17.1 Introduction

Construction Grammar and typology appear to be quite different approaches to syntactic analysis but, in fact, they share important fundamental assumptions. At the time of writing, however, the only major effort at combining the two approaches is Radical Construction Grammar (Croft Reference Croft2001, Reference Croft2022). Most Construction Grammar research has been done in a single language, usually one of the well-known European or East Asian languages, rather than as a cross-linguistic analysis. Conversely, most typological analysis is not done in an explicitly constructional approach, although informally most typological analysis is fairly straightforwardly interpretable as constructional. Integrating the two approaches is fruitful for both Construction Grammar and typology.

17.2 Parallels between Construction Grammar and Typological Theory

Perhaps the single most important tenet of Construction Grammar is that the basic unit of grammatical structure is a pairing of form and function, namely, a construction (a symbolic unit in Cognitive Grammar).Footnote ¹ ‘Form’ includes phonological, morphological, and syntactic form. ‘Function’ includes meaning but also categories described as ‘discourse-functional’, ‘pragmatic’, or ‘information structure’; we will use the term ‘information packaging’ here (see Section 17.6). This approach contrasts with most other approaches to grammar (apart from Cognitive Grammar).Footnote ² Those other approaches tend to be modular, that is, they are made up of distinct ‘levels’, each of which is self-contained and contains only one type of grammatical information: phonological, syntactic, semantic, etc. Fillmore, Kay, and O’Connor’s seminal paper in Construction Grammar (Fillmore et al. Reference Fillmore, Kay and O’Connor1988) argues that the syntactic, semantic, pragmatic, and even phonological and prosodic idiosyncrasies of particular constructions mean that highly general rules linking autonomous syntactic and semantic modules cannot capture the vast majority of a speaker’s knowledge about their language (see Section 17.9).

Constructions are basically the same type of unit, a form–function pairing, at all levels of complexity – from atomic units (words and morphemes) to phrases to complex sentences, and at all levels of schematicity – from completely substantive units such as It takes one to know one to completely schematic units such as the ditransitive construction [Sbj Verb Obj1 Obj2] (Goldberg Reference Goldberg1995, Reference Goldberg2006; Croft Reference Croft2001). In other words, lexicon and syntax are not treated as separate components of the grammar, and idioms and other multi-word expressions are syntactic structures, as well as schematic structures without any specific lexical content. Of course, there are differences between syntax and lexicon but the differences can be attributed to the parameters of complexity (atomic vs. complex) and schematicity (substantive vs. schematic, not to mention a mix of the two as in most idioms). In contemporary Construction Grammar, unlike traditional grammar, ‘construction’ covers all form–function pairings.

Typology starts from the same two assumptions but for different methodological reasons. Since Greenberg’s seminal word order paper (Greenberg Reference Greenberg and Greenberg1966), the primary basis for cross-linguistic comparison in typology has been the equivalence in function of the morphosyntactic forms compared across different languages. That is, typology begins methodologically with a pairing of form and function. Typology starts from equivalence of function because the great diversity of linguistic forms precludes using form as the starting point for cross-linguistic comparison (see Section 17.5). Both Construction Grammar and typology use the form–function pairing to get a handle on the high degree of grammatical variation: variation within languages for most applications of Construction Grammar, and variation across languages for typology.

Typology treats the form–function pairing as basic and the variation in formal encoding of the function as secondary. For example, for a desiderative/optative construction, one language may express this function syntactically, while another language expresses it lexically, with a single word; compare the single-word Mongolian expression in (1) with its periphrastic English translation (Janhunen Reference Janhunen2012: 155):

1. (1) ir-eesai

   come-des

   ‘If only (s/he) would come!’

Typologists are basically doing a constructional analysis, in the broad sense of ‘construction’, when they do a ‘typology of relative clauses’ or a ‘typology of secondary predication’. Like construction grammarians, typologists usually end up identifying a family of closely related functions and a family of morphosyntactic forms that express those functions, rather than a single function and a single form.

17.3 The Morphosyntactic Structure of Constructions in Typological Perspective

All human languages are general-purpose communication systems. They can be used to communicate any experience that a speaker wishes to convey in order to achieve her goals in interpersonal interaction. Yet each language is unique, using its own set of constructions (including words) that arose in the speech community’s culture over its history (Section 17.11).

Morphosyntactic typology seeks to understand the relationship between general-purpose communication and culture-specific linguistic forms by looking for patterns of forms encoding functions across languages. Due to the great diversity of morphosyntactic form across languages, the only formal structure that is common to all constructions across languages is the part–whole relation between a construction (the whole) and the roles defined for its elements (the parts; see Section 17.7). By coincidence, this is the basic structure of a construction specified by the Berkeley FrameNet Constructicon project (Fillmore et al. Reference Fillmore, Lee-Goldman, Rhomieux, Boas and Sag2012): the construction as a whole and its construction elements (CEs).

In addition, typological diversity leads one to the conclusion that constructions and the categories that a construction’s CEs define are language-specific – the result of historical processes in a speech community (Section 17.11). In fact, the categories that a construction’s CEs define are also specific to that construction (Section 17.7). Both of these conclusions are controversial, even among typologists (although much typological practice can be interpreted as compatible with these conclusions). The controversy arises in part because the language-specific nature of constructional form and CE categories means that another basis is required to compare languages and develop universals of grammar (Section 17.5). These conclusions, drawn from typological patterns, are the foundational hypotheses of Radical Construction Grammar (Croft Reference Croft2001).

17.4 Syntactic Categories, Distributional Analysis, and Constructions

Many non-constructional theories of syntax are built around systems of syntactic notation: a particular inventory of category labels for lexical and phrasal categories, as well as a particular inventory of types of relations between words and/or phrases in a sentence. Constructions are defined in terms of a particular configuration of categories or elements with their relations to each other. The theoretical assumption behind this notation can be called the ‘building block model’ (Croft Reference Croft, Trousdale and Hoffmann2013, 2024): Complex constructions are assembled from the ‘building blocks’ of the categories and relations in the inventory.

These inventories are intended to apply across all languages. They make up what is often called Universal Grammar. Some theoreticians allow languages to do without some of the categories and relations in the inventories (the ‘smorgasbord’ or ‘cafeteria’ approach to Universal Grammar). The universal applicability of the inventories of categories and relations can be called the ‘skeleton model’ (Croft 2024): All languages share the same skeleton of morphosyntactic categories and relations, even if some languages leave some of them out.

There exist several of these systems of syntactic notation, also called ‘frameworks’, because linguists do not agree on what the proper inventory of categories and relations needs to be in order to analyze the structure of all constructions across all languages. Also, individual frameworks change over time, which leads to difficulties in understanding the syntactic notation of older versions of the frameworks or frameworks that have been abandoned, as well as unfamiliar contemporary frameworks. Cross-linguistic diversity suggests that the assumption of Universal Grammar as a finite inventory of categories and relations should be discarded and that a framework-free grammatical theory (Haspelmath Reference Haspelmath2010) is more suitable for a model of grammatical structure that applies to all languages.

The problem that arises is that it is not always obvious that a particular category is the correct description of a role in a construction, or the correct description of a construction as a whole, in another language. A common problem is whether or not a language has a category that is labeled ‘adjective’. Many languages are said not to have adjectives. In some cases, linguists disagree as to whether a particular language has adjectives (Haspelmath Reference Haspelmath, Graf, Paperno, Szabolcsi and Tellings2012).

How are such problems supposed to be resolved? The answer, in contemporary terms, is supposed to be provided by syntactic argumentation. For example, (2)–(5) are intended to be arguments supporting the hypothesis that English has both adjectives and verbs:

1. (2) Heather sang.

1. (3) Heather is tall.

1. (4) the tall woman

1. (5) the woman that sang.

The first argument is that an adjective like tall takes a copula in (3), while a verb like sang does not, and in addition inflects for tense, as in (2). The second argument is that an adjective like tall is simply preposed to woman in (4), while a verb like sang is postposed and accompanied by the relativizer that in (5). These differences, and the unacceptability of switching the words around in (2)–(5) (*Heather talled; *the sang woman, etc.), make the case that English has both adjectives and verbs.

The facts in (2)–(5) are called arguments, evidence, tests, criteria in favor (or against) a syntactic analysis for a language (Croft Reference Croft2022: 9). This method of syntactic argumentation has a long history and in mid twentieth-century American structuralism was called distributional analysis. These different terms obscure the basic fact: Syntactic argumentation is essentially the occurrence (or non-occurrence) of a word or phrase in a construction. The constructions in (2)–(5) are described in (2′)–(5′), with a gap to indicate the role or CE being used to define the categories ‘adjective’ and ‘verb’ in (2)–(5):

1. (2′) Intransitive (Declarative): Heather __

1. (3′) Copular (Declarative): Heather is __

1. (4′) Adjective Modification: the __ woman

1. (5′) Subject Relative Clause: the woman that __

Distributional analysis supports the Construction Grammar principle that constructions are grammatical units, because it makes essential use of constructions. In fact, distributional analysis presupposes the existence and identity of constructions. We will return to this point in Section 17.7.

The issue with distributional analysis for cross-linguistic comparison is that categories such as adjective and verb are determined for English using English constructions. When we turn to another language, say, Lao, its categories are determined using Lao constructions, not English constructions. As a result, the syntactic argumentation does not carry over from English to Lao. One cannot simply assume that the categories in Lao are the same categories that are found in English, in this case, adjective and verb.

This would not be a problem if English, Lao, and all other languages had exactly the same set of constructions and the roles of those constructions defined exactly the same categories. Then one could safely say that the inventory is universal. But that is not the case; compare (6)–(9) from Lao to (2)–(5); all Lao examples are from Enfield (Reference Enfield2007), with page references given after the translation (numbers indicate tone, 0 = toneless):

1. (6)

   saam3	khon2	taaj3
   three	person	die
   ‘Three people died.’ (273)

1. (7)

   khuaj2	phen1	dam3
   buffalo	his	black
   ‘His buffalo is black.’ (252)

1. (8)

   lot1	dam3
   vehicle	black
   ‘black car’ (263)

1. (9)

   khon2	thii1	ñaang1
   person	rel	walk
   ‘the person who walks’ (246)

Lao uses seemingly the same construction for the translation equivalents of ‘die’ and ‘black’. Lao does have a copula construction but it is used for translation equivalents of English nouns:

1. (10)

   khaw3	pên3	nak0-hian2
   they	cop	ct.agt-student
   ‘They were students.’ (285)

Also, while *the sang woman, without a relativizer and preposed like an adjective, is unacceptable in English, Lao translation equivalents of English verbs may occur without a relativizer (and without a participial derivation, unlike the English translation):

1. (11)

   maa3	haw1
   dog	bark
   ‘barking dog’ (253)

Lao may have constructions similar to English (this is in fact not entirely clear; for example, there are two different copulas in Lao with different functions; Enfield Reference Enfield2007: 284–288). But the distribution of words in those constructions is quite different.

Nevertheless, common practice is to assume that categories such as adjective, verb, and copula in English can be carried over to other languages like Lao; and constructions are found that will fit what the analyst expects. Categories in another language, such as Lao, are equated with similar categories in English. To a lesser extent, constructions in another language are equated with similar constructions in English.

In practice, categories and constructions across languages are being equated by virtue of semantic similarity, despite the apparent use of distributional facts. This has been clear from the beginning of modern morphosyntactic typology, as shown by the following quote from Greenberg’s seminal word order paper:

But semantic similarity is not distributional analysis. Distributional analysis is language-specific – determined by roles in constructions in a specific language. Semantics can serve as the basis for cross-linguistic comparison, but it is not directly related to distributions within a language.

17.5 Constructions as Comparative Concepts

This problem has engendered a lively debate in typology on the nature of ‘comparative concepts’ (Haspelmath Reference Haspelmath2010), that is, concepts that are valid for cross-linguistic comparison (see also Croft Reference Croft2003: 6–19). From Greenberg (Reference Greenberg and Greenberg1966) onward, typologists have assumed that the basis for cross-linguistic comparison must be semantic. There are also properties of morphosyntactic form that are cross-linguistically valid, such as word order or the presence vs. absence of morphemes coding a function (Haspelmath Reference Haspelmath2010). However, these properties of morphosyntactic form are not distributional.

Construction Grammar provides a means to resolve this problem. Constructions, including words, are pairings of form and meaning. Hence, when a typologist is comparing, say, intransitive predication constructions across the world’s languages (Stassen Reference Stassen1997), they are comparing constructions that express the same function and looking at how they are similar or different in morphosyntactic form. Thus, ‘construction’ can serve as a comparative concept: any morphosyntactic form in any language that expresses a particular function (Croft Reference Croft2022: 17).Footnote ³

Typologists regularly use another type of comparative concept to describe different types of morphosyntactic form found across languages to express a function: ‘strategy’ (e.g., Keenan & Comrie Reference Keenan and Comrie1977; Givón Reference Givón1979; Stassen Reference Stassen1997). A strategy is a subclass of a construction: those constructions that express a particular function but employ a particular type of morphosyntactic form (Croft Reference Croft2022: 19).

Strategies can be classified into three broad types (Croft Reference Croft2022: 19–22). Encoding strategies are simply different ways of encoding function into morphosyntactic form, including word order or presence vs. absence of a morpheme (or morphemes) encoding a particular function. Examples include the word order strategy for the polarity question construction in English (Are you married?) vs. an overt question particle strategy in the polarity question construction in K’ichee’ (formerly known as Quiché) Mayan (Lā at culanic?; Mondloch Reference Mondloch1978: 21). These strategies, like all strategies, are defined in terms of cross-linguistically valid properties of form, not distributional facts.

A more complex type of strategy is a system of strategies for two or more constructions. In a system of strategies, the strategies are defined in terms of similarities and differences in morphosyntactic form between the constructions in a language. For example, alignment strategies are defined in terms of the relationship between the encoding of the arguments in a transitive construction to the one argument in an intransitive construction. In the accusative alignment of English, the transitive subject (labeled A by typologists) is encoded in the same way as the intransitive subject (S), and the transitive object (P) is encoded in a different way (the accusative). In the ergative alignment of Yuwaalaraay, it is the P argument that is encoded in the same way as the S argument (absolutive), and the A argument is encoded in a different way (the ergative; Williams Reference Williams1980: 36):

1. (12) The snake_nominative bit the man_accusative.

1. (13) The woman_nominative didn’t run.

1. (14)

   d̪uyu-gu	n̪ama	d̪ayn-Ø	yi:-y
   snake-erg	that	man-abs	bite-nfut
   ‘The snakeergative bit the manabsolutive.’

1. (15)

   wa:l	n̪ama	yinar-Ø	banaga-n̪i
   neg	that	woman-abs	run-nfut
   ‘The womanabsolutive didn’t run.’

The third and most important type of strategy (Section 17.10) is the recruitment strategy: The function of the construction in question is expressed by recruiting the morphosyntactic form of a related construction. For example, the presentation of an internal physical sensation such as hunger recruits the copular strategy of property predication in English (I’m cold) but in French it recruits the ‘have’ strategy of the French presentational possession construction (J’ai froid, lit. ‘I have cold’).Footnote ⁴ More generally, English recruits the form of the property predication construction, while French recruits the form of the presentational possession construction.

17.6 Constructional Function and Organization in Cross-Linguistic Perspective

Constructions and strategies as comparative concepts form the basis of typological analysis. Croft (Reference Croft2022) is a large-scale survey of constructions of the world’s languages and of the most common strategies used to express those constructions. The constructions are organized by function. The analysis of function is based on the hypothesis that function can be broken down into two dimensions: semantic content and information packaging.

The two-dimensional analysis of function originated in a typological analysis of parts of speech (noun, verb, adjective; Croft Reference Croft1991, Reference Croft2001). Semantics – nouns denote persons and things, verbs actions, and adjectives properties – has been treated as an unreliable guide for the analysis of parts of speech since, for example, a noun may denote an action (running) or a property (height). However, distributional analysis is not suited to the cross-linguistic analysis of parts of speech either, for the reasons given in Section 17.4.

For parts of speech, the information packaging dimension are the propositional acts (Searle Reference Searle1969; Croft Reference Croft1991) of reference, predication, and modification. These are of course categories that are already used in grammatical description, particularly for reference grammars of indigenous languages, but they are not usually considered to be information packaging functions. The nature of information packaging can be illustrated by the table of examples of propositional act functions and semantic classes in Table 17.1.

Table 17.1 illustrates three basic principles of information packaging. First, any semantic class can be packaged as reference, predication, or modification. Second, certain combinations are ‘privileged’ or prototypical: specifically, reference to objects, predication of actions, and modification by properties. Their special status is reflected in the minimal overt coding of the prototypical combinations in contrast to the non-prototypical combinations. Typologically, the universal pattern is that the non-prototypical combinations are encoded by at least as many morphemes as the prototypical combinations (Croft Reference Croft2003: 183–188). For instance, Lao conforms to the universal pattern even though property predication is zero-coded just like action predication, see examples (6) and (7). Third, the precise coding in a particular language is a convention of the speech community. It is a convention of Lao that it has recruited the action predication construction’s form (namely, zero coding) for property predication. Conversely, it is a convention of English that it has not done so, instead most likely recruiting the copula strategy of object predication (this convention is ancient in the history of English).

These properties of information packaging are precisely the properties of construal or conceptualization, a fundamental process in cognitive semantics (Talmy Reference Talmy1977; Langacker Reference Langacker1987: 116–137; Croft & Cruse Reference Croft and Cruse2004: chapter 3; Croft Reference Croft2012: 18). Construal of an experience serves the interlocutors’ goals in discourse. This is why any concept can be construed in almost any way. Construal is constrained by the nature of reality. This is why some construals are more common than others and encoded in a ‘privileged’ way. Finally, construal is constrained by cultural convention. This is why, for example, English and Lao use different strategies (zero vs. copula) for property predication, even though both languages have both strategies available; it is a fact of their linguistic history.

Croft (Reference Croft2007, Reference Croft2022) argues that all verbalization of experience involves some sort of information packaging of semantic content. In fact, syntax often conforms to information packaging more closely than the semantic content of the experience; hence the primary organization of the textbook (Croft Reference Croft2022) in terms of information packaging.

For example, Table 17.2 illustrates the information packaging options of degrees of salience represented by subject, object, and oblique argument encoding with respect to different semantic roles (Croft Reference Croft2022: 175, table 6.2). Again, any semantic role can be encoded as just about any grammatical role but there is a ‘privileged’ encoding of agent as subject, patient or theme as object, and other roles as obliques (this is the basic, ‘active’ or ‘direct’ voice). English does not typically encode agents as objects but the voice systems of Algonkian languages and Philippine Austronesian languages do.

The typological patterns of the argument encoding of different semantic types of events, different alignment systems, basic voice, and nonbasic voice (passive-inverse, applicative, and causative) can be largely accounted for by distinguishing semantic participant roles and degrees of topicality/salience of referents in those roles (Croft Reference Croft2022: chapters 6–9).

Another example is the construal of various semantic relations between events as a figure–ground or asymmetric relation (adverbial subordination) vs. a complex-figure or symmetric relation (coordination), illustrated in Table 17.3.

Yet again, either construal of the event-semantic relation is possible in English for all of the event relations, and they contribute to the explanation of their cross-linguistic variation (Croft Reference Croft2022: chapters 15, 17).

There are several major information packaging parameters. Most important is the reference-predication-modification distinction described above. Within reference, referents are packaged in terms of information status, particularly accessibility and identifiability. Within predication, dependent arguments are packaged in terms of topicality, as noted above. Within modification, modifiers serve to particularize the referent, that is, starting from the type description provided by the head noun, modifiers subcategorize the type, select an instantiation of the referent, and situate it in physical space and mental space (Croft Reference Croft2007, Reference Croft2022: 103–113). Finally, relations between events may be construed asymmetrically (a figure–ground construal) or symmetrically (a complex-figure construal; Croft Reference Croft2001: chapter 9). This list likely does not exhaust all information packaging construals. Nor are the different construals entirely discrete. There is typological evidence for a reference–modification continuum and a modification–predication continuum (Croft Reference Croft2022: 130–154, 443–448).

The two-dimensional analysis of function has proven useful to account for typological patterns and indicates that the function side of a construction consists of these two dimensions. Equally importantly, the analysis of function has important consequences for the organization of the constructicon, the inventory of constructions of a language. Most Construction Grammars primarily consider the organization of the constructicon in terms of a taxonomic hierarchy or lattice of the morphosyntactic structures of constructions. However, constructions are also organized in the constructicon in terms of function. In the approach described here, both semantic content and information packaging organize the constructicon.

Bybee (Reference Bybee1985: 118) proposes different strengths of connections between word forms in her typological, usage-based, network theory of the organization of the lexicon. The primary networks of lexical relations that she examines are inflectional paradigms and inflection classes. Her principles carry over to syntactically complex constructions in the constructicon, since the lexicon is part of the constructicon (Croft & Cruse Reference Croft and Cruse2004: 303f.). Relations between words may be phonological, semantic, or morphological. Phonological relations are the weakest but they are not nonexistent, as lexical competition in phonological space indicates. Semantic relations are much stronger, even in the absence of phonological relations. Bybee’s morphological relations involve parallel phonological and semantic relations, that is, morphological relations are symbolic relations.

For complex syntactic constructions, then, the strongest relations should be parallel morphosyntactic and semantic relations. In other words, constructions with similarly structured construction elements (CEs) whose semantic structure is parallel will have the strongest relations. This extension of Bybee’s model applies specifically to paradigmatic relations between constructions, organized by similarity of meaning, form, or both.

One example of a semantic relation leading to an analogical reformation of a syntactic paradigm is given in (16) (Croft & Cruse Reference Croft and Cruse2004: 320):

1. (16)

   	Declarative	Imperative	Prohibitive
   Verbal	a. He jumped.	c. Jump!	e. Don’t jump!
   Non-verbal	b. He is brave.	d. Be brave!	f. Don’t be cruel!
   			< g. Be not cruel!

The non-verbal prohibitive construction changed in form in the history of English to conform with the verbal prohibitive construction due to its close semantic similarity. This is despite the fact that the non-verbal prohibitive now has an otherwise anomalous structure: two auxiliaries in a row, the first of which (Don’t) is semantically inconsistent with a stative predicate. This also indicates that Don’t has been reanalyzed as a simple prohibitive morpheme. Another relevant factor is that the token frequency of non-verbal prohibitives is probably much lower than that of verbal prohibitives. The unique constructional form was weakly entrenched and thus was susceptible to replacement.

Bybee does not represent more schematic morphological constructions (e.g., a morphological template of inflections of a word such as English past [Verb-ed]) as separate nodes in a taxonomic network – the typical representation in Construction Grammar. Instead, she represents a more schematic construction as a pattern of similarity relations between individual word types such as the English regular past walked ~ talked ~ stalked or the partially productive irregular past snuck ~ struck ~ strung ~ spun ~ hung (Bybee Reference Bybee1985: 130). Bybee’s representation is characteristic of an exemplar-based model (Sections 17.8 and 17.11).

17.7 Linking Constructional Analysis in a Single Language and Cross-Linguistic Constructional Analysis

Cross-linguistic constructional analysis proceeds on the basis of comparative concepts (Section 17.5). Comparative concepts can be divided into two broad types. The first are constructions, which can be of any morphosyntactic form and are defined purely in terms of their function. The second are strategies, which are subtypes of constructions that are also specified to belong to a particular type of morphosyntactic form. The use of constructions and strategies as comparative concepts provides the basis for cross-linguistic comparison and the formulation of typological universals. Yet constructions and strategies are not defined by distributional analysis, the foundation of syntactic analysis in a single language. This poses a problem in relating typological universals of language to the grammars of individual languages.

Before we address this problem, we must take a closer look at distributional analysis of single languages. The same problems in using distributional analysis across languages also arise in distributional analysis in a single language. Syntactic categories that are defined distributionally, that is, via their occurrence in a role in a construction, are not identical from one construction to the next.

For example, an English adjective can be defined by four different constructions (Croft Reference Croft2022: 8):

1. (17)

   a.	Modification: a tall tree
   b.	Predication: That tree is tall.
   c.	Comparative/Superlative Inflection: tall-er, tall-est
   d.	Degree Modification: very tall, a little tall

But not all words usually described as adjectives occur in all four constructions (Croft Reference Croft2022: 10):

1. (18)

   a.	Modification: *an alive insect
   b.	Predication: *This chapter is entire.
   c.	Comparative/Superlative Inflection: *intelligent-er, *intelligent-est
   d.	Degree Modification: *a very even number

A somewhat more complex example is provided by the distribution of direct object phrases (Croft Reference Croft2001: 35f.). Most direct objects can occur in postverbal position without a preposition and as the subject of a corresponding passive:

1. (19)

   a.	Prepositionless postverbal phrase: Jack kissed Janet.
   b.	Passive subject phrase: Janet was kissed by Jack.

But some prepositionless postverbal phrases cannot occur as a passive subject and some passive subjects cannot occur as prepositionless postverbal phrases:

1. (20)

   a.	Jack weighs 160 pounds.
   b.	*160 pounds is weighed by Jack.

1. (21)

   a.	*Claude Debussy lived this house.
   b.	This house was lived in by Claude Debussy.

As with the cross-linguistic mismatches in distribution, the common approach is to ignore the distributions that do not match what the analyst expects. But this is empirically inadequate: The anomalous facts are simply ignored or some exception feature is introduced. This was called ‘methodological opportunism’ in Croft (Reference Croft2001: 30, 41). Nevertheless, the anomalous English facts fit into cross-linguistically more widespread phenomena, such as the association of passive subject with some degree of affectedness, thus disfavoring (20b) but favoring (21b).

The assumption behind methodological opportunism is the building block model of syntactic structure, where constructions are built up out of combinations of syntactic categories and relations that supposedly exist independently from constructions. But the distributional method that identifies syntactic categories and relations assumes we have already identified the constructions that are supposedly defined by those categories and relations. This is a circular argument. Radical Construction Grammar argues that instead of preserving the inventory of categories and relations, constructions should be treated as basic and categories and relations as derived from the roles in those constructions (see Croft Reference Croft2001: chapter 1 for a fuller discussion).

Distributional analysis is valid, but it presupposes the existence and identifiability of the constructions used in the analysis. The question arises: How does one identify constructions? Constructions consist of elements (CEs), the meaning of the CEs, and the meaning of the construction as a whole. A construction is defined by its form plus its function. Constructions are like any conceptual category as described in cognitive psychology and cognitive linguistics: Some constructional categories are more discrete than others; Constructional categories have more prototypical and more peripheral members; Constructional categories vary in their cue-validity (Croft Reference Croft2001: 51–53).

The distributional analysis of the syntax of a language is a gigantic many-to-many mapping between constructions on the one hand, and the words or larger units which serve as fillers for each role/CE in each construction on the other hand. There is no shortcut via a small inventory of categories and relations and a set of constructions built out of those categories and relations. For those grammarians who accept distributional analysis as the basic method for language-specific analysis of morphosyntactic form, the many-to-many mapping serves as the starting point for reconstructing grammatical analysis on a consistent within-language and cross-linguistic basis.

There is a crucial link between language-specific constructions and the distributions they define, and constructions and strategies and comparative concepts. The distributional mapping involves form–function pairings for both constructions and CEs. That is, function is part of the many-to-many distributional mapping. Function is also a comparative concept, namely, the basis for the comparative concept of constructions.

One can therefore compare constructions expressing the same function, such as property predication, across languages, not to mention different constructions expressing the same function in a single language. And one can compare the CEs, that is, the constructions (including words) that fill the equivalent roles in the construction, in terms of their function – for example, which property concept words fill the relevant role in the property predication construction in each language – not to mention different property predication constructions in a single language.

In other words, distributional analysis can be done across languages as well as within languages by using the functions of the constructions and the CEs to align constructions, both across and within languages. Strategies can also be used to align constructions that share morphosyntactic structures that can be validly defined across languages.

A classic example of this methodology is Keenan and Comrie’s (Reference Keenan and Comrie1977) typological analysis of noun phrase accessibility, that is, which argument roles of the predicate in a relative clause construction can be relativized, that is, serve as the head of the relative clause. Keenan and Comrie use a semantic definition of ‘relative clause’ to compare constructions across languages. Relative clause constructions are used to relativize different sets of argument roles across languages. Many languages have multiple relative clause constructions, each of which is used to relativize a set of argument roles. In many cases, the sets of argument roles for different relative clause constructions overlap. Keenan and Comrie discovered that the argument roles that can be relativized can be arranged in an Accessibility Hierarchy, namely subject < object < oblique < possessor (this is a modified version of the hierarchy, see Comrie Reference Comrie1989: 164). Keenan and Comrie propose universals of the distribution of relativizable argument roles for relative clause constructions. In addition, they propose universals of distribution for particular relative clause strategies. Specifically, they argue that the more explicitly the relative clause strategy encodes the relativized argument role, the lower on the hierarchy are the argument roles it can relativize (Comrie Reference Comrie1989: 163).

17.8 Methods to Analyze Cross-Linguistic Distribution Patterns and the Organization of Functions

There is so much variation in distribution within and across languages that typologists use a variety of techniques to visualize the patterns in the variation and, hence, infer patterns from that variation. The main technique used is the semantic map model. Another useful technique is multi-dimensional scaling.

The semantic map model represents functions as points or nodes in a network (graph structure; Croft Reference Croft2001; Haspelmath Reference Haspelmath and Tomasello2003). For example, each lexical concept can be represented as a node in the network. A construction’s distribution is the set of nodes (e.g., lexical concepts) that occur in the relevant role in the construction. The distribution of constructions across nodes is not arbitrary: Nodes representing concepts that occur in the same construction are semantically related. (This is not always true, for historical or accidental reasons; one must account for noise in the empirical data of cross-linguistic syntactic distribution.) If the distribution patterns of enough constructions are included in the analysis, a network can be constructed that represents semantic relations between concepts that are manifested in syntactic, morphological, or lexical distributions. Many typologists use the term ‘semantic map’ for both the conceptual network and the mapping of a construction’s distribution onto the conceptual network. Croft (Reference Croft2001: 92–98) distinguishes the two, using ‘conceptual space’ for the former and ‘semantic map’ only for the latter.

A very simple example of the semantic map model was developed to characterize ergativity (Comrie Reference Comrie and Lehmann1978; see Section 17.5). Typologists accommodate both ergative and accusative alignments in a single representation. A, S, and P roles are represented as distinct nodes. Accusative languages group A and S against P; ergative languages group A against S and P. The relations between A, S, and P are captured in the graph structure in Figure 17.1. The circles indicate the categories found in ergative languages (dashed) and accusative languages (solid).

Figure 17.1 Semantic maps of ergative and accusative systems in the same conceptual space of A, S, and P semantic role clusters

Typologists use semantic maps to account for variation in syntactic, lexical, and morphological categories; the maps replace universal essentialist building block categories such as ‘noun’, ‘verb’, and ‘adjective’, ‘subject’ and ‘object’, and so on. The principle linking universal semantic relations among concepts to constructions is the Semantic Map Connectivity Hypothesis: that a construction’s distribution must cover a connected subgraph of the conceptual space (Croft Reference Croft2001: 96).

The conceptual space – the graph structure linking nodes representing functions – is hypothesized to be universal, and so can function as a comparative concept. A semantic map – the connected subgraph representing the set of functions of a single construction in a single language, that is, its distribution – is language-specific. Much variation in categories is allowed but the categories can be compared across constructions and across languages by the ways in which they overlap in conceptual space. The structure of the conceptual space and the semantic relations between the concepts that are represented in the conceptual space are a major part of the functional organization of the constructicon for any language.

However, it is difficult to apply the semantic map model to a large number of concepts and/or a large number of languages. There is so much variation in construction distributions across languages, especially when there are a large number of concepts, that it is impossible to produce a coherent graph structure representation of the conceptual space manually. (There is now an algorithm to do it computationally, cf. Regier et al. Reference Regier, Khetarpal and Majid2013, but a large conceptual graph structure is still difficult to visualize easily.) Some linguists, including Levinson et al. (Reference Levinson and Meira2003) and Croft and Poole (Reference Croft and Poole2008), use multi-dimensional scaling to automatically generate conceptual spaces from large and complex datasets of cross-linguistic variation in the mapping from constructional form to meaning.

Multi-dimensional scaling (MDS) represents a conceptual space as a Euclidean geometrical space, that is, a continuous space, rather than a graph structure of discrete nodes. A function is a point in the space. Conceptual relations between points in the same space are represented directly as Euclidean distance. MDS (unlike principal component analysis or correspondence analysis) reduces all the variability of the data to a fixed number of dimensions. The most useful number of dimensions is determined by fitness statistics. Thus patterns in complex variation in distributions are easy to visualize.

The use of MDS allows one to plot points representing a much larger number of much more fine-grained situation types than a small number of crude semantic categories like ‘property’ and ‘action’ or ‘transitive agent’ and ‘transitive patient’. Examples include the Bowerman–Pederson set of 73 spatial relations pictures (Levinson & Wilkins Reference Levinson and Wilkins2006: 570–575) or Dahl’s (Reference Dahl1985) tense–aspect questionnaire with 250 sentence contexts. For example, Figure 17.2 presents a spatial model of the Bowerman–Pederson spatial relations pictures (Croft Reference Croft2010a: 9, figure 4). Each point represents a type of spatial relation represented by a Bowerman–Pederson picture. The position of each point in the figure is determined by the distributional pattern of the adpositions used to express the spatial relation. The closer two points are, the more times the two spatial relations are co-expressed by adpositions within and across languages. As expected, co-expression reflects semantic similarity of the spatial relations illustrated by the Bowerman–Pederson picture.

Figure 17.2 Spatial model of Bowerman–Pederson adposition meanings

(data from Levinson et al. Reference Levinson and Meira2003, analysis from Croft Reference Croft2010a)

When data on the cross-linguistic expression of these large sets of fine-grained situation types are plotted by MDS, it challenges the assumption behind the graph structure representation that semantic categories are discrete. If semantic categories were discretely conceptualized in language use, one would expect points representing instances of those classes to clump together in tight clusters. But that is not what is generally found (Croft Reference Croft2010a). The only example of a tight cluster in Figure 17.2 are the four spatial relations labeled NEAR-UNDER, all superimposed on one point in the spatial model. But the common pattern is that points are distributed across broad areas of conceptual space. Although Figure 17.2 gives suggestive names for regions of the conceptual space (ON/OVER/ON-TOP; ATTACHMENT; IN), this is an oversimplification. The ‘clusters’ are spread out in semantically significant ways and there are spatial relations found spatially – and, hence, semantically – between the ‘clusters’. Some spatial relations are more or less on their own (pictures 53 and 64 in Figure 17.2).

These patterns indicate that speakers are sensitive to fine-grained situation types and make grammatical distinctions throughout the conceptual space. Moreover, the conceptual space really is continuous: The dimensions of the space are straightforwardly interpretable. For example, the ‘IN’ part of the conceptual space of spatial relations, based on data from nine languages in Levinson et al. (Reference Levinson and Meira2003), is a continuum based on degree of enclosure of the figure by the ground (Croft Reference Croft2010a: 12f.).

17.9 Syntactic, Semantic, and Symbolic Relations in Constructions

Typology also has implications for the internal morphosyntactic structure of constructions. In Section 17.3, it was stated that cross-linguistic evidence indicates that the internal morphosyntactic structure of a construction is simply a set of part–whole relations between the construction as a whole and the CEs that make up the construction. In addition, there are also the symbolic relations between the construction’s form and its function, and each CE’s form and its function. This hypothesis has not been a major topic of discussion in typology. Representations of syntactic structure are not typically encountered in either typological studies or the reference grammars that serve as the empirical basis for typological research.

Most formal syntactic theories, and some Construction Grammars, posit syntactic relations between CEs rather than part–whole relations between each CE and the construction as a whole. These syntactic relations are represented in different ways: as constituents, as dependencies between words, and as sometimes quite complex combinations of relations. Construction Grammars include symbolic relations but modular theories do not, since syntax and semantics reside in separate modules.

Thus, there are three possible types of relations involving a construction’s morphosyntactic form: (i) part–whole relations from a CE to the construction as a whole (called ‘roles’ to distinguish them from other relations); (ii) syntactic relations between CEs; and (iii) symbolic relations from a construction to its meaning and from the construction’s CEs and their meanings.

What typological evidence is there which would argue for the elimination of syntactic relations? First, arguments for constituency or dependency relations and for categories such as head and dependent or argument and adjunct are based on distributional analysis (Croft Reference Croft2001: 185–190 and chapter 7). Distributional analysis is language-specific and so cannot serve as the basis for comparative concepts of syntactic relations. Other evidence against syntactic relations comes from considering models in which syntactic relations are preserved but either symbolic relations or roles (part–whole) relations are eliminated. Here we briefly summarize the basic types of evidence; see Croft (Reference Croft2001: chapter 5) for a fuller discussion.

The elimination of symbolic relations takes us back to the modular model of grammar: There is syntactic structure, including relations, in one module, and semantic structure in another module. Even in a modular model, there must be some linking or interface rules between modules to get the syntax–semantics mapping right. Typically, these rules appeal to general iconic principles that match a syntactic structure to the corresponding semantic structure.

However, there are many examples of non-iconic mappings in grammatical constructions around the world that undermine general linking rules or lead to complicated workarounds in comparison to simply positing direct symbolic relations between CEs and semantic structure (Croft Reference Croft2001: 206–220). For example, in many languages, the quantifier ‘floats’ away from the referring phrase it modifies and, instead, modifies the verb, as in Akimel O’odham (formerly known as Pima; Munro Reference Munro, Cook and Gerdts1984: 273):

1. (22)

   hegai	’uuvi	’o	vees	ha-ñeid	hegam	ceceoj
   that	woman	3.aux	all	them-see	those	men
   ‘The woman saw all the men.’

One could argue that ‘all’ describes not just a collection of individuals (the men), but alternatively describes a collective or summative act of seeing. Thus, there is an iconic mapping between vees ‘all’ as a verbal modifier and its meaning. However, this argument does not apply to the following example (Munro Reference Munro, Cook and Gerdts1984: 275):

1. (23)

   vees	ñei	’ant	heg	heñ-navpuj	ha-maakaika
   all	see	1sg.aux	art	my-friends	their-doctor
   ‘I saw the doctor of all my friends.’

In (23), I see just one person, the doctor; it seems implausible that seeing a single doctor can be construed as a collective or summative event.

Other non-iconic cases of form–function mapping include ‘possessor ascension’, where like quantifier float, the semantic possessor is a dependent of the verb but the possessor has nothing to do with the action; anomalous agreement, where a word agrees with a referent that has no direct relationship to the concept denoted by the agreeing word; and ‘raising’ (24a), ‘tough-movement’ (24b), ‘passive’ (24c), and ‘clause collapsing’ (24d) constructions, where a referent that is a participant in a dependent event but has nothing to do with the main clause event is expressed as a dependent argument (in boldface in the examples) of the main clause event.

1. (24)

   a.	Donna seems to have left early. [agent of leaving, but Subject of seems]
   b.	This nail is hard to pull out. [patient of pulling out, but Subject of is hard]
   c.	Teresa is rumored to have spiked the punch. [agent of spiking the punch, but Passive Subject of be rumored]
   d.	The rain continues to fall. [undergoer of falling, but Subject of continues]

These cases are numerous enough to show that symbolic relations offer a better analysis than syntactic relations. This is of course the basic argument for Construction Grammar: There are many specialized and idiosyncratic constructions that cannot be captured by general rules of syntactic form and general rules linking form to function.

The other alternative is the elimination of syntactic roles, leaving only syntactic and symbolic relations. This would be a Construction Grammar model that lacks syntactic roles but retains syntactic relations. It is not clear if any current variant of Construction Grammar is an instance of this model but it is a logical possibility.

Here, the problem is that syntactic relations impose more structure than actually exists in some cases (Croft Reference Croft2001: 220–233). A syntactic relation implies that there are two CEs between which the relation holds. Sometimes there are more than two elements, which are strictly ordered, such as the three arguments of the ditransitive construction in Kilivila (Senft Reference Senft1986: 110):

1. (25)

   eseki	luleta	yena	guyau
   he.give	his.sister	fish	chief
   ‘The chief gives his sister the fish.’

There are four possible ways to link together the three CEs in the absence of the part–whole role. If one had roles linking the arguments to the construction as a whole, then there is a single analysis of verb, indirect object, direct object, subject as roles in the ditransitive construction. Other ordering phenomena such as second position auxiliaries are also simpler to represent in terms of the role of the auxiliary in the clause as a whole.

The greater problem is the absence of one of the units in a syntactic relation. This is common with ‘flagging’ (also called ‘case marking’) and ‘indexation’ (also called ‘agreement’). A flag – adposition or case affix – is said to encode the syntactic relation. For example in (26), with encodes an oblique syntactic relation between signed and a fountain pen:

1. (26) She signed the document with a fountain pen.

But there are many constructions in many languages in which the flag is optional or absent. For example, in differential object marking, some object phrases take an adposition while others do not, or take it variably. In Spanish, the so-called ‘personal a preposition’ occurs with the object phrase if the P referent is human and definite or specific indefinite, but is absent otherwise (Comrie Reference Comrie1989: 134):

1. (27)

   a.	El director busca …
   	‘The director is looking for …
   		el carro. [‘car’; inanimate, definite]
   		al empleado. [‘clerk’; human, definite]

If the syntactic relation is encoded by the flag, then the relation ‘appears’ and ‘disappears’ when the flag appears or disappears. The simpler analysis is that the object phrase is a CE related to the construction as a whole; it does not matter whether or not the flag is present.

Indexation poses the most serious problem. Indexation is traditionally called ‘agreement’ because the traditional analysis is that the so-called agreement marker agrees with another element in the construction. This is true in many cases, but in many other cases, the element that the CE supposedly agrees with is simply not there, as in K’ichee’ Maya (28)–(29), from Mondloch (Reference Mondloch1978: 27), or it is there only optionally, as with the Kanuri object index in (30) (Hutchison Reference Hutchison1981: 139):

1. (28)

   chewe’k	k-ē-be	lē	achixab	pa	tinamit
   tomorrow	ipfv-3pl.abs-go	the	man.pl	to	town
   ‘Tomorrow, the men go to town.’

1. (29)

   c-Ø-u-c’ulu-ba	(lē	r-alc’u’āl)
   ipfv-3sg.abs-3sg.erg-marry-off.tr	(the	3sg.poss-child)
   ‘(He) marries (his child) off.’

1. (30)

   nyɪ̀-à	(nzú-)rú-kə́-nà
   2sg-assoc	(2sg-)see-1sg-pfv
   ‘I saw/have seen you.’

Again, the simpler analysis is that the ‘agreeing’ CE is related to the construction as a whole and also has a symbolic link to its referent. For this reason, many typologists have replaced the term ‘agreement’ with ‘indexation’, which implicitly indicates the symbolic relation.

These cross-linguistic phenomena all present evidence that syntactic relations are not necessary and, in fact, require the positing of syntactic structure which is not there in many constructions. Nor can syntactic relations plus general linking or interface rules capture what symbolic relations in a construction easily do. The result is a much simpler syntactic structure: the construction as a whole, plus the CEs and their relations to the construction as a whole. Most important, however, are the symbolic relations of the construction as a whole and its individual CEs to a rich representation of the construction’s function. These symbolic relations are of course what is most distinctive about Construction Grammar.

17.10 Constructions, Typology, and Diachrony

Both Construction Grammar and typology include language change, that is, diachrony, in their theoretical scope. Indeed, for many typologists, ultimate explanations for typological universals are largely diachronic (Greenberg Reference Greenberg1979). Language change in Construction Grammar is too large a topic to cover here. Instead, I will focus on developments in typology that converge on a constructional approach to processes of morphosyntactic change and what they imply about the structure and organization of constructions.

Recruitment, one of the general types of strategies described in Section 17.5, is defined in fundamentally dynamic terms: Speakers recruit the morphosyntactic form of another construction to express a particular function. Recruitment is in fact the fundamental morphosyntactic strategy; encoding strategies and systems of strategies are further stages after recruitment.

A speaker recruits a new form to express a related function. As time goes on, the recruited form adapts to its new function. For example, in French, the presentation of a physical sensation allows for degree modification, which is not possible for the source construction of presentational possession (Croft Reference Croft2001: 115):

1. (31)

   J’ai	très	froid
   I.have	very	cold
   ‘I’m very cold.’

1. (32)

   *J’ai	très	une	voiture
   I.have	very	a	car

If the form of the source construction is fused or eroded, as in many grammaticalized constructions, then the source of the recruited form is difficult or impossible to identify and we think of it as simply a way to encode the function of the construction. For example, if it weren’t for orthography, we might not be able to identify the source of the spoken form of the going to future construction of English:

1. (33)

   a.	I am traveling to deliver the letter.
   b.	I am going to deliver the letter.
   c.	I’m gonna deliver the letter.
   d.	I’m [ǝnǝ] deliver the letter.

If the morphosyntactic form of the source construction is replaced, then the source of the recruited strategy cannot be identified. For example, the word order inversion of the polarity question strategy (Will he leave?) is a relic of the verb inversion (Leave he?) of earlier stages of English, but the strategy was replaced for verbs in general by the do strategy (Did he leave?; Bybee & Thompson Reference Bybee, Thompson, Juge and Moxley1997).

The situation is more complex for a system of strategies, where a variety of processes lead to similarities and differences between the forms of the two constructions. For example, Chung (Reference Chung and Li1977: 15f.) suggests the following historical scenario for the emergence of ergative alignment in Pukapukan:

1. (34)

   na	patu	mātou	i	te	tamaiti
   pst	hit	we	acc	the	child
   ‘We hit the child.’

1. (35)

   kai-na	loa	na	tamaliki	e	te	wui	aitu	pau
   eat-pass	emp	the.pl	children	agt	the	pl	spirit	done
   ‘The children were all eaten by the spirits.’

1. (36)

   lomilomi	ai	e	tana	wawine	ma	na	tana	lua	tulivae	ia
   massage	prn	erg	his	woman	and	the.pl	boy	two	knee	that
   ‘The wife and the children massaged his two knees.’

Example (34) illustrates the older active transitive construction; the A argument follows the verb and, in turn, is followed by the P argument using the accusative preposition. Example (35) is the passive construction, which forms a system with the active transitive construction in that the encoding of the P argument of the passive (the ‘passive subject’) is the same as the A argument of the intransitive but the verb form is different (it has the passive suffix -na) and the A argument is also expressed differently than in (34), using the oblique preposition e. Example (36) shows both loss of the passive suffix from the verb and a change in word order. The change in word order is possibly by analogy to (34), or simply an adaptation to its new function, indicating that the oblique e-marked argument is now construed as a core argument of the transitive verb (the ergative).

Attested recruitment strategies are another important piece of evidence for the structure of the conceptual space. Recruitment is diachronic evidence for relations between points in the conceptual space (van der Auwera & Plungian Reference van der Auwera and Plungian1998). Croft (Reference Croft2022) describes many different recruitment strategies from which a space of constructional changes can be developed.

The best-studied set of diachronic paths in conceptual space are the semantic changes in grammaticalization (major overviews include Heine et al. Reference Heine, Claudi and Hünnemeyer1991; Hopper & Traugott Reference Hopper and Traugott2003; Lehmann Reference Lehmann2015). Kuteva et al. (Reference Kuteva, Heine, Hong, Long, Narrog and Rhee2019) gives an extensive inventory of semantic changes in grammaticalization, although intermediate steps in the paths of change are not generally given.

Grammaticalization theorists originally focused on the emergence of a single morpheme to encode a ‘grammatical function’, typically an inflectional category. Example (33) illustrates this process for the development of be going to into a future marker in English. Most grammaticalization theorists now consider grammaticalization to be a phenomenon affecting constructions as a whole (e.g., Bybee Reference Bybee and Tomasello2003: 146; Himmelmann Reference Himmelmann, Bisang, Himmelmann and Wiemer2004). The development of an inflectional category marker involves not only the narrowing of possible periphrastic expressions, say of futurity, to a single one (be going to), but also the schematization of the category with which the inflectional marker is associated (for example, it’s gonna rain is not an agentive process, unlike the purpose clause originally associated with be going to). Hence, at the very least, grammaticalization involves both the recruitment of a word or phrase to become a morpheme encoding a grammatical inflectional category, and the generalization of the word or phrase to which that inflectional category applies. (Generalization, of course, is just successive recruitment of additional words/phrases to combine with the incipient inflectional morpheme.)

Grammaticalization is a cover term for diachronic processes that lead to complex constructions containing new grammatical inflections. It is taken to contrast with lexicalization, another cover term for processes that lead to new lexical items. Less attention has been paid to processes that lead to constructions containing new word-formation morphemes (this could be called ‘derivationalization’; see Brinton & Traugott Reference Brinton and Traugott2005: 51; Traugott & Trousdale Reference Traugott and Trousdale2013: 160–177). Traugott and Trousdale generalize this entire set of diachronic processes as constructionalization, the emergence of new constructions.

Grammaticalization research seems to have shifted decisively in a constructional direction. A constructional form is recruited from a neighboring function in conceptual space, to be used to primarily convey a grammatical (inflectional) meaning. The form increases in frequency of use in its new function. Its distribution in other constructions and the distributional patterns of the categories/roles that make it up adapt to its new function. Its conventionalization leads to phonological reduction and erosion and greater rigidity and restriction in syntactic structure.

17.11 Constructions as Lineages Defining Populations

In the preceding section, recruitment is treated as a shift between the morphosyntactic forms of two discrete functions: The form conventionally used for one function is recruited for a semantically similar function. The two functions are assumed to be close in conceptual space. (There is also recruitment for figurative uses such as metaphor and metonymy; these are different yet grammatically relevant dimensions of conceptual space.) In Section 17.8, however, it was argued that multi-dimensional scaling analyses indicate that conceptual space is more continuous and that speakers are sensitive to very fine-grained distinctions between situations.

This hypothesis is further confirmed by an analysis of verbalization, the process of morphosyntactically expressing experiences, in a set of twenty English Pear Stories narratives (Croft Reference Croft2010b; data from Chafe Reference Chafe1980). These narratives are verbalizations of a short film that speakers watched and then were asked to describe to an experimenter. When one looks at multiple verbalizations in a controlled setting, it is obvious that there is immense variation in verbalization: No one describes the same scene in exactly the same way. This fact is of course not surprising, but its consequences for the representation of grammar have not been drawn out fully.

The analysis of variation in verbalization in the Pear Stories indicates that alternative morphosyntactic constructions used by speakers are potential incipient sources of grammaticalization and lexical semantic change (Croft Reference Croft2010b). For example, the basic English verb for application events is put. Speakers used a number of verbs other than put to describe scenes in the Pear film with application events, including throw, toss, and stuff. These verbs are common etymological sources of ‘put’, for example Ancient Greek bállō ‘throw’ > Modern Greek vazo ‘put’, Latin mittere ‘let go, throw’ > French mettre ‘put’, and Old English potian ‘thrust, push’ > Modern English put (data from Buck Reference Buck1949). Croft (Reference Croft2010b) argues that grammaticalization and lexical semantic change originate in variation in verbalization.

A close comparison of the Pear Stories scenes also shows that subtle differences in scenes are systematically represented by frequency differences in the use of variant forms. For example, in scenes where an event that is not intended by a human participant is verbalized, events that are more likely to be under the control of the person are more likely to verbalize the person as subject in the argument structure construction, while events less likely to be under the control of the person are more likely to be verbalized with another participant as the subject, in a gradient scale (Croft Reference Croft2010b: 30, Reference Croft2021: 264f.).

Typological comparison and verbalization in a single language indicate that very fine-grained distinctions between situations being verbalized are grammatically relevant, determining both variation and change in the frequency of use of different constructions for subtly different scenes. That is, a speaker’s grammatical knowledge must include knowledge of a very fine-grained set of situation types, arranged along continuous dimensions of conceptual variation. The mapping between form and function – that is, a construction – can be conceived of as a probability distribution of constructions across conceptual space (Croft Reference Croft2021: 271). These probability distributions are inferred from the frequency distributions of alternative forms – by the speaker learning and using language, not just by the linguist analyzing the patterns after the fact. Finally, language change involves a gradual shift of these probability distributions over time, even over the speaker’s lifetime.

This analysis of constructions can be situated in a broader theory of language change rooted in language use. Every experience that is verbalized is unique (Croft Reference Croft2007: 348f.). Of course, speakers construe an experience for the purposes of communication (see Section 17.4), but this is not easy:

In other words, every time I speak, I am recruiting morphosyntactic constructions (words and complex structures) that have been used for prior experiences that are not identical to the one I am currently trying to verbalize. The success of my communication is always uncertain, and that uncertainty leads to variation in verbalization and can lead to gradual shifts in verbalization choices that may lead to language change (Croft Reference Croft2000: 99–114).

Croft (Reference Croft2000) develops this basic observation into an evolutionary framework for language change. Another way of describing language use is that every time I speak, I am replicating words and complex constructions (and also sounds) that I have heard or previously used myself. In the evolutionary framework, language use does not simply involve a set of tokens that a speaker stores, in an exemplar-based representation of grammatical constructions which represents subtle differences in the meaning expressed in a particular occasion of use as well as subtle variations in formal structure. The grammatical construction has a temporal structure as well, defined by the chains of replications of the construction, which are called lineages (Croft Reference Croft2000: 32–34, Reference Croft2021: 284f.; see also Croft Reference Croft and van Lier2024a and Reference Croft2024b).

The result is that a language-specific grammatical construction is actually a population (in the biological sense; Hull Reference Hull1988) of tokens of the construction that are defined by the history of uses of that construction, that is, intertwining lineages of replications. The English word cat is a noun not primarily because the concept is conceptualized in a particular way but because speakers of English replicate cat and other English nouns in a particular constructional role when they replicate a construction in an utterance such as I don’t wanna see the little cat (Santa Barbara Corpus). The English category noun is the population defined by those intertwining replications (see Croft’s as yet unpublished research).

This is another difference between language-specific categories and constructions on the one hand, and comparative concepts on the other. Language-specific categories and constructions are historical entities, embodied in utterances and bounded in space and time by their use in the speech community. Of course, each occasion of use is unique in terms of form and function but they are bound together by the speaker’s act of replication in communicative interaction. In contrast, comparative concepts are ahistorical, essentialist categories of formal structure, meaning, and construal that linguists use to analyze the historically situated instances of constructional use and their evolution, diversification, and extinction over time.

In an exemplar-based model, a speaker has knowledge about a construction in her language based on the experiences she has verbalized herself as well as the verbalizations of her interlocutors in her lifetime. They represent fragments of the lineages of the construction population in the speech community but they form the basis of the speaker’s future verbalizations – that is, replications – using the construction.

17.12 Conclusion

The tremendous amount of morphosyntactic diversity in the world’s languages indicates that only minimal assumptions can be made about the morphosyntactic structure of constructions. The morphosyntactic structure of constructions consists of only the construction as a whole, its CEs, and the part–whole relation, or role of CEs in constructions. CEs are defined by their distribution in the construction they belong to, that is, categories defined by CEs are construction-specific.

Instead, construction function and the symbolic relations between a construction and its CEs and their functions, are the locus of explanation for typological universals of constructions. Constructions can be compared cross-linguistically by functional equivalence and also by the morphosyntactic strategies that they employ. Functions are organized in a largely continuous conceptual space, and the inventory of constructions is organized by relations in the conceptual space.

The morphosyntactic relations among constructions is the result of the strategy of recruiting a morphosyntactic form from a functionally related construction. Every experience that is verbalized by a speaker is unique, so every act of replicating a construction recruits a previous use of that construction and, hence, construes the current experience as an instance of the previous experience. A ‘construction’ is actually a population of utterances using that construction. In an exemplar-based constructional model, a speaker’s knowledge of a construction is the population of uses of that construction that she has been exposed to, including the experience that the construction verbalizes, and the social-interactional goal the construction was used to achieve. This knowledge forms the basis of her future uses of that construction to verbalize future experiences and achieve future social-interactional goals.

A crucial aspect of language contact is the fact that it takes place primarily in the mind of the bilingual speaker. The recognition that linguistic knowledge involves more than lists of individual words and general rules for their combination can accordingly enhance our understanding of the mechanisms by which contact affects language. Conversely, documentation of contact effects can contribute to our appreciation of the nature of linguistic knowledge.

The notion ‘construction’, introduced by Charles Fillmore in the 1980s as a “learned pairing of form and function” has continued to evolve (Östman & Fried Reference Östman and Fried2004). The focus of much early work was on semantics and syntax, but more recent work has expanded areas of inquiry. As detailed by Fried (Reference Fried, Alexiadou and Kiss2015: 975): “Form in constructions may refer to any combination of syntactic, morphological, or prosodic features, and meaning/function is understood in a broad sense that includes reference to lexical semantics, event structure, diathesis, pragmatics, and discourse structure.”

Constructions range along a continuum from short and fully substantive, like single words, through idiomatic phrases whose meanings may not be precisely compositional, mixed strings of substantive elements and variables, and ultimately to fully schematic sequences. Speakers know more about constructions than their forms and denotations: They know about their linguistic and extralinguistic contexts of use, their semantic and social connotations, their pragmatic effects, their frequencies in speech, and more – all elements of the construction. It is not surprising that such knowledge about one language might impact knowledge of another in the mind of the bilingual.

In contact situations, we see the replication of constructions of all lengths and degrees of abstraction, as well as all, most, or just some constructional features. In what follows, Section 18.1 discusses substantive constructions and calques, Section 18.2 partially filled schemata, and Section 18.3 fully schematic constructions. Section 18.4 raises the issue of the length and scope of constructional knowledge, Section 18.5 looks at prosody, Section 18.6 at frequency, and Section 18.7 at contexts of contact.

18.1 Fully Substantive Copying

Perhaps the most obvious effects of language contact are the copying of the smallest, most concrete type of constructions: loanwords. In the simplest cases both the substance and the basic meaning are replicated. Languages indigenous to California, for example, contain terms copied from Spanish for material items and concepts introduced during the eighteenth and nineteenth centuries. Barbareño Chumash, a Chumashan language of coastal California, contains such words as kamisa ‘shirt’ (from Spanish camisa), kapé ‘coffee’, (café), katu ‘cat’ (gato), kayina ‘chicken’ (gallina), lunes ‘Monday’ (lunes), mula ‘mule’ (mula), papas ‘potatoes’ (papas), tyenta ‘store’ (tienda), and more. In some cases frequent collocations were copied as single units, as in alipyelnu ‘hell’ (al infierno ‘to hell’), alamisyon ‘mission’ (a la misión ‘to the mission’), and lamesa or alamesa ‘table’ (la mesa ‘the table’, a la mesa ‘at the table’). The copies were then used as basic nouns. The noun ʔalamesa ‘table’ can be seen in (1) in a context typical of native nouns, with the dependency marker hi, distal demonstrative hoʔ, and article l=. In Barbareño there is no specification of direction or location. (Barbareño material cited here comes from the notebooks of speaker Mary Yee and the fieldnotes of John Peabody Harrington in his work with her. Numbers following the initials JPH refer to reel and frame of the microfilm copies.)

1. (1)

   Barbareño noun: Mary Yee, speaker
   Qʰiliʔetemesus				hi	Luwisa
   k-qili-etemes-us				hi	Luisa
   1sg.sbj-hab-be.opposite-ben.appl					dep	name
   ‘I used to sit across from Luisa’
   hi	klek’en	hi	hoʔlʔalamesa.
   hi	k-lek’en	hi	hoʔ=l=ʔalamesa
   dep	1sbj-sit	dep	dist=art=table
   ‘at the table’

Spanish verbs were copied into Barbareño in their infinitive forms, as in iskiliwil ‘to write’ (escribir) and pilil ‘to fry’ (freir), or with additional material, usually a, as in ʔalel ‘to read’ (a leer) and ʔalasál ‘to pray’ (a rezar). They were then treated like simplex verb roots for derivation and inflection.

1. (2)

   Barbareño verb: Mary Yee, speaker, JPH 59.407
   […]	ʔim’éči	siyqiliʔalasál.
   	ʔi=m’eči	s-iy-qili-ʔalisal
   	cleft=always	3sbj-pl-hab-pray
   ‘[After dinner] they used to pray.’

English of course contains vast numbers of words copied from French, but there are also multi-word constructions, as in It has a certain je ne sais quoi, He just exudes a wonderful joie de vivre, and They made plans for a rendez-vous in the garden at midnight. English speakers who use these constructions, like the French speakers from whom they were copied, are not assembling them online as they speak, borrowed word by borrowed word: je ‘I’ ne ‘not’ sais ‘know’ quoi ‘what’. The constructions were brought into the language as wholes and are recognized as such by speakers. English speakers who use the term joie de vivre lit. ‘joy of to.live’ = ‘zest for life’ know that it is a lexical item, while joie de manger ‘joy of to.eat’ is not.

Both the form and meaning of constructions may be replicated in the target language relatively intact. Often the forms are adjusted to some degree to fit the phonological structure of the recipient language. Sounds in the donor language not present in the target language may be replaced by similar sounds that are, as in the Barbareño kwaltu ‘room’ (from Spanish cuarto), lelos ‘clock’ (reloj), pospolo ‘match’ (fósforo), and pyesta ‘fiesta’ (fiesta). Syllable structures may be adjusted by the simplification of consonant clusters and/or the insertion of epenthetic vowels, as in the Barbareño ʔilesya ‘church’ (iglesia), lumiku ‘Sunday’ (domingo), pali ‘priest’ (padre), pilisala ‘blanket’ (fresada), pilixolis ‘beans’ (frijoles), puluta ‘fruit’ (fruta), sumlelu ‘hat’ (sombrero), tiliwu ‘wheat’ (Spanish trigo), and sululalu ~ suldalu ‘soldier’ (soldado).

The meanings of copied forms also do not always match those of their sources precisely. The Barbareño term wulewu ‘sheep’ is close in meaning to its Spanish source borrego ‘lamb’ but not exact. Numerous languages indigenous to North America have copied terms for ‘work’ from European languages. In the replicating languages, however, the term often refers specifically to work done for someone else, apparently a result of the situations in which the terms were replicated. A Barbareño verb root for ‘work’ is -talawax, from Spanish trabajar.

1. (3)

   Barbareño verb ‘work’: Mary Yee, speaker, JPH 59.643
   ʔakimpi	huktalawaxánuswun	hi	l’usʔustliyá:kuʔ	…
   ʔakimpi	hu=k-talawax-an-us-wun	hi	l=ʔus-ʔustliyakuʔ
   during	rem=1sbj-work-ipfv-ben.appl-3pl.obj	dep	art=rdp-Austrian
   ‘When I worked for the Austrians, …’

A possible explanation for the semantic narrowing is that in earlier times there was not a fundamental conceptual distinction between work and leisure in many of these cultures.

Replicated terms have not always come directly from their obvious sources. The region immediately to the north of present-day San Francisco, hundreds of miles away from Chumash territory, is home to languages of the unrelated Pomoan family. Central Pomo contains a noun t̯áwhal with a similar semantic narrowing.

1. (4)

   Central Pomo ‘work’: Frances Jack speaker, p.c.
   Čá:č	ʔel	šdihduwan	tí:kʰet̯’	t̯áwhal	yhét̯ač’kawʔkʰe.
   ča:č	ʔel	š-di-hduwan	ti:-kʰet̯’	táwhal	yhe-t̯-ač’-ka-ʔkʰe
   person	the	pulling-take-hab	3r-ben	work	do-mult.ev-pl.ipfv-caus-irr
   ‘They would haul people in to do the work.’

The Central Pomo were not themselves in direct contact with the missions but their neighbors to the south were. In 1817, the Spanish established a mission at San Rafael and among the people they recruited were Southern Pomo. The Southern Pomo term for ‘work’ is t̯áwhal. In 1823, the Spanish established Mission San Francisco de Solano in Sonoma and their recruits included Wappo people (Bean & Theodoratus Reference Bean, Theodoratus and Heizer1978: 299). The term in Wappo, a language unrelated to the Pomoan languages, is t̯áwhal’ (Sawyer Reference Sawyer1965: 112). The same form t̯áwhal was copied into the neighboring Lake Miwok, a Utian language unrelated to either Pomoan or Wappo, as both a noun and verb (Callaghan Reference Callaghan1965: 133). Lake Miwok also contains a derived benefactive applicative stem t̯awhálne ‘work for someone’. All of these languages distinguish dental and alveolar apicals, and these forms all preserve the Spanish dental.

Copied forms can also show semantic widening. The Central Pomo term for ‘Indian’ or ‘Indigenous’ is hínt̯il, used both as a noun referring to persons and as a modifier, as in hínt̯il čanú ‘Indian language’. The Wappo term is hínt̯il’ with the same meaning (Sawyer Reference Sawyer1965: 55); the source is the Spanish gentil ‘gentile’.

A foundational feature of Construction Grammar is that speakers know more about constructions than their forms and meanings. Among other things, they know about their parts of speech and how they should fit into larger linguistic structures. The French rendez-vous originated as a verb, the imperative form of se rendre ‘surrender’ or, with the preposition à, ‘go to’. But as Stéphane Robert notes (p.c.), it is now more often used in French as a noun meaning ‘appointment’ or ‘meeting’. When it was copied into English, its lexical categorization as a noun was carried along with it: a rendez-vous.

But the syntactic status of constructions is also not always replicated perfectly. A copy of the Spanish prepositional phrase a caballo ‘on horseback’ appears in Barbareño as a verb stem ʔakawayu ‘ride a horse, be on horseback’.

1. (5)

   Barbareño verb: Mary Yee, speaker, JPH 59.468

   K’akawayu.

   k-ʔakawayu

   1sbj-ride.horse

   ‘I went on horseback / I rode.’

From the outset (as in Fillmore Reference Fillmore and Cole1974/1981), Construction Grammar has recognized the importance of the linguistic and extralinguistic contexts in which constructions are used. In their special issue Constructions and Frames, Enghels and Sansiñena observe the blossoming of work on discourse within a Construction Grammar framework:

In situations of language contact, contexts of use are often replicated along with forms at all levels of structure. The Barbareño misa ‘Mass’ and pali ‘priest’ are used in the same religious contexts as their sources misa and padre in Spanish. The French construction je ne sais quoi occurs in contexts like that in (6).

1. (6)

   French: Stéphane Robert, speaker (p.c.)

   Il y a dans cette sauce un je ne sais quoi que je n’aime pas trop.

   ‘There’s a certain something in this sauce that I don’t really like.’

English speakers who use the construction je ne sais quoi use it in similar contexts: They are unlikely to say something like I found a je ne sais quoi to mean ‘I found something but I don’t know what it is’. Stéphane Robert (p.c.) observes that in French je ne sais quoi has a certain literary flavor; in the spoken language one might be more likely to say simply quelque chose ‘something’. She suggests that with je ne sais quoi “le locuteur insiste plus nettement sur le fait qu’il n’arrive pas à préciser sa sensation ni surtout à en définir les raisons: il y a donc du mystère dans l’affaire!” [‘The speaker is indicating that s/he cannot be more precise about his/her feeling or the reasons behind it: there is thus some mystery to the matter!’] For those speakers who use it, such connotations have been transferred into English along with the construction, perhaps enhanced in part by the fact that it is recognizably French.

Similarly, Garbell (Reference Garbell1965: 175–177), cited in Coghill (Reference Coghill and Grant2019: 511), reports that the Jewish Azerbaijani dialect of Northeastern Neo-Aramaic shares many idiomatic expressions with Kurdish and Zeri, among them “formulae used regularly in specific contexts, such as telling a story or expressing thanks, congratulations, or condolences.”

But precise meanings, connotations, contexts of use, and even frequency are not always replicated precisely. French speakers easily mention un rendez-vous chez le dentiste for a dental appointment, but the mention of a rendez-vous with the dentist by an English speaker would carry special connotations. The copy of the construction differs in both connotations and frequency, likely due at least in part to the fact that English contains more general native terms appointment and meeting. (French un rendez-vous avec le dentiste could have connotations closer to that in English.)

Constructions are also copied with everything but their forms. In calques, the components of a construction in the model language are replicated, piece by piece, with counterparts in the target language. Piirainen (Reference Piirainen2012), cited in Koptjevskaja-Tamm and Liljegren (Reference Koptjevskaja-Tamm, Liljegren and Hickey2017: 212f.), assembled a list of European formulaic expressions in seventy-three linguistic varieties spoken in Europe, seventeen non-European languages, and Esperanto. Among these calques are formulas for leave-taking, such as French au revoir, German auf Wiedersehen, Swedish på återseende, Russian do svidanija, and Finnish näkemiin. The precise semantics, grammatical status, linguistic and extralinguistic contexts of use, and social meanings were copied without the forms. It should be noted that this is not schematic copying: Specific morphemes appear in the copies.

Replicated phrasal constructions ranging in length from idioms of a few words to entire proverbs and interactional routines can be seen in linguistic areas throughout the world. Ameka (Reference Ameka, Aikhenvald and Dixon2006, Reference Ameka and Goddard2011), cited in Koptjevskaja-Tamm and Liljegren (Reference Koptjevskaja-Tamm, Liljegren and Hickey2017: 211), lists such interactional routines among West African languages, primarily in the Volta Basin, such as proverbs, good-night wishes (‘sleep/lie well’), thanking, and leave-taking expressions. Hayward (Reference Hayward, Mendel and Clauci1991, Reference Hayward, Renfrew, McMahon and Trask2000) describes them in an Ethio-Eritrean area, Smith-Stark (Reference Smith-Stark, MacKay and Vásques1994) in Mesoamerica, Matisoff (Reference Matisoff and Saxene2004) in Southeast Asia, and Koptjevskaja-Tamm and Liljegren (Reference Koptjevskaja-Tamm, Liljegren and Hickey2017) in the Greater Hindu Kush, among others.

18.2 Partially Filled Schemata

An important contribution of Construction Grammar is the recognition of partially schematic constructions, in which some but not all elements are lexically specific. A well-known example is the X-er the Y-er (Fillmore et al. Reference Fillmore, Kay and O’Connor1988: 505). Constructions of this type are also often copied from one language into another. Particularly common are clause combining constructions, consisting of a specific conjunction and a schematic configuration of clauses. In many cases, particularly in the absence of a strong written tradition, the recipient language had no parallel native construction: Prosody alone was sufficient for expressing relations among ideas (Mithun Reference Mithun, Haiman and Thompson1988).

Hellwig (Reference Hellwig and Grant2019: 444) describes phrasal and sentential conjunctions in Goemai, a Chadic language of Nigeria. She notes that many have been copied from Hausa, including kó ‘or’, àmmá ‘but’, kàfin ‘before’, sai ‘then/only’, hár ‘until/even’, sábò n‘doe ‘because of’, and kódàshikè ‘even though’. She provides the example in (7).

1. (7)

   Goemai ‘but’: Hellwig (Reference Hellwig and Grant2019: 444)
   Ní	góe	mis,	àmmá	ní	góe	jáp	bá.
   3sg.i	comit	man.sg	but	3sg.i	comit	child.pl	neg
   ‘[…] she has a husband but she doesn’t have children’

Souag (Reference Souag and Grant2019: 462) cites numerous clause connectors copied from Arabic into Berber, among them Chenoua lukan ‘if’ (irrealis) from law kān ‘if it were’ and ɣir ‘when, as soon as’ from ɣayr ‘only’. Citing Chaker (Reference Chaker1989), he notes that “some of these loans, such as bəlli ‘that’ or u ‘and’ fill positions that were previously left empty in Berber, increasing the incidence of syntaxis at the expense of parataxis.”

An area well known for such contact effects is Mesoamerica. Suárez (Reference Suárez1983: 136) lists common constructions that contain function words from Spanish, among them como ‘as, like’ in Northern Totonac, Ostuacan Zoque, Highland Chontal, Isthmus Zapotec, and Pokomán; con ‘with’ in Northern Totonac, Highland Chontal, and Mezquital Otomi; cuando ‘when’ in Ostuacan Zoque and Pokomán; de ‘of, from’ in Northern Tepehuan, Highland Chontal, Mezquital Otomi, Isthmus Zapotec, Tetelcingo Nahuatl, and Huave; hasta ‘till, until, even’ in Northern Totonac, Ostuacan Zoque, Highland Chontal, Huichol, Mezquital Otomi, Isthmus Zapotec, Tetelcingo Nahuatl, and Huave; lo or de que ‘that which’ in Ostuacan Zoque, Mezquital Otomi, Isthmus Zapotec, Tetelcingo Nahuatl, Pokomán, and Huave. A number of distinct language families are represented among these languages: Uto-Aztecan, Totonac-Tepehua, Otomanguean, Tequistlatec-Jicaque, Huave, Mixe-Zoque, and Mayan. Additional Spanish function words copied into the various languages include mas quei ‘although’, o ‘or’, para ‘for, in order to’, pero ‘but’, por ‘because of’, porque ‘because’, pues ‘then’, y ‘and’, sino ‘but’, casi ‘almost’, mientras ‘while’, entonces ‘then, therefore’, desde ‘since, from’, también ‘also’, todo ‘all’, and cada ‘each’.

The replication process of such constructions is sometimes more complex than may first appear. Gutiérrez-Morales (Reference Gutiérrez-Morales2008) describes constructions in Sierra Popoluca, a Mixe-Zoque language of Mexico, apparently copied from Spanish. One involves a reason marker porke ‘because’. A number of others, however, contain a Spanish element plus something else. A conditional marker si’iga ‘if’, for example, apparently consists of the Spanish si ‘if’ plus some other element.

1. (8)

   Sierra Popoluca: Gutiérrez-Morales (Reference Gutiérrez-Morales2008: 172)
   Porke		mi-mumnaktöng
   because		2sg.abs-get.sick.pfv
   ‘“Because you got sick
   dyam,			dyam		a-put.
   no.longer			no.longer		1sb.abs-leave.ipfv
   I did not leave”
   an-nö’,
   1sg.erg-say.prf
   I told her,
   “Si’iga	mi-ñök-pa-m					nök-s-ö-m.”
   if	2sg.abs-go-ipfv-already					go-now-ipfv-already
   “If you want to leave, go!”’

Until the 1970s, Sierra Popoluca speakers were not generally bilingual in Spanish. There was, however, earlier contact, including intermarriage, over an extended period with speakers of Mecayapan Nahuatl, a Uto-Aztecan language. Gutiérrez-Morales (Reference Gutiérrez-Morales2008) notes that older Sierra Popoluca speakers remember meeting people who spoke both languages. He shows that what appear to be Spanish loans were actually replicated from Nahuatl; they show phonological adjustments that would not have been necessary in Sierra Popoluca, and even fragments of Nahuatl morphology, including a Mecayapan Nahuatl complementizer iga. A number of function words copied from Mecayapan Nahuatl, including the conditional si’iga ‘if’, contain it.

As with fully substantive copying, not all features of partially schematic constructions are necessarily replicated intact. Sakel (Reference Sakel, Matras and Sakel2007) describes grammatical copying from Spanish into Mosetén, spoken in the Bolivian Andes. At present all Mosetén speakers are bilingual in Spanish. Sakel reports that Spanish conjunctions have been copied into Mosetén, appearing between juxtaposed clauses as in Spanish. But their functions and contexts of use are not perfectly isomorphic:

She notes that pero expresses not only contrast but also a change in topic in Mosetén.

Many subordinating conjunctions are also borrowed from Spanish. The conjunction pajki ‘so that’ (Spanish para que) serves a broader function than its Spanish source, marking not only purpose as in Spanish, shown in the first clause in (9), but also causation, as in the second clause.

1. (9)

   Mosetén: Sakel (Reference Sakel, Matras and Sakel2007: 573)
   Khin’-ki-ra’		ti-ksi			paj-ki-ra’		tsin	dyaidye’chhe
   now-contr-irr		bring-3pl.obj			for-contr-irr		1pl	stranger
   ‘And now he will bring them [here], so that we will marry them
   bae’-ja’	paj-ki			kï’-yä’-mïn		na’-en’
   live-1pl.sbj	for-contr			size.dim-adess-assoc		born-1pl.sbj
   because we [i.e., our babies] are born [too] small,
   yi-nä	khä		jike-in.
   say.m.sg-foc	well		past-pl
   so he says.’

Sakel notes that the Spanish reason marker porque was not copied. The marker si ‘if’ serves a narrower function than its Spanish source, only marking if-clauses that provide alternatives, as in Si mi rai’se’ jäe’mä ji’chhaeyiti khaei’si’ pheyakdye’ o rai’se’ chhibanmi jäe’mä miñidye’in … ‘If you want to study our language, or want to know about the medicines, …’. Otherwise a native enclitic =yä’ is used: Mi’ra’ wenchhïshän=yä’ […] ‘If he comes back again, [I’ll give him his money]’.

Calques, the replication of constructions with the substitution of native morphemes for their counterparts in the donor language, are also not uncommon in partially schematic constructions. Central Pomo, indigenous to the Clear Lake area of Northern California, contains two negative constructions, one with a clause-final negator t̯ʰin, the other with čʰów (Mithun Reference Mithun, Krasnoukhova, Crevels and van der Auwera2021a).

1. (10)

   Central Pomo negation: Florence Paoli, speaker (p.c.)
   Mú:t̯u	t̯’a:	ʔa:	yá:q’	t̯ʰín.
   3sg.pat	guess	1sg.agt	know	neg
   ‘I guess maybe I don’t know her.’

1. (11)

   Central Pomo negation: Salome Alcantra, speaker p.c.
   ʔa:	mu:l	čanó-č-’	čʰów=ya.
   1sg.agt	that	speak.sg-sml-pfv	neg=direct.experience
   ‘I didn’t say anything.’

The negative marker t̯ʰin occurs in an assortment of contexts: generic statements, irrealis constructions, commands, constituent negation, and fragments. The negative čʰow is used primarily to negate realis clauses, though there is some variation among speakers. The distribution suggests that the čʰów-construction might be an innovation in just this context, leaving the older t̯ʰin-construction as a residue.

The Pomoan language family consists of seven languages: Northeastern Pomo, Eastern Pomo, Northern Pomo, Southeastern Pomo, Central Pomo, Southern Pomo, and Kashaya. The negative construction based on t̯ʰín can be traced back to Proto-Pomoan, but the one with čʰów has a counterpart only in Eastern Pomo, not the most closely related. There is still evidence within both of these languages of the source of this newer construction: It is a complement construction with a matrix verb ‘not exist, be absent, lack’:

1. (12)

   Central Pomo verb ‘not exist’: Frances Jack, speaker (p.c.)
   Yahwí:	čʰó-w.
   thank.you	not.exist-pfv
   ‘There’s no [word for] thank you [in her language].’

This verb root undergoes the same derivational and inflectional processes as other verb roots.

1. (13)

   Central Pomo verb ‘not exist’: Frances Jack, speaker (p.c.)
   Mé:n	ʔi-n	ʔe	yá-ʔkʰe	sṭú:	ʔel	čʰó-č-an.
   so	be-same	art	1pl-dat	basket	art	not.exist-sml-ipfv.sg
   ‘That is why our basketmaking is disappearing.’

In Eastern Pomo, the basic negator is the cognate verb kʰúy, which also persists with its original meaning ‘not exist, be absent’. As a negator it has evolved further than its Central Pomo counterpart: It now occurs in all types of negative constructions and can appear as a suffix.

To the east and south of the Pomoan languages is the unrelated Wappo. The Wappo negative marker -lahkʰiʔ bears no similarity in form to the Central Pomo or Eastern Pomo markers, but its source is still clear: It is a stative verb consisting of a verb root la- ‘lack, be absent, be missing’ and stative aspect suffix kʰiʔ (Radin’s transcription is preserved here).

1. (14)

   Wappo absence: Radin (Reference Radin1924: 6.47)
   Oni	lá‘ki’.
   they	were.not.there
   ‘They were not there.’

The combination now appears as a verbal suffix.

1. (15)

   Wappo negation: Radin (Reference Radin1924: 4.6)
   Métewelalo’kwen	mémisi	tó’ta’	opáok	mestalá‘ke’.
   when.he.came.back	his.wife	he.whipped	food	not.had.she.cooked
   ‘When he returned he whipped his wife because she had not cooked his food.’

Immediately to the north of the Pomoan languages is Yuki, possibly distantly related to Wappo. Yuki contains a clause-final negative marker tąl/tan, which resembles neither the Pomoan nor the Wappo markers. Its source also still occurs on its own as a verb meaning ‘not exist’ (Balodis Reference Balodis2016).

Immediately to the east of these languages are the unrelated Wintun languages. In the Wintu language, negation is expressed with a negative suffix -mina on the verb, reinforced with a clause-initial negative particle ʔelew. The source of the negative suffix is a verb root min ‘not exist’, with indicative suffix -a (Pitkin Reference Pitkin1984: 121). Other languages in the family also show negative constructions based on markers originally meaning ‘not exist’.

Also within the area is the more recently arrived Lake Miwok language, unrelated to Pomoan, Wappo, Yuki, or Wintuan; it is a member of the Utian family. The Lake Miwok negative marker is hel:a (Freeland Reference Freeland1947: 41), completely different in form from the negative markers in the other languages. But here, too, its source is clear. It is a verb ‘be all gone, fail to be in a certain place [be absent], lack’ (Callaghan Reference Callaghan1965: 30). Utian languages outside of the area use entirely different forms for negation.

Modern negative constructions in all of these languages have thus developed from complement constructions with matrix verbs meaning ‘not exist’. The languages are spoken in a recognized linguistic area, around Clear Lake. Their speakers have traditionally lived in small communities characterized by exogamy. Negative constructions are known to erode quickly because of their high frequency but to be renewed quickly because of their important function. Multilingual speakers in this area apparently renewed their weakening negative constructions on the model of constructions in their neighbors: ‘S not exist’ > not S.

Calques on partially filled constructions are also common elsewhere. Hollenbach (Reference Hollenbach1995: 186f.) describes reason clauses in Mixtec languages of Mexico based on terms for ‘foot’, ‘stomach’, and ‘nape’ followed by a complementizer.

1. (16)

   Jamiltepec Mixtec: Johnson (Reference Johnson1988: 130, cited in Hollenbach Reference Hollenbach1995: 186f.)
   Kwahan	ra	nunduva	chaha	cha	vátyí	kwātyi	ra.
   go.cont	3sg.sbj	Oaxaca	foot	comp	exist.cond	sin	3sg.poss
   ‘He went to Oaxaca because he is guilty.’

Olguín Martínez shows that this construction was copied into the unrelated Huasteca Nahuatl.

1. (17)

   Huasteca Nahuatl: Olguín Martínez (Reference Olguín Martínez2022: 283)
   Chotl	para	ni-mits-maki-li-k	yekah	ti-choca-k.
   foot	comp	1sg.sbj-2sg.obj-hit-appl-pfv	thus	2sg.sbj-cry-pfv
   ‘Because I hit you, you cried.’

Along similar lines, Packendorf (Reference Packendorf and Hickey2020: 672f.) describes new finite subordinate clause constructions in various languages of Siberia. Like their Russian models, they consist of a subordinator and dependent clause but speakers did not borrow the Russian subordinators. Instead, they replicated the polysemy of the Russian subordinators, using indigenous adverbials as complementizers and interrogative pronouns as relativizers.

18.3 Fully Schematic Copying

Replication of fully schematic constructions is also common in contact situations. Johanson et al. (Reference Johanson, Csató, Karakoç and Hickey2020: 566f.) mention such copied constructions in a number of Turkic languages. While in most Turkic languages relative clauses precede their heads, for example, in Irano-Turkic varieties they generally follow, on the model of Persian. In Standard Turkish, genitives generally precede their heads but in Rumelian dialects, spoken in Bulgaria and Turkish Thrace, they follow.

1. (18)

   West Rumelian Turkish: Johanson et al. (Reference Johanson, Csató, Karakoç and Hickey2020: 566)
   baba-si	Alí-nin
   father-3sg.poss	Ali-gen
   ‘the father of Ali’

Similar copying can be seen in Lithuanian Karaim under the influence of Lithuanian, as in (19a), though the shift has not affected some highly frequent expressions, like that in (19b):

1. (19)

   Lithuanian Karaim: Johanson et al. (Reference Johanson, Csató, Karakoç and Hickey2020: 566)
   a.	orun-lar-ï		karay-lar-nïn
   	place-pl-3sg.poss		Karaim-pl-gen
   	‘the places of the Karaims’
   b.	biz‘-n‘in‘	ata-mïz
   	1pl-gen	father-1pl.poss
   	‘our father’

Ossetic, a language of the Iranian branch of Indo-European spoken in the Caucasus, is considered part of the Caucasian linguistic area. It contains two constructions for marking recipients in ditransitive constructions, one with dative case and the other with allative case. Belyaev (Reference Belyaev and Grant2019) reports that the dative construction is more general:

1. (20)

   Ossetic recipients: Belyaev (Reference Belyaev and Grant2019: 485)
   Farašt	ɜfšəmɜr-ə	šɜ=xo-jə	χɜməs-ɜn / ??χɜməs-mɜ
   nine	brother-gen	their=sister-gen	Kh-dat / Kh-all
   rat:-oj.
   give.pfv-past.3pl
   ‘Nine brothers gave their sister to Khamits [dat] (in marriage).’

He notes that if the allative were used in (20), then “according to some native speakers, the only available interpretation is that the brothers gave (‘lent’) their sister to Khamits for a limited period of time, as opposed to giving her to him in marriage” (Belyaev Reference Belyaev and Grant2019: 486). He cites Daniel et al. (Reference Daniel, Molochieva, Khalilova, Malchukov, Haspelmath and Comrie2010), who observe that this opposition is typical of East Caucasian languages, such as Akhvakh:

1. (21) Akhvakh recipients: Daniel et al. (Reference Daniel, Molochieva, Khalilova, Malchukov, Haspelmath and Comrie2010)

   Besanoda	ʁurus:i		milica-s:ʷ-a		o-x:-ada,
   hundred	ruble		policeman-m-dat		n-give-pfv
   ‘I gave a hundred rubles to the policeman [dat] (as a bribe),
   boq’:e:doda		ʁurus:i-la		di	imo-l:ira		o-x:-ada	zikira
   forty		ruble-and		1sg.gen	father-all		n-give-pfv	zikr
   b-el-urula.
   n-lead-inf
   and forty rubles to my father [all] to organize the zikr (a religious ritual).’

The dative in the first clause of (21) indicates that the money was transferred permanently, while the allative in the second marks a temporary transfer: The zikr in Daghestanian practice involves redistributing the money among the worshippers.

18.4 Construction Scope

As noted, a central feature of Construction Grammar has always been its integration of form and meaning at different levels but earlier work tended to focus on syntax. Work on discourse with Construction Grammar frameworks has, however, blossomed in the twenty-first century. Some examples are Östman (Reference Östman, Östman and Fried2005, Reference Östman, Bücker, Günthner and Imo2015, Reference Östman2020), Östman and Trousdale (Reference Östman, Trousdale, Hoffmann and Trousdale2013), Nikiforidou et al. (Reference Nikiforidou, Marmaridou and Miros2014), Fried (Reference Fried, Enghels and Sansiñena2021), and additional papers in Enghels and Sansiñena (Reference Enghels and Sansiñena2021b), among others. Patterns of construction replication at the discourse level can also be seen in situations of language contact.

Tail–head linkage, in which the final element of one clause or sentence is repeated at the beginning of the next, is known to be pervasive in certain areas of the world. De Vries (Reference De Vries2005) identifies some of its major functions in languages of the South Pacific, indicating referential coherence and thematic continuity or discontinuity (depending on the type). Næss (Reference Næss and Grant2019) describes relations between two distantly related languages of the Reef Islands in the southwest Pacific, Vaeakau-Taumako, part of the Polynesian branch of the Central/Eastern Oceanic subgoup of Austronesian, and Äiwoo, part of the Temotu subgroup of Oceanic. Though tail–head linkage is common in Vaeakau-Taumako narrative, it is not reported for any other Polynesian language.

1. (22)

   Vaeakau-Taumako tail–head linkage: Næss (Reference Næss and Grant2019: 637)

   Thai

   langi

   na

   ko

   nhoh-noho

   na

   ko

   hano

   one

   day

   dem

   incp

   rdp-stay

   dem

   incp

   go.sg

   loa

   ko

   kaukau.

   emph

   incp

   bathe

   ‘One day she went to have a bath.

   Ko

   hano

   ko

   kaukau

   na,

   incp

   go.sg

   emph

   incp

   bathe

   She went to have a bath,

   iako

   te

   pakhola

   ko

   ne-ho

   ko-i

   to-a.

   conj

   sg.sp

   giant

   incp

   go.sg-down

   incp-3sg

   take-tr

   and a giant came and took her.’

It resembles a similar structure in the neighboring Äiwoo.

1. (23)

   Äiwoo tail–head linkage: Næss (Reference Næss and Grant2019: 637)
   Ikâ		lâto		ki-dâa=to=wâ,			uule=kâ
   heron		then		ipfv-float=incep=dist			drift=dist
   ‘The heron started drifting,
   mo	lâ		ki-dâ=kâ.
   and	dist		ipfv-float=dist
   he drifted slowly.
   I-da=kâ,					lâ	i-de-to=to		ngä	nye-lägä=kâ.
   pfv-float=dist					dist	pfv-wash.up-go.in=cs		loc	place-dry=dist
   He drifted until he washed up in a dry place.’

18.5 Prosody

Prosody has long been considered a property of constructions, but greater attention has been paid to this area in more recent work. Fried and Östman (Reference Fried, Östman, Fried and Östman2004: 20) note that “the interpretation of a particular linguistic expression may depend on the prosodic contour conventionally associated with it.” Gras and Elvira-García (Reference Gras and Elvira-García2021) provide an overview of the literature and ask whether prosodic structures should be considered constructions in themselves or features of other constructions, citing among others Fried and Östman (Reference Östman, Östman and Fried2005), Marandin (Reference Marandin and Schönefeld2006), Sadat-Tehrani (Reference Sadat-Tehrani2008), Ogden (Reference Ogden, Barth-Weingarten, Reber and Selting2010), Nikiforidou et al. (Reference Nikiforidou, Marmaridou and Miros2014), Kaltenböck (Reference Kaltenböck, Kaltenböck, Keizer and Lohmann2016), and Ward (Reference Ward2019). Prosodic features of constructions can be transferred in situations of language contact as well.

Effects of language contact on both discourse structure and prosody can be seen in a construction that is pervasive in Central Pomo. This is a kind of copula construction, used by speakers to draw attention to the importance of a point. An important idea is introduced in one prosodic line or intonation unit, then reiterated in the next, with replication of the prosodic contour. The pattern is not limited by genre and occurs equally in monologue and conversation. The second line typically differs in some slight detail or details from the first, such as a reordering of words and/or distribution of particles.

1. (24)

   Central Pomo couplet structure: Salome Alcantra, speaker (p.c.)
   Meṭ’	mu:l	wá:yʔi	yhé-:n	ʔe;
   such	that	earlier	do-ipfv.sg	cop
   ‘She did that kind of thing before;
   mu:l	wá:yʔi	mu:l	yhé-:n.
   that	earlier	that	do-ipfv.sg
   she was doing that before.’

1. (25)

   Central Pomo couplet structure: Frances Jack, speaker (p.c.)
   Hkúm	síw	ṭ’á:yṭ’o,	qʰó:-y;
   kill.pfv	almost	once	two-ord
   ‘Almost killed him once, twice;
   qʰó:-y	ʔdóma	t̯í:t̯o,	hkúm	síw.
   two-ord	quot	3r.sg.pat	kill.pfv	almost
   almost killed him twice, he said.’

There might be a slight grammatical difference between the two lines, such as active versus passive voice, or packaging as an independent versus dependent clause, as in (26):

1. (26)

   Central Pomo couplet structure: Frances Jack, speaker (p.c.)
   Čá-:l	hlí-hč’iwač’;
   house-to	pl.go-hab.pl
   ‘They’d go home;
   t̯í-ya=:kʰe		ča,	ʔ=mída		čá		hlí-č’i-n.
   3r-plposs		house	cop=there		house		pl.go-ipfv.pl-same.simultaneous
   going there to their home.’

Often the second line adds elaboration:

1. (27)

   Central Pomo elaboration: Salome Alcantra, speaker p.c.
   Meṭ’	maná-:č’-ka-ya-w;
   such	pay-caus-pass-pfv
   ‘He was made to pay;
   mé:n	kúy	maná-:č-ka-ya-w.
   so	immediately	pay-caus-pass-pfv
   right away they made him pay.’

1. (28)

   Central Pomo elaboration: Florence Paoli, speaker p.c.
   Qʰá	čʰó-w	ʔe;
   water	lack-pfv	cop
   ‘No water;
   máya-l	čá-w	qʰá	čʰó-w;
   2pl.pat	house-in	water	not.exist-pfv
   you have no water in the house;
   qʰa	čʰó-h-duw.
   water	lack-sml-hab
   run out of water sometimes.’

1. (29)

   Central Pomo elaboration: Frances Jack, speaker pc.
   T̯i:	só-:	yó-w=ʔkʰe	hí-h-du-w;
   3 r.sg.agt	east-to	go-pfv=irr	said-sml-ipfv-pfv
   ‘He said he was going east;
   Sacramento=lil	wá-:n	ʔdóma	t̯i:.
   name=toward	sg.go-ipfv.sg	quot	3r.sg.agt
   he said he was on his way to Sacramento.’

Strikingly similar discourse patterns appear in the English of second-language speakers, with similar slight alterations in form and the same prosodic profile as in their Central Pomo speech.

1. (30)

   Central Pomo English: Salome Alcantra, speaker (p.c.)

   No one home;

   wasn’t at home.

1. (31)

   Central Pomo English: Salome Alcantra, speaker (p.c.)

   Outside they were;

   faucet was outside I guess;

   it was out front–their house.

Such patterns raise the issue of whether including them within our understanding of constructions will ultimately be useful, leading to higher-level insights, or dilute the concept. One likely reason for the fact that they were seldom discussed in early work on Construction Grammar is that rich documentation of unscripted spontaneous speech, with high-quality sound, was less accessible than it is now.

18.6 Frequency

There has been some discussion about whether frequency should be included among the features of constructions. Frequency is clearly something that speakers know, even if that knowledge is not conscious. And it is well recognized as a major force shaping grammar, as amply illustrated throughout the work of Bybee, Hilpert, and many others (cf. also Chapter 6 in this volume). Bybee (Reference Bybee, Hoffmann and Trousdale2013: 2) notes that “Constructions can be viewed as processing units or chunks – sequences of words (or morphemes) that have been used often enough to be accessed together.” She observes further that “such chunks or conventionalized sequences have a tendency over time to develop special pragmatic implications that can lead to special meaning. They can also develop idiosyncrasies of form in a variety of ways.” Work by Hilpert also puts frequency at the forefront, as in his careful tracing of changes in the meaning of English may through subtle shifts in the relative density of token frequencies in different contexts (Hilpert Reference Hilpert2016). Frequency can also play a role in contact effects.

Harris (Reference Harris, Milroy and Milroy1993: 175f.), cited by Hickey (Reference Hickey and Hickey2017: 240), describes effects of Celtic languages on English topicalization constructions. All contain native cleft constructions. They are much more frequent in Irish and Scottish English than in other forms of English, however. It was not the abstract construction itself that was copied but, rather, its frequency and range of uses. In particular, it no longer conveys the same marked information structure. Speakers of those varieties of English are likely to say such things as It’s to Dublin he’s gone today or It’s her brother who rang up this morning.

Passive constructions are common cross-linguistically but they differ in their uses and frequencies. The North American Pacific Northwest is home to three distinct language families: Wakashan, Chimakuan, and Salishan. In the Wakashan languages, core arguments are identified only by pronominal enclitics attached to the initial predicate. Lexical nominals, which may or may not be present, carry no case marking, and constituent order does not distinguish grammatical role. Examples of the enclitics in Nuuchahnulth, a language of the Southern branch of the family, can be seen in (32); third persons are unmarked:

1. (32)

   Nuuchahnulth pronominal enclitics: Nakayama (Reference Nakayama2003)
   waɬšiʔaƛ=s.	‘I went home.’		(Reference Nakayama2003: 195)
   wikʔaqƛ=s suutiɬ wiiqhap.	‘I will not harm you.’		(Reference Nakayama2003: 383)
   waɬsaap’at=s.	‘They sent me home.’		(Reference Nakayama2003: 167)
   ʕuḥšiƛ=s.	‘I cried.’		(Reference Nakayama2003: 166)
   ʔiič’imʔaƛ=s.	‘I am old.’		(Reference Nakayama2003: 451)
   huḥtakšiƛʔaqƛic=k.	‘You will know.’		(Reference Nakayama2003: 640)
   naʔuuqsʔaqlic=k sušiɬ.	‘You will accompany me.’		(Reference Nakayama2003: 510)
   ʔayaʔaq’atʔic=k ʔaʔataɬʔat.	‘Many will question you.’		(Reference Nakayama2003: 503)
   etc.

The enclitics correspond to both English subjects (‘I went home’) and objects (‘They sent me home’), so this is not a nominative/accusative system. They occur in transitives (‘I will not harm you’) and intransitives (‘I went home’), so it is not an ergative/absolutive system. They appear with semantic agents (‘I went home’) and semantic patients (‘I cried’), so it is not an agent/patient system. They appear with events (‘I went home’) and states (‘I am old’), so it is not an active/stative system. It is a hierarchical system. Two principles determine which participant is encoded in the enclitic:

1. (33)

   Nuuchahnulth hierarchical system:
   a.	First and second persons have priority over third:	1, 2 > 3
   b.	If only first and second persons are involved,
   	the semantic agent has priority over the patient:	A > P

Ambiguity is avoided by the use of a suffix -’at, visible in waɬsaap’at=s ‘they sent me home’ (as opposed to ‘I sent them home’) and ʔayaʔaq’atʔic=k ʔaʔataɬʔat ‘many will question you’ (as opposed to ‘you will question many’). The -’at suffix is a passive marker which functions as in other languages to convey topicality distinctions. Demoted agents of passives may be identified as obliques or not mentioned at all. But passivization has become obligatory with certain combinations of participants, though it has not penetrated the grammar completely: Imperatives can contain both a first and second person enclitic.

In the other two languages of the Southern branch of the Wakashan family, Nitinaht and Makah, first and second persons similarly have priority over third for enclitic marking (1,2 > 3). If a third person acts on a first or second person, a passive suffix -’it appears on the verb, cognate with the Nuuchahnulth -’at. The system has not penetrated these languages quite as far: Both first and second persons can be identified by enclitics in any clause.

The pattern does not occur in the Northern branch of Wakashan, though there is a mechanism in Kwak’wala, the closest geographically to Nuuchahnulth, for avoiding first-person objects. Subjects are identified by clause enclitics and objects by verb suffixes. But there are no suffixes for first-person objects. A periphrastic construction is used in their place, based on the verb gá:x ‘come’: a sentence like ‘He carried me away’ is rendered literally ‘he carried away coming’ (Boas Reference Boas1947). The bulkier form of this construction suggests that it is a more recent addition to the grammar, replacing an original object suffix. The other North Wakashan languages, Heiltsuk and Haisla, contain full sets of pronominal subject clitics and object suffixes, which are used in all combinations.

Directly to the south of the Wakashan family is the Chimakuan family, with two languages, Chemakum and Quileute. Documentation of Chemakum is scant, but records of Quileute provide ample evidence of the system there (Andrade Reference Andrade and Boas1931, Reference Andrade1933, Reference Andrade1953). Subjects are indicated by pronominal enclitics on the predicate, objects are indicated by verbal suffixes. In transitive combinations, object suffixes precede the subject enclitics:

1. (34)

   Quileute transitives: Andrade (Reference Andrade1933: 239)
   kwatola-swo=li	‘I am trying it on you’
   try.on-you=I
   kwatola-sta=litš	‘you are trying it on me’
   try.on-me=you

The language contains a number of passive markers, none of which were copied from Wakashan. But as in Wakashan, they are used to manage topicality and also to maintain a person hierarchy. The system has not penetrated the language quite as fully, however; it is obligatory only for second persons: 2 > 3.

To the east of the Wakashan and Chimakuan languages are the Salishan languages. The northernmost Salishan languages, Nuxalk (Bella Coola), Comox, and Sechelt, show no restrictions on person combinations. To the south of these along the coast, closer to Nuuchahnulth, are Squamish and Halkomelem, which prioritize just second persons over third (2 > 3). When a third person acts on a second, passivization is required. To the south of these are the North Straits language, with several dialects, and Klallam. The Saanich dialect of North Straits prioritizes just second persons over third (2 > 3) like Squamish and Halkomelem. The other North Straits dialects and Klallam show more extensive hierarchies, matching those of their Southern Wakashan neighbors to the west. In the Sooke and Lummi dialects, and Klallam, both first and second persons are given priority over third (1,2 > 3). If a third person acts on a first or second person, the clause must be passivized. Still further to the south, the Lushootseed, Upper Chehalis, and Tillamook languages show no restrictions. The Interior Salish languages, spoken to the east, similarly show no restrictions.

It is unlikely that the hierarchical alignment pattern shared by languages of these three families was transferred instantaneously. A more probable scenario is that the frequency of passives (a construction already present in the languages), perhaps in particular contexts, spread gradually before crystallizing in the categorical systems seen today. The likely pathway of spread can be seen by comparing the systems in the modern languages. A tendency to cast second persons as the single core argument of clauses would have increased over time, and, in languages near the epicenter of the area, been extended further to first persons; further details are in Mithun (Reference Mithun, Matsumoto, Oshima, Robinson and Sells2007a).

Frequency replication is undoubtedly a major factor underlying structural parallelisms among neighboring languages all over the world. Languages and cultures vary in the frequency with which the source of information is specified, for example, and accordingly, in the mechanisms by which it is conveyed. It can always be expressed syntactically: I heard that the neighbors are going to move, but a number of languages contain clitics or affixes for this purpose. Especially common are markers that indicate indirect evidence along the lines of ‘it is said’, ‘so they say’, ‘one says’, ‘reportedly’, etc., often termed ‘hearsay evidentials’. A Yup’ik example is in (35), a comment Mr. Charles addressed to his mother as they were about to record a conversation. The hearsay enclitic =gguq indicated that he had simply been told of the plan:

1. (35)

   Central Alaskan Yup’ik: George Charles, speaker (p.c.)
   Piciatun-gguq	tuai	qanqatalliniukuk.
   pi-ciq-atun=gguq	tuai	qaner-qatar-llini-u-kuk
   do-in.any.manner-aeq=hearsay	well	speak-imm.fut-apparently-intr.ind-1du
   ‘Well then the two of us are apparently going to speak.’

The source of information was not the main point of his statement, and =gguq is much more frequent in Yup’ik speech than ‘one says that’ in languages like English.

Languages vary in their inventories of evidential distinctions. Some have just one or two, Central Pomo contains seven: one for hearsay, one for inference, one for privileged or general knowledge, one for (non-verbal) auditory evidence, one for direct evidence, one specifically indicating that the speaker knows an event occurred because s/he did it himself/herself, and one indicating that the speaker knows of a situation because s/he was significantly affected (Mithun Reference Mithun, Brzech, Schultze-Berndt and Bergqvist2020a).

Evidential systems occur in languages all over the world but they are not distributed evenly. Aikhenvald (Reference Aikhenvald2004: 288–299) lists some areas in which their development has been traced to contact. One of the earliest to be recognized was the Balkans, where the Slavic languages Bulgarian, Macedonian, and Serbian; the Romance languages Daco-Romanian and various dialects, including Aromanian; and Albanian all mark basic evidential distinctions, while their relatives outside of the area do not. Turkic languages have been cited as the source of evidential systems across Central Asia, including the Iranian languages Tajik and Kurdish; the Indo-Aryan Dardic languages Kalasha and Khowar; the Ugric languages Khanty, Mansi, and the Samoyedic languages; and Yukaghir. Evidentials have been documented in geographically contiguous languages in the Southern Highlands Province of New Guinea and in four contiguous languages in Central Australia. They occur in neighboring but genealogically unrelated languages in a number of areas of North America: the Northwest Coast, the Great Basin, the Plains, part of the Southwest, and Northern California. They appear in languages from different families indigenous to Mexico: Chinantecan, Mixtecan, Uto-Aztecan, and Mayan. And rich systems have been described for two major areas of South America: the Andes and Amazonia. For the most part, the forms of the markers themselves do not match.

The Vaupés region of northwest Amazonia is an especially interesting contact area, home to dozens of languages belonging to four distinct families: Tukanoan, Arawak, Nadahup, and Kakua-Nukak (Epps & Michael Reference Epps, Michael and Hickey2017). The evidential system of Tariana, an Arawak language, is described by Aikhenvald (Reference Aikhenvald2003, Reference Aikhenvald2004):

Epps and Michael (Reference Epps, Michael and Hickey2017) highlight the role of discourse in motivating such developments (Hup is a member of the Nadahup family):

18.7 Contexts of Contact

The kinds of constructions that are replicated and the constructional features that are transferred are not entirely random; they are shaped at least to some extent by the nature of the contact situation. Constructions that include substance are more likely to be replicated in situations of language maintenance, where speakers bring aspects of another language into their first, as in the Barbareño lunes ‘Monday’ from Spanish lunes. There may be minor differences in substance, as speakers integrate the sounds into their native phonological systems, as in the Barbareño liyawlu ‘devil’ from the Spanish diablo. There may be narrowing of meaning, as in the Wappo t̯áwhal’ ‘work as an employee’ from the more general Spanish trabajo ‘work’, or widening, as in the Wappo hínt̯il’ ‘Indian’ from the Spanish gentil ‘gentile’. As is well known, substance is less likely to be replicated in situations of language shift, where second-language speakers are actively working to replicate those aspects of the new language of which they are most conscious. Learners, especially adults, are more likely to carry abstract schematic constructions or features of which they are less conscious from their mother tongue into the new one (Thomason & Kaufman Reference Thomason and Kaufman1988).

Even when bilinguals are skilled in both of their languages, other factors can affect the copying of constructions. An important one is speaker consciousness of the source of constructional features. Bilinguals are more likely to be aware of the origin of a particular word or phrase than of an abstract schema or a constructional feature. Such differences can be significant where the languages in contact differ in relative prestige or where there are particular cultural attitudes toward language mixing. Speakers may be more likely to replicate substantive constructions from a language with high prestige (however that is defined) than from one with lower prestige, particularly where language mixing is viewed as a mark of competence. Where language mixing is viewed negatively, speakers may consciously avoid using recognizable words or phrases from one of their languages when speaking the other. They may, however, replicate fully schematic constructions and constructional properties of which they are less conscious, such as precise semantic scope, subtle connotations, appropriate linguistic and non-linguistic contexts of use, and frequencies.

François (Reference François2011) provides dramatic examples of such phenomena. He describes seventeen closely related Oceanic languages on the northernmost islands of Vanuatu. The languages have a common origin but they are spoken by distinct communities in constant interaction:

He provides an example of the substantive differences but structural parallelisms across six of the languages. All of the languages contain a time focus category which explicitly anchors an event in time with an explicit contrast with an earlier time: ‘[do] then, and not earlier’. When used in an irrealis context, it functions as a delayed future (36). François sums up the situation as ‘one grammar, 17 lexicons’.

1. (36)

   Vanuatu lexical diversity with structural parallelism: François (Reference François2011: 223)
   Hiw	titə	takə	tɔɣəkesə,	pa	tɔkʷɛ
   Löyöp	i͡ɛn	pɛj	wulwuli͡ɛs,	pɛ	ʧɛk͡pʷɛ
   Vera’a	ɣindɪ	mak	ɔrɔrɔ,	mba	ɣitaɣ.ʔɪn
   Volow	ɣɪn	ŋɡ͡bʷaja	sisɛɣ,	mba	maŋɡʊh
   Lakon	ɣɪʧ	lak	ʔɔːʔɔ:,	la	tamalɛ
   Mwerlap	ɣɛ͡an	kʷɛr	ɔrɔr,	nɛn	tɪkʷɪtɛ͡a
   	1inc.pl	TmFoc	play	but	not.yet
   ‘Yes, we will (eventually) play, but not yet.’

Contact-induced structural parallels without matching substance have been described for a number of other regions as well. Epps and Michael (Reference Epps, Michael and Hickey2017: 940) report that in the Vaupés, despite high levels of multilingualism, language is viewed as an emblem of identity. There is, accordingly, no code-switching and low levels of lexical borrowing, but pervasive calquing of lexical items and larger constructions, congruence of semantic categories, and profound effects on grammatical structures.

As noted, most Indigenous communities in California have been relatively small and exogamy common, resulting in long traditions of multilingualism. But there is also a tradition of speaking the language of the place where one is situated. Here, too, loanwords are relatively rare but deeper structural parallelisms are pervasive (Mithun Reference Mithun2007b, Reference Mithun, Stolz, Vanhove, Otsuka and Urdzu2012, Reference Mithun and Hickey2020b, Reference Mithun, Krasnoukhova, Crevels and van der Auwera2021a, Reference Mithun, Koptjevskaja-Tamm, Schapper and Ameka2021b, Reference Mithun, Adamou and Matras2021c, Reference Mithun, Zariquiey and Valenzuela2022).

Particularly interesting is the social situation of the Äiwoo and Vaeakau-Taumako of the Reef Islands described by Næss (Reference Næss and Grant2019). The Äiwoo community is in a larger, more fertile area and Äiwoo speakers outnumber Vaeakau-Taumako speakers by about ten to one. But up until about a century ago, Vaeakau-Taumako speakers were key actors in a trade network that connected all the islands in the region, resulting in greater wealth and prestige. Interestingly, replication of lexicon has gone primarily in one direction (into Äiwoo) but replication of structure in the other (into Vaeakau-Taumako). Næss points out that while structural replication normally takes place under conditions of language shift, there is no evidence of shift here. Because of the prestige and economic power of the Vaeakau-Taumako derived from their role in trade, the Äiwoo were bilingual in Vaeakau-Taumako, the language of intergroup communication. The larger numbers of Äiwoo second-language speakers of Vaeakau-Taumako resulted in the spread of their second-language variety into the speech of the first-language Vaeakau-Taumako speakers. Næss notes that similar patterns were described by Slater for the variety of Chinese spoken in China’s Gansu Province:

18.8 Conclusion

Construction Grammar provides a useful framework for the discussion of the effects of language contact. Many contact phenomena involve the replication of either whole constructions or some constructional properties. Like constructions in general, copied constructions range along a continuum from fully substantive to fully schematic. In situations of language contact, substantive elements of constructions in the model language are also often replaced by their counterparts in the replica language, producing calques. Various features of the copied constructions, such as the phonological form, prosody, exact denotation, special connotations, appropriate linguistic and extralinguistic contexts of use may or may not be replicated precisely, as well as frequency. Both the full and partial replication of such features provide additional evidence that constructions are more than simple combinations of basic form and meaning, in line with Bybee’s position that linguistic meaning is not separable from encyclopedic knowledge, that words and constructions are represented in memory along with the situations they have been associated with in experience and their frequencies (Bybee Reference Bybee, Hoffmann and Trousdale2013: 60f.).

Of course contact situations are not static. Individual and group competence in each of the languages in contact can shift over time, as can social dynamics between cultures in contact. It would not be surprising if terms copied into California languages from Spanish ‘work’, first used only for employment, came to refer to work in general through more intensive contact with English. Languages themselves constantly evolve. Once a construction has been copied, it is not necessarily inert. The French rendez-vous was copied into English as a noun like the model but it is now also used in English as a verb. Similarities across languages may not have been copied in their current state: They may be the descendants of constructions or features of constructions copied earlier, which subsequently evolved. Negative constructions in languages of the Clear Lake area of Northern California can be traced to replicated complement constructions with matrix verbs ‘not exist’, but in Eastern Pomo, Wappo, and Wintu, they have since evolved into verb suffixes. The shared hierarchical alignment systems of the Southern Wakashan, Chimakuan, and some Coast Salish languages may not have been copied as such, but instead be the result of frequential copying of the native passive constructions present in each. Finally, parallel constructions in two languages may not be the result of direct copying. Speakers of one language may have replicated a construction from another language which was itself copied from a third, like the Sierra Popoluca conjunctions, which at first appear to be from Spanish but in fact came to them via Mecayapan Nahuatl.

Though the focus of earlier work on Construction Grammar was on single languages, there is increasing recognition of the fact that a good proportion of the world’s population is multilingual. Increasing attention is being paid to the role of constructional properties in contact situations. Some important contributions along these lines are those of Höder (Reference Höder, Braunmüller and Gabriel2012, Reference Höder, Afarli and Mæhlum2014a, Reference Höder2014b, Reference Höder, Braunmüller, Höder and Kühl2014c, Reference Höder, Boas and Höder2018, Reference Höder2019), Hilpert and Östman (Reference Hilpert and Östman2014a), edited volume Hilpert & Östman (Reference Hilpert and Östman2014b), and Boas and Höder (Reference Höder, Boas and Höder2018, Reference Boas and Höder2021) and papers in their edited volumes. Höder and colleagues have been developing a framework they term diasystemic Construction Grammar, which proposes that multilingual speakers can form generalizations across constructions in their languages, whereby some but not all features may be shared. Part of speaker knowledge of such constructions is their contexts of use, which may shift over time.

In sum, the recognition of constructions can be a useful tool for understanding contact phenomena: Much of what is replicated in contact situations are constructions or constructional features. Conversely, examination of the kinds of units that are replicated in these situations can shed further light on the nature of linguistic knowledge.

Abbreviations

abs = absolutive, adess = adessive, aeq = aequalis, agt = grammatical agent, all = allative, appl = applicative, art = article, assoc = associative, ben = benefactive, caus = causative, com = comitative, comp = complementizer, cond = conditional, cont = continuative, contr = contrastive, cop = copula, cs = change of state, dat = dative, dem = demonstrative, dep = dependency marker, dim = diminutive, dist = distal, du = dual, emph = emphatic, erg = ergative, fut = future, gen = genitive, hab = habitual, imm = immediate, incp = inceptive, inf = infinitive, ipfv = imperfective, irr = irrealis, loc = locative, m = masculine, mult.ev = multiple event, neg = negative, obj = object, ord = ordinal, pass = passive, pat = grammatical patient, pfv = perfective, pl = plural, quot = quotative, r = coreferential, rdp = reduplicant, rem = remote, sbj = subject, sg = singular, sml = semelfactive, tr = transitive.

19.1 Introduction

Since no two speakers have had exactly the same experience with language, it naturally follows from the usage-based perspective on language acquisition and grammatical representation embraced by most strands of Construction Grammar that different speakers of one and the same language will to some extent have different grammars. More specifically, their constructicons can differ in terms of (i) which constructions are included in the grammatical network, (ii) the formal or semantic properties of shared network nodes, (iii) the taxonomic or non-taxonomic links structuring the network, and (iv) the representational strength of these nodes and the links between them. Work by, most prominently, Dąbrowska (Reference Dąbrowska, Dąbrowska and Divjak2015a, Reference Dąbrowska2020) has shown that part of this interpersonal variation in grammar is truly individual, in the sense of being due to different individual minds processing and storing usage events slightly differently and coming to partially different generalizations, for instance representing basically the same grammatical knowledge at different levels of abstraction or schematicity. Still, since speakers operate and interact in different social contexts, it also goes without saying that part of this interpersonal diversity in grammar is lectally structured, resulting in ‘lectal variation’, a cover term adopted in work by Geeraerts and others in the field of Cognitive Sociolinguistics to refer to synchronic language-internal variation along geographical, social, ethnic, generational, gender, etc. lines (e.g., Geeraerts Reference Geeraerts, Tabakowska, Choinski and Wiraszka2010). In other words, the constructicons of speakers interacting in largely the same social and regional spaces may be assumed to be more similar in content and organization than the constructicons of speakers from widely different sociolinguistic backgrounds.

In spite of this, lectal variation has for a long time been largely neglected in construction-based research, as it has been in Cognitive Linguistics at large. As Hilpert (Reference Hilpert, Säily, Nurmi, Palander-Collin and Auer2017: 218) puts it, “construction grammar has traditionally focused on the linguistic knowledge of a single, idealized speaker,” aiming at providing a plausible model of the way in which this ideal speaker, viewed as a cognitive agent, stores and processes grammatical information, while mostly abstracting away from differences between groups of speakers and from the possible effects of social factors on language use and grammatical representation. Diessel’s (Reference Diessel2019) monograph on the grammar network in usage-based Construction Grammar is a case in point. In an introductory section on speakers’ linguistic decision-making processes, Diessel (Reference Diessel2019: 25) observes that “there is ample evidence that linguistic decisions are also influenced by social factors such as prestige and group identity,” but he goes on to say that such effects fall outside of the scope of his study, in which “the focus of analysis is on cognitive processes, including cognitive processes of social cognition.”Footnote ¹ Without intending to disparage Diessel’s work, this serves as an example of the predominantly non-socially oriented focus of much of the work in Construction Grammar.

The present chapter offers an overview and discussion of recent work that diverges from this general trend and which does combine a focus on issues of (synchronic) lectal variation with a construction-based view of grammar. The chapter is structured as follows. Section 19.2 elaborates on the inherent theoretical significance of lectal variation for usage-based Construction Grammar. Then, Section 19.3 presents and briefly evaluates four major strands in construction-based research on lectal variation and Section 19.4 discusses a number of challenges and open questions. The chapter ends with a brief conclusion in Section 19.5.

Before we move on, a note is in order about a number of related topics and domains that this chapter does not address, or addresses only partially. First, we will be concerned with synchronic lectal variation and not with patterns and principles of constructional change, apart from a brief discussion in Section 19.3.4 on the role of social factors in the propagation of constructional changes (but see Noël & Colleman Reference Noël, Colleman, Wen and Taylor2021 on diachronic Construction Grammar). Second, issues of multilingualism and language contact will be mentioned only in passing, in Section 19.4.3. Third, beyond such contact contexts we will also not be concerned with interlinguistic grammatical variation more generally and will therefore not deal with variation that is related to cultural differences between language communities either. Fourth, construction-based approaches to the study of genre will not be addressed (see, e.g., Hoffmann & Bergs Reference Hoffmann and Bergs2018 and Nikiforidou Reference Nikiforidou and Vergaro2018 for introductions to this sub-field). Finally, this chapter will not cover the role of constructions in conversation, which is the focus of the emerging sub-field of interactional Construction Grammar (cf. the kind of social Construction Grammar advocated in, e.g., Ziem Reference Ziem, Ziem and Lasch2015). Interested readers are also referred to Chapters 12, 14, and 20 for further discussion and references.

19.2 The Theoretical Import of Lectal Variation

The importance of the study of linguistic variation for the further theoretical development of usage-based models of grammar is argued for at length in a number of articles and book chapters by Geeraerts, Kristiansen, and associates on the theoretical and methodological underpinnings of Cognitive Sociolinguistics (Geeraerts Reference Geeraerts, de Mendoza Ibáñez and Cervel2005, Reference Geeraerts, Tabakowska, Choinski and Wiraszka2010, Reference Geeraerts2016; Kristiansen & Dirven Reference Kristiansen, Dirven, Kristiansen and Dirven2008; Geeraerts et al. Reference Geeraerts, Kristiansen, Peirsman, Geeraerts, Kristiansen and Peirsman2010; Geeraerts & Kristiansen Reference Geeraerts, Kristiansen, Dąbrowska and Divjak2015; etc.). This is a field of study aimed at the integration into Cognitive Linguistics of methods, concepts, and research questions from (variationist) sociolinguistics. Alternatively, looked at from the reverse perspective, its overall aim can be considered to be the integration into sociolinguistics of the study of variation in the kinds of linguistic phenomena that are of central theoretical import in Cognitive Linguistics, which prominently include constructions as linguistic means for conceptual construal.

Geeraerts and Kristiansen (Reference Geeraerts, Kristiansen, Dąbrowska and Divjak2015: 367–370) emphasize that such a convergence with sociolinguistics should not be considered a mere peripheral extension of Cognitive Linguistics, but that it perfectly aligns with the overall objective of the entire Cognitive Linguistics enterprise, which they characterize as a ‘recontextualization’ model of linguistics. That is, an integral part of the theoretical development of Cognitive Linguistics and related functional approaches has consisted in the reversal of the decontextualization trend that characterizes generative theory formation, namely in reversing the preference of form over meaning/function, of system over use, and of unity/homogeneity over diversity.

Not all of these recontextualizations have (yet) run their full course, though. In spite of the usage-based view of grammar having become dominant in Cognitive Linguistics generally around the turn of the millennium, and specifically in (cognitive) Construction Grammar, many studies are still conducted at the highly abstract level of ‘a language’. In the words of Geeraerts et al. (Reference Geeraerts, Kristiansen, Peirsman, Geeraerts, Kristiansen and Peirsman2010: 6), this is a level of granularity that “Cognitive Linguistics, to the extent that it takes the claim that it is a usage-based approach to language and cognition, cannot afford to work with.”Footnote ² After all, as was already observed in Section 19.1, a usage-based view naturally implies the existence of variation in the system. Speakers do not enter into communicative interactions with all other speakers of the ‘same’ language, but they communicate in specific networks differing along regional, social, generational, etc. lines. The grammars that these different sets of communicative interactions feed into will thus differ along all of these variational dimensions. Ignoring these lectal differences in favor of an idealized conception of a homogeneous grammar is at odds with the theoretical assumption of a dialectic relation between grammar and usage.

In addition to this theoretical motive, Geeraerts and Kristiansen (Reference Geeraerts, Kristiansen, Dąbrowska and Divjak2015: 370f.) point out that there is also a practical methodological reason for taking lectal variation seriously. As long as linguists draw their empirical data from the samples of real language use compiled in text corpora, as they have every reason to do in a usage-based model, lectal variation is automatically introduced in the data set, irrespective of whether the corpora used are controlled for certain demographic factors. This means that researchers always have to keep their eyes open for the potential impact of lectal variables on the patterns observed in the corpus data: “Even if the analysis of lectal variation is not the primary concern of the investigation, filtering out lectal effects requires an analysis of variation. Methodologically speaking, an awareness of variation is thus indispensable for a data-oriented usage-based analysis” (Geeraerts & Kristiansen Reference Geeraerts, Kristiansen, Dąbrowska and Divjak2015: 371).

The importance of social factors for usage-based theorizing has also been stated outside of Cognitive Sociolinguistics, that is, by authors who do not explicitly align themselves with this particular approach. Schmid (Reference Schmid2016), for instance, points out that social (and pragmatic) factors obtain logically prior to cognitive factors even:

Hence, an adequate model of grammatical knowledge cannot leave out such social factors (Schmid’s own Entrenchment-and-Conventionalization Model is briefly discussed in Section 19.3.4). Other influential studies which make the case for a more systematic ‘social turn’ in Cognitive Linguistics are Croft (Reference Croft, Evans and Pourcel2009) and Harder (Reference Harder2010).

This rationale for the necessity of attention to lectal variation in Cognitive Linguistics generally also pertains to Construction Grammar, or at least to the usage-based strands which have largely dominated the field over the last two to three decades. In fact, many of the existing case studies in Cognitive Sociolinguistics investigate lectal variation in exactly the kinds of linguistic items that construction grammarians are most typically interested in, that is, fully or partially schematic patterns with a certain degree of productivity and internal complexity (see Section 19.3 for examples). As has been observed by several authors, a research focus on patterns of lectal variation is highly compatible with a view of grammar as consisting of a structured network of conventionalized form–meaning pairings at different levels of complexity and schematicity. Leino and Östman (Reference Leino, Östman, Fried and Boas2005: 210), for instance, point out that constructions are “generalizations which language users make on the basis of the linguistic input they receive. And since speakers receive different inputs, the constructions in their ‘internal grammar’ may – and will – be somewhat different.” Similarly, Östman and Trousdale (Reference Östman, Trousdale, Hoffmann and Trousdale2013: 478) state that “constructions … are assumed to be shared by a network of speakers, and different subnetworks may have different conventions” and that “the hypothesis that language is a constructional network would suggest that knowledge of dialect vocabulary, and knowledge of dialect syntax, should be organized around the same general principles.” Hollmann (Reference Hollmann, Hoffmann and Trousdale2013) as well has pointed to the affinity between Cognitive Sociolinguistics and Construction Grammar and the importance of integrating social factors in research on constructional variation.

19.3 Four Research Strands

19.3.1 Constructions in Specific Lects

The first line of work we will introduce in this chapter consists of studies primarily aimed at the analysis of form–meaning pairings in specific lects. Examples include (i) a number of case studies by Hollmann and Siewierska on selected constructions in Lancashire English, for example, Siewierska and Hollmann (Reference Siewierska, Hollmann, Hannay and Steen2007) on the ‘alternative’ Theme-Recipient ditransitive construction illustrated in (1) below; (ii) De Clerck and Colleman (Reference De Clerck and Colleman2013) and Norde et al. (Reference Norde, De Clerck, Colleman, Boogaart, Colleman and Rutten2014) on specific degree modifier constructions with a limited geographic distribution in Dutch, including the construction with massa’s ‘masses’ in (2a), which is typical of (a western subregion of) Belgian Dutch, and the one with (me) een partij ‘(me) a lot’ in (2b), which is typical of informal Netherlandic Dutch; (iii) Yerastov (Reference Yerastov2015) on the [be done NP] construction that is found in Canadian English and some US dialects, exemplified in (3); and (iv) the study by Blevins (Reference Blevins, Boas and Höder2018) of several progressive constructions in Texas German, including the tun + infinitive construction illustrated in (4).

1. (1) She gave it the man. (Siewierska & Hollmann Reference Siewierska, Hollmann, Hannay and Steen2007: 86)

1. (2)

   a.

   Maar

   dat

   van

   die

   prophecy

   vind

   ik

   wel

   maar

   but

   that

   of

   that

   prophecy

   find

   I

   prtc

   prtc

   massa’s

   belachelijk

   hoor.

   masses

   ridiculous

   prtcFootnote 3

   ‘But I think the prophecy thing is bloody ridiculous, you know.’

   (internet example cited in Norde et al. Reference Norde, De Clerck, Colleman, Boogaart, Colleman and Rutten2014: 208)

   b.

   Hot moddefokking DAMN!

   Dat

   is

   me

   toch

   een

   partij

   vet,

   zeg!

   that

   is

   me

   prtc

   a

   lot

   cool

   prtc

   ‘Hot motherfucking damn, now that’s totally cool!’

   (internet example cited in Norde et al. Reference Norde, De Clerck, Colleman, Boogaart, Colleman and Rutten2014: 208)

1. (3)

   … all other teams were done their playoffs and preparing for the tournament anywhere from one to three weeks prior to the OHL Cup.

   (example from the Canadian Newsstand database cited in Yerastov Reference Yerastov2015: 160)

1. (4)

   Sie	tun	das	wegnehmen.
   they	do	it	away.take
   ‘They are taking it away.’
   (example from an elicitation experiment cited in Blevins Reference Blevins, Boas and Höder2018: 98)

Apart from their focus on constructions belonging to specific non-standard varieties (or to a specific national variety of a pluricentric language) rather than on nodes from the constructional networks of ‘English’, ‘German’, or ‘Dutch’, such studies are not fundamentally different in approach from other case studies in Construction Grammar. Mostly on the basis of aggregated data from a corpus made up of texts by (many) different speakers deemed representative of the lect under investigation, they try to uncover the formal and semantic properties of the constructions in question as they are used in the variety and as they are assumed to be stored in the mental constructicon of the idealized native speaker of the variety.Footnote ⁴

The overall theoretical import of such case studies is twofold. First, they confirm that the construction-grammatical machinery is well equipped to handle linguistic phenomena from non-standard language varieties and, in doing so, they show that analysts who restrict themselves to the abstract level of the homogeneous standard language miss out on part of the relevant data and run the risk of positing unwarranted generalizations. For instance, the existence in varieties such as Lancashire English of a (more or less) productive alternative double object pattern in which a pronominal Theme precedes the Recipient NP shows that, unlike what is often assumed on the basis of standard (especially standard American) English, the dative alternation cannot always be reduced to a binary choice between two constructions that differ in both the NP versus PP encoding of the Recipient and the order of the Theme and Recipient phrases. In varieties such as Lancashire English, there is a third constructional option (see Gerwin & Röthlisberger Reference Gerwin, Röthlisberger, Zehentner, Röthlisberger and Colleman2023 for a quantitative study of the dative alternation in British English dialects in which this alternative pattern is included as a third possible outcome next to the canonical double object construction and the to-dative). Nor would an exclusive focus on standard English allow consideration of the hypothesis drawn from the pattern in (1) that, in Lancashire English and similar varieties, the factors determining the absence or presence of prepositional marking on the Recipient do not fully coincide with the factors determining the relative order of the Theme and Recipient phrases. In any event, Siewierska and Hollmann (Reference Siewierska, Hollmann, Hannay and Steen2007: 98) emphasize that “the form–function mapping in ditransitives in regional dialects should not necessarily be expected to conform to that of the standard variety.” Reflecting on such instances of dialect variation (and historical change) in constructions, Goldberg (Reference Goldberg2019: 45) points out that these corroborate the fundamental assumptions of the usage-based model: “The fact that these differences emerge in systematic ways within a given dialect and differ across dialects implies that learners’ knowledge of constructions must be shaped in dynamic ways by the language that is witnessed.”

The second argument for the theoretical importance of case studies of the kind presented in this sub-section is that they demonstrate the added value that can be gained from bringing a construction-based angle to the investigation of (well known as well as hitherto unexplored) cases of grammatical variation, that is, they show that the field of (variationist) sociolinguistics also stands to benefit from the adoption of analytical concepts from Construction Grammar. The analysis of non-standard Dutch degree modifier constructions illustrated in (2), for instance, crucially relies on the identification of subschemas at different levels of abstraction in the taxonomic hierarchy of Dutch Quantifier and Degree Modifier constructions, along the lines set out by Traugott (Reference Traugott, Bergs and Diewald2008) and Traugott and Trousdale (Reference Traugott and Trousdale2013). That is, there is an abstract Degree Modifier schema in Dutch that has attracted new slot fillers from various lexical sources over time, often via the prior development of a quantifier use, and the synchronic result of this is a hierarchy with all kinds of different subschemas which may be found in certain dialects only, such as [massa’s <Adjective>] and [een partij <Adjective>], both meaning ‘very <Adj>’.

The examples given above all concern constructions that are typical of classical geographically defined varieties, that is, specific local dialects or national varieties, though this geographic variation of course intersects with other kinds of variation. The massa’s and een partij degree modifier constructions, for instance, are not only geographically but in all probability also stylistically and generationally marked. The data suggest that they are typically used by younger speakers in informal registers which sanction the display of a certain degree of linguistic extravagance or expressivity, such as internet messages and discussion boards.

For an altogether different example of a lectally specific construction, we can turn to Hoffmann’s (Reference Hoffmann2015) analysis of particular British English football chants as partially schematic constructions with a conventionalized form and function. An instance is the pattern in Figure 19.1, sung to the tune of a well-known religious hymn, where the (two-syllable) name of a random ‘bad’ football team is to be filled in and which serves the function of mocking the opposing team and their supporters by comparing them to a side with a reputation of underachievement.

Figure 19.1 Hoffmann’s (Reference Hoffmann2015: 278) representation of the Are you X in disguise construction, slightly adapted

The construction in Figure 19.1 is different from those in (1)–(4) not only in that it is a kind of idiom with an open slot rather than a more prototypically ‘grammatical’ construction, but also in its association with a much more narrowly circumscribed social community and setting, namely that of football fans in a stadium. Even if the analysis of football chants as constructions may at first sight come across as rather stretching the notion of a community grammar, there is no denying that the pattern in Figure 19.1 captures a regularity in the communicative behavior of the members of a particular social group.

19.4 Challenges and Open Questions

This section reviews a number of challenges and open questions in the constructionist study of lectal variation. In spite of the progress that has been made in the course of the past fifteen years or so, the interest in lectal variation within Construction Grammar is still a relatively young development and many questions remain, many more than can be included in this short format. The selection is partly a personal one, inspired by research interests of our own, and even for the three sets of questions that have made the cut, our discussion necessarily has to remain brief.

19.5 Conclusion

One of the directions in which the field of Construction Grammar has been expanding over the past decade and a half is a move away from the near-exclusive focus on selected form–meaning pairings from present-day standard (American) English that characterized the foundational publications of the 1980s and early 1990s. Recent years have seen an increase in construction-based work on issues of language change and (to a somewhat lesser extent) on topics in the area of synchronic lectal variation. The latter development, focused on in the present chapter, is related to the rise to prominence of Cognitive Sociolinguistics, a new sub-field aimed at the integration of methods, concepts, and research questions from Cognitive Linguistics on the one hand and (variationist) sociolinguistics on the other. As a “maximalist” theory (Fried & Östman Reference Fried, Östman, Fried and Östman2004: 24; see also Chapter 20) that explicitly aims to account for the entirety of language, it seems only natural that Construction Grammar should have turned its attention to the investigation of patterns of language-internal linguistic diversity. As we have stressed throughout this chapter, the existence of variation in the grammar which is structured along geographic, social, generational, etc. lines is predicted by the fundamental theoretical assumptions of usage-based grammar. However, constructionist theorizing about the grammatical representation of lectally structured variation can be said to be in its infancy still and many questions remain. In the above, we have tried to outline a number of possible avenues along which we believe further progress can be made.

20.1 Introduction

One of Charles J. Fillmore’s central tenets for Construction Grammar is that it should be usage-based, that is, consistent with what we know about cognition, about social interaction, and about cultural context. These aspects of human behavior constitute the guiding principle on which to base an adequate model of grammar of any language – and of language generally. Fillmore’s principle can be sustained in one of two ways, either (i) hoping that, ultimately, one’s grammar will be in accordance with what we know and keep learning about cognition, interaction, and culture, or (ii) developing grammar, and especially Construction Grammar, in such a way that discourse-pragmatic aspects are taken into account from the very start. In this overview, the latter approach is taken as the sine qua non.

In the forty-year history of Construction Grammar, the issue of approaching the interplay and connections between Construction Grammar and aspects of discourse pragmatics can be and has been approached in basically three ways. One – the standard – approach is to primarily see Construction Grammar (henceforth, CxG) as a framework for how phonology, morphology, lexical semantics, and in particular syntax can and should be handled. Issues of sentence grammar and propositional semantics that are not seen as following directly from fairly crude combinatory features are in this alternative covered by what is generally known as linguistic pragmatics and information structure, that is, deixis, presupposition, speech acts, conventional implicatures, factivity, topic and comment, etc. The general approach here is to investigate what is needed of language function (however defined) in order to explain features that get their manifestation in the structure of (a) language. Prominent early work in this approach are the studies on information structure in Lambrecht (Reference Lambrecht1996) and Michaelis and Lambrecht (Reference Lambrecht1996). In their approach, with reference to standard construction-grammatical attribute–value matrices (AVMs), the category, that is, the attribute [pragm] ‘pragmatics’, is treated on a par with attributes like [syn] ‘syntax’ and [sem] ‘semantics’, and filler-possibilities, that is, values like [topic] and [focus], are possible choices under [pragm] – comparable to values like ‘noun’ [n] and ‘verb’ [v] of the attribute ‘word class’ [cat] (on notations, see further Section 20.3). Other, more intricate aspects of discourse pragmatics are, however, largely left to other approaches, theories, and models. That is, there are systematic accounts within sub-areas of linguistic discourse-function but the very question of what function is, is either left unanswered in this approach, or subsumed under a more general understanding of what ‘meaning’ is.

A discourse approach that goes in a somewhat opposite direction is when scholars take Fillmore’s notion of usage-based to mean ‘usage-determined’. In these approaches, any structure found in language is to be seen as grounded in and being due to – if not dependent on – participants’ communicative behavior and general human interaction. This is the approach taken in studies inspired by Paul Hopper’s Emergent Grammar (Hopper Reference Hopper, Aske, Beery, Michaelis and Filip1987), further developed by Bybee (Reference Bybee1998), Du Bois (Reference Du Bois2014), and Thompson (Reference Thompson, Ono, Laury and Suzuki2019) (cf. also the discussion in Auer & Pfänder Reference Auer and Pfänder2017), which all take a strictly inductive perspective based on data. In principle, this approach is not restricted to interactional phenomena but applies to communication generally: not only to dialectal and sociolectal aspects influencing what becomes part of the grammar of a language, but also to sociocultural aspects, which all create the emergence of grammar. In this approach, features explicable as being parts of an emerging construction always have discourse-pragmatic causes. However, since studies in these approaches are reluctant to aim at formalizing or sometimes even systematizing their findings into grammatical descriptions, they do not adhere to another of Fillmore’s basic tenets of CxG, namely that it should be a generative grammar in the original sense of this word as (i) being applicable beyond a corpus and (ii) in principle being formalizable. Laudable as this approach to context is, it refrains from building a theory or model that could be applied in a more deductive manner to new data.

Both of these approaches have many followers and practitioners and their working hypotheses are well founded and clearly add to our understanding of constructions as being the basic element of analysis for linguistic description.

In contradistinction to these two, this chapter will present a third approach, namely, Construction Discourse, which seeks to take the very notion of ‘discourse function’ to be as important as form and meaning. This third approach to language function within construction-grammatical studies is one where CxG methodology and systematization is extended beyond the single sentence and beyond paragraphs as combinations of sentences.

As long as the focus of a grammatical description is on what is traditionally regarded as ‘(sentence) grammar’, it is feasible to retain the generally acceptable definition of a construction as a form–meaning or a form–function pair(ing); with constructional descriptions in this sense being ultimately built on the traditional, Saussurean linguistic understanding of language. Work on discourse and pragmatics especially during the last thirty years has, however, considerably changed our conception of what language really is, showing in particular that discourse function is not something outside of grammar, but rather clearly ‘inside’ grammar – ultimately relying on the Morrisian (Morris Reference Morris1938) tripartite distinction within the study of language and semiotics between form, meaning, and discourse function (what Morris referred to as syntactics, semantics, and pragmatics). Once issues of discourse and discourse-pragmatics are brought in, constructions need to be seen as ‘form–meaning–discourse constellations’.Footnote ¹

One might argue that it is futile to stipulate that every construction (including morphological constructions, or the Subject-Predicate construction) will in some sense have all three dimensions as distinct sets of constraints. But as with all constructions, not every attribute–value specification will be actualized. In the early days of CxG, a similar argument was put forth against the definition of constructions as form–meaning pairings, suggesting that it is difficult to see how a basic Subject-Predicate construction has a ‘meaning’ dimension with distinct sets of constraints. We have since learned that this is not a valid argument against CxG. In parallel, the Construction Discourse approach does posit that all three dimensions are potentially relevant for all constructions; but – again following Fillmore’s line of argumentation (e.g., against using an abundance of ‘bars’ in X-bar theory) – if there is nothing to specify with respect to a particular attribute, CxG simply leaves that attribute–value specification out of the description.

In this overview I will be using the term ‘discourse pragmatic(s)’ as a general denominator for discourse level approaches to the study of language in a constructional framework. Issues of Frame Semantics and FrameNet are clearly crucially important also for discourse-pragmatic approaches to constructions, but since these are addressed elsewhere in this volume (see Chapters 1 and 3), I will not deal with them in this chapter.

20.2 Construction Discourse: An Overview

At the heart of the Construction Discourse (henceforth, CxD) conception of grammar are the following three features of constructions. First, in CxD, the ‘D’ is to be seen as equally rigorous and systematic as the ‘G’ of CxG: If a grammatical description of a language is a systematic approach to that language and a description of how that language works – that is, a grammar is an abstraction of what we see, hear, and experience as language – then everything that influences such an abstraction has to be specified in the description. Once discourse aspects are accepted as influencing and being part of how language works, these must be specified in the same rigorous manner as traditional sentence-grammatical abstractions like noun and subject are specified. In CxD, AVMs are used in the Fillmorean and Firthian spirit of ‘function implies choice’, and the overarching CxD idea (with respect to notations and possible formalizations) is that the same way of representing linguistic findings can be used throughout in descriptions of language. Thus, in addition to traditional linguistic facts, issues related to genres, to implicit anchoring, to responsibility, and to ideology are adequately accounted for in CxD in a similar format as nouns and subjects are described.

In all constructional approaches to language, constructions are generalizations over instances of the (interactional, cognitive, cultural) resources that we have available to us and that we utilize as we communicate. But once discourse-pragmatic aspects are brought in, the ontology of constructions also has to change. In what I have above called the standard approach to discourse in CxG, constructions are seen as generalizations over meaningful syntactic structures; in the emergentist approach, constructions do not exist beforehand but emerge (if at all) on the basis of the frequencies with which patterns are used. In CxD, the notion that constructions are prototypes is made its central characteristic: There are constructions, but they take on different shapes in different contexts (depending on users, cultural and dialectal context, etc.). This is thus the second important point to note with respect to CxD analyses. In constructional approaches generally, the declarative, active, positive, transitive clause has no special status. In addition to this, in the CxD conception, the constructional account does not have to be ‘complete’. That is, not every piece of occurrence (including ‘mistakes’, memory lapses, interruptions, etc.) in language is licensed by a construction – albeit that an apparent mistake may well suggest that a construction is in the making, emerging, so that it can at some future point in time be said to license what might at first be, precisely, a mistake. All this speaks in favor of seeing constructions as cognitive, interactive, and sociocultural prototypes that take on different shapes in different contexts (depending, e.g., on users and on cultural, interactive, and dialectal contexts).

Finally, most importantly and as a direct consequence of what has been said above, we may reiterate what was already pointed out in Section 20.1: The often-stated view that constructions are form–meaning or form–function pairs gives way in CxD to a conception where constructions are form–meaning–discourse constellations. In the subsequent sections, an overview of some of the discourse-functional features will be presented as attributes and (some of) their possible values.

20.3 Establishing Discourse Attributes and Their Values in the Wellerism construction

In this section I will give an example of a step-by-step argumentation of how features of context are arrived at and how they can be formalized into ‘discourse attributes’ in CxD (Section 20.3.1). In the brief additional sections, I will show how this system of discourse attributes can be referred to when variation and variability in language are encountered (Section 20.3.2) and I will give some further indications of how ‘peripheral’ structures are handled in CxD (Section 20.3.3). The section will thus in effect show how CxD is implemented in concrete grammatical analyses of constructions.

As noted in Section 20.1, it has become customary in constructional analyses to specify characteristics of elements in a construction with the help of AVMs. In a traditional proposition-based or phrasal construction, we need to specify the value of an attribute like ‘part of speech’ [cat], for example, that it is a ‘noun’ [n]. In constructional analyses, the notations for attributes and their possible values are typically presented within square brackets; thus, in this case, the expression ‘the part of speech is noun’ can be presented as [cat n], to be read as ‘a word’s value with respect to the attribute cat is n’. Further, a distinction is made between internal and external attributes and their values: An internal [cat n] can for instance be presented as embedded in an external ‘noun phrase’ [np] construction, giving us [np [cat n]].Footnote ² This type of notation is recursive and is also fully employed in CxD.

For presentational purposes, I refer to what I consider to be the ‘outermost’ external features as ‘discourse attributes’ in the present overview.

20.3.1 Wellerisms: Explicating the CxD Methodology

When you as a linguist encounter a piece of discourse like (1), which might not be readily comprehensible (even for the participants in a given interaction) in its particular context (especially when there is no Captain nor any spitting visible), and you want to figure out when, where, how, and why an expression like this can be used, you first of all need to gather further data – both similar kinds of expressions and also information about the context in which (1) occurred:Footnote ³

1. (1) “It all comes back to me now,” said the Captain, as he spat into the wind.

On the basis of our knowledge of spitting and the effects of a wind blowing against our face, there is clearly a humorous aspect to (1). More importantly, though, we should notice that (1) is no longer ‘the same kind’ of expression if the order of its elements were to be changed into (2a) or into (2b).

1. (2)

   a.	The Captain said, “It all comes back to me now,” as he spat into the wind.
   b.	As he spat into the wind, the Captain said, “It all comes back to me now.”

Expressions like (1) are called Wellerisms and they have a very set format. They start with some kind of quotation, followed most often by the finite verb say in the past tense, said, which, in turn, is followed by a specification of who uttered the quotation (here: the Captain); additionally, this sequence is often followed by an indication of the situation in which the quotation was uttered. For an ordinary grammatical description, this would hardly pose any challenges: the Captain is the subject, said is the predicate, finite verb, the quotation is the direct object, and the as-clause is a (temporal) adverbial. But this analysis would – from a grammatical point of view – equally well be a description of (2a) and (2b), which, as we noted, and as native speakers would agree, are not Wellerisms.

On this basis alone, we should be able to suspect, or even conclude, that (1) is not just an instance of an ordinary Subject-Predicate construction. The particular word-order constellation with its specific function, together with the use of the past tense said should make us entertain the possibility that here we have to do with a specialized manner of speaking; more precisely, that maybe there is something in the grammatical inventory of English that we might want to call a Wellerism construction that sanctions (in CxG terminology: that ‘licenses’) expressions like (1). Thus we may hypothesize about a Wellerism construction that licenses a special type of constructs.

If and when we come across and collect further examples like that in (1) and if we furthermore approach data with an open mind (i.e., without preconceived ideas about what a ‘sentence’ should look like), we find a recurring pattern of the very word said followed by the Subject who performed the saying. If (1) and similar expressions were examples from a previously unknown language, we might even want to say that expressions like said the Captain (and thus the construction that licenses such constructs) is a single constituent – a string of language that as such has a near-constant form in a Wellerism. In fact, Wellerisms suggest that the very notion of a ‘constituent’ needs to be revised if one’s grammar is to be truly based on actually occurring language data.

This is an important point because CxG has on and off been accused of being a ‘notational variant’ of structuralist or minimalist accounts of grammar. The idea of proposing that finite_verb+subject is one constituent is clearly not a notational variant of mainstream grammatical descriptions. But the question for CxG and CxD is of course not one of constituency as such; what is important is that there is a construction with its own internal structure, a Wellerism construction. A simplistic description of the parts of a Wellerism construction is given in Figure 20.1 in a fairly ordinary construction grammatical format.Footnote ⁴

Figure 20.1 The structural elements of the Wellerism construction

Once we have established the possibility that there is a Wellerism construction (in English and possibly in language in general), we need to specify what it is that makes the Wellerism construction different from, say, the Subject-Predicate construction. If it is not a matter of structural aspects (i.e., a mere change in word order will not be sufficient; cf. I’m going to work, said Dad, as he drove away), we have to focus on discourse aspects. And one feature that is clearly definitional of the Wellerism construction that licenses (1) is what we already alluded to and what can be specified as [humor]. That is, any predication can in principle be expressed in the format of a Wellerism (cf. [3]), but not all language users might see the humor in a structure that is only ‘dressed up’ as a Wellerism. For instance, an expression like (3a) might be dressed up as the wannabe Wellerism in (3b), but for most non-linguist addressees this will be an ‘ordinary’ OVSA sentence. However, for a pragmaticist and especially for a conversation analyst, (3b) is indeed a humorous (bordering on outrageous) rendering couched in the Wellerism mold.Footnote ⁵

1. (3)

   a.	In his lecture today, Joe Linguist said, “A text is simply a collection of sentences.”
   b.	“A text is simply a collection of sentences,” said Joe Linguist in his lecture today.

Thus, [humor] seems clearly to be an element of discourse that needs to be specified for a proper understanding of a Wellerism. That is, more generally, [humor] needs to be taken into account in a complete description of the CxD-grammar of a language. But [humor] might not be a primary feature of discourse; that is, when we start building a model or a theory of what major aspects of discourse context need to be taken into account, we will realize that [humor] is one type of how we show our [involvement] in our way of talking. And further analyses will show that [involvement] ([inv] for short) is one aspect of how we ‘implicitly anchor’ what we want to communicate, that is, how we subconsciously communicate something ‘between the lines’ of what we explicitly say. We thus end up with the following characterization of where humor fits in, with [implicit_anchoring] (when necessary, abbreviated as [ia]) being one of the central discourse attributes needed in any linguistic characterization: [implicit_anchoring [involvement humor]].

But humor is clearly not only something that expresses one aspect of one’s personal involvement in the discussion at hand. Admittedly, we can be humorous in our personal thinking (especially if we have ‘a rich inner life’) but typically, humor also involves the expression of humorous circumstances to a communication partner, that is, in an interaction where aspects of politeness (and impoliteness) are central. The expression of humor is also dependent on the culture and context when and where an attempt at being humorous is practiced. One simply cannot joke about everything with everybody, nor in all cultural settings. But in all these cases, we communicate – with humor – something between the lines of what we say or write. We implicitly anchor our message with humor. There are thus three major values that the discourse attribute [implicit_anchoring] has to cater for: [involvement] (as we saw above); interaction-politeness ([i_p] for short); and contextual and cultural coherence ([coh] for short).

All three of these [implicit_anchoring] values – [involvement], [interaction_politeness], and [cultural_coherence] – can be further specified (i.e., given further values) and made relevant in the specification of the discourse attributes of a construction. But again following Fillmore’s conception of CxG, if a possible specification in the model has no value for the description of a construction, there is no need to force an AVM specification in the description of a construction. This is analogous to descriptions of syntax; for example, if the word class of an element in a construction is irrelevant, that description is simply left empty or specified as ‘anything goes’, that is, in technical terms as [].

In order to further, albeit very briefly, explicate and illustrate how some additional discourse attributes and their values are established, this section will take a closer look at the Wellerism construction, that is, when, where, how, and why this construction can be used as a resource. In the spirit of engaging languages other than English – and thus also to counter the early critique of CxG as a grammar for English – I will here focus on a set of examples from a dialect of Swedish in Finland, the Solv dialect, where Wellerisms are actively and productively used.

The examples in (4) are actually occurring Wellerisms in Solv.Footnote ⁶ (See also Östman Reference Östman, Bücker, Günthner and Imo2015.)

1. (4)

   a. Fösst gaangon me tejje böksona, saa Pära-Freedrik, tå an too se in fyllo å hadd nyykåstyymin påsse.

   ‘First time with these trousers, said Pära-Fredrik, as he got drunk in his new suit.’

   b. No fåår dö pikk de å raak de å int entå, saa fłikkon åt ryssin.

   ‘You can make the cross sign and you can shave, but I still won’t make love to you, said the girl to the Russian.’

   c. He va bara ååvosiidon, saa fłikkon, tå on breend e hååł på förkłe sett.

   ‘It was only the reverse side, said the girl as she burned a hole in her apron.’

   d. - Å epłen haar dö?! - No så föbanna, saa Jokkas-Saandär. ‘- And you do have lots of apples!’ ‘- You’re damned right about that, said Jockas Sander.’

The examples in (4) illustrate (some of) the kinds of usage where Wellerisms are employed in Solv Swedish. Briefly, people are mentioned by name, connecting one’s discourse to people who live or have lived in the village. Typical habits in the village are narrated in a fragmentary fashion and recounted (like getting drunk every so often, example [4a]); there is a sense of humor (and not seldom of stupidity, cf. example [4c]) expressed in all the examples. Additionally, by using a Wellerism you are allowed to use swearwords (cf. föbanna in [4d]), which is typically looked down upon in the village, and Wellerisms also allow you to be xenophobic and chauvinistic (4b).

An explication of the discourse attribute [implicit_anchoring] will have to make reference to all three of its major values in a full description of the Wellerism construction.Footnote ⁷ The village-history identity aspect mentioned above is represented under [cultural_coherence] and needs to be given a further specification that makes clear that the Wellerism directly taps into the identity of people who are living or have lived in the community. In a very simplistic manner, this can be represented as [identity [communal]]. The whole sequence of specification for the Wellerism construction with respect to cultural coherence can thus be represented as in (5).

1. (5) [implicit_anchoring [cultural_coherence [identity [communal]]]]

The specification [identity [communal]] is the result of analyses of a large amount of discourse and communication, properly argued for in ongoing, as yet unpublished research. Briefly, alternative specifications to that of [identity], that is, alternative values of the attribute [cultural_coherence], include those of [ideology], [intervention], and [interdiscursivity]. And alternative values for [communal], that is, as possible values for [identity], we find [institutional] and [individual] among others. In this overview of how CxD works, such alternative attributes and values are not dealt with in detail.Footnote ⁸

As for the [implicit_anchoring] value [interaction_politeness], the Wellerism construction needs to be specified as responsive, since it is hard to imagine a situation where a Wellerism would be used out of the blue in an interaction. Responsivity will be specified under the [interaction_politeness] value [sequentiality], and the values for [sequentiality] are precisely those of [responsive] and [projective], with the latter referring to cataphoric reference in a discourse. Some of the values for [sequentiality] under [interaction_politeness] are [cooperation], [tact], and [equality_type]. And the last mentioned of these ([equality_type]) is clearly also applicable in a full understanding and description of the Wellerism construction as a generalization over examples like those in (4). The general characteristic feature of what was mentioned above as Wellerisms offering the opportunity for xenophobic and chauvinistic expressions is labeled [intolerance] in CxD. Xenophobia is exemplified in (4b) above, and a more direct chauvinistic expression is illustrated in the Solv Wellerism in (6).

1. (6) Akt öögona fö knappan, saa gobbin, tå an sloo tjelnjen me böksona.

   ‘Watch out for the buttons, said the man, as he hit his woman with the trousers.’

We have seen that there is a grain (and sometimes an excessive amount) of humor involved in Wellerisms – see also (1) – and this gets specified under the involvement attribute as [involvement [humor]]. In addition to [humor], some of the other values for [involvement] are those of [appraisal] and [affect], and the characterization [humor] can be further specified as [funny], [irony], or [sarcasm]. But in the present case – that is, in relation to a proper specification of the Wellerism construction – there is no need to specify [humor] further. In other words, a specific kind of humor in a Wellerism is not definitional of the Wellerism construction.

The discussion above has focused on the discourse attribute [implicit_anchoring] as a central attribute of the discourse attributes (in the sense of constructions being form–meaning–discourse constellations) that are needed to give a full contextual account of when, how, and why Wellerisms can be felicitously used. The description is based on examples from one language, the Solv dialect of Swedish in Finland, spoken in a dialect community where Wellerisms are in actual usage. Future studies will show the extent to which these attributes and values are applicable to the use of Wellerisms in other language communities. Figure 20.2 gives a general overview of the attributes and values of the discourse attribute [implicit_anchoring] that have been mentioned so far. This is not the full picture of this discourse attribute but showcases in a more accessible format only the relations between the attributes and values that have been discussed.

Figure 20.2 A partial overview of the values of the discourse attribute [implicit_anchoring]

Figure 20.3 is a rendering in the AVM-format (with the abbreviations introduced) of the discussion above of the [implicit_anchoring] attribute and its values relevant for the Wellerism construction.

Figure 20.3 The values of the [implicit_anchoring] attribute relevant for a detailed description of the Wellerism construction

I have gone through the values of the [implicit_anchoring] attribute in some detail in order to show how a CxD argumentation leads to establishing attributes and values. Other discourse attributes will also be relevant and necessary for a full description of the contextual specification of the Wellerism construction. I will not go through the argumentation for these discourse attributes in similar detail, especially not with respect to the values of attributes that are available in a full-fledged theoretical description of the workings of CxD.

In its present conception, there are twelve major discourse attributes, comparable to the [implicit_anchoring] attribute discussed in relation to the Wellerism construction. The twelve discourse attributes are the following:

• discourse prosody;

• text-level semantics;

• application (with the values naming and referring);

• discourse-level information structuring;

• (syntactic) construction packaging;

• grounding;

• non-verbal;

• speech community;

• sociolinguistic variables;

• discourse pattern;

• implicit anchoring; and

• functional relations.

In addition to [implicit_anchoring], the discourse attributes [discourse_pattern], [functional_relations], and [speech_community] will be especially relevant for a complete description of the Wellerism construction. I will briefly mention some of their values in the discussion below.

With respect to the external attribute [discourse_pattern] ([dp] for short), Wellerisms need to be context-characterized both as narratives and as fragments. The narrative aspect is clearly seen in (4a), which introduces a person in the village community and tells the story of him getting drunk in his new suit. But Wellerisms are furthermore fragmentary narratives which, by their very nature, not only require that a substantial amount of information be left out, but also that this cultural and contextual information be adequately added by one’s interlocutors in the community, that is, by the targets of one’s production of a Wellerism.

In the spirit of CxD, all this needs to be systematized and indeed formalized in order to function as a reliable account of which discourse factors influence the general characterization of the Wellerism construction. The attribute [discourse_pattern] covers issues of text and discourse and it has three main values that specify text and discourse from different perspectives: [discourse_frame] ([d-frame] for short), [activity_mode], and [text_type].

The [discourse_frame] focuses on the cognitive dimension of text and discourse: A text or discourse will call forth a particular image of what it (typically) looks and sounds like.Footnote ⁹ The second value under [discourse_pattern] caters for the sociocultural and interactional aspects of text and discourse, which are often talked about as registers or genres (Bhatia Reference Bhatia1993). Borrowing from Linell (Reference Linell2001), this value is called [activity_mode].Footnote ¹⁰

The third attribute is called [text_type] and is directly relevant for a proper characterization of the Wellerism construction. This attribute has the actual structure of text and discourse as its focus, that is, the message as text. Three main values as further attributes are distinguished: [overtly_strategic], [fragment], and [linguistic landscape] ([ll] for short).Footnote ¹¹ The [overtly_strategic] value takes the following further values: [narrative], [instructive], [descriptive], [expository], and [argumentative]. And further values under [narrative] are [eventive], [interpretive], and [illustrative].

Formalized in CxD fashion, the discourse-functional specifications in (7) emerge as relevant for the formal description of the Wellerism construction. A further specification of [narrative] as [illustrative] may be needed if further data so require.

1. (7)

   $\begin{array}{l}\left[\mathrm{d}\mathrm{p}\right[\mathrm{t}\mathrm{e}\mathrm{x}\mathrm{t}\_\text{\_}\mathrm{t}\mathrm{y}\mathrm{p}\mathrm{e}[\mathrm{o}\mathrm{v}\mathrm{e}\mathrm{r}\mathrm{t}\mathrm{l}\mathrm{y}\_\text{\_}\mathrm{s}\mathrm{t}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{g}\mathrm{i}\mathrm{c}[\mathrm{n}\mathrm{a}\mathrm{r}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{v}\mathrm{e}]]]]\\ \left[\mathrm{d}\mathrm{p}\right[\mathrm{t}\mathrm{e}\mathrm{x}\mathrm{t}\_\text{\_}\mathrm{t}\mathrm{y}\mathrm{p}\mathrm{e}\left[\mathrm{f}\mathrm{r}\mathrm{a}\mathrm{g}\mathrm{m}\mathrm{e}\mathrm{n}\mathrm{t}\right]\left]\right]\end{array}$

The discourse attribute [functional_relations] ([fr] for short) caters for the dynamic aspects in language and is built up on a similar basis as how Lyons (Reference Lyons1968) discussed sense relations (antonymy, hyponymy, etc.) in the lexicon. One central value of [functional_relations] is that of [negotiability] with the following main values: [responsibility], [accountability], [ethics], [appropriation], and [complicity]. The important aspect for the Wellerism construction is that of [responsibility] since by using a Wellerism you can indicate that you are not yourself responsible for the quotation (cf. [phon [quotation]] in Figure 20.1) you are mentioning when you use a Wellerism. This comes in handy in a community like the Solv Swedish one, where swearing is not considered acceptable; if you happen to swear, you can thus rapidly make your expletive followed by an addition that somebody else had used that word: ‘said X when Y’. Then, naturally, you yourself have not cursed. Details of the attribute [responsibility] are dealt with in Östman (subm.) and, more generally, in Solin and Östman (Reference Solin, Östman, Östman and Solin2016) and Ädel et al. (Reference Ädel, Nyström Höög, Östman, Ädel and Östman2023). A CxD summary of these prior results shows that [responsibility] has the four major values specified in (8), with some of their further values indicated.

1. (8)

   [linguistic]:	[explicit] and [implicit]
   [cause]:	[moral] and [legal]
   [expectation]:	[individual] and [collective]
   [agentivity]:	[in-focus] and [out-of-focus]

In order to further specify the importance of the collective Solv community and its views (e.g., on swearing), the Wellerism construction also needs to specify that there are negative expectations (e.g., with regard to swearing) from the collective community. All this is most clearly illustrated in (4d). Thus, in order to capture the function and use of Wellerisms for the purpose of saying something that one does not want to be responsible for saying, the systematization in (9) is required.

1. (9) [fr [negotiability [responsibility [expectation [collective [neg]]]]]]

The discourse attributes and their potential values discussed above are specifications that need to be incorporated into a full CxD description of Wellerisms – at least for Solv Swedish. Further detailed investigations of Wellerisms in other languages will show whether these are general features of the Wellerism construction in language.

One very important aspect in relation to the description in terms of attributes and values is that a ‘thread’ of attributes-and-values is not to be seen as a choice that excludes other choices within a specific discourse attribute. The metaphor is rather that the end-points of the threads (i.e., the ultimate values) are like synapses that can be activated to different degrees once they have been established as being part of the discourse attributes and values of a construction. In this respect, Firth’s dictum of ‘meaning implies choice’ cannot be directly used for pragmatic, functional choices. Rather, the principle must be ‘discourse function implies a set of simultaneous choices’.

The prototype for the Wellerism construction with the discourse attributes and their values added is given in Figure 20.4.

Figure 20.4 Prototypical Wellerism construction with discourse attributes and values

The values are there ‘in the background’ as potential functions in the prototypical Wellerism construction. In every reference to the Wellerism construction, some aspect is typically more highlighted than others. But since there is always the possibility of misunderstanding, such mix-ups can directly be explained (and thus cognitively understood and managed) by reference to the prototypical construction.

20.3.2 Accounting for Variation in CxD

Since CxD sees constructions as prototypes, there have to be ways of showing how non-prototypical cases are handled.Footnote ¹² In CxD, it is precisely with the help of the discourse attributes and their values that variations of the prototype can be described. In Figure 20.4, a convention with pointed brackets is used: < … >. This is an arrangement introduced in Leino and Östman (Reference Östman, Östman and Fried2005) to indicate variable possibilities. A sequence of the form ‘attribute < x; y >’ indicates that an attribute can have different values, x or y. The different values can be in free variation, as seems to be the case with the choice between the allative (all) and the ablative (abl) case co-occurring with the themes of perception verbs in Finnish, as shown in (10), taken from Leino and Östman (Reference Östman, Östman and Fried2005: 194):

1. (10)

   a.	Tuo	haisee	pahalta.
   	that	smell.3sg	bad. abl
   b.	Tuo	haisee	pahalle.
   	that	smell.3sg	bad. all
   	‘That smells bad.’

But the different values can also be dependent on discourse variation, and if that dependence is a systematic feature, this can be marked in a construction and especially in particular constructs under consideration, with reference to discourse attributes and their values.

For instance, in relation to what I have called the situation_specification in the Wellerism construction (Figure 20.1), there is variation as to whether this part of the Wellerism construction is expressed as a temporal clause such as when/as, cf. (4a), (4c), (6), as a prepositional phrase (4b), or whether there is no such specification (4d). There is also at least one further possibility, exemplified in (11) from standard Swedish.

1. (11) “Det löser sig,” sa han som sket i badkaret.

   ‘“It will solve itself/dissolve,” said he, who shat in the bathtub.’

In (11), the situation_specification is presented as a relative clause. Östman (Reference Östman, Herlin, Kalliokoski, Kotilainen and Onikki-Rantajääskö2002) shows that expressions with relative clauses are much more preferred in Swedish in Sweden, whereas a temporal clause structure such as (4a), (4c), or (6) is the typical one used in Swedish in Finland (and in Finnish). In order to capture differences like these, reference can be made to the discourse attribute [speech_community] ([sc] for short) and especially to its value and further specification [size [regional]].Footnote ¹³ Regions can then be further specified – in this case as [SweSwe] ‘Sweden Swedish’ and [FinSwe] ‘Finland Swedish’. These values can then be linked in the customary constructional manner to features in the internal structure of a construction, as shown in Figure 20.5.

Figure 20.5 The Wellerism construction with specifications about the variability between Sweden Swedish and Finland Swedish

In a similar vein, Östman (Reference Östman2021) shows how to deal with variation in the prototypical TatT construction, exemplified by There are boys, and then there are boys. Typically, the final phrase is in focus in a TatT construction but this can, for example, be turned around in advertisements: There is UC-Berkeley, and then there are universities. This variation can be handled and specified with reference to [grounding] and a switch in Figure and Ground.

Detailed analyses of the variational possibilities of prototypical constructions can also result in the necessity to rethink the internal attributes and values. Thus, in their discussion of names and name phrases, Östman and Mattfolk (Reference Östman, Mattfolk, Heikkola, Paulsen, Wojciechowicz and Rosenberg2021) show that names are not only a matter of lexical semantics but also one dependent on values of the discourse attribute [application]. Their way of handling the variation between Paris as a proper noun and Paris as a common noun in examples like (12a–b) is not to mark one as [proper +] (12a) and the other as [proper −] (12b); the differences go much deeper:

1. (12)

   a.	We always go to Paris in the spring.
   b.	Tromsø is the Paris of Northern Europe.

On the basis of detailed analyses of name phrases, Östman and Mattfolk suggest that expressions like (12a) are licensed by the Name-Phrase construction, with special reference to the discourse attribute [application [unique]], whereas (12b) is licensed by the discourse feature [application [trite]]. Irrespective of further analyses of these structures (e.g., that Paris in (12b) is necessarily followed by a prepositional phrase), both uses share semantic and discourse features as lexical and phrasal constructions. In their further analysis, the authors then suggest that both the Name-Phrase construction and the Determination construction (cf. the NP in other models) inherit a more general Nominal construction.

That is, a thorough discourse-pragmatic investigation of – in this case – the variation in nominal expressions shows that the basic structure of nominal phrases might have to be re-evaluated. This is in line with Fillmore’s idea that one’s grammar does not have to be complete and that further and more thorough investigations will reveal patterns that might at first sight seem random.

20.4 Going Further and Beyond

The overview of CxD in this chapter has of necessity been abstract, and an arsenal of seemingly arbitrary terms and categories have been introduced. It has been an attempt to pull together information dealing with discourse and pragmatics (if not languaging generally) that have been published and discussed, with a further attempt at gathering them together into a systematic scheme. To me, this is the very point of constructional approaches, that is, that they have to cover everything related to language. Irrespective of whether the reader is comfortable with the CxD formalism or not, the aspects of discourse and context that have been briefly dealt with in this chapter are aspects that any constructional approach to discourse has to take into account and relate to. Further, since the approach in CxD is very much in opposition to minimalist approaches, CxD can with good reason be called a ‘maximalist’ approach to language. The presentation in this chapter has been couched in CxD terms precisely in order to show the systematicity that can be accomplished with respect to context and discourse pragmatics.

As should have become clear, there are two complementary facets of discourse-pragmatic function in relation to the discourse attributes discussed: discourse function as context and discourse function as what takes place implicitly in communication, that is, between the lines of what is communicated. Put differently, discourse-function-as-context explicates the set of contextual attributes that restrict (or permit) the usage of a linguistic expression, whereas discourse-function-as-implicitness highlights the co-effectiveness and co-occurrence of values on each of the discourse attributes. As noted above, in terms of how the system works, values on the discourse attributes are to be seen as comparable to synapses in the brain that get activated simultaneously and in a synergetic manner. Further, in terms of linguistic ontology, the discourse attributes are discourse-pragmatic frames. In this respect, Fillmore’s notion that Construction Grammar and Frame Semantics are sister theories has been collapsed into one model in CxD.

It is a pertinent question to ask what is gained by having a single system and thus presenting everything in one construction (as in Figure 20.5), rather than having information in several constructions (and frames) related through and organized in a network. The CxD answer to this is – again, pertaining to Fillmore’s original theoretical insights – that we should as much as possible avoid having ‘interfaces’ or other kinds of bridge-building sub-theories. Future research will show whether this is a good strategy or not; but for the moment, this is the CxD approach.

There are undoubtedly further aspects of discourse that have not been covered by the twelve discourse attributes that at present constitute the backbone of CxD. I will mention some of them here and note that they may have to be added to the present set-up of discourse attributes, or that interpretations of some of the existing ones can be changed. They are all, however, aspects that need to be taken seriously in future research from a discourse-constructional perspective. Some of these issues have been dealt with extensively in other disciplines, so there is ample information to work with in order to address them in future studies.

One such aspect is that of polyphony, heteroglossia, and intertextuality. My present view is that these aspects can be specified under [implicit_anchoring] and more precisely under [cultural_coherence]. But these matters are clearly in need of further systematic and coherent overall analyses.

A further aspect that requires attention are more or less artistic aspects of linguistic expression. Again, this is an area that requires further study but provisionally, it will need some indication under [discourse_pattern [text_type]]. For instance, shape poems, a.k.a concrete poems or calligrams, and – more generally – aspects of geometry, for example, issues related to the Golden Section, need to be addressed in a full-fledged attempt at dealing with discourse.

As noted above, the manner of presenting discourse-pragmatic aspects in this chapter is just one way of doing it but one that is in line with the basic Fillmorean vision for constructional analyses. The synopsis in this chapter has attempted to give a basic overview of what aspects need to be systematically taken into account in relation to how constructions function in discourse and, more precisely, what discourse constructions have to take into account in order to provide a full-fledged understanding of constructional analysis.