1. INTRODUCTION
The three Mainland Scandinavian languages, Danish (DA), Norwegian (NW) and Swedish (SW),Footnote 1 have two basic ways of forming passive sentences: one morphological, by adding an s-suffix to the verb (as in (1a)), and one periphrastic, which involves an auxiliary, typically bli(ve), and the past participle of a main verb (as in (1b)) (see Grammatik over det Danske Sprog (Hansen & Heltoft Reference Hansen and Heltoft2011), Norsk referansegrammatikk (Faarlund, Lie & Vannebo Reference Faarlund, Lie and Vannebo1997), Svenska Akademiens grammatik (Teleman, Hellberg & Andersson Reference Teleman, Hellberg and Andersson1999, henceforth SAG).
(1)
Both passive forms are productively used in everyday language, albeit to different extents and in a different distribution among the three languages (Laanemets Reference Laanemets2012). The use of either one of the two structures appears to depend on a variety of lexical, semantic, syntactic and pragmatic factors (Engdahl Reference Engdahl1999, Reference Engdahl2001, Reference Engdahl, Lyngfelt and Solstad2006; Enger Reference Enger2001; Laanemets Reference Laanemets2004, Reference Laanemets and Fryd2009, Reference Laanemets2012).
Traditional grammar descriptions (Mikkelsen Reference Mikkelsen1911, Hansen & Heltoft Reference Hansen and Heltoft2011 for Danish; Western Reference Western1921, Hansen Reference Hansen1991, Faarlund et al. Reference Faarlund, Lie and Vannebo1997 for Norwegian; Thorell Reference Thorell1973, Teleman et al. Reference Teleman, Hellberg and Andersson1999 for Swedish) and previous single-study research (e.g. for Danish, Heltoft Reference Heltoft, Jörgensen, Platzack and Svensson1994, Heltoft & Falster Jakobsen Reference Heltoft, Jakobsen, Engberg-Pedersen, Fortescue, Harder, Heltoft and Jakobsen1996) have provided a set of distinctions to explain the differences in use between the two passives (see also Laanemets Reference Laanemets2012).
The following distinctions have been proposed to account for the different uses in Danish and, to a lesser extent, Norwegian. The early grammarians (Mikkelsen Reference Mikkelsen1911, Western Reference Western1921) linked the distinction of the two passives to Aktionsart. In their view, the morphological passive is preferred to describe ongoing events, whereas the periphrastic passive mostly describes the completion of events or the inception of events. Heltoft & Falster Jakobsen (Reference Heltoft, Jakobsen, Engberg-Pedersen, Fortescue, Harder, Heltoft and Jakobsen1996) argued against this, claiming that Aktionsart only secondarily determines the use of either one of the passive forms (see also Engdahl Reference Engdahl1999, Enger Reference Enger2001). In these latter accounts the morphological passive is related to rules/commands, and to general events or norms which exist independently and objectively of the speaker. The periphrastic passive is associated with specific events or specific rules, which describe a subjective mood of the speaker (Heltoft & Falster Jakobsen, Reference Heltoft, Jakobsen, Engberg-Pedersen, Fortescue, Harder, Heltoft and Jakobsen1996). To illustrate these different uses, Heltoft & Falster Jakobsen (Reference Heltoft, Jakobsen, Engberg-Pedersen, Fortescue, Harder, Heltoft and Jakobsen1996:203) give the following examples in (2) and (3).Footnote 2
(2)
(3)
The use of the s-passive in (2) is said to indicate a rule or a general norm that exists independently of the speaker, whereas the use of the bli-passive in (3) appears to signal a specific event or rule that is subjectively anchored.
However, these straightforward distinctions cannot readily be extrapolated to Swedish (Enger Reference Enger2001, Engdahl Reference Engdahl, Lyngfelt and Solstad2006). One major difference is that Danish and Norwegian only allow for morphological passivization with simplex tenses and that strong verbs can only take the morphological passive form in the present tense. Apart from these two grammatical restrictions, different semantic differences seem to apply as well (for Danish see Zola Christensen & Christensen Reference Zola Christensen and Christensen2005:122–124; Hansen & Heltoft Reference Hansen and Heltoft2011:Chapter V:651; and for Norwegian see Faarlund et al. Reference Faarlund, Lie and Vannebo1997:513).
With respect to Swedish, several factors have been suggested in the literature to account for the choice of either one of the passive forms. However, it is generally acknowledged that no single factor fully determines the use of the two passive forms. Building on the probabilistic grammar approach to syntactic alternations (see Bod, Hay & Jannedy (Reference Bod, Hay and Jannedy2003) for a collection of studies working in this framework and Bresnan & Ford (Reference Bod, Hay and Jannedy2010) for an application of this approach to the English dative alternation), we hypothesize that the passive alternation is simultaneously motivated by multiple factors.
Using a corpus sample of N = 1,197 passive sentences, we examined the influence of the following six factors that have been suggested in the literature: Subject Animacy, Subject Person and Subject Number, the presence/absence of a Modal Verb and an Av-phrase (‘by’-phrase), and Aktionsart. A logistic regression analysis was used to evaluate the effect of these factors on the use of these two passive forms.
The paper is structured as follows. Section 2 describes the Swedish passive forms and some possible factors that may influence which passive form is chosen. Section 3 outlines our corpus design, data annotation based on the factors and statistical method. The results are presented in Section 4. Section 5 discusses the results and concludes with some suggestions for future research.
2. THE SWEDISH PASSIVE ALTERNATION
This section focuses on the alternation between the morphological s-passive and the periphrastic bli-passive (‘become’-passive) in Swedish. The vara-passive (‘be’-passive) and få-passive (‘get’-passive) are not discussed because the present study is only concerned with the processual passive and not with the static passive as reflected by the two auxiliaries mentioned (see e.g. SAG IV:393, Larsson Reference Larsson2012). Processual passives emphasize the action/event, whereas static passives focus on the condition/state (see e.g. the examples mentioned in SAG IV:393: Patienten blev hela tiden övervakad av vår personal vs. Patienten var hela tiden övervakad av vår personal ‘The patient is continuously monitored by our staff’ with focus on the process and the situation, respectively). Moreover, most of the previous studies on the Swedish passive alternation which we relate to, have paid most attention to the choice between the bli-passive and the s-passive and little to no attention to the static passive forms. Furthermore, it should be noted that to the best of our knowledge only few studies are available on the Swedish passive alternation.
The s-passive is formed by adding -s to the verb form, which may additionally be inflected for tense. In contrast, the bli-passive is formed by adding the copula bli ‘become’ to the participle of the sentence verb, which agrees with the subject in number and gender. The sentences in (4), taken from SAG (IV:360), show an active sentence, the s-passive and bli-passive, respectively.
(4)
According to SAG, almost every transitive verb can occur in both passive forms. However, a quantitative analysis by Laanemets (Reference Laanemets2004) shows that the s-passive is by far the most frequent alternate and that it has a wider distribution in terms of different tenses and different verb types (see also Engdahl Reference Engdahl2001:87). In Laanemets’ (Reference Laanemets2004) corpus of 562 passive sentences, 91% occurs with an s-passive and only 1% with a bli-passive (the remaining 8% sentences were vara-passives). The frequencies reported on in the large-scale corpus study in Laanemets (Reference Laanemets2012) show a similar distribution: 97% s-passives versus 3% bli-passives in the Swedish newspaper corpus containing 19.7 million words and 90.4% and 9.6% respectively in the corpus containing fiction data with almost 2.5 million words (the få- and vara-passives were not included in this study) (Laanemets Reference Laanemets2012:92).
With regard to the wider distribution of the Swedish s-passive, Laanemets’ corpus study further reveals that the morphological passive can occur in all tenses (i.e., the present, past and perfect tense). This finding is notably different from Danish and Norwegian, where the use of the s-passive is restricted to the present tense and simple past (Laanemets Reference Laanemets2004:97f.).Footnote 3 According to Engdahl (Reference Engdahl, Lyngfelt and Solstad2006:29–30), the differentiation of the two passive forms along the lines of tense in Danish and Norwegian (i.e. s-passive for simple tenses and bli(ve)-passive for perfect tenses) fits well with the claim that general events are preferably expressed by using the s-passive and particular events by using the bli(ve)-passive. Indeed, general events are usually described by means of simple tenses and particular events by means of perfect tenses. However, this claim does not hold for Swedish, because the s-passive is found in all tenses.
Regarding the s-passive in Swedish, SAG additionally mentions that certain intransitive verbs can occur in the passive voice, but nearly always in the form of an s-passive, as is illustrated in the passivization of the active sentence Folk pratade mycket vid mötena ‘People talked a lot at the meetings’ (SAG IV:360).
(5)
In addition to the association between intransitive verbs and s-passive, SAG also lists other preferences. For example, verbs with experiencer or recipient roles, such as troddes ‘was thought’ and hördes ‘was heard’, mostly take an s-passive (SAG IV:397, Engdahl Reference Engdahl2001:88). These restrictions are, however, not connected with the influencing factors regarding the passive alternation. As Engdahl (Reference Engdahl1999:8) justifiably points out, ‘[i]n order for there to be a choice of passive forms, both the bli-passive and the s-passive must be available for a given lexical item’. Engdahl (Reference Engdahl1999) found, for example, that frequent verbs, such as tala ‘speak’, stänga ‘close’ and öppna ‘open’, never occur with the bli-passive in the parole corpus.Footnote 4 Engdahl (Reference Engdahl, Lyngfelt and Solstad2006:36) adds that bli-passives with those verbs are not grammatically impossible, but just ‘somewhat odd’.
In order to be able to assess the effect of a number of influencing factors, it is imperative to examine verbs that occur in both passive forms. Engdahl (Reference Engdahl1999:24) provides a non-exhaustive list of ten verbs that meet this requirement: acceptera ‘accept’, avbryta ‘interrupt’, behandla ‘treat’, bjuda ‘invite’, erbjuda ‘offer’, framföra ‘present’, mörda ‘murder’, råna ‘rob’, rädda ‘save’ and välja ‘choose’. In the present study, we selected three verbs from the list to test the effect of the influencing factors (see below).
Previous research, some of which is already mentioned, has come up with a number of possible influencing factors, including syntactic, lexical and semantics ones (see e.g. Engdahl Reference Engdahl2001). The syntactic factor put forward in Engdahl (Reference Engdahl2001:88) is associated with the fact that there are two syntactic constructions which can only contain morphological passives and not periphrastic ones, i.e. constructions with raising verbs (e.g. Räntan antas vara 10 procent ‘The interest is assumed to be 10%’) and impersonal passives (e.g. Det bjöds på kaffe ‘Coffee was served’). Lexically, it appears that for certain verbs slightly different meanings are instantiated depending on which passive forms they take. For instance, the verb anta means ‘accept’ or ‘pass (a law)’ with the s-passive (see (6)), but implies a more specific meaning of ‘being admitted to an education program’ in the bli-passive (see (7)) (Engdahl Reference Engdahl2001:87).
(6)
(7)
As for semantics, one factor is related to Aktionsart. The s-passive is said to be used to express ongoing and/or repeated events, whereas the bli-passive would be reserved to completed and/or temporary events. As a consequence, bli-passives seem to focus on the inception of a new state (8) or the result of a changed state (9), whereas the s-passive seems to focus on the event itself (10) (Engdahl Reference Engdahl, Lyngfelt and Solstad2006:27).
(8)
(9)
(10)
A second semantic factor, suggested by Sundman (Reference Sundman and Karlsson1983) and elaborated by Engdahl (Reference Engdahl, Lyngfelt and Solstad2006:27), is associated with the subject of the passive. According to this account, the bli-passive is preferred when the subject of a passive sentence can control or influence the situation to a certain degree. Following SAG, Engdahl additionally links this control over the situation with animacy, as is illustrated in (11), where the animate participant, ‘the representative’ controls the efforts s/he makes in trying to get reelected (taken from Engdahl Reference Engdahl1999:7).Footnote 5
(11)
Engdahl (Reference Engdahl, Lyngfelt and Solstad2006) further claims that control of the subject can be implied even if the subject is not expressed, as in imperative sentences, as in (12).
(12)
Example (12) is interpreted as ‘avoid putting yourself in a situation where you are likely to get robbed’. If bli-passives are assumed to be associated with animate subjects that are in control over a certain event (either explicit or implied), then logically s-passives will presumably be used with inanimate subjects, as well as with impersonal subjects as in example (13) taken from Engdahl (Reference Engdahl1999:11).
(13)
In (13) the impersonal subject can obviously not exert any influence on the situation In impersonal passives, present-day Swedish only allows for the s-passive (SAG IV:359–404: Engdahl Reference Engdahl, Lyngfelt and Solstad2006:24: Lyngfelt Reference Lyngfelt, Jóhannesson, Larsson, Magnusson Petzell, Malmgren, Rogström and Sköldberg2010:178), even though the use of bli-passives is possible in a few restricted cases (see Engdahl Reference Engdahl, Lyngfelt and Solstad2006:39; Rawoens & Johansson Reference Rawoens and Johansson2014).Footnote 6 Clearly, personal subjects and impersonal subjects reflect different levels of control. In the same vein, local (first and second person) and non-local (third person) subjects, as well as singular and plural subjects, might reflect different levels of control, which in turn might have a different effect on the choice of either one of the passive forms. The latter differentiation according to person (local vs. non-local) and number (singular vs. plural) has hitherto not been looked at, but is examined in the present study.
Another motivating factor that has been suggested in the literature is the role of the expressed agent by means of an Av-phrase (‘by’-phrase). In passive sentences the agent in the av-phrase can be omitted, either because the reference to the agent is unclear, unknown, or simply because the specific agent is unimportant or irrelevant to mention. After all, agent demotion is what characterizes passivization. In her sample of the parole corpus, Engdahl (Reference Engdahl, Lyngfelt and Solstad2006:37) found that the agent in the av-phrase is only expressed in about 10–15% of the passive sentences, however, without any clear differentiation between s-passives and bli-passives. Laanemets (Reference Laanemets2012:124ff.), on the other hand, found a preference for the bli-passive.
Finally, it has been observed that the passive alternation can be associated with the use of modal/temporal auxiliaries. This has been especially pointed out for Danish where the presence of a modal verb appears to influence the choice of the passive form, in that a subjective interpretation or prediction motivates the use of the s-passive, and a non-subjective reading the use of the bli-passive (Heltoft & Falster Jakobsen Reference Heltoft, Jakobsen, Engberg-Pedersen, Fortescue, Harder, Heltoft and Jakobsen1996, Hansen & Heltoft Reference Hansen and Heltoft2011).
(14)
(15)
According to Heltoft & Falster Jakobsen (Reference Heltoft, Jakobsen, Engberg-Pedersen, Fortescue, Harder, Heltoft and Jakobsen1996:209) the use of the bli-passive in (14) is associated with a subjective interpretation, while the s-passive in (15) signals a non-subjective interpretation. With respect to Swedish, Engdahl (Reference Engdahl1999) observed some similar tendencies, but at the same time she found that the auxiliaries kan ‘can’ and ska ‘shall/will’ are actually more commonly followed by an s-passive than by a bli-passive (with a ratio of >1000 s-passives: around 60 bli-passives); the auxiliary vill ‘want to’, however, seems to be more commonly connected with bli-passives. The choice of the passive forms does not seem to be so strictly associated with a (non-)subjective interpretation as in Danish. Nevertheless, it seems to be the mere presence or absence of a modal/temporal auxiliary already influences the choice of the passive form used.
To sum up, previous scholarship on the passive alternation in Danish, Norwegian and Swedish has revealed several factors that seem to motivate the use of one of the two passive forms. However, a multifactorial analysis of the different factors based on a representative corpus sample of actual language data remains wanting. As a result, it remains unclear which factors are actually involved and how strong their respective influence is on the speaker's choice of options. The next section elaborates on the methodology that we used to answer these questions.
3. METHODOLOGY
3.1 Corpus
Starting from the list of verbs provided by Engdahl (Reference Engdahl1999), we selected three verbs taking both passive forms: acceptera ‘accept’ (N = 400), behandla ‘treat’ (N = 399), välja ‘choose’ (N = 398). We specifically chose these three verbs and not the others from Engdahl's list because they had the highest number of bli-passives in the ‘The Swedish Language Bank,’ språkbanken.Footnote 7 To illustrate this: the bli-passive forms of välja (i.e. blir vald/t/a, blev vald/t/a etc.) yielded over 4,000 attestations, while the bli-passive forms of döda ‘kill’ (not included in our study) yielded only 198 observations.
Using Korp, we searched the written subcorpora of Språkbanken for the following forms in present tense and thus retrieved more than 50,000 attestations (see Table 1).Footnote 8
Table 1. Search terms for the verbs acceptera ‘accept’, behandla ‘treat’ and välja ‘choose’.
Since it was practically unfeasible to annotate all attestations manually, we selected the first 200 s-passives and the first 200 bli-passives of each verb from the random list provided by KORP. Our final corpus sample consists of N = 1,197 attestations – three sentences, which turned out to be false positives, were removed. Using a case-control research design, we selected an (almost) equal proportion of bli- and s-passives for each verb, as indicated in Table 2.
Table 2. Number of observations by passive form and verb.
The examined proportion of passive forms implies that we overestimate the number of bli-passives (recall that, in reality, the s-passive is much more frequent than the bli-passive). As a downside to this sample design, however, we cannot evaluate the effect of the particular verbs under analysis (notice that, given the number of observations in Table 2, the relationship between verb and passive form is fully independent). As a consequence, no conclusions about the effect of the verbs will be drawn in this study.
3.2 Predictor variables
The next paragraphs discuss the operationalization of each factor under analysis and state the associated hypotheses.
3.2.1 Subject Animacy: ‘animate’ vs. ‘inanimate’
Humans, animals and organizations/institutions were regarded as animate. In (16) landet ‘the country’ refers to a country's citizens and was regarded as ‘animate’. Inanimate subjects are expected to occur with the s-passive, and logically, animate subjects are expected to prefer the bli-passive.
(16) Sudans regering måste inse att landet måste samarbeta med FN för att bli helt accepterat i det internationella samfundet.
‘The Sudanese government needs to understand that the country has to collaborate with the UN to become completely accepted in the international society.’
(Europarl svenska)
3.2.2 Subject Person: ‘local’, ‘non-local’, ‘other’
Local was defined as the first- or second-person (singular or plural), while non-local as the third person (singular or plural). The label ‘other’ was used for cases with no clear contextual reference. The sentence in (17) illustrates a third person singular inanimate subject of the s-passive.
(17) Amöbadysenteri behandlas med Metronidazol.
‘Amoebic dysentery is treated with Metronidazol.’
(Svenska Wikipedia)
We hypothesize that local subjects will be associated with the bli-passive, given that local subjects are always animate, whereas non-local subjects involve both animate and inanimate arguments. This is in line with the hypothesis that the difference between local and non-local subjects is associated with different levels of control (see Section 2). When no referent was explicitly mentioned, as in (18), we relied on the context in order to label the subject. Note that the agreement on the past participle in this example reveals a reference to an animate referent.
(18) Det är svårt att bli fullt accepterad i det thailändska samhället, menar hon.
‘She thinks it is difficult to become fully accepted in the Thai society.’
(GP2012)
3.2.3 Subject Number: ‘singular’ vs. ‘plural’
The presence or absence of a plural marker was used to determine the category of this variable. When the subject was ‘other’, a value for singular or plural is of course missing. We expect that the effect of this variable will be in line with that of Aktionsart (see below), in the sense that ongoing general events are often expressed through the use of quantitatively undefined or plural arguments (SAG IV:328, 337). We hypothesize that ongoing events are associated with the s-passive. The same relation is expected to apply when plural subjects are involved.
3.2.4 Modal Verb: ‘present’ vs. ‘absent’
The presence or absence of a modal/temporal auxiliary (kan, ska, måste, vill) was used to determine the category of this variable.Footnote 9 This means that the factor Modal Verb does not involve a semantic/functional interpretation (in terms of subjective and non-subjective interpretation, see above), but a formal differentiation between the presence or absence of a modal verb. The presence of modal/temporal auxiliaries is expected to be associated with the use of an s-passive, an example of which is provided in (19), where the auxiliary kan ‘can’ precedes the s-passive. This expectation follows from Engdahl's (Reference Engdahl1999) observation that Swedish does not strictly agree with the Danish pattern and instead uses a modal verb more frequently with s-passive than with bli-passive.
(19) Men, tillägger han, en direkt avsikt att döda kan inte accepteras.
‘But, he adds, a direct intention to kill cannot be accepted.’
3.2.5 Av-phrase: ‘present’ vs. ‘absent’
The presence or absence of an agent expressed by means of an av-phrase was used to determine the category of this variable. First, we expect that av-phrases will only be expressed to a minimal extent, i.e. in all probability to the extent of 10–15% observed in Engdahl (Reference Engdahl, Lyngfelt and Solstad2006:37). An example with an av-phrase is given in (20).
(20) Ledamotens värdighet härrör från att han väljs direkt av folket.
‘The members’ dignity originates from that he is elected directly by the people.’
Secondly, we expect that the presence of an av-phrase will more likely yield the use of the s-passive. This hypothesis runs counter to the finding by Laanemets (Reference Laanemets2012:124ff.) that the use of an av-phrase prefers a bli-passive. However, our reasoning is that the focus shifts from the subject to the ‘by’-agent, which implies a decreased control of the subject over the situation, which would in turn motivate the use of an s-passive. This reasoning is thus in line with findings by Kirri (Reference Kirri1975), Silén (Reference Silén1997) and Engdahl (Reference Engdahl, Lyngfelt and Solstad2006).
3.2.6 Aktionsart: ‘completed vs. ‘ongoing’
To determine the Aktionsart of the passive sentence we relied on the distinction between ‘completed’ and ‘ongoing’ and its respective definitions in SAG (IV:324): ‘if an event implies an event with a natural endpoint, then the event is called completed. After the endpoint the event can no longer take place. An event without a natural endpoint is called ongoing’.Footnote 10 The SAG (IV:398–399) further describes that s-passives are used with ongoing events, which do not have a clear beginning or ending (as in (21), while bli-passives are used with completed events, especially when these have a clear outcome (as in (22)).Footnote 11
(21)
(22)
It is important to note here that Aktionsart is associated with a combination of different factors, including the number and the (in)definiteness of the arguments present in the sentence, as well as accompanying adverbials. (See in this regard Hansen (Reference Hansen2008), who discusses the role of phrasal adverbials and Husband (Reference Husband, Cuervo and Roberge2012), who discusses the influence of quantization and definiteness.)
The SAG (IV:328) further relates the distinction between ongoing and completed events to unlimited vs. limited quantities and/or amounts of time. As illustrated in the examples under (23) (all taken from SAG), a limited quantity and/or a limited amount of time tends to refer to a completed event, while an unlimited quantity and/or amount of time generally indicates an ongoing event.
(23)
Apart from the time adverbials within the brackets, these examples also reveal an interplay with criteria such as number and definiteness/specificity. Plural arguments (Maj skrev brev) thus point to ongoing events (see predictor ‘Subject Number’ above), while definite or specific arguments (Vi drack kaffet; Maj skrev ett brev) point to completed events.
These criteria were used in the annotation process. In the case of specific and definite arguments, the Aktionsart was annotated as ‘completed’. The specific argument ett manuscript in (24) and the definite argument klagomålet in (25) signal completed events, which have a resultative end state: i.e. the acceptance of a certain manuscript and the treating of a certain complaint.
(24) Frågan om när ett manuskript blir accepterat eller inte i en viss tidskrift är naturligtvis av stor vikt för författarna.
‘The question of when a manuscript is accepted or not in a particular journal is of course of great importance to the authors.’
(Medicinska texter)
(25) De har som uppgift att hjälpa konsumenten så att klagomålet mot den utländska näringsidkaren blir behandlat.
‘It is their job to assist the consumer so that the complaint against the foreign trader is treated.’
(Finlandssvenska texter)
In the case of plural arguments, the Aktionsart was annotated as ‘ongoing’. Such plural arguments are unspecific and point towards general events or norms, which have neither a clear beginning nor a clear end state. The plural arguments in (26)–(28) do not involve specific ‘friends’, ‘violations’, or ‘girls’, and as a result the respective propositions indeed indicate general events or norms.
(26) Vänner kan inte väljas enbart rationellt, då är det fråga om någonting annat än vänskap.
‘Friends cannot be selected purely rationally, then it is a question about something else than friendship.’
(Finlandssvenska texter)
(27) Kränkningar av svenskt territorium kan aldrig accepteras.
‘Violations of the Swedish territory can never be accepted.’
(Tidningstexter)
(28) Oftast tjejer blir accepterade för dom är alltid snälla med mig, kram på er!
‘Girls are usually accepted because they are always nice to me, hug on you!’
(Bloggmix)
In case both criteria were combined, i.e., a combination of plural and definiteness, the Aktionsart was annotated as ‘completed’. In (29), for example, ‘the girls’ refer to a specific group of girls, in contrast to (28), where ‘girls’ are referred to in a general sense. The proposition in (29) is completed, as that specific group of girls will get accepted.
(29) Men tjejerna som börjar blir automatiskt accepterade.
‘But the girls who start become automatically accepted.’
(Tidningstexter)
Following the SAG, we want to emphasize that the association between bare plurals and ongoing events on the one hand and definiteness/specificity and completed events on the other hand, is to be understood as a tendency and not a clear-cut rule, which fully squares with the probabilistic approach that we subscribe to. The use of certain adverbials can change the Aktionsart.
(30) Jag hoppas verkligen att denna strävan blir allmänt accepterad inom alla EU-institutioner, även här i parlamentet.
‘I truly hope that this pursuit is generally accepted in all EU-institutions, even here in parliament.’
(Europarl svenska)
In (30) the use of allmänt ‘generally’ induces the interpretation of an ongoing event, even though the argument denna strävan ‘this pursuit’ is specific and comes with a definite article.Footnote 12
As mentioned above, the classical expectation with regard to Aktionsart is that s-passives are used with ongoing events, and bli-passives with completed events (SAG IV:389–399; see Mikkelsen Reference Mikkelsen1911 for Danish and Western Reference Western1921 for Norwegian). Engdahl (Reference Engdahl, Lyngfelt and Solstad2006) casts doubts, however, about the weight of Aktionsart. According to Engdahl, other factors, such as animacy, have a stronger effect. The same argument was made for Danish by Heltoft & Falster Jakobsen (Reference Heltoft, Jakobsen, Engberg-Pedersen, Fortescue, Harder, Heltoft and Jakobsen1996) who ascribe more weight to the subjective/objective mood of the clause (see also Enger Reference Enger2001). Our multivariate analysis enables us to weigh the effect of this variable in relationship to the other influencing factors under analysis.
3.3 Statistical analysis
Logistic regression modelling is a widely used statistical method to analyze dichotomous outcomes (in our case, s-passive vs. bli-passive). The statistical model is now also commonly applied in corpus-linguistic research (e.g. Bresnan et al. Reference Bresnan, Cueni, Nikitina, Baayen, Boume, Kraemer and Zwarts2007 which has become a textbook example in quantitative linguistics).Footnote 13 The logistic regression model is flexible, in that it may include both continuous and categorical variables and the results are relatively easy to interpret.
With six predictor variables (excluding interactions), the logistic regression model can be formulated as in (i):
\begin{equation}
\log _e \left( {\frac{{\pi _i }}{{1 - \pi _i }}} \right) = \beta _0 + \beta _1 X_1 + \beta _2 X_2 + \cdots + \beta _6 X_6\end{equation}
whereby πi equals the probability of an s-passive and 1 − πi equals the probability of a bli-passive. Then, the ratio πi/(1 − πi) = the odds of an s-passive. X 1 − 6 indicates the six predictor variables and β1 − 6 their six coefficients, which basically express the mathematical relationship between the outcome variable and the predictor variables. The (linear) logistic regression model in (i) results from the logit transformation of the (non-linear) logistic response function in (ii):
\begin{equation}
\pi _i = \frac{{{\rm exp}(\beta _0 + \beta _1 X_1 + \cdots + \beta _6 X_6 )}}{{1 + {\rm exp}(\beta _0 + \beta _1 X_1 + \cdots + \beta _6 X_6 )}}\end{equation}
In the next section, which presents the results of our statistical data analysis, we will show how this logistic response function in (ii) can be used to estimate the probability of an s-passive based on a particular combination of factors, and consequently also for each specific observation.
4. RESULTS
We present the results of our statistical data analysis in two consecutive parts. First, by way of introducing the data, we present the bivariate results, which evaluate the association between the outcome variable and the predictor variables separately. Then we present the results of the logistic regression analysis, which evaluates the simultaneous effects of the predictor variables.
The bivariate distributions of all the predictor variables under analysis are given in Table 3. The statistical significance of these distributions was tested by means of chi-square tests. Cramér's V coefficient was used to estimate the effect sizes. Row percentages and standardized residuals are also reported. The latter allow for a more accurate interpretation of the effects than the row percentages; standardized residuals larger than |2| indicate a significant effect.
Table 3. Bivariate statistics.
*28 missing values
Following this analysis, we have strong evidence that the choice of passive form is significantly associated with four predictor variables: Subject Animacy and Subject Person, Modal Verb and Av-phrase. The observed tendencies are that the s-passive is positively associated with an inanimate and non-local subject, the presence of a modal verb, and by the use of an av-phrase. Based on these bivariate statistics, no evidence is found for Aktionsart and Subject number. Note that the distribution is strongly skewed in the case of Subject Person and Av-phrase: the large majority of observations includes a non-local subject and lacks an av-phrase. Finally, the bivariate results further suggest that Subject Animacy and Modal Verb yield the strongest effect.
Table 4 gives the model estimates of our logistic regression analysis. Model estimates, except for Odds Ratios and their 95% Confidence Intervals, were computed with functions from Harrell's (Reference Harrell2013) rms package.Footnote 14
Table 4. Logistic regression model estimates (success = s-passive).
s.e. = standard error of the estimated coefficient
Based on our logistic regression model we have evidence for the effect of five predictor variables: Subject Animacy and Subject Number, Av-phrase, Aktionsart, and Modal Verb; Subject Person is not retained as significant in this multivariate model.
The direction of the effects corroborates what we hypothesized, except for Aktionsart: all else being equal, the s-passive is more probable than the bli-passive when the Subject is inanimate and plural, when there is an av-phrase and modal verb present, and when a completed Aktionsart is expressed. According to this model, the presence of a modal verb yields the strongest impact effect on the choice of passive construction. The odds ratio estimate indicates that the odds of a sentence with a modal verb being used with an s-passive are about nine times the odds of a sentence without a modal, with a 95% confidence interval of 6.07–13.77. The second strongest effect is associated with the animacy of the Subject. Here the odds of an animate Subject to be used with an s-passive is nearly seven times that of an inanimate Subject. Notice further that although Aktionsart and Subject Number are significant, their estimated impact effects are very low; an odds ratio of slightly over 1 basically means that there is only a small preference for the s-passive when a plural Subject is used or when a completed action is expressed.
The impact effects associated with each factor may also be visualized in terms of probabilities (i.e., on a scale from 0 to 1), as in Figure 1.Footnote 15
Figure 1 Predicted probabilities and 95% Confidence Intervals for the significant variables of the logistic regression model in Table 4.
The plots nicely illustrate the opposite effects that the different levels of the variables have on the choice of s-passive. Thus, with an av-phrase present, the probability of an s-passive is on average approximately 80%.Footnote 16 Notice again the small probability differences for Aktionsart and Subject Number.
Finally, we can estimate the predicted probability of each observation or combination of factors by means of the logistic response function given in (ii). By way of illustration, we compute the probability of an s-passive based on three different combination of factors.
The predicted probability of an s-passive form of a sentence with a singular, animate Subject, without an Av-phrase, with a completed Aktionsart but without a Modal Verb, is estimated to be 18%. A probability lower than 50% basically means that the bli-passive is actually predicted for this particular combination of factors. The predicted probability is obtained by solving the equation in (iii). The estimated coefficients associated with each variable (see Est. Coeff. in Table 3) are multiplied either by 1 or by 0 depending on which level we are interested in. Thus, ‘inanimate’ = 1 vs. ‘animate’ = 0, ‘present’ = 1 vs. ‘absent’ = 0, etc.
Probability of s-passive
\begin{equation}
\begin{array}{l}
= \displaystyle\frac{{\exp \left( { - 1.789*1 + 1.9147*0 + 1.393*0 + 0.300*1 + 2.213*0 + 0.579*0} \right)}}{{1\! +\! \exp \left( { - 1.789*1\! +\! 1.9147*0\! +\! 1.393*0 + 0.300*1 + 2.213*0 + 0.579*0} \right)}} \\
= 0.18
\end{array}\end{equation}
Now, changing the status of Subject Animacy to ‘inanimate’ and that of Modal Verb to ‘present’ (two variables for which we found a large impact effect), we find that the probability shifts to 93% (see equation (iv)). Thus, according to our model the use of the s-passive is very likely, given the specific combination of factors.
Probability of s-passive
\begin{equation}
\begin{array}{l}
= \displaystyle\frac{{\exp \left( { - 1.789*1 + 1.9147*1 + 1.393*0 + 0.300*1 + 2.213*1 + 0.579*1} \right)}}{{1\! +\! \exp \left( { - 1.789*1\! +\! 1.9147*1\! +\! 1.393*0 + 0.300*1 + 2.213*1 + 0.579*1} \right)}} \\
= 0.93
\end{array}\end{equation}
If we also change Aktionsart to ‘ongoing’, then the probability slightly lowers to 91% (see equation (v)), which reflects the low impact effect associated with Aktionsart.
Probability of s-passive
\begin{equation}
\begin{array}{l}
= \displaystyle\frac{{\exp \left( { - 1.789*1 + 1.9147*1 + 1.393*0 + 0.300*0 + 2.213*1 + 0.579*1} \right)}}{{1\! +\! \exp \left( { - 1.789*1\! +\! 1.9147*1\! +\! 1.393*0\! +\! 0.300*0\! +\! 2.213*1\! +\! 0.579*1} \right)}} \\
= 0.91
\end{array}\end{equation}
The next section discusses our findings in comparison with the previous research on the passive alternation.
5. DISCUSSION
The present study of three verbs acceptera ‘accept’, behandla ‘treat’ and välja ‘choose’ that alternate between the s-passive and the bli-passive investigates the simultaneous effects of six factors as well as the relative strength of each factor separately.
The results show that the strongest effect is obtained for the factor Modal Verb. This finding corroborates earlier studies on the passive alternation. Engdahl (Reference Engdahl1999), for example, assumed that particularly the modal/temporal auxiliaries kan, ‘can’ and ska, ‘will’ are preferably associated with the s-passive. Since our corpus not only contains the modals mentioned, but also includes other modal/temporal auxiliaries as måste ‘must, have to’ in (31), we can conclude that the presence of any modal/temporal auxiliary motivates the use of an s-passive.
(31) Det är en funktionell del av gemenskapen och den måste behandlas som en sådan.
‘It is a functional part of the community and it must be treated as such.’
(Europarl svenska)
Engdahl (Reference Engdahl1999:19) and Laanemets (Reference Laanemets2012:170) note, however, that the modal vill ‘want to’ may be an exception, as it seems to be more common to use vill with the bli-passive. Our data confirms this: of the 38 bli-passive sentences with modal verbs, 13 include vill, whereas of the 236 s-passive sentences with modal verbs, only one involves vill (32).
(32) Om EU vill accepteras av sina invånare måste de uppleva att vi arbetar för dem.
‘If the EU wants to be accepted by its citizens they need to experience that we work for them.’
(Europarl svenska)
It would of course be worthwhile to investigate the influence of the different modal verbs in closer detail, but since this was the first investigation using a probabilistic approach, we limited the factor Modal to the distinction present vs. absent, leaving more fine-grained distinctions for further research.
Other strong effects were obtained for the factors Animacy and Av-phrase. The finding for Animacy is in line with the literature, which has repeatedly stated that inanimate subjects tend to occur with the s-passive whereas animate subjects appear to prefer the bli-passive. With respect to the factor Av-phrase, our results first of all corroborate the observation that the use of an av-phrase is rare; only 9% of our observations include an Av-phrase. We further found that the use of an av-phrase is associated with a preference for the s-passive. This contradicts Laanemets’ (Reference Laanemets2012) observation that the av-phrase is associated with a preference for the bli-passive. As explained in our discussion of the variables in Section 2, we believe that the preference for the s-passive may be explained by the fact that the focus shifts from the subject to the ‘by’-agent, which is thus attributed more control of the situation compared to the subject, as is illustrated in (33).
(33) Presidenten väljs av parlamentet och utser sedan en statsminister vars regering måste godkännas genom en omröstning i Saeiman.
‘The president is elected by the parliament and then appoints a prime minister whose government must be approved through a re-election in Saeiman.’
(Bloggmix)
Regarding agents, it can furthermore be noted that the ‘by’-agent may be expressed in alternative ways, for instance through using adverbials, as inom alla EU-institutioner and i parlamentet in (34). However, we did not label such instances as av-phrase. This use of alternatives with regard to expressing the ‘by’-agent is something future studies could examine in more detail.
(34) Jag hoppas verkligen att denna strävan blir allmänt accepterad inom alla EU-institutioner, även här i parlamentet.
‘I truly hope that this pursuit is generally accepted in all EU-institutions, even here in parliament.’
(Europarl svenska)
The remaining factors that yielded an effect were Subject Number and Aktionsart. Our analysis suggests that their effect is relatively low. With regard to Subject Number the use of the s-passive is found to be associated with plural subjects. Possibly, the use of plurals coincides with the expression of general events or general norms (see SAG IV:328–329), as in (35), which squares with the finding that the s-passive is associated with a general event or norm (Engdahl Reference Engdahl1999, Reference Engdahl, Lyngfelt and Solstad2006).
(35) I övriga trafikslag har dessa risker kunnat minimeras genom ett systematiskt säkerhetsarbete och tydliga regler där inga överträdelser accepteras.
‘In other types of transport these risks have been minimalized through systematic security work and clear rules where no violations are accepted.’
(Finlandssvenska texter)
Our results for the factor Aktionsart corroborate Engdahl's (Reference Engdahl, Lyngfelt and Solstad2006) observation that the influence of Aktionsart is very weak. This finding is in line with the observations of the Danish passive use discussed by Heltoft & Falster Jakobsen (Reference Heltoft, Jakobsen, Engberg-Pedersen, Fortescue, Harder, Heltoft and Jakobsen1996). However, in contrast to the expected pattern, our results suggest that the s-passive is associated with a completed event, rather than with an ongoing event. In (36), for example, we had expected the use of an s-passive, because the sentence expresses an ongoing event. The same applies to (34) above, where an ongoing event, which is particularly reinforced by the use of allmänt ‘generally’, is used together with a bli-passive. At this stage, we have no explanation for this finding.
(36) Men – är det inte BRA om det blir accepterat även för killar att vara lucia, och att ha vilka kläder man vill?
‘But – is it not GOOD if it is accepted that even boys are Lucia, and that they dress the way they want?’
(Bloggmix)
Finally, Subject Person is the only factor that did not appear to be significant. We related this factor to the notion of control − a local subject would reflect a higher level of control and accordingly prefer a bli-passive. No evidence was found for this influence.
6. CONCLUSIONS AND FUTURE PROSPECTS
In the context of the special issue, we have taken a corpus-linguistic and probabilistic approach to examine a contemporary research question in Nordic linguistics. Previous research on the Swedish passive alternation provided us with several factors that potentially motivate the choice of either one of two passive forms. A statistical evaluation of the effects of the proposed factors had hitherto not been conducted. Aiming at filling that gap, the present study performed a statistical analysis on the various factors that can influence the choice of the morphological or periphrastic passive starting from a data set of three verbs that can alternate. Our study thus adds to earlier research in that it allows for a more accurate analysis of the simultaneous effect and relative strength of each factor on the speaker's choice of one of the passive forms.
Our study only included three verbs, which may raise questions about the generalizability of our findings. Taking into account the semantic diversity of the three verbs in the study, we could conjecture that the findings are applicable more generally and more widely than the study itself. This is merely speculative, however, and more research is required.
Two directions of future research include examining a larger corpus sample and testing the statistical model(s) with a psycholinguistic experiment. A larger corpus sample should include a larger number of verbs, which will shed more light on possible preferential differences between the verbs. With a larger corpus sample one could also retain the proportional difference between the two passive forms. As mentioned, the fact that we used a case-controlled design overestimates the probability of an s-passive.
In addition to a larger corpus study, one could apply the ‘100-split task’, a psycholinguistic test originally developed by Bresnan (Reference Bresnan, Featherston and Sternefeld2007) in the context of the English dative alternation. The 100-split task is a grammaticality judgment experiment that aims to capture the probabilistic preferences of speakers. In this test, participants are asked to ‘rate the naturalness of alternative forms as continuations of a context by distributing 100 points between the alternatives’ (Ford & Bresnan Reference Ford, Bresnan, Krug and Schlüter2013:300). The average score for each sentence is then compared to the predicted probabilities estimated by a statistical model. Earlier tests on the dative alternation show a remarkable correlation between the scores of the participants and the statistical models, which suggests that probabilistic models of language variation accurately capture the intuitions of the speakers. Together with the results of an observational corpus study, experimental research provides strong converging evidence for the probabilistic approach to syntactic variability.
ACKNOWLEDGEMENTS
The authors wish to thank two anonymous Nordic Journal of Linguistics reviewers for their valuable comments and suggestions and Sara Van Herreweghe for her help with the data retrieval. Gudrun Rawoens also expresses her gratitude to the foundation Petra och Karl Erik Hedborgs stiftelse for a research stay at Stockholm University in 2014.
