Indirect evidentiality in Dutch and German : a contrastive corpus study of the seem-type verbs schijnen and scheinen

The present study deals with the seem-type verbs schijnen and scheinen in Dutch and German. On the basis of an in-depth analysis of spoken and written corpus material, the construction types these verbs typically appear in as well as their function and meaning are analysed. As seem-type verbs often develop into evidential markers (this is the case in e.g. English, French and Spanish), I will particularly concentrate on evidential uses (and the syntactic patterns that are associated with those uses). The study will lay bare important differences between German, Belgian Dutch and Netherlandic Dutch regarding both verbs. Moreover, the distinction between spoken and written language will be shown to play a crucial role with respect to the construction types found. Finally, the fact that the verbs exhibit different constructional preferences will be linked to different semantic properties as well.

1 Introduction: German scheinen vs. Dutch schijnen Both German and Dutch have a verb whose original semantics has to do with 'shining', i.e. giving light, which has developed into an evidential 1 marker with meanings more or less comparable to those of English seem (e.g.Aijmer 2009), i.e. the speaker using 1 Of course, evidential markers coding the source of information or knowledge have a strong bearing on the marking of epistemic authority, stance and validity (Mushin 2013).In this paper, however, I will concentrate on the evidential semantics of the seem-verbs.This focus does not imply a dismissal of their epistemic contribution nor of their crucial functions in actual discourse.
evidential schijnen/scheinen indicates there is indirect evidence for the proposition that is asserted.
(1) Het schijnt dat ze ziek is. it seem.prs.3sg that she ill is 'It seems that she is ill' (2) Sie scheint krank zu sein.she seem.prs.3sgill to be.inf 'She seems to be ill' Just like English seem, both verbs pattern in a wide variety of construction types: they are not only found in combination with an infinitive preceded by te/zu, but can also function as copular verbs, in adverbial and parenthetical constructions or as matrix verbs in impersonal constructions followed by a complementizer (Dutch dat, German dass 'that' or German als (ob) 'as if').Although these superficial observations make it tempting to equate German scheinen with Dutch schijnen, a number of previous studies (especially Vliegen 2011a-b;Van Bogaert & Colleman 2013;Van Bogaert & Leuschner 2015) have revealed remarkable differences that preclude such an equation.The main differences between the two verbs pertain to their preferences for particular constructional patterns, the specifics of their evidential meaning and their overall frequency, details of which will be presented in section 2.
Apart from the general differences between Dutch schijnen and German scheinen, Van Bogaert & Colleman (2013) have shown that Dutch schijnen is used differently in Netherlandic Dutch than in Belgian Dutch.Van Bogaert & Colleman's study focuses on the hearsay particle ('t) schijnt, which is said to be a typical feature of spoken Belgian Dutch.Other differences between Belgian and Netherlandic Dutch are only hinted at in their paper but are not given a systematic treatment.Another goal of this paper is therefore to bring out the differences between Belgian and Netherlandic Dutch schijnen.
Finally, the findings of Van Bogaert & Colleman (2013) also point to the relevance of register and mode (i.e. the distinction between speech and writing) as a possible parameter influencing the use of Dutch schijnen, as the particle ('t) schijnt is said to mainly occur in informal spoken (but hardly in written) language.By the same token, Cornillie's (2007) study of the Spanish seem-type verb parecer shows that it behaves very differently in spoken than in written language.It can indeed be hypothesized that an evidential seemtype verb develops other, more interactional uses in spoken language (e.g.serving the expression of the speaker's point of view and stance) than in written language (in which such interactional uses are less natural).I will therefore also compare spoken to written data (not only for Dutch, but also for German) in order to find out whether and how register and mode influence the use of German scheinen and Dutch schijnen.
Thus, the main empirical aims of this paper are threefold.First, I want to bring out the main general differences between German scheinen and Dutch schijnen, whereby I will focus on differences in semantics and constructional variability.Second, I will zoom in on the differences between Belgian Dutch schijnen and Netherlandic Dutch schijnen.And third, I will try to establish to what extent register and mode (more specifically, the difference between spoken, more colloquial and written, more formal language) licenses the use of particular construction types.
The empirical basis of this study is provided by corpus data, taken from both written and spoken corpora, on the one hand, and the results of a number of earlier studies (mainly Vliegen 2011a-b; Van Bogaert & Colleman 2013;Van Bogaert & Leuschner 2015 for Dutch and Askedal 1998;Diewald 2000Diewald , 2001;;Diewald & Smirnova 2010 for German) on the other.The remainder of the article is structured as follows.In section 2, I will discuss the results of previous studies and introduce the categories that are relevant for this study.In section 3, I will introduce the corpora that build the empirical basis for the study, the results of which will be presented in section 4. Finally, section 5 will provide a short summary and hint at explanations for the various findings.

State of the art
Previous studies (especially Vliegen 2011a, a corpus study based on a Netherlandic Dutch and a German newspaper corpus) have shown that German scheinen and Dutch schijnen differ with respect to their overall frequency, their semantics and their constructional preferences.These three aspects will be discussed in the following sections.

2.1
The frequency of German scheinen vs. Dutch schijnen Vliegen (2011a, 239) notes that Dutch schijnen is considerably less frequent than its German counterpart.Vliegen's statement is based on the absolute frequency of both verbs in a Netherlandic Dutch and German newspaper corpus: to find a more or less similar number of tokens (995 occurrences of schijnen, 724 occurrences of scheinen), Vliegen had to search a full year's edition of the Dutch newspaper NRC and only a month of the German newspaper Frankfurter Rundschau.Automatic frequency counts in the German 16 billion word corpus deTenTen and the 2 billion word Dutch counterpart nlTenTen (via https://the.sketchengine.co.uk) confirm the frequency differences, though: whereas German scheinen has a frequency ratio of 199.55 per million words, Dutch schijnen occurs only 30.63 times per million words. 2 2 It should be noted that these figures do not provide any information regarding regional variation nor with respect to register or mode.These will be dealt with in the main part of this study.

The semantics of German scheinen vs. Dutch schijnen
Undoubtedly, both verbs can have evidential meaning, i.e. they can express that the speaker (writer) has particular evidence for the asserted proposition, in the sense of Boye (2010Boye ( , 2012)), who speaks of 'epistemic justification'.Important in Boye's account is the fact that evidentials (just like epistemic verbs) semantically scope over propositions, i.e. over meaning units with a truth value, which means they can be epistemically evaluated.
With regard to the evidence, a general distinction can be made between 'direct' and 'indirect' types of evidence, referring to the (non)-immediate accessibility of the evidence (see e.g.Willett 1988).With direct evidence, the speaker typically has direct sensory access to evidence that is indicative of a particular proposition (e.g. the speaker sees that John is playing football and says: "I see that John is playing football"), whereas with indirect evidence, the asserted proposition is either based on inference or mediated information (hearsay) (e.g. the speaker sees that both the football and John are gone and infers: "John must be playing football").There seems to be a consensus in the literature that seem-type verbs, when used evidentially, typically indicate that there is indirect evidence -either in the form of inference or hearsay.
However, a number of issues blur this picture.First of all, it is not altogether clear in which construction types seem-type verbs possess evidential meaning.That the verbs are evidential when used in combination with a to-infinitive, seems to be beyond doubt (Cornillie 2007;Aijmer 2009;Diewald & Smirnova 2010 all discuss evidential uses of their (Spanish, English, German) seem-type verbs in this syntactic environment). 3hether the same holds for copular seem, for seem in matrix-verbs constructions (e.g.German es scheint, dass; Dutch het schijnt dat 'it seems that') or in adverbials (e.g.German wie es scheint, Dutch naar het schijnt 'as it seems') is much more controversial.Cornillie (2007, 16f.), for instance, does not pay any attention to the evidential potential of copular parecer (which accounts for more than 30% of all occurrences in his data sample), whereas Aijmer (2009, 82) clearly regards copular seem as potentially evidential, as it can express "inference on the basis of perceptual evidence".Diewald & Smirnova (2010) concentrate on the scheinen zu + infinitive construction and hardly take into account that scheinen can have evidential meaning in other construction types.As a copular verb, in the complement construction and when used parenthetically, scheinen is said to convey the meaning of "(visual) impression, to be seemingly, to appear (as)" (Diewald & Smirnova 2010, 180).In this paper, however, I side with Whitt (2015Whitt ( , 2017) and take the evidential potential of all constructional environments into account, especially those in which the seem-type verb clearly has propositional scope. 4  A second issue adding to the complexity of a semantic characterization of seem-type verbs is the fact that they can either express inference, hearsay or can be vague with respect to these values.Very often, only the wider discourse context can help decide which meaning is intended.For German scheinen, it seems to be generally accepted that the verb only expresses inference (e.g.Diewald & Smirnova 2010), whereas Dutch schijnen allows both inferential and hearsay interpretations, whereby the inferential meaning seems to be typical of "more formal and written styles" (Van Bogaert & Colleman 2013, 495). 5According to Wiemer (2010, 115), the hearsay meaning has gradually developed out of the inferential reading, which is said to be a development with cross-linguistic validity.Classification difficulties especially arise in cases in which an inference is based on hearsay, on the one hand, and in cases in which direct perception is at stake, on the other.Consider example (3), taken from the spoken German FOLK-corpus (see section 3 for more detailed information on this corpus).One could either argue that scheint expresses that the speaker infers that the father has a girlfriend (on the basis of what the son Gabriel has said) or that German scheint in this example is in fact reportive, as no real inferencing is involved -since the son has explicitly spoken about the girlfriend.
( As a copula, scheinen can invite an inferential reading; in this case, propositional scope is semantically coerced (see Whitt 2015Whitt , 2017;;Boye 2012).5 Koring (2012, 876) restricts the evidential meaning of Dutch schijnen to its reportive function, as it indicates that "the speaker has only indirect evidence for the embedded proposition in the form of hearsay evidence".On the basis of a corpus study of spoken Belgian Dutch material only, Van Bogaert & Colleman (2013, 294) conclude that "[s]chijnen is chiefly used as a hearsay evidential, indicating that the speaker is reporting what s/he has heard from others".
The following example illustrates the 'direct perception' use of German scheinen, which Diewald & Smirnova (2010) Diewald & Smirnova 2010, 190) In my view, however, it is not so much the fact that visual perception is involved, but rather the fact that the dependent infinitival clause designates a state of affairs rather than a proposition that renders this use of scheinen non-evidential: the speaker describes a particular impression that is not construed as referring to something in reality.In this impression reading of scheinen, epistemic justification is not a relevant issue, hence this use is not evidential.Note that such impression uses easily lead to non-factive, only-appearance readings when the context makes clear that a given impression does not correspond to reality, as in the following example (5), here in combination with a to-infinitive. 6  (5) Eine Note that such make-believe readings are also explicitly discussed by Diewald & Smirnova (2010, 184).They are regarded as a persistence effect of the original appearance meaning of the verb.
That visual evidence can give rise to an inferential interpretation which the speaker seems to accept as true, is shown by the following example: On the basis of what s/he observes (and this is presented as evidence in the following sentence), the speaker infers that Reus is not really upset by the breakup of his relationship.
( And finally, the explicit presence of a dative experiencer thwarts the evidential potential of a seem-type verb.According to Diewald & Smirnova (2010, 184), uses of scheinen with a dative experiencer cannot be labelled 'evidential', since the verb denotes "the perspective (often the visual perspective) of an observer who is part of the described scene".For the comparable Spanish me parece que 'it seems to me' construction, Cornillie (2007, 38) advocates a similar analysis, whereby the construction is said "not [to] rely on evidence but [to have] a subjectifier function" and as such "renders the conceptualizer's opinion based on pure knowledge or on inference" (see also Nuyts 2004).In example (8), therefore, scheinen cannot be labelled evidential as it combines with a dative experiencer mir.Vliegen (2011b) points out that German scheinen easily combines with a dative experiencer in almost every construction type, even in combination with an infinitive, a possibility that is more or less ruled out for the Dutch construction.In Dutch, only the copular construction still allows the presence of an experiencer (see also section 4.1).
Let's take stock: seem-type verbs can be used both non-evidentially and evidentially.In the latter use, they semantically scope over a proposition and have inferential (both German scheinen and Dutch schijnen) or reportive (only Dutch schijnen) meanings or are vague.The obligatory propositional scope of evidentials is the reason why widescope uses of the seem-type verb, i.e. as auxiliaries, in complement constructions or as parentheticals lend themselves naturally to an evidential interpretation.Cases in which the seem-type verb invites a non-factive, 'only appearance' reading or in which it is accompanied by an experiencer are categorized as non-evidential.

Constructional preferences of German scheinen vs. Dutch schijnen
It is well known that seem-type verbs appear in a wide array of constructional patterns.Askedal 1998, for instance, discusses no less than 15 different so-called Satzmuster 'clause patterns' in which scheinen occurs.In section 2.3.1,I will first lay out these patterns, thereby limiting myself to seven main types (based on proposals in Askedal 1998;Diewald 2000Diewald , 2001;;Diewald & Smirnova 2010 for German and Vliegen 2011ab; Van Bogaert & Colleman 2013 for Dutch).In section 2.3.2,then, I will briefly sketch the main differences between schijnen and scheinen with respect to their constructional preferences.

A common constructional inventory for scheinen and schijnen
The seven constructional patterns associated with scheinen/schijnen are the following: (i) scheinen/schijnen as a main verb (ii) scheinen/schijnen as a copular verb (iii) + INF=to be (iv) + INF (v) scheinen/schijnen as a matrix verb in complement constructions (vi) in adverbial/parenthetical constructions (vii) particle use of (ʼt) schijnt/scheints The types (i) and (ii) are the most simple constructional patterns.(i) is the original main verb use of scheinen/schijnen with the meaning 'to give light, to shine' (example 9); (ii) refers to the copular use of both verbs, whereby they combine with nominal, adjectival, prepositional or participial complements (examples 10 and 11).( 9 Although Diewald & Smirnova (2010) regard types (iii) and (iv) as subtypes of the same construction, I will (at least initially) keep them apart, as the INF=to be construction in (iii) is argued to have developed out of the copular construction, to which it still bears a clear semantic affinity (Diewald & Smirnova 2010, 179); for this reason, Diewald & Smirnova label the construction type in (iii) the compound copula construction.It is a kind of bridging construction between the less grammaticalized copular use in (ii) and the full infinitival construction in (iv): "it served as a kind of catalyst triggering the development of a more general constructional pattern scheinen & zu-infinitive with all types of infinitives" (Diewald & Smirnova 2010, 260).
The construction types (v) to (vii) are on the whole less frequent than the infinitival construction (see for instance Askedal 1998, 54), which is probably also the reason why they are not treated in much detail in Diewald & Smirnova (2010).In type (v), the seemtype verb functions as the matrix verb in a complement construction introduced by the 'factive' complementizer dass/dat 'that', the comparative complementizer als (ob) 'as if' (only possible in German) or a zero complementizer (again only possible in German).Whereas Diewald & Smirnova (2010) do not consider the evidential potential of this construction type, Whitt (2015Whitt ( , 2017) ) argues that complement clause constructions are equally capable of expressing evidential meaning as zu-infinitive constructions (see also De Haan 2007, 141 for a similar claim).In fact, with a 'factive' that-complement, the seem-type verb can easily be interpreted as scoping over a proposition, so that it can develop into an evidential marker expressing that the proposition in its scope is either inferred or reported.And indeed, the Dutch complement construction het schijnt dat naturally invites an evidential (hearsay) interpretation (see section 4.3).An evidential (inferential) analysis could also be proposed for the following German example involving es scheint, dass.Note, however, that this is not the default interpretation of the German complement construction, as will become clear in section 4.3.What unites adverbials and parentheticals furthermore is that they typically evoke a wide-scope reading, i.e. they pertain to the entire proposition.So, in ( 22) and ( 23) adverbial naar het schijnt 'as it seems' and parenthetical schijnt het 'it seems' have the entire proposition (that she has told everything or that in America costs are enormous) in their scope.Finally, type (viii) includes the strongly grammaticalized particle uses of ('t) schijnt and scheints.As a particle, (typically Belgian) Dutch ('t) schijnt is syntactically and prosodically integrated in the clause, in contrast to its parenthetical 'counterpart' (zo) schijnt het.This also holds for German scheints, which arose via a different grammaticalization path than its Belgian Dutch counterpart and has not acquired the same degree of formal grammaticalization as Dutch ('t) schijnt.Other differences between particle and parenthetical use have been described by Van Bogaert & Colleman (2013) and Van Bogaert & Leuschner (2015).

Different preferences: state of the art and some open questions
In spite of the common constructional inventory, previous analyses have shown that Dutch schijnen differs from its German counterpart with respect to its preference for particular construction types.In fact, at least four differences have been described so far.First, the copular use is found to be much more popular in German (20.7% of all scheinen occurrences in Vliegen's (2011a) sample are copular) than in Dutch (copular constructions account for only 6.1% of all occurrences of schijnen, Vliegen 2011a, 239f.).A second difference pertains to the frequency and form of the complement construction.According to Vliegen (2011a, 239f.), the infinitival construction is the one with the highest frequency in both German and Dutch.Complement constructions are infrequent in Vliegen's data, especially in German (0.7%), but also in Dutch (3.8%).However, in Van Bogaert & Colleman's (2013) corpus study based on spoken Belgian Dutch, the complement construction accounts for 33.8% of all occurrences of schijnen, and even tops the infinitival construction in terms of frequency (Van Bogaert & Colleman 2013, 493).So, at least in Belgian colloquial Dutch, the complement construction is a much more important construction type than the one in German.It remains to be seen, then, to what extent the complement construction occurs in spoken Netherlandic Dutch, on the one hand, and written Belgian Dutch, on the other.Note also that German complement constructions can be introduced by 'comparative' als ob 'as if'' and by zero, two options that are ruled out in present-day Dutch.A third difference pertains to the particle use of ('t) schijnt and scheints, whereby the latter has not only arisen via a different developmental path, but is also considerably less grammaticalized than its Belgian Dutch counterpart (Van Bogaert & Leuschner 2015, 113).Finally, attention has been drawn to the fact that German scheinen licenses uses with a dative experiencer in almost every constructional pattern, whereas this option is severely restricted for Dutch schijnen (Vliegen 2011a, 240).All these differences clearly point to a higher degree of formal grammaticalization of Dutch schijnen as compared to its German counterpart: Dutch schijnen generally evokes a wide-scope reading pertaining to a proposition, hardly allows the presence of a dative argument (with an experiencer role) and is more strongly grammaticalized in its particle use as well.To what extent regional differences, register and mode influence these general constructional preferences will be discussed in the following sections.

Resources and methods
The empirical basis of this contrastive analysis is made up of a number of corpora.One of the general problems I faced was to find more or less comparable language material for both written and spoken (Belgian and Netherlandic) Dutch and German, with a focus on standard language.
For Dutch, the CONDIV-corpus (for the written part of this study) and the Corpus of Spoken Dutch (Corpus Gesproken Nederlands or CGN) were used.The CONDIVcorpus is an electronically available 47 million word corpus that has the advantage that it incorporates both Belgian and Netherlandic Dutch material to a more or less equal extent. 7The register ranges from very informal to highly formal written language.For the purpose of the present study, I only included the newspaper material (which is published in 1998).It should be noted that this part of the corpus is somewhat unbalanced given that the Belgian newspaper corpus (12 796 556 words) contains almost three times as many words as the Dutch part (4 791 281).One of the major drawbacks of the CONDIV corpus is the fact that it is not lemmatized, so that I had to search for the individual word forms of schijnen: 1 st person present tense singular schijn, 2 nd and 3 rd person singular present tense schijnt, plural present tense and infinitive schijnen, singular and plural past tense forms scheen and schenen and the past participle geschenen.For the Netherlandic Dutch part, 146 forms of schijnen could be collected (frequency ratio: 30.4 per million words), whereas the Belgian Dutch part contained 292 forms of schijnen (frequency ratio: 22.7 per million words).

7
For more information on the CONDIV-corpus, see Grondelaers et al. 2000.
The Corpus of Spoken Dutch (CGN) consists of about nine million words collected between 1998 and 2004 and comprises a Belgian (CGN-VL) and a Netherlandic Dutch (CGN-NL) component.The corpus is regionally unbalanced since the Belgian Dutch data account for only about a third of the data. 8The corpus is lemmatised, so that it can easily be searched for all occurrences of schijnen in general, i.e. in all tenses and person forms.In absolute terms, schijnen occurred 502 times in the Netherlandic Dutch part of the CGN and 399 times in the Belgian Dutch part.The relative frequency of Dutch schijnen is therefore clearly higher in spoken (with values between 88 per million words for Netherlandic Dutch and 122 per million words for Belgian Dutch) than in written language.For this study, I only included occurrences of schijnen in the a and d components of the CGN, consisting of spontaneous (and hence informal) face-to-face interaction (component a) and telephone dialogues (component d), yielding 236 usable instances for Netherlandic Dutch schijnen and 186 instances for Belgian Dutch schijnen.A number of unclear occurrences and repetitions were manually omitted.
For the German written part of this study, I made use of three different corpora: the ZEIT-corpus provided by the DWDS (Digitales Wörterbuch der deutschen Sprache, see www.dwds.de)(as a representative of a high quality national newspaper) and two regional newspapers, the (Northern German) Hamburger Morgenpost (HMP) and the (Southern German) Nürnberger Zeitung (NZ), both part of the Deutsches Referenzkorpus (DeReKo) provided by the Institut für deutsche Sprache.For all three newspapers, only recent material published in 2013 was collected. 9For every subcorpus 75 random occurrences of scheinen (again in all tense and person forms) were taken into account.
Finally, finding comparable corpora for informal spoken German proved not to be easy.The Datenbank für gesprochenes Deutsch (DGD) provides a number of online corpora of spoken German, most of which are either older than the Dutch material (dating from the 1960s or 1970s, like the König-corpus or the Dialogstrukturencorpus) or contain rather specific language varieties (highly colloquial, dialectal etc.).For this reason, I only made use of the FOLK-corpus (Forschungs-und Lehrkorpus für gesprochenes Deutsch) containing spoken material recorded between 2005 and 2012.This corpus, however, yielded only 25 usable instances of scheinen.I added all instances of scheinen (in all mood, person and tense forms) from the Korpus Gesprochene Sprache on the DWDS-website (www.dwds.de), the majority of which stem from the famous German television talk show 'Literarisches Quartett' which was broadcast between 1988 and 8 The Belgian part contains 3261628 words, the Netherlandic Dutch part 5654644 words; see http://lands.let.ru.nl/cgn/doc_Dutch/topics/version_1.0/overview.htm).

9
As such, a 15 year time-lap between the Dutch written material and the German written material exists, but I take the material to reflect present-time usage.
2001.The majority of the German spoken data is therefore considerably less informal than the Dutch spoken data.
To summarize, table 1 provides an overview of the various corpora and the number of instances that were taken into account for the analysis (the first number between brackets).However, as I was not interested in the main verb uses of schijnen and scheinen, these were manually omitted as well.The figures in bold face refer to the actual number of tokens that were analysed in detail.
For each instance, the syntactic pattern the verb appeared in was analysed as well as its meaning.The results of the syntactic and semantic analysis will be presented in section 4.

Findings
Table 2 gives an overview of the distribution of the different constructional patterns the verbs occur in.Patterns occurring over 15% are highlighted.It is immediately clear that considerable differences exist between German and Dutch: whereas the copula construction is a dominant construction type in the two German samples (with frequency values over 30%), it is much more peripheral in the Dutch samples, especially in the spoken ones.Table 2: Distribution of construction types of scheinen and schijnen Table 2 also brings out the relevance of distinguishing between written and spoken language, on the one hand and between different varieties of Dutch, on the other.With respect to the former distinction, the distributional differences between the Belgian Dutch written sample and the spoken one are most striking: the infinitival construction is the most frequent construction type in the written Belgian Dutch sample (57.8%), but is quite marginal in the spoken one (6.5%), the opposite applies to the complement construction, which is peripheral in the written Belgian Dutch sample (6.5%), but appears to be the most frequent construction type in the spoken Belgian Dutch one (41%).By the same token, the frequency of the complement construction and the particle use of schijnt is clearly higher in the spoken Netherlandic Dutch sample (12.7% and 7.1% respectively) than in the written one (8.7% and 0% respectively).With respect to differences between the two Dutch varieties as such, the table shows that Belgian and Netherlandic Dutch clearly differ from one another regarding the presence of parenthetical/adverbial constructions, which are found clearly more often in Belgian Dutch (written: 17.3%; spoken: 24.3%) than in Netherlandic Dutch (written: 5.8%, spoken: 1%).
Table 2 also points to something else: the distributional differences between the to-INFconstruction and the INF=to be construction do not seem to be systematic.So, although German scheinen is still frequently used as a copula, it does not combine more often with INF=to be (i.e. as a so-called compound-copula) than with another infinitive, at least not in the written German sample.And mutatis mutandis, whereas Dutch schijnen hardly occurs as a copula, it is the INF=to be which is the most frequent construction type in the spoken Netherlandic Dutch sample.It thus seems reasonable to conflate both infinitival construction types into one construction, i.e. scheinen/schijnen + to-INF, which brings out the general dominance of the infinitival construction even more (again, with the exception of the Belgian Dutch spoken sample).This is done in figure 1, which visualizes the distributional patterns of table 2. As was noted earlier, uses of scheinen with a dative experiencer are not classified as evidential, but mainly function as hedges or stance markers, typically used when speakers express their personal opinion, as in ( 24), in which scheinen occurs in a parenthetical construction, and ( 25), where scheinen is the matrix verb in a complement construction with the complementizer als 'as if' (see also examples 8, 20 and 21 above, all taken from the spoken German corpus).
( In the following subsections, I will address each construction type in somewhat more detail.The copular use of scheinen is quite frequent in both spoken and written German.In Dutch, on the other hand, copular use of schijnen is still possible, but is clearly on the decline.This is also evidenced by the fact that copular use in Dutch occurs somewhat more often in the written sample -which is traditionally regarded as more conservative -than in the spoken one.The verb lijken 'be similar to, seem' seems to have taken over the copula function (Vliegen 2011a, 241).

Scheinen/schijnen as a copula
(26) maar de nieuwste methode schijnt beloftevol, ik but the newest method seem.prs.3sgpromising I verwacht er véél van expect it a lot of 'But the newest method seems promising, I expect a lot of it' (CONDIV-VL) When used without an experiencer, the copular use of scheinen/schijnen can often be argued to be inferential, i.e. the verb can be interpreted as semantically scoping over a proposition that has been inferred by the speaker.( 27 Especially in past tense uses, 10 however, copular scheinen easily invites non-factive readings, as in the following example: although it seemed certain that the president was going to be voted out of office, he eventually wasn't.It should also be noted that in a clear majority of the copular uses in the spoken German corpus (51/66), a dative experiencer was present, in which case the copula construction cannot be classified as evidential either.

scheinen/schijnen +to-INF
As visualized in figure 4, the infinitival construction is the central construction type in every sample, with the exception of the Belgian Dutch spoken sample, in which the complement construction is the most frequent one.Note that the values for to-INF in Netherlandic Dutch are overall clearly higher (about 78%) than in German (between 52 and 59%).Semantically, one can distinguish between German and Dutch as German scheinen in the to-INF construction generally expresses inferential evidentiality, whereas Dutch schijnen in the to-INF construction can express both inferential and reportive evidentiality.In most cases, these uses can be kept apart, as the context makes clear that the speaker either draws an inference herself or refers to a reportive source.Figure 5 shows that the reportive reading is the most frequent one in the spoken samples (S-NLDUT and S-BDUT); in the written samples (W-NLDUT and W-BDUT), on the other hand, the Fig. 5: Reportive vs. inferential meaning in the to-INF construction in NL/B Dutch inferential reading of schijnen is still more prominent.This ties in with the expectation that written language is more conservative so that the later reportive meanings of schijnen prevail in spoken language.
Examples ( 29) and (30) illustrate inferential uses of Dutch schijnen.In (29), schijnen is used in a weather forecast, whereas in (30), the speaker infers something on the basis of the parties' election programmes.
( In examples ( 31) and (32), schijnen indicates hearsay.In (31), the speaker has probably been informed that the lost package has been sent to Arnhem (it would be strange to infer concrete information like this), whereas in the most natural interpretation of (32), the speaker must have read or heard that two-headed pigs sometimes stay alive -against all odds.
(31 highly uncommon is 'It seems (i.e.I've heard/read) to occur more than once, but Rudy, the two-headed piglet is alive and stays alive, which is highly unusual' (CONDIV-VL)

scheinen/schijnen as matrix verb in a complement construction
The complement construction comes in three variants: introduced by the 'factive' complementizer dat/dass 'that', by the comparative complementizer als (ob) 'as if' or by zero. Figure 6 gives an overview of the relative distribution of the complement construction in the respective samples.
Complementation by means of dat is the most frequent constructional pattern in the Belgian Dutch spoken sample, in which it accounts for 41% of all occurrences.In general, the complement construction seems to be more typical of spoken than of written language: also in the Netherlandic Dutch and the German samples, the complement construction is found somewhat more often in the spoken (NLDUT 12.7%, GER 9.7%) than in the written samples (8.7%, GER 5.2%).34) and ( 35 or corpses of monsters perhaps even of people 'Nobody can be seen, there is no boat on the water, no fisher man on the side.It seems [i.e.It is said that] there are dead crocodiles lying on the bottom or corpses of monsters, perhaps even of people' (CONDIV-NL) A completely different picture emerges when we look at instances of the German complement construction.Apart from the fact that the pattern is considerably less frequent than in Dutch (German 32/427 (= 7.5%) vs. Dutch 117/656 (17.8%) and allows different complementizers (9 dass; 14 als (ob); 9 zero), the complement construction typically invites either an impression or make-believe reading (not only in combination with als (ob), but also with dass) or is used as a hedge in utterances expressing the speaker's personal opinion (in the latter case, typically in combination with a dative experiencer).Examples for make-believe readings are given in ( 36) and (37), the hedge 11 Only in the Netherlandic Dutch spoken sample there seems to be a small number of instances that invite an inferential or indifferent reading.Note that zero complementation (example 39) only occurs in the spoken sample and that 8 out of 9 zero complementation cases combine with a dative experiencer (as in 39).By the same token, 7 out of 8 dass occurrences in the spoken sample combine with a dative experiencer as well, again typically hedging the speaker's opinion.
Still, there are some instances in which the speaker does not so much compare a state of affairs to something else, but actually infers that a particular proposition holds, i.e. clearly refers to something in reality.In the following occurrence (introduced by comparative als) it can be argued that the speaker actually infers that the USA are less considerate with respect to financial criminals than Germany given the fact that Bernie Madoff was sentenced to 150 years of prison.
( A similar example involving dass-complementation was given in ( 16) above.It remains to be seen, however, whether and to what extent evidential (inferential) interpretations of scheinen in complement constructions become productive in German.

scheinen/schijnen in adverbial and parenthetical constructions
Adverbial and parenthetical construction types with scheinen/schijnen come in different forms including Dutch naar het schijnt and (zo) schijnt het, schijnt 't and German wie es scheint, will es scheinen, scheint mir, so scheint/schien es (mir) and wie mir scheint.They are infrequent in German and Netherlandic Dutch, but are often used in Belgian Dutch, occurring both in the written and spoken sample, as the following figure 7 makes clear.
In fact, it is the naar het schijnt construction that is responsible for the remarkable presence of this construction type in the Belgian Dutch sample: naar het schijnt accounts for about 90% of all instances that were classified as either adverbial or parenthetical (66/74).Belgian Dutch naar het schijnt 'as it seems' has evidential reportive meaning.I will not go into the other representatives of the adverbial/parenthetical category, as the number of instances is low.Suffice it to say that the majority of the German parenthetical instances combines with a dative experiencer and as such can be interpreted as hedges (see example (21) above).

The particle use of ('t) schijnt/scheints
The particle use is only found in the Belgian and (to a lesser extent) Netherlandic Dutch spoken samples.As this use has already been extensively discussed in the literature (see Van  The Netherlandic Dutch particle schijnt, on the other hand, has acquired a lesser degree of grammaticalization.This is evidenced, among other things, by the fact that it is considerably less frequent than its Belgian Dutch counterpart and can still occur in the past tense form, as in example (47).In fact, Van Bogaert & Colleman (2013) treat Netherlandic Dutch schijnt as a reduced form of parenthetical (zo) schijnt het rather than as a genuine particle, as it still mainly occurs in clause-final position.The fact that schijnt has arisen out of the parenthetical construction (zo) schijnt het also explains why schijnt always occurs without clitical ʼt.
( With respect to the German particle scheints Van Bogaert & Leuschner (2015, 115) argue that it is strongly associated with a particular register and genre: "southern, especially Austrian/Swiss, lowbrow journalese".It can express both inference and hearsay (Van Bogaert & Leuschner 2015, 111); as such, it seems to be one of the only uses of German scheinen compatible with a hearsay reading.

Conclusions
First of all, this paper has not only confirmed the differences with respect to the global constructional preferences that have been described for German scheinen and Dutch schijnen (section 2.3.2),but has also shown that the actual distribution of the constructional patterns is to some extent dependent on register and mode (i.e. the distinction writing vs. speech), on the one hand, and regional variation, on the other.Copular use of scheinen, for instance, is indeed much more prevalent in German than in Dutch, but copular uses are found to occur in written Dutch still considerably more often than in spoken Dutch (see fig. 3).Regional variation, on the other hand, does not crucially affect the distribution of the copular use of schijnen in Dutch.Another example: the occurrence of German scheinen with a dative experiencer is a typical feature of spoken German, whereas it hardly occurs in the written corpus (see fig. 2).This is only natural, given that the combination of self-reference (mir 'to me', the most frequent representative of the dative experiencer) and a seem-type verb lends itself very well to the expression of personal opinion, hedging, stance and the like, processes that are typical of spoken language.Or a third example: the distributional prevalence of the complement construction het schijnt dat 'it seems that' is a feature of Belgian spoken Dutch, i.e. it is dependent on both register and regional variation.This distribution is not repeated in the Belgian written sample nor in the Netherlandic Dutch spoken sample, in both of which the to-INF is the most frequent pattern (see table 2).A fourth example: the distribution of reportive vs. inferential readings in the Dutch to-INF construction is clearly dependent on the distinction speech vs. writing, as reportive readings are found to occur much more often in the spoken than in the (more conservative) written Dutch samples (see fig. 5).And finally, the adverbial naar het schijnt construction appears to be a regional Belgian Dutch feature, as it hardly occurs in the Netherlandic Dutch samples (see fig. 7).
A second issue concerns the possible functional motivation of these distributional differences.I would like to argue that it is no coincidence that in Belgian spoken Dutch (which is the variant in which schijnen acquires its highest frequency ratio, see section 3), schijnen 'prefers' to occur in typical wide-scope syntactic environments (complement het schijnt dat, adverbial naar het schijnt, particle ('t) schijnt), as these perfectly match the verb's development into a wide-scope hearsay evidential.In Netherlandic Dutch, schijnen mainly functions as an evidential that can have either inferential or reportive meaning, depending on the context.Typically reportive construction types (adverbial naar het schijnt, the complement construction het schijnt dat, the particle schijnt) occur, but with a much lower frequency than in Belgian Dutch.In German, then, scheinen can be considered an inferential evidential in only some of its uses.Wide-scope construction types (complement constructions, parenthetical constructions) are clearly more marginal in German than in Dutch.By contrast, non-evidential combinations with a dative experiencer (scheint mir), copular uses and non-factive/impression uses occur considerably more often than in Dutch.On the basis of these observations, a cline could be sketched whereby German scheinen is the least grammaticalized evidential, whereas (spoken) Belgian Dutch schijnen can be argued to have grammaticalized into a genuine hearsay evidential.Netherlandic Dutch schijnen, then, is somewhat in the middle.
A final question concerns the underlying reason for these differences in speed of development: why does Belgian Dutch schijnen stick out that much?It could be hypothesized that the presence of a dedicated reportive modal sollen in present-day German (Diewald 1999, 225ff.)renders the development of an alternative reportive marker scheinen somewhat superfluous.The Dutch reportive marker zou, on the other hand, is found to be used more frequently in Netherlandic Dutch than in Belgian Dutch (Timothy Colleman, pc), thus blocking to some extent the full development of the reportive function of schijnen in Netherlandic Dutch.It could further be hypothesised that language contact plays a role, in that Belgian Dutch is more strongly influenced by French (which has a reportive 'seem' marker in the form of the complement construction il parait que 'it seems that', see Thuillier 2004, 30) than Netherlandic Dutch.The remarkable dominance of the reportive complement construction het schijnt dat in spoken Belgian Dutch could thus be due to the influence of French.By the same token, Van Bogaert & Leuschner (2015) found most reportive uses of the German particle scheints in colloquial Southern German including Switzerland and Austria, which again -at least in the case of Switzerland -points to language contact with French as a possible explaining factor for the reportive use.More research, however, would be necessary in order to come up with a fully satisfying answer to this question.

Figure 3
Figure 3 visualizes the relative distribution of the copula construction in the different samples.
out of office of the conservative president seemed so certain' (ZEIT, ONLINE 2009/26/iran-familien_dwdsddc.xml) 10 Note that about 23% (i.e.16 out of 68 occurrences) of all copular uses of scheinen are in the past tense.

Fig. 4 :
Fig. 4: Relative (%) distribution of the to-INF construction in German and (NL/B) Dutch

)
Die Sonne scheint.In types (iii) and (iv), the seem-type verb functions as an auxiliary followed by infinitival te zijn/zu sein 'to be' (see examples 12 and 13) or by any other to-infinitive (illustrated by examples 14 and 15).But in the mean time globalisation continues and nothing seems to be able to stop it' (CONDIV-NL) 'The prisoners of war appeared to be in a hopeless situation: death in Chelm or killing in extermination camps' (ZEIT 2009/28/DOS-Demjanjuk_dwdsddc.xml2009) (13) De film Soldaat van Oranje, die heel goed schijnt te the movie SvO rel very good seem.prs.3sg to zijn be.inf 'The movie Soldaat van Oranje, which seems to be very good' (CGN-NL) It seems however that a lot of socialists have not yet realized this [i.e.their loss of support].Because they still act like giants' (HMP09/SEP.00003)When followed by a comparative complementizer (als ob), on the other hand, German scheinen typically does not scope over a proposition, but expresses that a particular state of affairs looks like, i.e. is similar to something else.Often, this impression contrasts with reality and as such blocks an evidential reading.It must be noted that present day Dutch schijnen does not allow the comparative complementizer alsof (and the associated impression reading) anymore.It almost seemed as if Porsche were able to print the money itself with which it bought VW.' (ZEIT 2009/22/Porsche_dwdsddc.xml) Construction type (vi) subsumes both adverbial and parenthetical uses of Dutch schijnen and German scheinen.Adverbial constructions (Dutch naar het schijnt, German wie es scheint) can in principle be distinguished from parenthetical ones in that the former trigger inversion in sentence-initial position (which is indicative of the fact that they are syntactically integrated in the clause and function as genuine first constituents), whereas parenthetical constructions are not.In (18) and (19), the finite verbs is and könnte occur in second position, thus indicating that adverbial naar 't schijnt and wie es scheint have first constituent status.As it seems a new man for the vacant manager position could be presented today.' (NUN13/JAN.00334)In non-initial position, however, it is sometimes difficult to keep them apart, as typical 'adverbial' constructions (e.g.German wie es/mir scheint) can behave like typical parenthenticals (e.g.German (so) scheint es/mir), for instance by being prosodically separated.The following examples illustrate this: 'adverbial' wie mir scheint is separated by commas (to indicate a pause) and takes up a position that could also be filled by 'parenthetical' so scheint es mir.

Table 1 .
Overview of the Belgian Dutch, Netherlandic Dutch and German corpora To me it seems all the time as if the Leubner-Saupe-model were completely clear to you' (FOLK E 00033 SE 01 T 01) In the meantime the postal inspection has stated that the package has never been offered to the printer's.De Vries: 'The package seems (i.e.so I've heard/read) to have been sent to Arnhem instead of Rotterdam'' (CONDIV-NL) rel hoogst ongebruikelijk is.
function is illustrated in (38) and (39), in which the speaker expresses some kind of disagreement with what has been previously said.
It seems as if people in the USA are less considerate towards financial criminals.The financial wizard Bernie Madoff was sentenced to an impressive 150 years of prison last week' (ZEIT 2009/29/Argument-Banker_dwdsddc.xml) It was superb it seemed (i.e.soI have heard/read)' (CGN-VL)An interesting observation pertains to the fact that Belgian Dutch naar het schijnt exclusively occurs in the present tense, which signals a high degree of idiomatization and fixation.By contrast, we find one past tense form among the five Netherlandic Dutch cases.Moreover, it is not clear whether Netherlandic Dutch naar het schijnt always has reportive meaning (in contrast to the Belgian Dutch cases that are all clearly reportive).The German counterpart wie es scheint occurs only twice, in both cases without reportive meaning (in example 43, wie es scheint seems to evoke an only-appearance interpretation).It might look as if Hemingway turns into someone like Ernst Jünger.But he wasn't like that.' (DWDS-LQ-1989-JB) Van Bogaert & Colleman (2013)ogaert & Leuschner 2015), I will not treat it in great detail here.Suffice it to say that the Belgian Dutch particle ('t) schijnt is a hearsay evidential with a high degree of formal grammaticalization.Van Bogaert & Colleman (2013)assume that it has arisen out of the impersonal complement construction het schijnt (dat) 'it seems'.The particle most often occurs with its original subject het 'it' (always reduced to 't), but occurrences without clitical 't are found as well.As was shown by Van Bogaert & Colleman, ('t) schijnt can occur in clause-final (44), medial (45) and even initial (46) position.
Loo do.inf seem.prs.3sg'but perhaps it won't take place because they are no longer allowed to do it in the Loo palace, it seems [i.e.I've heard/read]' (CGN-NL)