1. Introduction
Present-day Icelandic has one main form which functions as an expletive: það, formally identical to the 3sg.nt pronoun.Footnote 1 It is well known that expletive það is positionally restricted to the clause-initial prefinite position (e.g. Sells Reference Sells, Butt and King2005, Sigurðsson Reference Sigurðsson, Bhattachayra, Reuland and Spathas2007, Thráinsson Reference Thráinsson2007:312), as illustrated in (1).Footnote 2
(1)
The restricted positional distribution of expletive það is standardly interpreted as evidence that it does not qualify as a subject, since subjects in a Germanic verb-second language like Icelandic occur in the immediately postfinite position in contexts like (1b) (see Thráinsson Reference Thráinsson1979:480–481, Platzack Reference Platzack and Karlsson1983, Sigurðsson Reference Sigurðsson, Bhattachayra, Reuland and Spathas2007). Compare the expletive det in Swedish, which does behave like a subject, e.g. (2) (for details on the distribution of the Swedish expletive, see Falk Reference Falk1993 and Håkansson Reference Håkansson2017).
(2)
The clause-initial prefinite position in Icelandic is standardly assumed to be an information-structurally privileged position which is associated with topical constituents, henceforth topic position (e.g. Rögnvaldsson & Thráinsson Reference Rögnvaldsson, Thráinsson, Maling and Zaenen1990). As such, it has been claimed that the clause-initial expletive in examples like (1a) signals a topicless verb-second (V2) sentence (Zaenen Reference Zaenen1983, Rögnvaldsson & Thráinsson Reference Rögnvaldsson, Thráinsson, Maling and Zaenen1990, Sells Reference Sells, Butt and King2005). Það in such contexts is a topic position placeholder.Footnote 3
When one examines topicless impersonal sentences in Old Icelandic (1150-1350), expletive það is typically absent, rendering verb-initial (V1) structures, e.g. (3).
(3)
Previous studies claim that the expletive did not emerge in such contexts until c.1500 (Rögnvaldsson Reference Rögnvaldsson2002) and underwent a dramatic increase in frequency in the 19th century (Hróarsdóttir Reference Hróarsdóttir1998).
In other constructions – namely those with a clausal argument – það is however robustly attested in Old Icelandic (‘extraposition’, Faarlund Reference Faarlund1990, Rögnvaldsson Reference Rögnvaldsson2002). This applies both to constructions where the clausal argument is the subject of the matrix clause predicate, e.g. (4a), as well as those where the clausal argument is the object of the matrix clause predicate, e.g. (4b).
(4)
In contexts like (4), I assume that það has cataphoric reference to the clausal argument (henceforth cataphoric það) and is distinct from expletive það, which is neither cataphorically nor anaphorically referential.Footnote 5 Another property which distinguishes expletive það from cataphoric það is the fact that the expletive is indeclinable, whereas cataphoric það inflects for case. The example in (5) features the predicate geta ‘guess’, which takes a genitive argument as its object; cataphoric það has genitive case marking (þess).
(5)
In this paper, I present an account for the diachronic development whereby það spread from cataphoric contexts like (4) to contexts like (3) which lack a clausal argument. In this account, the early attested cataphoric það plays a role in the later emergence of expletive (non-cataphoric) það. The structure of the paper is as follows: Section 2 outlines the methodology of the corpus study. Section 3 presents findings for the status of cataphoric það in various constructions types in the history of Icelandic and discusses a structurally ambiguous construction, using Lexical-Functional Grammar (LFG) for the formal analysis. Section 4 presents an account for the development whereby expletive það emerged on the model of earlier cataphoric contexts and compares this with previous findings for the rise of það in presentational constructions (Booth Reference Booth2018). Section 5 concludes the paper.
2. Methodology
The basis for this study is data from the Icelandic Parsed Historical Corpus (‘IcePaHC’, Wallenberg et al. Reference Wallenberg, Ingason, Sigurðsson and Rögnvaldsson2011). IcePaHC contains approximately 1,000,000 words, from 61 text extracts spanning 10 centuries (1150-2008), thereby covering all attested stages of Icelandic. IcePaHC thus allows one to examine change across the centuries which many studies focusing on data from specific periods do not capture (e.g. Hróarsdóttir Reference Hróarsdóttir1998, Rögnvaldsson Reference Rögnvaldsson2002).
The IcePaHC annotation follows the Penn treebank format established for historical English (e.g. Kroch & Taylor Reference Kroch and Taylor2000, Santorini Reference Santorini2010) and is compatible with the CorpusSearch query language (Randall Reference Randall2005). All content is lemmatised, part-of-speech tagged and annotated for constituent structure, with additional tagging for certain grammatical functions (e.g. subject, object). For further information on IcePaHC, see Rögnvaldsson et al. (Reference Rögnvaldsson, Sigurðsson, Ingason and Wallenberg2012). The corpus does have some limitations: the texts included represent a very small sample of attested historical Icelandic, and certain genres are over-represented and others under-represented. Moreover, the ‘narrative’ genre which is dominant in the corpus comprises mainly saga-style texts up to the 19th century but modern novels after that, and thus cannot be considered a homogenous category across time. Some texts are also based on source texts in other languages. Nevertheless, the advantages offered by the syntactic annotation outweigh these issues, and as long as one keeps these limitations in mind, IcePaHC is a valuable source of data.
Both overt and ‘null’ expletives – constructions where expletive það could be expected but is not attested, e.g. (3) – are distinctively tagged in IcePaHC. For sake of time, I do not look beyond the tagged instances for overt or ‘null’ expletives which are not tagged as such in the corpus. However, a parallel manual investigation of expletives in Old Icelandic presented in Booth (Reference Booth2018) yielded comparable results to the IcePaHC findings for early texts, suggesting the IcePaHC tagging of expletives is reasonably reliable. Combining the overt and ‘null’ expletives gives a dataset which represents the total contexts in which the expletive could potentially occur (henceforth ‘expletive contexts’). This allows me to go beyond previous studies which only count the number of sentences where expletive það is present, without taking into account the overall number of contexts in which the expletive could plausibly occur (Hróarsdóttir Reference Hróarsdóttir1998, Rögnvaldsson Reference Rögnvaldsson2002). Impersonal constructions which qualify as expletive contexts were isolated via CorpusSearch queries (see Booth Reference Booth2018 for details). To make the investigation manageable, I restrict the study to matrix clauses.
The IcePaHC treatment of constructions with cataphoric það and a clausal argument is not consistent. Sometimes það is tagged as an expletive (‘ES’), sometimes as a referential pronoun (‘PRO’), depending on the subject properties of the clause. A variety of CorpusSearch queries are therefore required in order to capture all examples of cataphoric það (see again Booth Reference Booth2018). ‘Null’ instances of cataphoric það – contexts where it could be expected but is not attested – are also tagged. Combining the overt and ‘null’ instances thus gives a total dataset of potential cataphoric contexts. As with the expletive contexts, I only include matrix clauses in the cataphoric dataset.
Once the data was collected, I manually examined each example to check whether the tagging was correct. Once erroneous examples were removed, I tagged the remaining examples for additional properties relevant to the investigation: position of það if present; verb position; predicate type. The two datasets which form the basis of the study are shown in Table 1. I show the number of contexts for each dataset as a proportion of the total number of matrix clauses to give an idea of relative frequency across time stages.
Expletive and cataphoric contexts in IcePaHC
As shown in Table 1, I split the IcePaHC diachrony into five periods. Periodising the data in this way has the advantage of abstracting away from the fact that the individual texts in IcePaHC are not evenly distributed across time. Throughout the study, I compare the proportion of examples with það to those without, over the five periods. This allows me to assess the strength of preference for það in the various contexts historically.
3. Cataphoric Það in Diachrony
As already shown, Icelandic exhibits constructions where a cataphoric það co-occurs with a clausal argument, see (4). In this paper, I adopt an LFG analysis for such constructions which has been proposed by Berman et al. (Reference Berman, Dipper, Fortmann, Kuhn, Butt and King1998) for parallel constructions in German (es plus clausal argument). I return to the architecture of LFG in section 3.3; for a general account, see Bresnan et al. (Reference Bresnan, Asudeh, Toivonen and Wechsler2015). This analysis assumes that the cataphoric element and the clausal argument contribute to the same argument slot of the matrix predicate; the information contributed by each constituent unifies under the same grammatical function at LFG’s f-structure (the functional dimension). Footnote 6 For constructions like (4a), I assume that both það and the clausal argument map to the subj(ect) function. For constructions like (4b), I assume that það and the clausal argument map to the obj(ect) function.Footnote 7 By contrast, expletive það in constructions like (1a) is a topic position placeholder and is exclusively motivated for structural reasons; it does not map to a grammatical function. In LFG terms, expletive það is thus an element which is present at c-structure (the constituent structure dimension) but has no representation at f-structure (the functional dimension) (see Lødrup Reference Lødrup, Robert and Börjars2011).
3.1 Constructions with a clausal subject
In earlier Icelandic (1150-1750), cataphoric það in constructions like (4a) qualifies as a subject. Evidence for this comes from its positional distribution in such contexts. Firstly, in sentences with topicalization – where a non-subject constituent occupies the sentence-initial position – the dominant pattern is for cataphoric það to be present in the immediately postfinite position, e.g. (6). This is typical subject behaviour in a Germanic V2 language like Icelandic.
(6)
The IcePaHC data shows that a postfinite það is overwhelmingly present in topicalization contexts in the data for 1150-1750 (81.8%), see Table 2.Footnote 8 The frequency of postfinite það is roughly comparable with the frequency at which það occurs in the clause-initial prefinite position in sentences without topicalization (86.4%), see Table 3.
Frequency of postfinite það in constructions with a clausal subject and topicalization in IcePaHC, 1150-1750
Frequency of prefinite það in constructions with a clausal subject in IcePaHC, 1150-1750
The second piece of evidence which supports the subject status of cataphoric það in this context is the fact that it occurs in the immediately postfinite position in yes/no-interrogatives, another key positional property of subjects in a Germanic V2 language. An early example is shown in (7).
(7)
Yes/no-interrogatives which feature clausal subjects are very rare in IcePaHC; I have only found the single example in (7) in which a postfinite það is present. Similar examples can however be found in other Old Icelandic texts, e.g. (8).
(8)
Of course, the fact that there is no example in IcePaHC which lacks a postfinite það does not rule out the possibility that such structures could occur, but (7) and (8) are at least clear evidence that a postfinite cataphoric það is possible in yes/no-interrogatives.
The third piece of positional evidence which supports the subject status of cataphoric það comes from sentences like (9), which are V1 declaratives where það occurs in the immediately postfinite position.
(9)
In the IcePaHC data for 1150-1750, a postfinite cataphoric það is present at a frequency of 78.1% in such contexts, see Table 4.
Frequency of postfinite það in V1 declaratives with a clausal subject in IcePaHC, 1150-1750
V1 declaratives are a marked construction in Icelandic; they are still a feature of the present-day language but occur more frequently in earlier stages (e.g. Butt et al. Reference Butt, Bögel, Kotcheva, Schätzle, Rohrdantz, Sacha, Dehé and Keim2014). A particular type of V1 declarative is the narrative inversion construction (Platzack Reference Platzack1985, Sigurðsson Reference Sigurðsson, Maling and Zaenen1990) which has a topical subject – prototypically a pronominal – in the immediately postfinite position, as in the second sentence in (10).
(10)
The fact that the immediately postfinite position is available to prototypical subjects in the narrative inversion construction suggests that immediately postfinite cataphoric það in V1 declaratives like (9) can be analysed as a subject.
While the occurrence of cataphoric það in constructions with a clausal subject is already known for earlier Icelandic (Faarlund Reference Faarlund1990, Rögnvaldsson Reference Rögnvaldsson2002), the diachronic behaviour of this element has not been previously examined. The IcePaHC findings show a striking change which is visible in the data as of 1751 whereby það becomes increasingly restricted to the clause-initial prefinite position.Footnote 9 Firstly, there is a decrease in the frequency at which það occurs in the immediately postfinite position in sentences with topicalization, see Table 5. By the modern period (1901-2008), það is only present in 38.9% of instances. Secondly, in V1 declaratives there is also a decrease in the frequency at which það occurs in the immediately postfinite position in the periods after 1751, see Table 6. By contrast, clause-initial prefinite cataphoric það – which is ambiguously either a subject or a topic position placeholder – does not undergo a decrease, see Table 7. I interpret the approximately simultaneous decrease in það in two contexts in which it is a unambiguously a subject – together with the stable status of prefinite það – as indication that cataphoric það in constructions with a clausal subject is loosing its subject status and transitioning towards becoming a placeholder for the topic position, i.e. the same function that expletive það serves in impersonal constructions, see (1) above.
Frequency of postfinite það in constructions with a clausal subject and topicalization in IcePaHC, 1150-2008
Frequency of postfinite það in V1 declaratives with a clausal subject in IcePaHC, 1150-2008
Frequency of prefinite það in constructions with a clausal subject in IcePaHC, 1150-2008
This diachronic account is interesting for two reasons. Firstly, it offers an explanation for the synchronic status of cataphoric það in this context in modern Icelandic. It has been observed that cataphoric það in constructions with a clausal subject in modern Icelandic exhibits a mixed positional distribution, whereby its occurrence in a position in which it is unambiguously a subject (the immediately postfinite position) is possible, but dispreferred (see Rögnvaldsson Reference Rögnvaldsson2002; Thráinsson Reference Thráinsson1979; Thráinsson Reference Thráinsson2007:312–313). Rögnvaldsson (Reference Rögnvaldsson2002:12) highlights the acceptability of all three variants in (11). Það is typically present in the clause-initial position in sentences without topicalization, e.g. (11a); in topicalization contexts, it is both acceptable for það to be absent as in (11b), or present in postfinite position as in (11c). According to Rögnvaldsson, the variant without það (11b) is more common than the ‘exceptional’ type with a postfinite það in (11c).
(11)
Standard accounts assume that postfinite það in contexts like (11c) is a referential pronoun, while prefinite það can be either an expletive or a referential pronoun (as first proposed by Thráinsson Reference Thráinsson1979; see also Thráinsson Reference Thráinsson2007:312–313). Furthermore, Thráinsson (Reference Thráinsson1979:481) interprets the synchronic observation whereby it is possible for some speakers to ‘raise’ cataphoric það as evidence for the fact that it is developing from a placeholder for the clause-initial prefinite position (‘surface adjustment particle’) into a subject, e.g. (12).
(12)
The diachronic account I have proposed here offers an alternative explanation, claiming that cataphoric það optionally exhibits some subject properties in the modern language as a reflex of its older status as a subject. This account assumes a change for cataphoric það in the opposite direction to that suggested by Thráinsson (Reference Thráinsson1979:481): from subject to clause-initial prefinite position placeholder.
Secondly, from a broader cross-Germanic perspective, my diachronic account for ‘subject’ cataphoric það goes against standard accounts for the emergence of expletives in Germanic, which can be summarised as the prefinite first hypothesis, see (13) (e.g. Breivik Reference Breivik1990 and Ingham Reference Ingham2001 on English, Lenerz Reference Lenerz and Abraham1985 on German, Falk Reference Falk1993 on Swedish, Faarlund Reference Faarlund1990 on Norwegian; for more general overviews, see Haiman Reference Haiman1974, Richards & Biberauer Reference Richards, Biberauer, Dikken and Tortora2005, Silva-Villar Reference Silva-Villar1996.)
(13)
The Prefinite First Hypothesis states that the general pathway whereby expletives emerge in Germanic is that positionally restricted expletives first appear in the clause-initial prefinite position, and only late generalise to all positions and thereby reach ‘subject expletive’ status. In line with this assumed trend, it has been claimed that the initial rise of expletives in Germanic is conditioned by structural considerations concerning V2 (i.e. as a strategy to occupy the clause-initial prefinite position, e.g. Richards & Biberauer Reference Richards, Biberauer, Dikken and Tortora2005).
The account which I have proposed here for the development of cataphoric það in Icelandic challenges the Prefinite First Hypothesis. I have shown that cataphoric það had subject status prior to 1750, and only relatively late in the diachrony began to lose its subject status, transitioning to becoming a structural placeholder for the clause-initial prefinite position (i.e. a prefinite expletive). The proposed change thus operates in the opposite direction to that predicted by the Prefinite First Hypothesis: a prefinite expletive emerges from a subject element. Moreover, the Icelandic findings are in line with studies on Old High German (Axel Reference Axel2007) and Middle Norwegian (Kinn Reference Kinn2016) which have similarly challenged the Prefinite First Hypothesis; comparative investigations in this area would be interesting to pursue in future.
3.2 Constructions with a clausal object
In this part of the study, I investigate the diachronic status of cataphoric það in constructions with a clausal object, e.g. (4b) above. Such constructions are defined as those whose matrix clause predicate has active morphology and where the clausal argument would map to object for that predicate’s argument structure (I discuss mapping relations in section 3.3). Specifically, I examine a subset of such constructions which lack an overt subject in the matrix clause (impersonal constructions), e.g. (14).
(14)
Table 8.Frequency of prefinite það in constructions with a clausal object in IcePaHC, 1150-2008
Table 9.Frequency of postfinite það in constructions with a clausal object and topicalization in IcePaHC, 1150-2008
Assuming the unification analysis outlined at the beginning of Section 3 – whereby the cataphoric element and the clausal argument contribute to the same argument slot of the matrix predicate – in such contexts það ought to qualify as an object. However, the IcePaHC findings regarding the positional distribution of það in such contexts challenge this assumption. The first relevant observation is that prefinite cataphoric það in constructions with a clausal object is optional and shows no clear change throughout the diachrony, see Table 8. The second relevant observation is that það is virtually restricted to the prefinite position in these contexts; it is attested only very infrequently in the postfinite position in topicalization contexts, see Table 9. As an illustration of this positional restriction, compare the example with það in (15a) with the example without in (15b), which is taken from the same text.
(15)
Due to its positional restriction, it does not seem reasonable to analyse cataphoric það in these contexts as an object. Rather, the IcePaHC findings challenge the unification analysis outlined above and suggest that það instead qualifies as a structural placeholder for the clause-initial prefinite position. Moreover, an examination of the information-structural properties of the examples with clause-initial það reveals that such examples typically initiate a new discourse and exclusively express discourse-new information, i.e. lack a topic. An example was already shown in (15a). Further examples are shown in (16).
(16)
As such, cataphoric það appears to be a structural placeholder for the topic position when there is no appropriate constituent to occupy that position. This is the same function as that served by expletive það in impersonal constructions with no clausal object in modern Icelandic, see (1) above.
I have thus shown that the topic position placeholder það – generally assumed to be a relatively recent phenomenon in Icelandic (see Section 1) – in fact has a long history in the language and is already solidly attested in topicless impersonal constructions with a clausal object in Old Icelandic (1150-1350). In Section 4, I claim that this status of cataphoric það in early Icelandic serves as a model from which það emerges as a topic position placeholder in a wide range of impersonal constructions in later stages of the language.
3.3 Structurally ambiguous constructions
3.3.1 Mapping Theory within LFG
So far, I have shown that two types of cataphoric það were already established at an early stage of Icelandic:
1. Cataphoric það in constructions with a clausal subject, which – at least in earlier Icelandic (pre-1750) – qualifies as a subject.
2. Cataphoric það in constructions with a clausal object, which functions as a topic position placeholder.
I now introduce a third, structurally ambiguous type of construction where cataphoric það and a clausal argument co-occur, e.g. (17). The predicate in this type is a passive transitive which can take a clausal complement, e.g. segja ‘say’. I henceforth refer to such predicates as say-type predicates.
(17)
I argue that constructions like (17) are structurally ambiguous with respect to the grammatical function which is assigned to the clausal argument and allow for at least two analyses:Footnote 10
1. As a promotional passive construction, in which the clausal argument maps to the subj(ect) function.
2. As a non-promotional passive construction, in which the clausal argument maps to the obj(ect) function.
The parallel architecture of LFG offers many advantages for modelling this distinction between promotional and non-promotional passives, since different dimensions of linguistic information are handled at independent, interacting levels of structure (see e.g. Asudeh & Toivonen Reference Asudeh, Toivonen, Heine and Narrog2009, Bresnan et al. Reference Bresnan, Asudeh, Toivonen and Wechsler2015). There are three such levels which are relevant to this paper:
1. c-structure, which captures information about category and constituency.
2. f-structure, which captures information about grammatical functions, grammatical features and grammaticalised discourse functions.
3. a-structure, which captures information about the arguments required by a predicate and their thematic roles.
I adopt the view of a-structure assumed by Kibort (Reference Kibort, Butt and King2007; Reference Kibort, Butt and King2008; Reference Kibort, Butt and King2014) and developed as part of her revision of (Lexical) Mapping Theory, which is a tool for mapping between grammatical functions (at f-structure) and arguments (at a-structure), as I show below. The Kibortian view of a-structure is shown in (18). It consists of two levels of information: an ‘argument position’ level and a ‘semantic participant’ level.
(18)
The argument position level is the core component of a-structure. At this level, any predicate has access to the universal subcategorization frame in (19), from which it selects certain arguments (Kibort & Maling Reference Kibort, Maling, Butt and King2015:152). In this paper, I only deal with arg1 and arg2.
(19)
Each argument position in (19) is associated with an intrinsic syntactic classification expressed in terms of the binary features [± r(estricted)] and [± o(bjective)]. This serves to specify what type of grammatical function each argument position can in principle map to, as summarised in Table 10 (see Bresnan & Kanerva Reference Bresnan and Kanerva1989). obj θ is an abbreviation for secondary objects; obl θ abbreviates multiple oblique functions.
Decomposition of grammatical functions into features
Arg1 is specified as either [–o] or [–r], depending on the predicate type: for transitive and unergative verbs, arg1 is associated with [–o]; for unaccusative verbs, arg1 is associated with [–r]. In this paper I only deal with mappings for transitive predicates, so arg1 will always be [–o].
A transitive predicate like segja ‘say’ in (17) takes two arguments: arg1, Agent and arg2, Theme. The default argument-function mapping for a say-type predicate is as in (20).
(20)
Passivisation is an operation which results in a change in the default mapping between arguments and grammatical functions. In Mapping Theory, passivisation is understood as comprising two components: demotion and (potentially) promotion. When a transitive predicate like segja ‘say’ is passivised, the highest argument (already specified as [–o]) receives an additional specification that it must map onto a grammatical function which is also [+r]; the combination [–o,+r] results in specifying an oblique (see Table 10). The result is the mapping in (21). The second argument (arg2) – which by default maps to obj – remains specified as [–r] and can be promoted to subj (see Table 10).
(21)
Mapping Theory also allows for non-promotional transitive passives, which have the mapping in (22) (Kibort & Maling Reference Kibort, Maling, Butt and King2015:156). In a non-promotional passive, the first argument is still demoted to an oblθ , as in the promotional passive in (21). The point of difference is that in the non-promotional passive, the second argument is additionally specified as [+o] and is thus ‘blocked’ from being promoted to subj; it is forced to remain an obj (see again Table 10).
(22)
For constructions like (17), the two analyses thus differ in terms of the grammatical function which is assigned to the clausal argument: in the promotional passive analysis, the clausal argument maps to subject; in the non-promotional analysis, it maps to object.
3.3.2 Comparable analyses for Germanic
The structural ambiguity of such constructions has been previously acknowledged for other Germanic languages. Specifically, the possibility that they are open to either a promotional passive or non-promotional passive analysis is captured by Berman (Reference Berman2003:162–164) for parallel constructions in German and by Bennis (Reference Bennis1986) for Dutch. Berman (Reference Berman2003:162–164) presents an LFG analysis of parallel constructions in German with the predicate gesagt, the passive of sagen ‘say’, which also takes two arguments – an Agent and a Theme – which by default map to subject and object respectively. Berman argues that, in such constructions, the Theme straightforwardly maps to subject if it is a DP. If the Theme is clausal – as in the Icelandic construction in (17) – then it can map to either subject or object (Berman Reference Berman2003:162). When the clause maps to subject, the construction qualifies as a promotional passive; when the clause maps to object, it qualifies as a non-promotional passive on the terms outlined above.
In a different framework, Bennis (Reference Bennis1986:108) points out that in Dutch constructions with passive transitives which take a clausal complement, both het (it-type expletive) and er (there-type expletive) are possible, see (23) (examples are taken from the discussion of Bennis Reference Bennis1986:108 in Vikner Reference Vikner1995:229).Footnote 11
(23)
In examples with het like (23a), Bennis assumes that het is base-generated in the object position (i.e. within the VP) and is moved into the subject position. The embedded clause is assumed to be in an adjoined position, cf. (24a) (as presented in Vikner Reference Vikner1995:229). The sentence is thus analysed as a promotional passive, which in this particular framework involves movement of the constituent base-generated in object position (het) into subject position. In examples with er like (23b), Bennis assumes that the embedded clause occupies the object position and that er is base-generated in subject position (SpecIP), cf. (24b) (Vikner Reference Vikner1995:229). There is no movement of a base-generated object into subject position, and so the sentence essentially qualifies as a non-promotional passive.
(24)
This difference is in turn related to the claim that het has argument status (i.e. is a non-referential argument, a so-called ‘quasi-argument’, Chomsky Reference Chomsky1981:324–327), and can thus be base-generated in a theta-marked position (e.g. the object position of a passive). Er does not qualify as an argument (i.e. is a non-referential non-argument, a ‘true expletive’), and so it can be base-generated in the subject position of a passive, which is a non-theta-marked position. What is relevant to my analysis is the fact that the passive of a say-type predicate with a clausal argument is in this view open to both a promotional and non-promotional passive analysis. Moreover, since Dutch has both an it-type expletive and there-type expletive, this difference shows up on the clause-initial element (het versus er). Since Icelandic has only the it-type (það) and not the there-type (þar), the two possibilities do not show up in the same way. The distinction can however be recovered by the positional distribution of það, as I now show.
3.3.3 Findings for Icelandic
So far, I have claimed that constructions like (17) are structurally ambiguous and allow for the clausal argument to be analysed as a subject (promotional passive) or an object (non-promotional passive). In Section 3.1, I showed that cataphoric það in constructions with an unambiguous clausal subject frequently occurs in both pre- and postfinite position, and thus qualifies as a subject – at least in earlier stages. By contrast, in Section 3.2 I showed that, in constructions with an unambiguous clausal object, cataphoric það is virtually restricted to the prefinite position and thus behaves as a placeholder for the topic position.
With these findings in mind, I now examine the positional distribution of cataphoric það in the ambiguous construction type in (17), which is attested in IcePaHC, albeit relatively infrequently. The IcePaHC findings support the availability of the two analyses I have proposed. The results indicate that cataphoric það in this context occupies an intermediate position between the findings for constructions with an unambiguous clausal subject and the findings for constructions with an unambiguous clausal object with respect to its frequency in postfinite position. Overall, cataphoric það in the ambiguous type occurs in the postfinite position not as frequently as in the unambiguous clausal subject type (Table 5) but more frequently than in the unambiguous clausal object type (Table 9).
Frequency of postfinite það in structurally ambiguous constructions with a clausal argument and topicalization in IcePaHC, 1150-2008
I interpret this intermediate result as supporting my proposal that, in these ambiguous contexts, the clausal argument can be analysed as a subject (in the promotional passive analysis) – in which case það is preferred in postfinite position – or as an object (in the non-promotional passive analysis) – in which case it is dispreferred in postfinite position, in line with the general positional restriction observed above for cataphoric það in constructions with an unambiguous clausal object.
The corpus findings also reveal that cataphoric það occurs very frequently in the prefinite position in the ambiguous construction type, see Table 12. This finding is in line with the proposal made in this section. Regardless of whether the clausal argument is analysed as a subject or an object in such contexts, það is expected to be frequently present in the prefinite position on the basis of the results for unambiguous constructions with a clausal subject and those with a clausal object (Sections 3.1 and 3.2 respectively).
Frequency of prefinite það in structurally ambiguous constructions with a clausal argument in IcePaHC, 1150-2008
4. The Emergence of the Expletive
4.1 Say-type predicates as a bridging context
So far, I have shown that cataphoric það is already established in Old Icelandic as a topic position placeholder in topicless impersonal constructions with a clausal object, as in the examples repeated in (25).
(25)
As mentioned in Section 1, in Old Icelandic topicless impersonal constructions which lack a clausal object, það is by contrast overwhelmingly absent, rendering V1 sentences, e.g. (26); see also (3) above.
(26)
The IcePaHC results confirm that impersonal constructions which lack a clausal object in the earlier periods are predominantly V1 constructions which lack það, see Table 13. There is a dramatic increase in prefinite það which – at least in the corpus data – is visible in the period 1901-2008.
Frequency of prefinite það in impersonal constructions without a clausal object in IcePaHC, 1150-2008
Nevertheless, there are three early examples in the data prior to 1550 with a prefinite það which can be analysed as an expletive. I show these in (27).
(27)
All three of these exceptional examples with það have a say-type predicate (mæla ‘speak’, segja ‘say’). Say-type predicates are precisely those transitive predicates which participate in constructions with a clausal argument, i.e. the contexts where cataphoric það is already robustly attested in Old Icelandic, see (25) above. The examples in (27) – while scarce – are evidence that það can already appear in early Icelandic in constructions with say-type predicates which lack a clausal argument. Furthermore, there is a possibility that það in these examples is cataphoric – with reference to the proposition in the later discourse – though it is hard to be sure whether það is expletive or cataphoric here.
It is only in the last two periods – and specifically in texts as of c.1850 – that an unambiguously expletive það is attested in topicless impersonal constructions with other types of predicate, beyond the say-type. I show examples from 1850-2008 in (28), which exhibit expletive það occurring in impersonal constructions with a range of predicate types.
(28)
I suggest that the early examples of það in (27) are indication that constructions with say-type predicates and no clausal argument serve as a bridging context (e.g. Heine Reference Heine, Wischer and Diewald2002) in facilitating the spread of það to impersonal constructions with all types of predicate.Footnote 15 Moreover, since the status of það in (27) as either cataphoric or expletive is ambiguous, such examples appear to provide a bridging context from the unambiguous cataphoric to the unambiguous expletive type.
The examples in (28) thus represent the third and final stage in the diachronic development by which það becomes established in the clause-initial prefinite position in the full range of topicless impersonal constructions, see Table 14. Between stages 2 and 3, the spread of það is likely to have proceeded via lexical diffusion, but the IcePaHC data lacks the necessary detail to show the actual stepwise progression. More research of texts beyond IcePaHC would be necessary to show this development in greater detail.
The rise of það in Icelandic topicless impersonal constructions
4.2 A parallel development in presentational constructions
I now discuss how the emergence of the expletive in topicless impersonal constructions as shown in Section 4.1 has a parallel diachronic development in presentational constructions. Presentational constructions – which are ‘all new’ and hence topicless – are another context where það functions as a topic position placeholder in present-day Icelandic, e.g. (29).
(29)
In Old Icelandic, the expletive is absent in such constructions, rendering V1 structures, e.g. (30).
(30)
Booth (Reference Booth2018) conducted a corpus-based study of IcePaHC examining the status of the expletive in presentational constructions. I display the findings for prefinite það in such contexts in Table 15. The first examples with a prefinite það occur in the period 1351-1550. However, the frequency of það remains low from this point onwards, until the last two periods where a stark increase in það is visible in the corpus data for 1751-2008. I show examples from each period in (31).
(31)
It is striking that the increase in prefinite það in presentational constructions approximately coincides with the development shown for impersonal constructions in Section 4.1, whereby það generalises in its role as a topic position placeholder to all types of topicless impersonal construction. I interpret these two developments as representing one overall change – the establishment of það as a topic position placeholder across presentational and impersonal constructions. As shown above, in earlier stages of Icelandic the topic position was typically unoccupied in such contexts, rendering V1 structures. Strikingly, the growing establishment of this topic position placeholder coincides with the change involving cataphoric ‘subject’ það shown in Section 3.1, which also develops towards becoming a topic position placeholder at this point in the diachrony.
Frequency of prefinite það in presentational constructions in IcePaHC, 1150–2008
5. Conclusion
In this paper, I have presented a diachronic account for the development whereby the Icelandic expletive það emerged as a topic position placeholder in topicless impersonal constructions, on the model of an earlier restricted set of cataphoric contexts. I showed that there are two types of cataphoric það which are robustly established in earlier Icelandic (pre-1750):
1. A cataphoric það which co-occurs with a clausal subject and behaves positionally like a subject (Section 3.1).
2. A cataphoric það which co-occurs with a clausal object and behaves positionally like a structural placeholder for the topic position in topicless sentences (Section 3.2).
I also discussed constructions with passive transitive say-type predicates where cataphoric það co-occurs with a clausal argument. I claimed that such constructions allow for two analyses (promotional or non-promotional passive) and showed how this difference can be modelled with a version of LFG’s Mapping Theory developed by Kibort (Reference Kibort, Butt and King2007; Reference Kibort, Butt and King2008; Reference Kibort, Butt and King2014). Furthermore, I argued that the mixed positional distribution of það in such contexts supports the availability of the two analyses.
I showed that the overwhelming pattern in Icelandic prior to 1850 for impersonal constructions which lack a clausal argument is for það to be absent. However, sparse early examples with það were observed and found to share a particular property: having a say-type predicate (Section 4.1). Only in the latter half of the 19th century do the first examples of það with other predicate types appear, with það attested with a broad range of predicates in the data for 1850-2008. This led me to propose a pathway of change via generalisation, whereby það spread from a more restricted set of topicless impersonal constructions with a clausal object (cataphoric það) to all types of topicless impersonal constructions – including those which lack a clausal object and in which það cannot be cataphorically referential (expletive það). I argued that this change was facilitated by impersonal constructions with a say-type predicate, which act as a bridging context between the older context with a clausal object, and the innovative context with no clausal object.
Strikingly, three changes shown in this paper occur at approximately the same point in the Icelandic diachrony: the generalisation of það in its function as a topic position placeholder to all types of impersonal construction roughly coincides with the change whereby ‘subject’ cataphoric það begins to loose its subject status and transitions towards becoming a topic position placeholder, i.e. the same function as það in topicless impersonal constructions. Furthermore, the change whereby expletive það emerges in Icelandic presentational constructions – which are always topicless – also approximately coincides with the two changes involving cataphoric ‘subject’ það and topicless impersonal constructions. I interpret these three developments as representing one overall change – the establishment of það as a topic position placeholder in topicless sentences – occurring across three broad construction types (constructions with a clausal argument, impersonal constructions; presentational constructions). Furthermore, an IcePaHC-based study of other word order phenomena (V1 declaratives, subject position) by Booth et al. (Reference Booth, Schätzle, Börjars, Butt, Butt and King2017) observed significant changes towards reduced word order freedom occurring in Icelandic in c.1900, which was interpreted as a shift towards increased syntactic configurationality and the growth of left peripheral structure. The changes involving cataphoric and expletive það presented in this paper similarly point towards this broader development in Icelandic clause structure at this relatively late stage in the diachrony. The exact nature of the interaction between these various changes, I leave for future research.
Acknowledgements
The research for this paper was funded by an Arts & Humanities Research Council UK (AHRC) doctoral scholarship. Particular thanks go the editors of this issue and two anonymous reviewers for their feedback on this work. I thank audiences at the Manchester Forum in Linguistics (April 2018), Grammatik i Norden (May 2018), the 25th South of England LFG Meeting (May 2018) and the Workshop on Word Order in Scandinavian Languages at the University of Konstanz (December 2018) for comments on this work. For their valuable insights I am especially grateful to Kersti Börjars, Tine Breban, John Payne, Helge Lødrup, Joan Maling, Halldór Sigurðsson and Christin Schätzle.
