IntroductionFootnote ¹

In a whole range of variation phenomena, language users base their choices on the cognitive accessibility of their linguistic devices, thereby reducing the processing effort. ‘Support strategies’ are defined as the use of linguistic variants which mitigate such processing effort. They apply to contexts of variation and/or change in which language users have a choice between several grammatically correct and functionally overlapping variants. Here, a wide array of processing principles come into play, such as Rohdenburg's (Reference Rohdenburg1996) Complexity Principle (more explicit grammatical alternatives tend to be preferred in cognitively more complex environments, where ‘more complex’ implies, for instance, discontinuity, passivisation, length, subordination, deletion), Hawkins’ (Reference Hawkins2004) Domain Minimisation Principle (minimise the distance between dependents and heads of phrases) or Jaeger's (Reference Jaeger2010) Uniform Information Density Principle (the more probable a word is in its context, the less information it carries in that context; see Jaeger Reference Jaeger2010: 24).

The articles in this special issue provide empirical analyses which shed light on the underlying motivation for a whole range of apparently unrelated, heterogeneous and idiosyncratic preferences of language users on different levels of linguistic analysis. The contributors investigate strategies that are employed in language to aid processing by taking the cognitive accessibility of their linguistic devices into account.

Research of the last two decades in the area of grammatical variation has shown that grammatical variation is far more extensive than expected (see contributions in Fanego et al. Reference Fanego, López-Couso and Pérez-Guerra2002; Rohdenburg & Mondorf Reference Rohdenburg and Mondorf2003; Kortmann et al. Reference Kortmann, Schneider, Burridge, Mesthrie and Upton2004; Dufter et al. Reference Dufter, Fleischer and Seiler2009; Peters et al. Reference Peters, Collins and Smith2009; Rohdenburg & Schlüter Reference Rohdenburg and Schlüter2009; Maguire & McMahon Reference Maguire and McMahon2011) and that grammatical variation is determined by a complex network of largely interdependent factors from all levels of linguistic analysis that may interact synergetically or antagonistically (see multivariate studies modelling the interaction of factors).

This means that any theory aiming at descriptive adequacy and predictive power will need to come to grips with a multitude of highly diverse internal and external factors constraining the choices made by language users. This multitude and diversity of factors constraining the choices made by language users ‘can sharpen our eye for uncovering underlying principles and generalizations’ (Mondorf Reference Mondorf, MacWhinney, Malchukov and Moravcsik2014: 210).

As regards empirical approaches, one strand of research has been primarily concerned with finding and operationalising factors constraining variation. Such studies reveal an intricate interaction of highly diverse levels of linguistic analysis, challenging both strictly modular theories and those assuming a top-down processing in which phonology does not influence grammar (see Schlüter Reference Schlüter2005). Their research findings thus strongly advocate non-modular, usage-based approaches to explaining language variation. Additionally, multivariate analyses permit us to weigh the effects of several factors against each other, thereby assessing their relevance for language users’ choices (see Gries Reference Gries, Rohdenburg and Mondorf2003; Szmrecsanyi Reference Szmrecsanyi2005; Hilpert Reference Hilpert2008). While these analyses are restricted in terms of the number of factors they can fit into a model, they are not only able to show which factors affect choice decisions in language but also weigh their effect strengths – often arriving at an impressively high degree of predictive power. This special issue combines both monofactorial (see Lorenz, Eitelmann, Rohdenburg, Stange, Kaatari, this issue) and multifactorial (see Cheung & Zhang, Wiechmann & Kerz, Mondorf & Schneider, this issue) types of studies.

The existence of linguistic variation also gives rise to the question of why human language, which is otherwise a relatively economical system, affords the luxury of redundancy at all, i.e. several options that express the same function. Early conceptions of variation were based on the assumption that variation in the sense of layering results from an intermediate stage in which variants co-exist only for a brief period of time, namely for as long as new variants eventually replace older variants. However, studies involving long-term variation phenomena (see e.g. Stein Reference Stein1990 or Mondorf Reference Mondorf2009) demonstrate that variants continue to co-exist for centuries, an observation that leads to the question of what ultimately motivates linguistic variation.

This special issue bridges the gap between traditionally descriptive variation linguistics, historical linguistics and functional-typological approaches (see Comrie Reference Comrie1981; Givón Reference Givón1984, 1990; Croft Reference Croft1990; Bolinger Reference Bolinger1977) which are related to psycholinguistic and cognitive principles, with the primary concern to explain how language users base their choices on the cognitive accessibility of their linguistic devices. It brings together a range of empirical studies on grammatical variation that lend themselves to an explanation in terms of processing-based choices of variants.

The concept of ‘support strategies’, as introduced in the present issue, is based on the assumption that in a wide array of variation contexts speakers base their choices on the cognitive accessibility of their linguistic devices, thereby reducing processing effort. Interestingly, this does not necessarily mean that certain features are highlighted, e.g. by providing end-weight (see Pérez-Guerra's Reference Pérez-Guerra2016 account of the supremacy of end-weight vs syntactic constraints such as ‘complements-first’), by choosing a syntactic rather than morphological variant or by using more explicit variants. Support can also be brought about by selecting less salient options, e.g. attenuating semantic prominence by means of irrealis or attenuating lexical prominence by means of formulaic expressions. Support is thus effected by the processing-based choice of linguistic variants.

Support in this sense is conceptually limited to contexts of variation and/or change that allow for the choice between grammatically correct and functionally overlapping variants.Footnote ² From a diachronic perspective, the eventual survival of linguistic variants or their functional specialisation depends on the system-internal trade-off between the pros and cons of their use. This trade-off can also be observed in operation in variation contexts, as these contexts involve choices that are not overlaid by semantic requirements. In other words, variation contexts allow for the observation of this trade-off process in vivo, which may in some cases eventually give rise to scenarios of a developing division of labour or language change.

Support strategies manifest themselves on highly diverse levels of linguistic description, ranging from graphemics via phonology, morphology, lexicology and syntax to semantics and pragmatics. They are thus a cover term for various processing principles operative in grammatical variation, such as the Complexity Principle (which applies to variants that differ in terms of explicitness; see Rohdenburg Reference Rohdenburg1996), the Domain Minimisation Principle (which applies to heads of phrases and their dependents; see Hawkins Reference Hawkins2004), the Principle of Rhythmic Alternation (which aims at a perceptionally ideal sequence of stressed and unstressed syllables; see Schlüter Reference Schlüter2005) or the Horror Aequi Principle (which predicts a tendency to avoid certain identical adjacent grammatical elements (see Rohdenburg Reference Rohdenburg2003; Stange, this issue).

The term ‘support’ has first been established in a purely descriptive sense and only with reference to do-support in major English grammar books (see Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 133–4; Huddleston & Pullum Reference Huddleston and Pullum2002: 93–7). Extending this descriptive notion to a processing-based conception of support allows revisiting well-researched phenomena from a new angle. As an illustrative example, it is instructive to reconsider do-support, the historical development of which is (at least partially) processing-motivated. In a range of contexts, do-support can be described as a phonological support strategy, since do-support established itself first in those contexts in which the inflectional suffix -st added to the main verb, rather than do, would have resulted in a hard-to-pronounce consonant cluster: imaginedst /ndst/ thou? > didst /dst/ thou imagine? (see Stein Reference Stein1990). In other words, do-support serves as a phonological avoidance strategy in these contexts. Moreover, the distribution of do-support also reveals that historically it was largely motivated by factors that can be related to processing concerns (see Ellegård Reference Ellegård1953; Stein Reference Stein1990; Eitelmann, this issue). Do-support is systematically triggered in contexts that are cognitively challenging in terms of their semantics (i.e. the expression of irrealis in interrogatives and negation contexts), in terms of pragmatics (i.e. emphatic do) or in terms of syntax (i.e. preservation of SVO pattern in transitive verb constructions). Similarly, processing-related factors are at work in the introduction of tun-periphrasis in German dialects (see Erben Reference Erben, Engel, Grebe and Rupp1969; Rohdenburg Reference Rohdenburg1986). These uses are within our narrow definition of support.Footnote ³

The notion of ‘support’ has subsequently been extended to other variation phenomena, such as comparative alternation for which 26 processing-related factors have been shown to trigger more-support (see Mondorf Reference Mondorf2009). Similarly, the broader concept of analytic support (Mondorf Reference Mondorf, MacWhinney, Malchukov and Moravcsik2014) relates to functionally motivated choices between synthetic and analytic options:

• • English comparative alternation: fuller vs more full (see Mondorf Reference Mondorf2009)

• • English genitive alternation: the topic's relevance vs the relevance of the topic (see Rosenbach Reference Rosenbach2003)

• • English future tense alternation: will vs going to (see Szmrecsanyi Reference Szmrecsanyi2003)

• • English mood alternation: on the condition that he agree-∅ (subjunctive) / on the condition that he agrees (indicative) vs on the condition that he should agree (modal periphrasis) (see Schlüter Reference Schlüter2009)

• • Spanish future tense alternation: comeré vs voy a comer (see Lastra & Butragueño Reference Lastra and Butragueño2010)

• • German past tense alternation: Er brauchte Geld vs Er hat Geld gebraucht (see Jäger Reference Jäger1971)

Finally, in the case of enough-support (see Rohdenburg & Schlüter Reference Rohdenburg and Schlüter2009: 380), e.g. plenty nice enough, the addition of enough supports the weakly established plenty that derives via conversion from the noun plenty (see Rohdenburg & Schlüter Reference Rohdenburg and Schlüter2009: 371). In the same vein, the innovative use of oddly as a sentence adverbial is supported by the semantically bleached enough (oddly ∅ > oddly enough).Footnote ⁴

A whole range of seemingly diverse variation phenomena analysed in this issue can thus be consolidated regarding their explanation in terms of support strategies. They comprise:

• • end-weight as a determinant of do-support and self vs ∅ (Eitelmann)

• • formulaicity as a determinant of adverbial clause positioning (Wiechmann & Kerz)

• • (de-)transitivisation processes with causative bring (Mondorf & Schneider)

• • extracted vs non-extracted constructions (Rohdenburg)

• • possessive have vs have got (Lorenz)

• • pseudo-passive constructions, e.g. be sat vs be sitting (Stange)

• • complementiser that-deletion in adjectival complements (Kaatari)

• • comparative and superlative alternation (Cheung & Zhang)

This special issue thus takes a usage-based approach aimed to shed new light on the roles of support strategies in language variation and change. In the following we provide summaries of the contributions.

Eitelmann discusses the plausibility of regarding end-weight (as in, for example, Leech Reference Leech1983: 65) as a support strategy for grammatical variation phenomena in English. He introduces a more fine-grained notion of end-weight, differentiating between end-weight as (i) a short-before-long ordering principle, (ii) as a constructional trigger that makes language users prefer constructions that guarantee the longer syntactic constituent to be positioned at the sentential end, and (iii) a gravity principle that creates end-weight by reinforcing the sentential end by means of bulky elements. In order to isolate end-weight effects in grammatical variation, Eitelmann presents the results of two case studies. First, he investigates five semi-reflexive verbs accompanied by either the -self variant or the zero variant (I hide myself vs I hide) and shows that the radical decline of -self objects across time is delayed in final position, particularly in Late Modern English. Second, Eitelmann analyses 31 randomly selected verbs in Early Modern English declarative clauses containing either the finite verb (He smiled) or the do-supported counterpart (He did smile). He concludes that do-support is favoured in final position so as to balance out the weights of the verb phrase and that of the preceding syntactic constituents; in this context, it is conspicuous that monosyllabic verbs take do-support to a larger extent in final position. The case studies thus provide first evidence for end-weight functioning as an autonomous support strategy, according to which the more explicit variants (e.g. I hide myself and He did smile) save processing costs on the human parser.

Wiechmann & Kerz's article shows that the use of formulaic language is a support strategy which exerts a significant influence on users’ choices between competing linguistic variants. In this study, they explore Diessel's (Reference Diessel2008) database of sentences including either preposed or postposed temporal adverbial clauses, extracted from the British component of the International Corpus of English (ICE-GB), and try to account for the ordering of the temporal clause and the ‘main’ clause. Wiechmann & Kerz investigate five factors associated with the conceptual order of the clauses, namely (i) the degree of syntactic complexity of the adverbial clause (either simple or involving at least one subordinate clause), (ii) the semantics of the adverbial clause (purely temporal or temporal with implicit conditional, concessive, causal or purposive meanings), (iii) the length of the adverbial clause and (iv) formulaicity. As regards the latter factor, they focus on the frequency of formulaic sequences or n-grams in the adverbial clause, on the one hand, and, on the other hand, on pattern frequency indicators (i.e. use of subordinating conjunctions, lexical vs pronominal subjects and transitive vs intransitive clauses) which assess the degree of entrenchment of the adverbial clause. They conclude that all the variables except syntactic complexity are significant predictors of ordering and that the inclusion of formulaicity indicators improves the statistical model considerably. Assuming that preposed adverbial clauses imply a higher processing load since they require the overall planning of the utterances at the beginning, formulaicity compensates for this cognitively demanding ordering choice and thus constitutes a support strategy in English.

Mondorf & Schneider's article deals with contexts which reduce the clause's transitivity, i.e. Moderate Transitivity Contexts in the sense of Mondorf (Reference Mondorf, Dorgeloh and Wanner2010). Such environments serve as support strategies that alleviate processing in cases where the same verb can be used in several structures – (at least) one highly frequent and one marginal, possibly even on the verge of dying or losing relevance in the linguistic system. In this article the latter is illustrated by causative bring (bring X to-infinitive . . .), which has almost completely lost its ability to take fully fledged direct objects in Present-day English (^? I brought him to laugh). Causative bring accepts virtually only reflexives as direct objects. These reflexives further co-occur with modals (conveying non-factual meaning) and with negative polarity (also triggering a non-factual interpretation). These are detransitivised contexts that maintain the transitive syntactic frame while semantically weakening the causative force of the verb. Causative bring decreases significantly across time and finds a perfect support device in these environments which reduce the degree of transitivity and thus the effect of the verbal action. Thus, Mondorf & Schneider demonstrate in a corpus-based study that from the fifteenth to the twentieth century causative bring is specialising mostly to modal, negative and reflexive (detransitivising) contexts.

Rohdenburg's contribution contrasts two support strategies at work in extractions of postverbal elements like They have promised (that) they would tackle this problem (canonical) vs This is a problem (that) they had promised (that) they would tackle Ø (object extraction). He aims at resolving the alleged rivalry between Hawkins’ (Reference Hawkins2004) Domain Minimisation Principle and Rohdenburg's (Reference Rohdenburg1996) Complexity Principle. The Domain Minimisation Principle basically predicts that languages prefer their processing domains (such as filler–gap extraction environments) to be as simple as possible both morphosyntactically and semantically. The Complexity Principle holds that more explicit constructional options tend to be preferred in cognitively more complex environments (illustrated by extraction contexts in this study). Rohdenburg investigates 16 complement pairs in extraction contexts, among them that- vs infinitive clauses (This is a problem (that) they had promised {(that) they would tackle / to tackle}), subject-to-object raising (This is the man they alleged {was / to be} a fraud) and gerunds vs infinitives (It was a scene which he had dreaded {to see / seeing}). While some variation phenomena are accounted for by Domain Minimisation and are incompatible with the Complexity Principle, the opposite holds for others. This suggests that different support strategies are used for different syntactic variables. Rohdenburg concludes that there is a division of labour between the two processing principles in English. In extraction contexts, the two principles have non-overlapping territories, i.e. ‘the effects of the two antagonistic principles are found with virtually complementary ranges of complement types’ (Rohdenburg, this issue). The support strategy of explicitness aids processing in some constructions, while the support strategy of minimising domains prevails in others.

Lorenz's corpus-based study focuses on the emergence and early use of stative possessive have got from the sixteenth century onwards. First, Lorenz tries to justify the adoption of possessive have got as a pattern ‘preserver’ (or support strategy) which compensates for the increasing vocalic reduction of have and has to schwa and zero in contracted forms ’ve and ’s. To this purpose, he investigates contexts favourable to contraction such as those with vowel-final subject pronouns but the data show no connection between these environments and the emergence of have got. Second, Lorenz argues in favour of the semantic shift the perfective form of get has undergone from ‘have obtained’ to ‘possess’. Following Traugott & Dasher's (Reference Traugott and Dasher2002) Invited Inferencing Theory of Semantic Change, he claims that a past event of ‘obtaining’ invites the inference of ‘possession’. The analysis of the context types identified by Heine's (Reference Heine, Wischer and Diewald2002) model of invited inference leads Lorenz to conclude that the emerging implicature of possession is gradually conventionalised and subsequently leads to the reanalysis of have got with the meaning of stative possession. In fact, have got implies a previous ‘get’-event in Present-day English in the majority of the instances. Finally, Lorenz detects a side effect of have got, which he discusses as a syntactic support strategy: namely its function to maintain subject-verb-object word order in interrogative sentences, an order not preserved in the inverted alternative.

The notion of support strategy also features prominently in Stange's article on intra-dialectal variation between so-called ‘pseudo-passive’ constructions (see Klemola Reference Klemola1999), such as be sat with progressive meaning (active in semantics and passive in form) and progressive constructions like be sitting. Stange retrieves her data from the conversation section of the British National Corpus, and investigates the following factors: frequency, dialect, age, gender, the applicability of Rohdenburg's Complexity Principle and the Horror Aequi Principle (see, for example, Rohdenburg Reference Rohdenburg2003). First, Stange illustrates that be sat is two or three times less frequent than the progressive counterpart be sitting. Second, she finds that the use of pseudo-passives spreads from the North and the Southwest of England as well as from Lancashire. Third, age and gender are not significant factors as regards pseudo-passive/progressive variation. Fourth, Horror Aequi, which predicts that formally (near-)identical and (near-)adjacent elements are mutually avoided, justifies that be sat is favoured when an -ing form follows the construction (as in we were sat/ ^? sitting having a drink) but this tendency is not significant. Fifth, speakers who display variation show no pronounced preferences in cognitively ‘simple’ utterances but in cognitively complex environments such as hypotactic constructions (I don't know . . .) and non-assertive structures (negative, interrogative, conditional) and with infrequent lexical items they opt for be sitting significantly more frequently. This shows that be sat/be sitting variation is governed by the Complexity Principle and can thus be taken as another instance of support-based linguistic strategies.

Kaatari explores whether Rohdenburg's (Reference Rohdenburg1996) Complexity Principle and Jaeger's (Reference Jaeger2010) Uniform Information Density Principle correctly predict that-realisation and omission in linguistic contexts containing adjectives followed by that/Ø-complement clauses. The author distinguishes between tokens of extraposition (It is impossible that/Ø the principles are wrong) and of post-predicate complementation (I am convinced that/Ø the principles are correct). He investigates five factors which are claimed to determine that-retention and omission: collostruction strength as a measure of information density, the type and length of the subject in the that-clause, the type of matrix subject as well as separation between the adjective and the complement clause as determinants of complexity, and finally genre. He separately measures the relative explanatory power of these variables in extraposition and post-predicate complementation. Kaatari concludes that (i) that-omission is far more frequent in post-predicate clauses; (ii), as predicted by the Uniform Information Density Principle, formulaic expressions, such as I'm/am sure, in the matrix clause are a major trigger for that-omission in post-predicate clauses; (iii) the Complexity Principle correctly predicts that that is always retained where the adjective and the complement are separated; (iv) the Complexity Principle can also explain the high rates of that-omission when the subject of the complement clause is a pronoun and when the subject of the matrix clause of post-predicates is I.

Finally, Cheung & Zhang's article focuses on synthetic–analytic variation in comparatives (thicker vs more thick) and superlatives (thickest vs most thick). Repeating and extending Hilpert's (Reference Hilpert2008) multifactorial analysis, Cheung & Zhang are to be credited with being among the first to provide an in-depth analysis of both comparative alternation and superlative alternation. They explore the British National Corpus carrying out a logistic regression model with 17 phonological, syntactic and frequency-based variables. Unlike monofactorial approaches, the authors’ multifactorial models allow them to compare the explanatory power of individual factors in terms of effect strength. They conclude that there is a stronger syntheticity bias in comparatives than in superlatives. Moreover, analytic variants are more often used as a support strategy in cognitively complex environments with both comparatives and superlatives. This finding is in keeping with Rohdenburg's (Reference Rohdenburg1996) Complexity Principle and Mondorf's (Reference Mondorf2009, Reference Mondorf, MacWhinney, Malchukov and Moravcsik2014) notion of analytic support, i.e. the more explicit, easier-to-process analytic form is used as a compensation strategy in contexts which are cognitively more demanding. However, interestingly analytic support is used more often with comparatives than with superlatives – and there is one interesting reversal to the predicted trend: the more frequent an adjective, the more likely it is to take the synthetic -est variant (fullest) rather than the analytic most variant (most full). Cheung & Zhang also show that the variables that outrank other predictors are the phonological factors length (measured in the number of syllables) and -y as final segment. Syntactic factors, such as the position of the adjective and the presence of complements, are weaker predictors for both comparative alternation and superlative alternation. We are hence dealing with two morphosyntactic alternations that are primarily constrained by phonological factors.

All in all, the present volume subsumes an array of hitherto uncharted and unrelated variation phenomena under the notion of support strategies, to investigate these variation phenomena empirically and to offer new, more exhaustive explanations. This way, support strategies provide the common denominator for a series of variation phenomena which are investigated in Present-day English (see Wiechmann & Kerz, Cheung & Zhang, Stange, Rohdenburg, Kaatari, this issue) as well as from a diachronic perspective (see Eitelmann, Lorenz, Mondorf & Schneider, this issue). While these variation phenomena, at first glance, seem unrelated, they share the commonality that they are all affected by processing demands exerted by different types of cognitive complexity.