7.2.1 The Concept of Register

Register is a concept that traditionally exists at an intersection of several linguistic disciplines, such as sociolinguistics, discourse studies, literary and linguistic stylistics, text-linguistics, applied linguistics, and the study of language for specific purposes. All these disciplines share an interest in the study of language use, addressing linguistic variation with a focus on the discourse situation, for example its formality, purpose, preferences or traditions of style, or specific topics and audiences. With the concept of register, linguists thus aim to recognise, for instance, that ‘people speak differently depending on whether they are addressing someone older or younger, of the same or opposite sex, of the same or higher or lower status …; whether they are speaking on a formal occasion or casually, whether they are participating in a religious ritual, a sports event, or a courtroom scene’ (Ferguson Reference Ferguson, Biber and Finegan1994: 15).

For more than a decade now, the study of register variation has also been recognised as a discipline of its own. Aiming at the precise and systematic description of linguistic features associated with different situations of language use, register studies nowadays propose a systematic framework for the linguistic analysis of textual variation. The starting point is a comprehensive set of situational parameters that provide the template for register classification. Following Biber and Conrad’s (Reference Biber and Conrad2019) Register, genre, and style, this set comprises six major situational characteristics: the discourse participants and their relations, the channel, circumstances, and setting (i.e., time and place) of both language production and comprehension, and the purpose and topic of the discourse produced (Biber & Conrad Reference Biber and Conrad2019: 40). Based on this framework, registers are text varieties with specific linguistic characteristics arising from these core components of the discourse situation. For identifying and classifying registers, the properties of the situation are thus ‘more basic’ than their linguistic characteristics (Biber & Conrad Reference Biber and Conrad2019: 9), which are both frequent and pervasive because they are functional for the situation. This means that this understanding of register variation is closely associated with the frequency-based approach to (non-)canonicity laid down in the Introduction to this volume. However, since some functional associations are also described as non-basic patterns (e.g., as non-basic patterns of word order resulting from information structure), the approach also relies on theory-based assumptions.

Biber and Conrad’s approach highlights that linguistic features, which include non-canonical constructions, are conditioned by the context, not the other way round. For example, the real-time production mode of spoken, conversational registers is typically associated with the occurrence of forms of ellipsis. However, as the work by Biber, Wizner, and Reppen in Chapter 8 of this volume shows, structurally reduced clauses also occur in settings with mixed characteristics of speech and writing. The language of news broadcasts they deal with is mostly planned and scripted but received in real time, which are properties that account for the mix of reduction and complexity features they observe. Situational characteristics are thus logically prior to linguistic features, and register analysis focuses on such ‘functional associations’ (Biber & Conrad Reference Biber and Conrad2019: 10).

Registers can be explored at very different levels of specificity, ranging from quite general text varieties, like conversation, news, and academic prose, to more specific sub-registers, such as scripted face-to-face conversation, for instance the dialogue in TV shows (e.g., Quaglio Reference Quaglio2009), news in social media (e.g., Liimatta Reference Liimatta2019; see also Clarke Reference Clarke2022 or Scheffler et al. Reference Scheffler, Kern and Seemann2022), or subtypes of academic discourse, which can be as diverse as introductions to research articles or office hour consultations. Two studies in this part of the volume, Chapter 8 by Biber et al. and Chapter 9 by Pham, deal with more general registers (news, reviews) but ultimately look into more specific sub-registers (television news broadcasts; printed, spoken, and online reviews). Often, a complex interplay of situational factors is required for pinning down functional associations. Pham’s study of clefts in evaluative language, looking at reviews from different media, refers to the situational parameter of channel of communication, but also suggests that the parameter of purpose is particularly relevant.

It needs to be emphasised that this framework of register analysis does not easily deal with all register constellations. Problems of register definition arise from cases (1) where there is a lot of variation among texts within one register, (2) where texts from different registers share many characteristics, or (3) where texts possibly do not belong to a register at all (cf. Biber & Egbert Reference Biber and Egbert2023). An example of the first category that has recently been discussed in the literature is student academic writing (Goulart et al. Reference Goulart, Biber and Reppen2022; Biber & Egbert Reference Biber and Egbert2023: 10). The texts discussed in these studies typically involve more than one communicative purpose; they contain an almost equal amount of text passages that either explain or argue. In a similar vein, the register of conversation typically includes discourse varieties as diverse as joking around, engaging in conflict, or giving advice (Biber et al. Reference Biber, Johansson, Leech, Conrad and Finegan2021). A classic example of case (2), cross-register similarities, is the register of fiction, with novels, in particular, typically containing a mix of narration and speech. It has been an issue both of debate and empirical analysis in register analysis if fiction is one register or (at least) two (Egbert & Mahlberg Reference Egbert and Mahlberg2020). As for case (3), the issue of ‘texts-with-no-register’ is still open for discussion. Some scholars find that such texts are ‘more prevalent than we might believe’ (Biber & Egbert Reference Biber and Egbert2023: 18; also Biber & Egbert Reference Biber and Egbert2018) but, since corpora are usually based on pre-classified registers, corpus-based register analysis tends not to target this issue in particular. However, Biber and Egbert’s (Reference Biber and Egbert2018) study of web documents finds many ‘hybrid’ texts that users did not identify as belonging to a particular register, for example, due to their containing an inseparable mix of description and both personal and commercial persuasion. In experimental studies participants are typically not asked to produce a specific register (corresponding to a real-life situation). The study on particle placement by Günther in this part of the volume shows that the choice between continuous and discontinuous particle verbs results from cognitive complexity, thus highlighting that sentence- or discourse-internal parameters are also relevant for syntactic variation.

7.2.2 Non-Canonical Syntax and Discourse

Registers are text varieties (i.e., units of language use), which turns the syntax within them into utterances rather than mere units of grammar. As an utterance, a (canonical or non-canonical) sentence must be seen as tied to its discourse in two possible ways. Since discourse is, formally, any element larger than the sentence and, functionally, language use with a given purpose (Dorgeloh & Wanner Reference Dorgeloh and Wanner2023: 16), any construction is bound to its discourse by the surrounding text (the co-text) as well as by the discourse situation (the proper context).

As for the role of co-text, syntactic choices can be explained by factors such as information packaging, topic, focus, or processing load – all factors in which the text surrounding a construction plays a role. For example, Günther in Chapter 10 of this volume looks at particle placement in sentences in isolation. In longer texts, the placement of the object NP in front of or behind the particle also depends on the co-text since it will impact on the information status of the NP (Lohse et al. Reference Lohse, Hawkins and Wasow2004). In addition, aspects of the context are also relevant for particle placement; for example, there is an effect of speakers’ intentions, like emphasising what is important to them (Dehé Reference Dehé2002). Other non-canonical constructions are often primarily explained by context rather than the co-text, for instance, that-omission, or the passive voice. For example, zero-that clauses (I heard you were sick) are more likely in speech than in writing (Biber Reference Biber2012), while the passive voice, especially the long passive, is more common in writing (Biber et al. Reference Biber, Johansson, Leech, Conrad and Finegan2021: 929). However, properties of the matrix clause, such as a first- or second-person pronoun and certain verbs (e.g., verbs of sensation), also play a role in that-omission (Thompson & Mulac Reference Thompson and Mulac1991), and the givenness of information in the subject vs. the agent by-phrase is also relevant for the use of the passive voice. According to Biber et al. (Reference Biber, Johansson, Leech, Conrad and Finegan2021: 932), 90% of the by-phrases in long passives express new information, while Birner (1996, Reference Birner2018) shows that English long passives are subject to similar information-structuring constraints as, for example, inversion (e.g., Above his head hung a massive seagull with its beak open; Birner Reference Birner2018: e159). So, ultimately, we need to consider both co-text and context together to understand the use of non-canonical constructions as utterances within discourse.

7.2.3 The Study of Non-Canonical Syntax in the Context of Register

The relevance of discourse, in the form of co-text and context, motivates another core distinction for the study of syntax and the role register plays therein. There are two distinct approaches to exploring this role, which differ fundamentally in how register is conceptualised. One approach is rooted in so-called variationist linguistics and looks at syntactic variation as different ways of ‘accomplishing the same function’ (Szmrecsanyi Reference Szmrecsanyi2019: 277). For example, the variationist approach aims to understand the syntactic choice between that- and that-less complement clauses, or between active and passive voice, by focusing on ‘constraints’ that govern which variant will be chosen (Szmrecsanyi Reference Szmrecsanyi2019: 78). Register will be one crucial factor here, which means it is a predictor variable in that kind of work (Dorgeloh & Wanner Reference Dorgeloh and Wanner2023: 32–41).

Register has an alternative role in the so-called text-linguistic approach, which is based more directly on the register analysis framework. As we have explained in the previous section, this approach looks at the frequency of linguistic characteristics and explains them with reference to their ‘functional correspondences’ with the discourse situation (Biber & Egbert Reference Biber and Egbert2023: 4). For example, the passive is relatively more frequent in news and academic texts because these are texts that focus on events or generalisations rather than individual agents (Biber et al. Reference Biber, Johansson, Leech, Conrad and Finegan2021: 930). In that way, the approach uses syntactic characteristics for describing registers, which turns the role of register into one of a proper ‘object of investigation’ (Biber Reference Biber2012). Here, the description of registers is based on counting features, with a focus on those features that have a higher rate of occurrence in one register compared to others. A non-canonical construction thus becomes a register feature when it has a higher frequency (relative to text length or corpus size) in one specific register than in other registers. When certain features do not occur in any other register, they become register markers (Biber & Conrad Reference Biber and Conrad2019: 51). The studies by Biber et al. and Pham in this part of the volume mainly use the text-linguistic approach, presenting (normalised) frequencies across several registers, while Günther’s experimental study exemplifies the variationist approach, comparing the participants’ reactions to different syntactic variants.

Both approaches require usage-based evidence, typically in the form of corpus-based frequencies. These corpora need to be both representative and comparative (Biber & Conrad Reference Biber and Conrad2019: 10): As we emphasised in the discussion of register classification, register differences are often only gradient, since many registers are only more (or less) alike or different. For example, the study by Biber et al. in this part of the volume, dealing with the complexity of reduced structures in news broadcasts, touches on similarity with spoken discourse (reduction in general) but, in focusing on multiple reduction, there are also similarities with other written ‘economy’ registers. From a methodological viewpoint, syntactic features become register features if they are found to occur more frequently in the given register compared to at least one other (cf. Biber & Conrad Reference Biber and Conrad2019: 215).

The necessity for a comparative approach explains that corpus linguistics is the primary method in register studies. Depending on the specificity of the target registers, people either use freely available corpora or design their own to match a specific research question. Corpus size will vary depending on the frequency with which features occur. The study by Pham in this part of the volume looks at clefting, which is relatively rare, and uses a corpus of 310,000 words from six registers. By contrast, the TV news broadcast corpus compiled for Biber et al.’s study has a size of approximately 50,000 words covering three networks, further split into four text sections for an analysis of sub-registers. This corpus size is sufficient because reduced structures occur throughout the texts, which means they are not only more frequent than elsewhere but also pervasive in that register (cf. Biber & Conrad Reference Biber and Conrad2019: 9). As we pointed out earlier, experimental work typically lacks a specific register context, but instead usually strives for controlling for the co-text as variable. In this way, Günther in this part of the volume explores conditions of cognitive complexity and their effect on particle placement using the methods of a self-paced reading and a split rating task.

A conceptually and methodologically more complex approach to register analysis is the so-called multi-dimensional (MD) analysis, first introduced in Biber (Reference Biber1991). The method originally started out with a group of 67 lexical and grammatical features, which, by way of a computational factor analysis, were turned into a set of five dimensions of textual variation, identified in a large corpus of texts. Subsequently, these dimensions were labelled for reflecting certain functional properties of the texts, such as being ‘involved’ or ‘informational’, or ‘abstract’ or ‘non-abstract’. For example, as for some syntactic features we looked at before, Dimension 1 (involved vs. informational discourse) contains a higher density of that-deletion, while it is marked by a low density of agentless passives and of what is here called ‘deletions’ (Biber Reference Biber1991: 104–8). The main point of the approach is to provide quantifiable and generalisable descriptions of registers (Biber & Conrad Reference Biber and Conrad2019: 216), and the approach has established a broad research area over the last thirty years. The method goes beyond describing registers by individual features, but focuses on their patterns of co-occurrence. Within a dimension, syntactic characteristics are always part of a group of positive or negative features: for example, in an MD analysis of university registers, that-omission co-occurs with other ‘oral’ features such as contractions or wh-questions and with a negative dimension score for agentless passives (Biber & Conrad Reference Biber and Conrad2019: 228). In this way, the method manages to cover a larger set of linguistic features and to reveal similarities and differences across registers. Biber et al. (Chapter 8 in this volume) also discuss their findings on multiple reduction in the context of other ‘phrasal’ features belonging to the Dimension 1 characteristics of radio broadcasts.

7.3.1 Reduction

Grammatical reduction, the simplification of grammatical structure (and with it the reduction of number of words per utterance), is a pattern characteristic of spoken registers, particularly conversation. Biber et al. (Reference Biber, Johansson, Leech, Conrad and Finegan2021: 1037) list pronouns, other proforms, and ellipsis among the most common forms of grammatical reduction occurring in this register. The use of reduction here is mainly due to situational parameters. Firstly, since conversation happens in real time and is interactive, it draws significantly on cognitive resources. Grammatical reduction can be seen as a strategy to reduce the demand on those resources. Secondly, the shared situational context of a situation allows speakers to underspecify what is said explicitly and to rely on the shared experience of the situation (and often also on shared knowledge). As a result, conversation includes many utterances that are non-clausal units, which is one of the reasons that the syntactic make-up of conversations looks very different from that of written registers. In their chapter on ‘the grammar of conversation’, Biber et al. (Reference Biber, Johansson, Leech, Conrad and Finegan2021: 1065) found that 38.6% of units in a sample of American and British English conversation (1,000 turns, 5,369 words) were non-clausal. Below is an example that shows how they count speech units (separated by a double line) and classify them as clausal (<Cl>) or non-clausal (<NCl>).

In this example, all non-clausal units are treated the same, but in their contribution to this volume, Biber et al. make the distinction between non-clausal units without any syntactic structure, which include discourse markers like speaker C’s last utterance in the example above (Oh), and non-clausal units with internal syntactic structure, such as speaker A’s response The alligator in the ocean. They subsume the latter, as well as units that include a verb, under the category of ‘non-canonical reduced structures’. In their study, they look at such structures specifically in news broadcasts, looking into the parameters of the register that favour such expressions. Another example of reduced structures is the case of subject ellipsis, which is characteristic of spoken registers like conversation (Narimaya Reference Nariyama2004) and written registers like diary writing (Haegeman Reference Haegeman2013), texting (van Dijk et al. Reference Dijk, van Witteloostuijn, Vasić, Avrutin and Blom2016), and blog writing (Teddiman & Newman Reference Teddiman and Newman2007). Other ways to reduce or shorten an utterance include auxiliary contractions (hasn’t, we’ll) and acronyms. Both typically occur in registers that prize short forms due to high interactivity.

In most of these cases, it is clear which syntactic material has been omitted, and the non-canonical, reduced form can be considered a variant of the canonical, non-reduced form, which opens the path for a variationist design. However, as Biber et al. point out in their contribution to this volume, there are also reduced units for which it can be hard to say what the non-reduced form would have been, especially in cases where an utterance is constituted by a single noun phrase or adjective phrase. Structures reduced in this way do not lend themselves to a variationist approach, since there is no clear comparison of two variants. In example (1) above, we cannot say with certainty which canonical sentence C’s utterance Really? is the reduced version of. The variationist research design is best applied when it is clear which elements exactly have been reduced or deleted, as in the case of subject or object ellipsis. Since this does not hold for all of the data discussed here by Biber et al., their analysis follows the text-linguistic design.

7.3.2 Expansion

Non-canonical syntactic constructions that arise from making a canonical sentence longer and/or more complex include, for example, it-extraposition (2), it-clefts (3), wh-clefts (4), and left dislocation (5).

1. 1. a. It is bad to have such sharply diverging classes. (Corpus of Contemporary American English, COCA, News)

   2. b. To have such sharply diverging classes is bad.

1. 1. a. It was Democrats who killed the DREAM act. (COCA, Spoken)

   2. b. Democrats killed the DREAM act.

1. 1. a. What you should not do is prescribe carrot juice (COCA, Blog)

   2. b. You should not prescribe carrot juice.

1. 1. a. This guy, … he is an extra brand of crazy. (COCA, Spoken)

   2. b. This guy is an extra brand of crazy.

The distribution of these constructions is different from the reduced structures just discussed. They are found in all types of registers, but their uses are quite register-specific and, seen across registers, they are quite rare. For example, subject it-extraposition, as in (2a), is found to occur, at a rate of about two times per 1,000 words, in both academic and popular writing (Zhang Reference Zhang2015), while left dislocation, as in (5a), occurs almost exclusively in conversation (around two times per 10,000 words) and only ‘occasionally’ in fictional dialogue or written prose (Biber et al. Reference Biber, Johansson, Leech, Conrad and Finegan2021: 948). The choice of a longer or more complex form is also discussed as affecting processing. For example, psycholinguistic work on reading comprehension shows that readers utilise it-clefts to anticipate the way in which information will be packaged next (e.g., Alemán Bañón & Martin Reference Alemán Bañón and Martin2019).

Sentences that lend themselves to it-extraposition (esp. sentences with a subject clause and an adjectival predicate, as in (2b)) occur more often in writing than in spoken registers (Kaltenböck Reference Kaltenböck2005), but, from a variationist angle, it is true for both written and spoken registers that in the vast majority of syntactic environments that allow it-extraposition the construction is applied. The reasons for choosing the construction usually lie in the co-text. Based on 1,701 instances of examples extracted from the British component of the International Corpus of English (ICE), Kaltenböck showed that in almost three out of four occurrences of it-extraposition the extraposed subject clause contains new information, which aligns with general ideas about information packaging.

Register-based research on cleft sentences is usually built on a small set of examples, and researchers look at the syntactic category and the information status of the foregrounded element (e.g., Hedberg Reference Hedberg1990). In Chapter 9 of this volume, Pham takes a different approach, focusing on how the use of cleft sentences may be influenced by situational factors, particularly the communicative purpose of evaluation. She found that, while most cleft constructions themselves are evaluative (a concept that is not as easy to code as syntactic categories like ‘clausal’ above), clefts do not occur more frequently in texts that clearly have an evaluative purpose. She hypothesises that clefts, which are non-canonical sentences by syntactic criteria, could be considered ‘less non-canonical’ if one looks into the function of the construction. This shows again that a syntactically non-canonical pattern can be the functionally canonical one.

7.3.3 Placement Variation

English is considered a strict word order language in which the arguments of the verb (like agent, theme, goal) are mapped onto syntactic positions predictably. Non-canonical sentences include those in which the expected arrangement is not followed. For example, it is generally assumed that canonically a direct object will directly follow the verb (Huddleston & Pullum Reference Huddleston and Pullum2002: 247), as in (6a). Sentences in which the direct object occurs after an adjunct, as in (6b), a case of heavy NP-shift, are non-canonical.

1. 1. a. He praised me for mastering the trills quickly. (COCA, Fiction)

   2. b. He was by all accounts a prodigy and mastered quickly the hard-earned lessons that most shipwrights spent a lifetime accumulating. (COCA, Magazine)

Another example of placement variation are passive constructions, which break the expected alignment of agent and subject position, resulting either in a sentence in which the agent is not realised at all (short passive) or a sentence in which a non-agent is placed in subject position and the agent becomes an adjunct inside a by-phrase (long passive). Much of the research on the occurrence of the passive as a non-canonical construction is text-linguistic in nature. The use of passive constructions is known to vary considerably across registers; for example, they are a frequent and pervasive characteristic of academic writing (Biber et al. Reference Biber, Johansson, Leech, Conrad and Finegan2021). The reasons for that register specificity lie, to some extent, in the purpose and genre conventions of academic writing. Texts tend to be about processes, data, and discoveries rather than about the people who make them. The passive is one way to place a non-agent in the subject position of a sentence, that is, in the syntactic position most clearly linked to the discourse function of topic. Long passives are quite rare, even in academic writing. The main reason for that is that there is a long tradition, especially in the natural sciences, of presenting research as a disembodied endeavour. The passive provides the option to leave an agent unexpressed, and academic writing is a register in which this option is often considered desirable, especially if the agent is the author of the text, despite modern style manuals explicitly taking the position that the use of the first person (I will show instead of It will be shown) is clearer and intellectually more honest. As pointed out above, if passives do occur with by-phrases, the by-phrase tends to express new information or at least information that is less given than the subject (Biber et al. Reference Biber, Johansson, Leech, Conrad and Finegan2021: 933).

Like most syntactic constructions, non-canonical or not, the use of passive constructions can be examined from a variationist or a text-linguistic perspective. Biber et al. (Reference Biber, Johansson, Leech, Conrad and Finegan2021), for example, take a text-linguistic approach and measure the use of the passive by tokens per X number of words in a corpus. A different approach is chosen by Seoane (Reference Seoane, Dalton-Puffer, Kastovsky, Ritt and Schendel2006), who studies the frequency of passives from a variationist perspective, counting the percentage of transitive verbs realised in active vs. passive voice.

Like reductions, the use of placement alternations and the preference for a non-canonical pattern in a given discourse situation is also explained from a cognitive angle. Following the influential work by Hawkins (Reference Hawkins1994, Reference Hawkins2004), it is generally assumed that speakers, when given a choice, prefer constructions that can be processed more efficiently. Hawkins’ principle of ‘minimising domains’ allows him to calculate which structures are more efficient in that regard. Simply put, the question is how much syntactic material (the ‘domain’) must be processed before the overall structure of the verb phrase is determined. Changing the word order in a sentence can be a way to minimise that domain.

Building on Hawkins’ work, Günther (Chapter 10 in this volume) looks at word order variation in the case of particle verbs (look up the information/ look the information up). Generally, the continuous word order (verb-particle) is considered the canonical pattern, the one that minimises the ‘domain’ that must be processed to figure out the meaning of the verb. If that is the case, it is not immediately clear why the non-canonical pattern, the one that is supposed to be less efficient to process, is more common in spoken language, a register that is supposed to favour the minimising of processing burdens, and that the canonical form cannot be chosen at all if the object of the verb is a pronoun (*look up it). Günther makes the important observation that the cognitive load for language production is not necessarily the same as for language perception. She chooses experimental data to explore the difference between minimising one’s own cognitive load as speaker vs. reducing the cognitive load for the hearer.