact that most advocates of marriage equality viewed the opening of marriage to same-sex couples as the only way forward is also mirrored in the term Ehe fuer alle. Shifting to this term created the possibility to refer to the issue without using the term Homo-Ehe, whichas was seen abovewas associated with a very polarising and divisive debate: (7) Selbst in Deutschland kam die Homoehe erst durch, als sie in 'Ehe für alle' umbenannt worden war. (Even in Germany, the Homoehe was only introduced after it had been renamed as 'marriage for all'). (Die Welt, 5 July, 2017) This example highlights that Homo-carried negative connotations, so a shift in terminology was seen as helpfulnote that the phrase 'even in Germany' implies that Germany is seen as progressive country open to change. Also, by this point Homo-Ehe had become associated with the then-status quo (i.e. registered life partnerships, not marriage), and Ehe fuer alle clearly signalled that the opening of marriage to same-sex couples was the aim now.
However, as already implied above, the very use of the term Ehe has the potential to spark a new debate about the very definition of the term.
In contrast to the findings for Homo-Ehe, though, the fact that this might be a polarising issue is not reflected in the collocates of Ehe fuer alle in a straightfoward way (Table 7), since, e.g. gegen (against) is not included in the top 50 collocates at all (it ranks as number 56, n = 2). Rather there are semantically related clusters of collocates signalling opposition and support, respectively (e.g. Gegner, opponent; Opposition, opposition vs. fuer, in favour of; protestiert, protest(s); kaempfen, fight; Gleichberechtigung, equality; Oeffnung, opening), with a clear focus on those supporting marriage equality.
A close reading of the concordance lines (andin some casesthe broader context) revealed, however, that this is mostly due to the fact that Ehe fuer alle is the name of an initiative and hence closely connected to a particular stance. Often, the arguments brought forward against marriage equality are discussed explicitly but this is not reflected in collocates but the wider context: (8) Ihr wollt die Ehe für alle? Na gut, dann wird es aber nicht bei zwei Schwulen oder bei Lesben stehen bleiben. (You want 'marriage for all'? Okay, but then it will not stop at two gays or lesbians). (Die Welt, 30 June 2017) This implies yet another shift in the debate, which has moved from problematising Homo and marriage to viewing the re-definition of marriage to include same-sex unions as potentially dangerous, in particular because this is referred to as 'marriage for all '. 12 The redefinition of marriage and family as a 'slippery slope' Overall, following up the identification of keywords with a collocation analysis has helped to uncover statistically salient patterns and developmental shifts. However, as the last example has shown, the purely statistically driven, local analysis of keywords, collocates, and concordance lines is not sufficient, since not all semantically related collocates will be frequent enough to be listed, and sometimes a close reading of the whole text is needed in order to see how individual terms are used in the construction of an argument. Then again, shifting to an exclusively corpus-based perspective, especially with the aim of uncovering dominant discourses, might lead to a 'cherry-picking' approach. In order to avoid this while still being able to incorporate a close reading of selected texts, ProtAnt (Anthony & Baker, 2015 was used to identify the most prototypical texts overall, based on the number of keywords they share with a reference corpus. For corpus I, texts were divided into 6 batches (5 with 20 and 1 with 19 texts) in chronological order. For each batch, the top 2 articles were identified, using the whole of corpus I as a reference corpus. The same method was used for corpus II (4 batches overall, with 19 articles each, corpus II as reference corpus). This resulted in 20 articles overall (12 from corpus I, 8 from corpus II). In the next step, dominant themes were identified (using the categories proposed by Crehan & Rickenbacker, 2006, as a guide).
Whereas religion/morality and judicial activism were not salient at all, about half of all articles in both corpora dealt with (equal) rights. Family values (i.e. trying to 'protect' the institution of marriage) as well as the procreation/welfare of children were also found to be dominant themes in both corpora. These two themes are inextricably linked in both time frames but there are differences in the ways in which these issues are discussed.
Looking at corpus II, it is striking that the connection between marriage and family is seen as much less straightforward: Furthermore, it is acknowledged (in 3 out of the 8 texts) that homosexuals can and do (want to) have children, with the argument brought forward by those opposing marriage equality being that they should not have children: (13) Ich bin gegen die Adoption von Kindern durch zwei Männer, weil ich glaube, dass die Kinder darunter leiden. (I am against two men adopting children because I think the children will suffer). (Die Welt, 30 November 2016) Note that, at this point, the debate was about opening marriage to same-sex couples, entailing that they would gain equal rights on joint adoption. This is different from the situation in other countries, like the United States, where the ruling on same-sex marriage did not automatically include joint adoption rights.
Lastly, it is striking that in corpus II there is an explicit discussion about children being born into same-sex relationships and issues with the recognition of both partners as legal parentsa theme that is entirely absent from corpus I: (14) Ein Gesetz mit Haken: Auch nach Öffnung der Ehe für Homosexuelle können zwei Ehefrauen nicht automatisch gemeinsam Eltern werden (A law with a catch: even after the opening of marriage to homosexuals, two wives cannot automatically become parents together) (TAZ, 11 July 2017) Under the current German law of descent, if a child is born into a heterosexual marriage, both partners will be considered as legal parents. This is not the case in same-sex marriages, where the non-biological parent still has to go through the process of adoption in order to become a legal parent. Even though a reform of the law is currently underway, it remains to be seen whether a consensus will be reached on this particular pointand how the debate will be covered and constructed in the German print media.

Conclusion
Using corpus-assisted critical discourse analysis, this article has discussed some of the most salient keywords and discourses associated with the marriage equality debate in Germany, as covered by German national newspapers and magazines in two key periods (July 2000-August 2001 and May 2016-July 2017).
A main finding of the analysis of (statistically as well as ideologically) salient keywords and their collocates was that there has been a clear longitudinal shift in lexical choices (from Homo-Ehe in corpus I to Ehe fuer alle in corpus II). Whereas Homo-Ehe was associated with a separate legal institution for same-sex couples (sharing most but not all rights with the institution of marriage), Ehe fuer alle stands for opening marriage to same-sex couples (i.e. equal rights).
Furthermore, close reading of 20 prototypical texts from both corpora, selected using ProtAnt, has revealed that the three themes equal rights, family values (i.e. re-defining/protecting marriage) and procreation/welfare of children are prominent throughout. The 'slippery slope' argument is made repeatedly, tying in with findings on same-sex marriage debates in other countries (e.g. Bachmann, 2011;Findlay, 2017;Van der Bom et al., 2015).
However, idiosyncratic longitudinal trends and shifts in the ways in which arguments against marriage equality are framed can be identified as well. In contrast to corpus I, for example, it is generally acknowledged in corpus II that same-sex couples can and, in fact, often do have children. It has been suggested that this leads to opponents shifting the focus of their argument from procreation as a key purpose of marriage to children's welfare, which is also seen to be at stake since the opening of marriage is associated with joint adoption rights for same-sex couples.
Overall, this study has shown how corpus-based and corpus-driven methods can be combined in order to provide converging and complementary evidence on how the marriage equality debate was represented in the German print media. Notes 4. The final search was for Lebenspartnerschaft* OR verpartner* OR Homo-Ehe* OR Homoehe* OR Schwulenehe* OR "schwule* w/15 Ehe*" OR "lesb* w/15 Ehe*" OR "Ehe für alle" OR "homosexuell* w/15 Ehe*" OR "gleichgeschlechtlich* w/15 Ehe*" OR "schwul* W/15 heirat*" OR "gleichgeschlechtlich* w/15 heirat*" OR "lesb* w/15 heirat*" OR "homosexuell* w/15 heirat*" OR "schwul* w/15 hochzeit*" OR "lesb* w/15 hochzeit*" OR "homosexuell* w/15 hochzeit*" OR "gleichgeschlechtlich* w/15 hochzeit*". 5. While only 34 hits had to be discarded for corpus I, the number was significantly higher for corpus II (133)this is mostly due to the fact that a lot of articles in the latter period dealt with marriage-equality debates and recent legislative changes in other countries, especially the USA, France, and Ireland. 6. Since shorter articles (with fewer than 500 words) were excluded, it can only be concluded that TAZ has the most consistent coverage within the corpora but not necessarily the most consistent coverage overall. I would like to thank one of the reviewers for pointing this out. 7. While pitching the two corpora against each other would allow for the identification of shared keywords via lockwords (i.e. keywords with an effect size close to zero), this would not allow for an analysis of shifts in keyness. 8. http://ota.ox.ac.uk/desc/2467. 9. Even though it is highly likely that the analysis of unique keywords would yield interesting results as well, this is beyond the scope of this paper. 10. Furthermore, there are three cases in which the phrase 'Initiative Ehe fuer alle' (the initiative [called] marriage for all) is used without inverted commas. 11. All collocation analyses are based on a minimum collocate frequency of 2 and a window of +/−3 (left/right of the target item), using log-likelihood. 12. Findlay (2017, p. 19) refers to this belief that 'to change the definition of "marriage" to include same-sex couples will be either illogical or abhorrent' as the 'discourse of linguistic rigidity'. 13. In his analysis of debates in the House of Lords, Findlay (2017, pp. 20-21) found this to be a prominent argument as well.