10.2.1 Constructs and Corpora

First of all, we need to maintain a distinction between empirical ‘constructs’ on the one hand and theoretical notions such as ‘constructions’ on the other. An (empirical) construct is an actual linguistic expression, such as a word, phrase, or sentence. Constructs can be observed and collected in a ‘corpus’. Example (1) offers a small corpus of constructs from Franz Kafka’s Metamorphosis (translated by Edwin and Milla Muir).

1. (1) As Gregor Samsa awoke one morning from uneasy dreams he found himself transformed in his bed into a gigantic insect. He was lying on his hard, as it were armor-plated, back and when he lifted his head a little he could see his dome-like brown belly divided into stiff arched segments on top of which the bed quilt could hardly keep in position and was about to slide off completely. His numerous legs, which were pitifully thin compared to the rest of his bulk, waved helplessly before his eyes.

10.2.2 Constructional Analyses and Analysis Banks

Everything beyond constructs “is part of a theory, invented by a linguist” (Dowty Reference Dowty, Bunt and van Horck1996: 12). A first theoretical concept is that of a ‘constructional analysis’ (of a construct), which represents a linguist’s model of a construct similar to how phonemes represent a linguist’s analysis of actual sound patterns in vocal languages or actual signs in signed languages. For instance, while example (1) contains the actual construct a gigantic insect, a linguist may ‘model’ that construct in a more abstract way in order to identify generalizations about a language, such as using a phrase structure tree (example [2]). A widely used representation in the constructional literature is the ‘boxes-within-boxes’ notation (Fried & Östman Reference Fried, Östman, Fried and Östman2004) of example (3), in which the tree structure of example (2) is represented as nested boxes, with the outer box representing a parent node (also called a ‘Mother’ node) and with the inner boxes representing its children nodes (also called ‘Daughters’). Each box also has a ‘feature structure’ (consisting of pairs of features and values) for adding more fine-grained linguistic information, such as the features ‘cat’ (category) and ‘lxm’ (lexeme), and values such as ‘np’ (noun phrase) or the lexeme gigantic. As we will see later in this chapter, constructional analyses are often more complex than tree structures, so instead of collecting tree structures in a treebank, constructional analyses can be stored in a ‘constructional analysis bank’.

1. (2)

   

1. (3)

   

By far most of the constructional analyses in the literature are ‘pen-and-paper’ insights performed by linguists themselves. These analyses are almost never presented with all of their details, either because the construction grammarian is using informal diagrams, or because the full analysis involves too much information to fit within the space limits of a publication. Alternatively, an analysis can also be constructed by a computational construction grammar. In this case, a linguist typically uses a computational tool for operationalizing constructional knowledge, which is then applied for automatically analyzing or producing constructs. The linguist can then verify whether the analysis corresponds to their expectations, which they can use either to validate their theory or for identifying which parts need further refinement. The computational model can then be shared as supplementary materials to a publication, allowing other researchers to verify and reproduce its analyses, or be made available as a web demonstration.

The term ‘constructional analysis’ does not really occur in the literature in the same sense as used in this chapter. In fact, many construction grammarians never clearly distinguish between what they call a ‘construction’ and what is called here a ‘constructional analysis’ and what we defined earlier as a ‘construct’. It is therefore common to find statements such as that the constructional analysis of example (3) is ‘an instance of a Determiner-Adjective-Noun construction’, or that the construct she gave her son a present is ‘an instance of the Ditransitive construction’. As we will see later, such phrasing is confusing and it would be more precise to say that a constructional analysis such as example (3) is accounted for by a combination of several constructions, including not only the Determiner-Adjective-Noun construction, but also lexical constructions that contribute information about the words a, gigantic, and insect.

Unsurprisingly, the more formal constructional approaches make the distinction clear between constructions and constructional analyses. In CxG and Sign-Based Construction Grammar (SBCG), a constructional analysis is called a ‘modeling-domain object’ (Kay Reference Kay2002), or a ‘linguistic object’ (Sag Reference Sag, Boas and Sag2012). In Fluid Construction Grammar, a constructional analysis is called a ‘(final) transient structure’ (Steels Reference Steels2017).

10.2.3 Constructions and Constructicons

Obviously, it is impossible to establish an exhaustive corpus of all the constructs of a language “because all languages allow an indefinitely large number of sentences” (Haspelmath Reference Haspelmath, Heine and Narrog2009: 290), hence it is equally impossible to develop an exhaustive constructional analysis bank. Descriptive linguistic theories therefore try to describe recurrent patterns of language usage in a way that accurately predicts the linguistic behavior of the members of a linguistic community. Traditionally, such recurrent patterns have been called ‘constructions’. When used in this sense, a linguist might say for instance that the English language employs the ‘Progressive Construction’ for expressing ongoing events using the pattern ‘Auxiliary-BE + V-ing’ as exemplified in the sentence I am learning how to develop computer games.

Since the Chomskyan revolution in the 1950s–1960s (Chomsky Reference Chomsky1957, Reference Chomsky1965), cognitive linguistic theories have emerged that aim to account for the linguistic knowledge that a language user needs for producing and comprehending novel constructs. Chomskyans assume that such knowledge takes the form of a generative device that applies a small set of abstract rules (Chomsky Reference Chomsky1981, Reference Chomsky1995). In this view, constructions exist only as the ephemeral side-effects of more fundamental operations and are therefore inconsequential for linguistic theorizing. As a result, there has been a wide gap between descriptive and cognitive linguistic theories. Indeed, the linguist Norbert Hornstein has attempted to rebrand descriptive linguistics as ‘languistics’, keeping the term ‘linguistics’ exclusively for theories that focus on mental models of linguistic competence (Leivada Reference Leivada2014). The major innovation of Construction Grammar, then, was to reconceptualize the traditional notion of a construction as the basic unit of linguistic knowledge (Fillmore et al. Reference Fillmore, Kay and O’Connor1988). In other words, for a language user to ‘know their language’, they simply have to know its constructions (Fried & Östman Reference Fried, Östman, Fried and Östman2004), which could finally bridge the gap between descriptive and cognitive linguistic theories.

All flavors of construction grammar, except for SBCG, therefore assume that a language user’s linguistic inventory is a structured network that exclusively consists of constructions (Diessel Reference Diessel2019). Since the construction inventory does not sharply distinguish between lexical and grammatical constructions, it is called a ‘constructicon’. SBCG is the odd one out by distinguishing between ‘listemes’ (which roughly corresponds to the ‘lexicon’) and phrasal constructions, though it still offers a uniform way to model the syntax–lexicon continuum (Michaelis Reference Michaelis, Busse and Moehlig-Falke2019). Fluid Construction Grammar, which supports any kind of constructional analysis, assumes a single constructicon by default but allows its users to distinguish lexical from grammatical constructions if they wish to do so.

10.2.4 The Relation between Constructions and Constructional Analyses

The relation between constructions on the one hand, and constructional analyses of constructs on the other, is different depending on the constructional flavor. As a rule of thumb, constructions can be thought of as (partially) schematic pieces of conventionalized linguistic information (or ‘linguistic patterns’), while a constructional analysis is always specific to the construct it represents. For example, the English Determiner-Adjective-Noun pattern in example (4) abstracts away from the specific analysis of example (3) by leaving the lexemes of the inner boxes unspecified. This pattern is therefore not only compatible with the construct a gigantic insect, but also with any other construct that follows the same pattern, such as his numerous legs.

1. (4)

   

Most construction grammarians will say that constructions can be combined with each other for producing or comprehending utterances, but it is almost never spelled out exactly how constructions combine. In fact, most constructional flavors, including those that subscribe to the tenets of usage-based linguistics (Diessel Reference Diessel and Aronoff2017), uphold some distinction between competence and performance (Chomsky Reference Chomsky1965) in practice and only focus on constructions as pieces of linguistic knowledge without modeling how that knowledge should be processed.

The strongest competence-performance distinction is found in the so-called ‘constraint-based’ construction grammars, particularly SBCG, which aims at processing-neutral competence models (Sag & Wasow Reference Sag, Wasow, Borsley and Börjars2011; Sag Reference Sag, Boas and Sag2012). In a constraint-based grammar, constructions are defined as ‘partial descriptions’ of well-formed structures or, put more precisely, as ‘constraints’ on what the grammar will accept as well-formed structures. An intuitive way to think about an SBCG grammar is to compare it to the list of building regulations that a safety inspector must check in order to approve a house, without having to build the house themselves (which is the job of the contractor). Likewise, SBCG constructions are formulated as constraints that can be checked independently from each other in order to estimate the degree of well-formedness of a construct, without having to be combined to actually form the construct (which is the job of the processor).

A second kind of construction grammars are ‘usage-based’ (Diessel Reference Diessel and Aronoff2017), which includes, among others, Radical Construction Grammar (Croft Reference Croft2001), Cognitive Construction Grammar (Goldberg Reference Goldberg2006, Reference Goldberg2019), and a lot of work in Berkeley Construction Grammar (see further below). Usage-based construction grammarians assume that language learning and usage have a profound impact on linguistic structure and therefore typically address their attention to questions of language acquisition (Diessel Reference Diessel2004; Goldberg et al. Reference Goldberg, Casenhiser and Sethuraman2004; Lieven Reference Lieven2009), language change (Croft Reference Croft2000; Fried Reference Fried2009), and language processing (Goldberg Reference Goldberg2019). However, such studies will typically use verbal explanations for the role of constructions in production and comprehension but rarely offer an explicit processing model of how constructions can be combined with each other, as lamented by Bod (Reference Bod2009).

(Berkeley) Construction Grammar offers a versatile approach that can be classified as constraint-based, but which has also been fruitfully applied for usage-based studies (e.g., Fried Reference Fried2009), contrary to the common misunderstanding that CxG is non-usage-based (see, e.g., Goldberg Reference Goldberg2006; Hoffmann Reference Hoffmann2022). In fact, CxG is the only theoretical approach in our comparison that proposes a detailed model of how two constructions can be combined with each other: ‘Kay unification’ (Kay & Fillmore Reference Kay and Fillmore1999; Kay Reference Kay2002).Footnote ¹ Unification is a complex concept that has many different instantiations (Knight Reference Knight1989), but for our purposes it suffices to understand it as a way to combine two constructions with each other as long as they do not contain conflicting information. In CxG, a construct is therefore said to be licensed if there exists a set of constructions that can be unified with each other to form the constructional analysis of the construct. However, just like the other constructional flavors, CxG does not propose a processing model for actually computing a constructional analysis.

Linguists who are nevertheless interested in developing models of constructional language processing can do so using Fluid Construction Grammar (Steels Reference Steels, Daelemans and Walker2004, Reference Steels2017), which is a special-purpose programming language for implementing construction grammars.Footnote ² FCG can be downloaded both as open-source software, or as part of the FCG Editor, an integrated development environment for construction grammars (van Trijp et al. Reference van Trijp, Beuls and Van Eecke2022). In FCG, constructions are used for mapping meanings onto constructs (‘production’, or ‘formulation’ in FCG terminology), or vice versa, for decoding a construct into its underlying meaning (‘comprehension’). Details of processing can be inspected through an interactive web interface and modified with the help of a grammar configurator. In FCG, constructions are not combined directly with each other but operate on a structure called a ‘transient structure’, which contains all the information about a construct, starting from the meaning to express in production or the form to analyze in comprehension. The word ‘transient’ highlights the fact that this structure changes over time as more constructions contribute information, until a ‘final transient structure’ is achieved, which corresponds to what I call ‘constructional analysis’ in this chapter. We will return to these matters in Section 10.5.

10.3.1 The Distributional Method Revisited

One of construction grammar’s oldest problems has been how to identify a construction based on a corpus of empirical evidence. For a long time, it was popular to posit the existence of a construction as soon as a pattern could be found whose form or meaning could not be reduced to the sum of its parts, or which could not be formed through the composition of other constructions. Goldberg formulated this view as follows:

Such a definition implies that there is no room for redundancy in the constructicon, which was indeed the prevalent view in early Construction Grammar and which is still true for SBCG. Nowadays, however, most construction grammarians argue that any pattern that is frequent enough can be stored as a construction, even if doing so would be redundant. That brings us back to the question how a construction can be identified.

The constructional theory that has been most explicit about how constructions can be identified is Radical Construction Grammar, which grew out of a thorough re-examination of what kind of syntactic argumentation is warranted for describing linguistic structures (Croft Reference Croft2001, Reference Croft, Östman and Fried2005, Reference Croft, Boye and Engberg-Pedersen2010). Most importantly, Croft formulates his methodological arguments in a way that does not presuppose a constructional approach, but in a way that approaches the empirical data as neutrally as possible (outlined in more detail in Chapter 17). Too often in the past, linguists have wittingly or unwittingly described languages from the viewpoint of well-studied languages such as Latin or English, or they have applied unwarranted assumptions from a linguistic framework (Haspelmath Reference Haspelmath, Heine and Narrog2009). To avoid such pitfalls, Croft (Reference Croft, Boye and Engberg-Pedersen2010: 314) argues that the ‘distributional analysis’ remains “the most important and soundest methodology in syntactic analysis,” provided that it is modified to relate utterances to actual usage events. Let us first look at how the distributional analysis can be used for examining the occurrence or non-occurrence of items in the ‘slots’ of a construction before addressing the question of how to identify the constructions themselves, using a couple of simple examples from Croft (Reference Croft2001: 12), repeated here in examples (5)–(8).

1. (5)

   a.	Jack is cold.
   b.	*Jack colds.

1. (6)

   a.	Jack is happy.
   b.	*Jack happies.

1. (7)

   a.	*Jack is dance.
   b.	Jack dances.

1. (8)

   a.	*Jack is sing.
   b.	Jack sings.

In the (a) sentences in examples (5)–(8), the words cold, happy, dance, and sing follow the inflected copula be. This is acceptable for cold and happy, but not for dance and sing. In the (b) sentences, there is no copula and the four words appear with the suffix -s. These sentences are acceptable for dance and sing, but not for cold and happy. Cold and happy therefore have the same distribution, while dance and sing share a different distribution.

Now consider examples (9)–(10) in which happy and cold appear after a determiner and before a noun, which again confirms their shared distribution pattern.

1. (9) a cold winter

1. (10) a happy person

Again, cold and happy have the same distribution pattern. Traditionally, such data have been used as evidence for the existence of the part-of-speech adjective as an abstract category. However, a group of words that Boyd and Goldberg (Reference Boyd and Goldberg2011) call a-adjectives resist occurring in a prenominal position, as shown in examples (11)–(12).

1. (11)

   a.	the boy is afraid
   b.	??the afraid boy

1. (12)

   a.	the man is asleep
   b.	??the asleep man

The empirical record thus shows that words such as afraid and asleep do not share the exact same distribution pattern as cold and happy, which is problematic for traditional approaches that classify all four words as adjectives. As noted by Croft (Reference Croft2001), a common ‘solution’ to such problems is methodological opportunism whereby the linguist decides in an ad hoc manner to ignore certain data or criteria to fit their analysis, even if it means positing abstract categories that are not borne out by the evidence. Another common way to handle these problems is to introduce additional theoretical apparatus for saving the analysis (also see the discussion on Active vs. Passive in Section 10.3.2), for which no empirical grounds exist either.

Croft (Reference Croft2001) argues that the problems of the distributional analysis stem from the fact that most linguistic theories are ‘reductionist’. That is, most linguistic theories try to reduce complex phenomena to ‘atomic’ (i.e., irreducible) building blocks such as parts of speech that are more manageable. Construction grammarians, on the other hand, assume that the ‘primitive’ units of language (constructions) are inherently complex themselves. ‘Primitive’ here means that constructions cannot be explained in terms of their components because the whole construction is greater than the sum of its parts. RCG takes this constructions-as-the-primitives-of-language assumption to its logical conclusion and, therefore, argues that a construction’s components are construction-specific and should be defined in terms of their role within a specific construction.

From a theoretical perspective, it may seem counterintuitive to give up atomic categories for more complex primitives. However, it is well known from other fields that more sophisticated building blocks ultimately lead to simpler solutions for complex and large-scale problems. A good example is architecture: You can easily build a dog house using basic materials such as a sheet of plywood, clams, and screws, but those materials would never scale to support the size of a house for humans. Likewise, it is easy to account for declarative sentences in English using simple rules such as phrase structure rules, but trying to apply them to more complex structures or typologically different languages rapidly leads to convoluted analyses that require additional mechanisms such as transformations, lexical rules, or global principles – all of which can be discarded when constructions are used (van Trijp Reference van Trijp2014, Reference van Trijp2015). Indeed, the constructional approach “provides a uniform model of grammatical representation and at the same time captures a broader range of empirical phenomena” (Croft Reference Croft2001: 17) – all the while using a simpler scientific metalanguage than other approaches.

RCG has the simplest scientific metalanguage of all constructional flavors. All that is needed for describing a language is a taxonomy of constructions and all that is needed for describing the internal structure of constructions is part–whole relations: “The syntactic structure of constructions consists only of their elements (which may also be complex …) and the roles that they fulfill in the construction” (Croft Reference Croft2001: 5). Categories are therefore not global pieces of knowledge but construction-specific, so every construction can be described in its own (often quirky) way, which avoids problems such as leaky abstractions. Returning to examples (5)–(12), RCG would therefore not introduce an abstract category such as ‘adjective’, but simply posit separate constructions for constructs such as John is happy and for constructs such as a happy person. Both constructions have their unique slot for elements that can play a modifier role and both have their own conditions on which elements may fill that slot. While the first construction happily allows the so-called a-adjectives to play the modifier role, the second one is more restrictive.

So far we have only focused on the occurrence and non-occurrence of elements in the slots of a construction but we haven’t answered the question yet how constructions themselves can be identified. Croft (Reference Croft2001: 236f.) writes that constructions have a “structural Gestalt” that can be recognized as a pattern by language users. For example, the English Passive has several unique cues that set it apart from other constructions, such as the combination of auxiliary-be and the past participle, and a by-phrase with the Agent or Source of the action.

Since RCG is a cognitive theory of language just like the other flavors, it offers a theory about the linguistic knowledge that language users need to have for producing and comprehending constructs. However, its careful treatment of data and simple scientific metalanguage also makes RCG a good starting point for a descriptive theory of a language without making claims about cognitive plausibility, and Croft (Reference Croft2001: 59–61) provides a brief overview of how a reference grammar in RCG would look like.

10.3.2 The Collostructional Analysis and Its Extensions

While RCG caters for linguists who need to approach an undescribed language and therefore need to discover in an unbiased way which constructions exist, most empirical studies in the construction grammar community typically start by already positing a construction, and then use the ‘collostructional analysis’ for characterizing its meaning in a more precise and objective way (Stefanowitsch & Gries Reference Stefanowitsch and Gries2003; see also Chapters 6 and 7). The name ‘collostructional’ is a combination of ‘construction’ and ‘collocational’, the latter being a term from corpus linguistics that designates a series of forms that co-occur more frequently than would be expected by chance. The collostructional analysis can thus be seen as a kind of collocational analysis that assumes the existence of a construction and then identifies which forms are associated with the slots of the construction. The underlying assumption is that a construction will attract semantically compatible lexemes, while repelling others. The collostructional analysis is in principle compatible with any constructional flavor, but has in practice mostly been carried out within the framework of Cognitive Construction Grammar (Goldberg Reference Goldberg2006, Reference Goldberg2019). Just like RCG, CCG is usage-based but with a stronger focus on the psychological aspects of language learning (Tomasello Reference Tomasello2003; Diessel Reference Diessel2004; Lieven Reference Lieven2009) and processing (Boyd & Goldberg Reference Boyd and Goldberg2011; Goldberg Reference Goldberg2019).

The collostructional analysis has been extended in several areas. For instance, diachronic collostructional analysis (Hilpert Reference Hilpert, Allan and Robinson2012) tracks changes in collocational patterns over time using temporally ordered corpora. Another popular extension, called ‘distinctive-collexeme analysis’, is geared towards the investigation of pairs of semantically similar constructions, also known as ‘alternations’ (Gries & Stefanowitsch Reference Gries and Stefanowitsch2004). One of the most famous ‘alternating’ pairs in English is the Active versus Passive voice, as illustrated in examples (13)–(14).

1. (13)

   The boy	kicked	the ball.
   SUBJ	Verb	OBJ

1. (14)

   The ball	was kicked	(by the boy).
   SUBJ	Aux-be Verb-ed	byAgent

A distinctive-collexeme analysis can help refine an analysis of both constructions, or compare competing theories. Here we can contrast a ‘lexicalist’ versus a ‘constructional’ approach. In lexicalist and other mainstream approaches, the English Passive has typically been treated as a derivational rule, which is essentially a syntactic operation that turns an active verb into a passive one. In a constructional approach, the Passive Construction is a first-class citizen just like the Active Construction, and therefore has its own semantics.

Gries and Stefanowitsch (Reference Gries and Stefanowitsch2004: 108) note that only a few studies exist that have attempted to characterize the meaning of the Passive construction independently from the Active, but find a proposal by Pinker (Reference Pinker1989: 91), who defines the meaning of the Passive as “X is in the circumstance … for which Y is responsible,” with X referring to the Passive subject and Y to the agentive by-phrase. Gries and Stefanowitsch (Reference Gries and Stefanowitsch2004: 108) argue that if the Passive is essentially a syntactic derivation (as hypothesized in the lexicalist approach), one would not expect to find any strong differences in which verbs would fit the Active or Passive sentence types. If, on the other hand, the Passive Construction has its own semantics (as hypothesized by the constructional approach), one would expect to find many distinctive collexemes.

The results of their study show that the constructional approach makes the correct prediction. The most distinctive collexemes for the Active Construction (e.g., have, think, get) are verbs whose direct objects are not easily construed as patients (e.g., stative verbs). The most distinctive collexemes for the Passive Construction (e.g., base, concern, involve) confirm the definition of Pinker (Reference Pinker1989: 91) and “overwhelmingly encode processes that cause the patient to come to be in a relatively permanent end state” (Gries & Stefanowitsch Reference Gries and Stefanowitsch2004: 109). They conclude that “the distinctive-collexeme analysis shows that passive voice is a construction in its own right with its own specific semantics.”

10.4.1 Scenes and Frames

The central slogan of Frame Semantics is that meanings are relativized to scenes (Fillmore Reference Fillmore and Zampolli1977; see also Chapter 1). Scenes can be grounded in the real world through our sensorimotor apparatus, or they can be cognitive scenes that are evoked by constructs. For instance, a verb such as to cook activates an entire cooking scene such as the one depicted in Figure 10.1. Cooking is an extremely rich experience, but people can learn to construct and use a ‘semantic frame’ to make sense of what is happening. A semantic frame is a skeletal representation of recurrent experiences with open slots (called ‘frame elements’) that need to be filled in. According to the English FrameNet database (Baker et al. Reference Baker, Fillmore and Lowe1998), the Cooking_Creation frame involves at least twelve frame elements, including prominent ones such as Cook and Produced_food, but also roles such as Container, Ingredients, Place, and Recipient. Example (15) provides some constructs that evoke the frame.

1. (15)

   a.	[Shecook] is preparing [vegetable soupproduced_food] in [an orange potcontainer].
   b.	[Shecook] loves cooking in [her new kitchenplace].
   c.	[Shecook] loves cooking for [her guestsrecipient].
   d.	[The vegetable soup produced_food] is [slowlymanner] cooking.

Figure 10.1 How a Cooking-frame can be used for making sense of the activity in the picture

(photo by Jason Briscoe on Unsplash)

As can be seen, the relation between a semantic frame and its morphosyntactic encoding in language is indirect: Only a handful of frame elements are actually realized in a construct, none of them is guaranteed to appear, and the same frame element may play a different grammatical role depending on the construction in which it occurs. Semantic frames are also not tied to a single lexical unit (e.g., both preparing and cooking may evoke the cooking frame in example [15]), and vice versa, a lexical item may evoke different frames (e.g., one can also prepare a presentation, which usually involves little to no cooking).

Different constructional flavors provide different solutions for handling the indirect relation between meanings and constructs, which to a large extent depends on how strongly the flavor is embedded in a broader theory of cognition. To compare the different flavors to each other, we will take the diagram in Figure 10.2 as a starting point. This figure, based on Fried (Reference Fried2005), shows the different steps that an analysis (potentially) has to address in order to explain the different ways in which semantic frames can be expressed and understood. The figure comprises four mapping processes, with each mapping adding a layer of abstraction: (i) between scenes and semantic frames, (ii) between the frame elements (FEs) of semantic frames and semantic roles (SRs), (iii) between semantic roles and grammatical functions (GFs), and (iv) between grammatical functions and their surface forms. As stated before, all constructional flavors assume a Frame Semantics approach so they agree on the first step in which scenes are conceptualized in terms of semantic frames. We will now go through the different steps for each constructional flavor.

Figure 10.2 Constructional approaches to argument realization compared to each other by breaking down production and comprehension into four potential steps

10.4.2 Berkeley Construction Grammar

10.4.3 Cognitive Construction Grammar

Mapping between Frame Elements and Semantic Roles

The most popular constructional approach to argument structure is proposed by Cognitive Construction Grammar (Goldberg Reference Goldberg1995; Goldberg et al. Reference Goldberg, Casenhiser and Sethuraman2004), which borrows many insights from CxG and is therefore very similar in spirit. Just like CxG, CCG assumes a frame-semantic approach to lexical meaning. A lexical construction such as cook, therefore, evokes the Cooking_Creation frame and its frame elements (or ‘participant roles’ in Goldberg’s terminology). Goldberg uses a more informal representation that includes the most salient frame elements in boldface, as shown in example (19).

1. (19)

   COOK < Cook

   Recipient

   Processed_food … >

One important difference between CCG and CxG is that CCG does not include a valence feature in lexical constructions. Instead, CCG posits the existence of more abstract ‘argument structure constructions’ such as the Ditransitive Construction. Crucially, an argument structure construction expresses argument structures in very much the same way as lexical constructions express semantic frames. Goldberg (Reference Goldberg1995: 51) writes that the semantics of the English ditransitive construction is ‘X CAUSES Y TO RECEIVE Z’, which is expressed as an argument structure frame in example (20). This pattern is also repeated in the diagrammatic representation of the Ditransitive Construction in Figure 10.3.

1. (20) CAUSE-RECEIVE < agt rec pat >

Figure 10.3 The English Ditransitive Construction

(adapted from Goldberg Reference Goldberg1995: chapter 6)

As can be seen in example (20) and on top of Figure 10.3, the CAUSE-RECEIVE frame has three semantic roles (or ‘argument roles’ in Goldberg’s terminology): agent (agt), recipient (rec), and patient (pat). Verbs can combine with argument structure constructions if they can, at a minimum, provide frame elements that can be ‘fused’ with the semantic roles of the Argument Structure Construction that are printed in boldface (Goldberg Reference Goldberg1995: 51). Fusion can be compared to the notion of ‘thematic fit’: The FE Cook best fits the agent role, the FE Processed_food best fits the patient role, and so on. The verb may optionally supply frame elements for the other semantic roles in regular font, or the argument structure construction can supply the semantic role itself and thereby augment the verb’s argument realization pattern. This process is called ‘coercion’ (Goldberg Reference Goldberg1995) or ‘type shifting’ (Michaelis Reference Michaelis2004).

The choice for specifying semantic roles in argument structure constructions rather than in the valence of lexical constructions is more than simply a notational variant. First of all, CCG’s argument structures are ‘Gestalt structures’ just like constructions are. That is, semantic roles can only be understood in terms of the argument structure patterns they occur in. This means that the fusion of verb-specific frame elements with more abstract semantic roles is only possible if (part of) the verb’s semantic frame can be construed as an instance of the more abstract argument structure frame, which ensures a close semantic alignment between the verb and the argument structure constructions that it may combine with. In Berkeley CxG, on the other hand, semantic roles are somewhat disconnected from actual semantic frames so they could easily be replaced by any formal symbol for linking frame elements to grammatical functions, which makes them almost meaningless. Secondly, CCG’s argument structure constructions and the fusion process make it in principle possible to have construction-specific semantic roles and grammatical functions, while the CxG approach always implies the existence of categories that are shared across multiple constructions. CCG thus incorporates many CxG insights, while its Gestalt structures and possibility for construction-specific categories align it more with RCG.

10.4.4 Radical Construction Grammar

RCG is close in spirit to CCG because it involves argument structure constructions as well. However, the RCG approach does not involve abstract semantic roles (or argument roles). Instead, Croft (Reference Croft, Butt and Geuder1998: 31) argues that generalizations across different semantic frames should be based on a ‘causal chain of force-dynamics’. A force-dynamic analysis focuses on the transmission of force relationships between frame elements. Example (21) illustrates the force-dynamic relations between the participants in the construct Bill cut the vegetables with a knife for Carol. This construct describes a scene with the following causal chain of subevents: (i) Bill acts on a knife (he takes it); (ii) the knife acts on the vegetables; and (iii) the vegetables cause an effect in Carol in the sense that she benefits from them being cut for her. As can be seen, the transmission of force (indicated by an arrow) is an asymmetric relation between an ‘initiator’ and an ‘endpoint’.

1. (21)

   

The mapping from a causal chain to a construct can be described using a small set of universal (in the Greenbergian sense) linking rules (Croft Reference Croft2012: 221), which I will explain here in a simplified manner for convenience’s sake. First, verb meaning is represented as ‘profiling a segment’ of the causal chain, which is indicated in example (22) by solid arrows. The initiator of this segment maps onto the subject, while the endpoint (if different from the initiator) maps onto the object. This segment is called the ‘verbal profile’.

1. (22)

   

Besides subject and object, there are two types of ‘oblique’ participants: those that precede the object in the causal chain (called ‘antecedent oblique’; knife in the current example) and those that follow the object (called a subsequent oblique; Carol in the current example). Most languages tend to treat each kind of obliques differently. For instance, the English preposition with only introduces antecedent obliques, while prepositions such as for and to only occur with subsequent obliques.

As stated before, argument structure constructions in RCG do not involve semantic roles but, instead, they indicate how the verbal profile should be construed. For instance, when the verb cut combines with the Active Transitive Construction, the verbal profile delimits the segment going from Bill to vegetables in example (22). The English Passive Construction, on the other hand, implies a different construal, where the initiator of the verb profile is vegetables (and therefore maps onto subject), hence Bill becomes an antecedent oblique, leading to constructs such as The vegetables were cut by Bill. The profiled causal chain for the Passive Construction is shown in example (23).

1. (23)

   

In sum, RCG shares several assumptions with CxG and CCG. It assumes a Frame Semantics approach, and argument realization is handled through constructions that exist on an equal footing instead of treating one kind of construction (such as the Passive) as a derivation of another one (such as the Active). RCG also shares with CCG the existence of argument structure constructions that express argument realization patterns in a more holistic way than CxG’s linking constructions do. One important difference, however, is that argument structure constructions do not express more abstract argument structure frames but, rather, the causal chains and verb profiles as construed by language users.

10.4.5 Sign-Based Construction Grammar

Sign-Based Construction Grammar (SBCG) is considered to be the most remote from the other flavors of construction grammar because it proposes a lexicalist rather than a constructional approach to argument realization, which means that argument realization is mostly handled through constraints in the lexicon rather than through the combination of lexical and grammatical constructions. For educational purposes (and for constraints of space), I will explain the SBCG account of Sag (Reference Sag, Boas and Sag2012) in a simplified way that does not reflect all of the technical details but instead focuses on the essence of the approach.

Just like the other constructional flavors, SBCG includes event-specific frame elements in the definition of lexical constructions. However, frame elements are not mapped onto semantic roles or event profiles but are directly associated with a syntactic structure. For instance, a verb such as to cook is typed as a transitive verb whose default argument realization pattern is directly specified in a feature called ARG-ST (argument structure), as shown in example (24). In other words, this feature states that the verb needs to be combined with two noun phrases to make a well-formed sentence. Just like we saw for CxG before, each argument is indexed (here: the subscript letters i and j) for linking the argument to a frame element (which here would link the index i to Cook and the index j to Produced_food).

1. (24)

   [ARG-ST

   < NPi, NPj >]

The arguments in the feature ARG-ST are ordered according to their rank in the Accessibility Hierarchy of Keenan and Comrie (Reference Keenan and Comrie1977). The first argument therefore maps onto subject, the second onto direct object, and so on. In order to handle multiple argument realization, SBCG must therefore modify the ARG-ST list, which is achieved through ‘lexical rules’ (called ‘lexical constructions’ in SBCG terminology). Lexical rules are templates that take a lexical entry or lexeme as ‘input’ and that create a new lexeme as ‘output’. For instance, the Passive Lexical Rule – illustrated in example (25) – modifies the ARG-ST list of a lexeme by turning the object into the subject and by turning the subject into a by-phrase (or omitting it). Additionally, it changes the morphology of the verb. The newly formed passive lexeme thus requires an NP and optionally a by-phrase in order to make a well-formed sentence.

1. (25)

   

10.5.1 Common Myths about Formalizing Construction Grammars

10.5.2 The Boxes-within-Boxes Notation

The most recognizable formalization of constructions is the ‘boxes-within-boxes’ notation of Construction Grammar (Fried & Östman Reference Fried, Östman, Fried and Östman2004), which was briefly introduced in Section 10.2.2. In early CxG, constructions were considered to be similar to the (sub)trees generated by phrase structure grammars but spanning wider ranges than local tree configurations and specifying detailed linguistic information through feature structures. Example (26) shows a constituent tree on the left and the boxes-within-boxes notation on the right.

1. (26)

   

Due to this original conception of constructions as (sub)trees and Paul Kay’s work on the CxG formalism (Kay Reference Kay2002), there exists a common misunderstanding – even among construction grammarians – that CxG’s boxes-within-boxes notation always represents phrase-structural relations. In reality, CxG has developed a much more sophisticated notion of a construction as a multi-dimensional structure that does not need to involve phrase-structural relations. As explained by Fried and Östman:

An example of both the versatility of the boxes-within-boxes notation, as well as the misunderstanding of what it is used for, can be found in Osborne and Gross (Reference Osborne and Gross.2012) who have proposed to repurpose the notation for representing a sentence’s ‘dependency structure’. As illustrated in a simplified way on the right in example (27), the outer box now represents a syntactic head and the inner boxes represent its dependents. The example also illustrates how each box can be enriched with feature–value pairs. The left of example (27) shows a traditional dependency graph representation.

1. (27)

   

Osborne and Gross (Reference Osborne and Gross.2012) present dependency structures as an alternative to constituent structures but that misses the point of constructions as multi-dimensional structures (Fried & Östman Reference Fried, Östman, Fried and Östman2004; van Trijp Reference van Trijp2020). Instead of arguing that one structure should be preferred over the other, both constituent and dependency structures can be considered as alternative views on the same structure. Indeed, when Charles J. Fillmore was awarded the ACL Lifetime Achievement Award, he said that “anyone who looks closely at syntax knows that … you can never represent everything about a sentence in a single diagram” (Fillmore Reference Fillmore2012: 708).

The boxes-within-boxes notation can therefore be used for highlighting different views, and different languages may require different views. For instance, it is common practice to describe word order in languages such as Dutch and German in terms of ‘field topology’ rather than phrase-structural relations (Haspelmath Reference Haspelmath, Heine and Narrog2009). As illustrated by Fried & Östman (Reference Fried, Östman, Fried and Östman2004) and van Trijp (Reference van Trijp2020), such an analysis can be captured using ‘ordering constructions’, which here would use the same boxes-within-boxes notation for emphasizing the part–whole relations between a clause and its topological fields. The same notation is also amenable to representing morphological part–whole relations (Booij Reference Booij2010), and so on. The boxes-within-boxes notation is therefore a general way to represent any kind of part–whole structure, which Croft (Reference Croft2001: 5) calls a construction’s ‘meronomic structure’. Any analysis formulated in RCG can therefore be straightforwardly expressed in a more precise way using CxG’s boxes-within-boxes notation and the feature structures within those boxes (Croft Reference Croft2001: 48f.).

10.5.3 Fluid Construction Grammar and Its Interactive Web Interface

One inconvenience of the boxes-within-boxes notation is that we can only see one view at a time and that each view treats a particular dimension (e.g., phrase structure) as more basic than other linguistic information, which is represented using feature–value pairs. Fortunately, a solution to this inconvenience exists in the formal approach of FCG (van Trijp Reference van Trijp2020). Let us illustrate the approach using the construct a gigantic insect and assume for the sake of exposition that our constructional analysis involves two overlapping dimensions: constituent structure and head-dependency structure. Example (28) shows the head-dependency structure on top and the constituent structure below.

1. (28)

   

Example (29) shows a first but inadequate attempt at representing this multi-dimensional structure using the boxes-within-boxes notation. Here, constituent structure is represented using boxes, while the head-dependency structure is described as a feature–value pair. More specifically, the constituents a and gigantic are indexed as #1 and #2, respectively. These indices are then used in the feature ‘dependents’ in the insect box in order to ‘point to’ the noun’s dependents.

1. (29)

   

Even though the notation in example (29) succeeds in simultaneously representing the construct’s constituent structure and head-dependent structure, it has opted for a nested feature geometry in which constituent structure is seen as the most basic structure. That is, all information that is represented as feature–value pairs can only be accessed through the constituent structure. This goes against the principle of CxG that “none of the layers is seen as ‘more basic’ than any other” (Fried & Östman Reference Fried, Östman, Fried and Östman2004: 19). The solution in FCG is to describe all dimensions as feature–value pairs and to organize features as a flat list of units, as shown in example (30).

1. (30)

   

A ‘unit’ can be considered as a region in the multi-dimensional space, similar to identifying city names on a map. The names of the units do not matter (they might as well have been indices such as #1, #2, #3) as long as they are unique, and here I used names that are easier to remember and interpret for a human linguist. Each unit is represented as a box of feature–value pairs, and unit names can be used for pointing to other units. The NP-unit therefore contains a feature called ‘constituents’ that points to the units that represent information about these constituents, while the insect-unit contains a feature called ‘dependents’, which points to the units that represent information about the unit’s dependents. Of course, a flat list of units with feature–value pairs is extremely difficult for human readers to interpret. FCG, therefore, has an interactive web interface (van Trijp et al. Reference van Trijp, Beuls and Van Eecke2022) that allows its users to switch between different views by specifying which features represent a view, as illustrated in Figure 10.4.

Figure 10.4 The same constructional analysis (‘transient structure’ in FCG terminology) shown from the viewpoint of its constituent structure on the left and its head-dependency structure on the right

(van Trijp et al. Reference van Trijp, Beuls and Van Eecke2022)

One important distinction between FCG and other constructional flavors is that in the latter, there is no real difference between the representation of a construction and a constructional analysis: A construction is simply more schematic than the analysis. However, this implies a relatively sharp separation of competence (the inventory of constructions) and performance (a processing model that needs to ‘know’ how to use the constructions efficiently for producing and comprehending constructs). In FCG, on the other hand, constructions are informative as to how they should be processed. More specifically, FCG constructions divide their units of feature–value pairs into a ‘conditional pole’ and a ‘contributing pole’. The conditional pole specifies the conditions under which it is appropriate for a construction to apply in either production or comprehension. For instance, the Insect-Construction will require the lexeme insect to be present in a construct during comprehension, while a Goldbergian Ditransitive Construction (cf. Figure 10.3) requires a predicate that can at least supply two frame elements that can be fused with the semantic roles agent and patient in production. If these conditions are satisfied, the information of the construction’s contributing pole is released. This information is then either contributed by the construction itself, or unified with information that was already contributed by other constructions. For instance, the Insect-Construction may add the meaning of the word insect in comprehension, or the Ditransitive construction may merge its recipient role with one of the predicate’s frame elements and map it onto the grammatical function indirect object in production. Successful application of a construction leads to a new transient structure that, in turn, may trigger other constructions to apply, until the final constructional analysis is achieved (for more detailed explication, see Chapter 21).

FCG’s web interface allows researchers to inspect the interaction between constructions in great detail. The open-source version of FCG is part of a larger software suite called Babel (Loetzsch et al. Reference Loetzsch, Wellens, De Beule, Bleys and van Trijp2008), which includes a system of monitors that allow the behavior of construction grammars to be traced at larger scales, a multi-agent framework for modeling language as a complex adaptive system (Steels Reference Steels, Schoenauer, Deb, Rudolph, Yao, Lutton, Merelo and Schwefel2000; Beckner et al. Reference Beckner, Blythe, Bybee, Christiansen, Croft, Ellis, Holland, Ke, Larsen-Freeman and Schoenemann2009), and a learning module for implementing language learning strategies (Van Eecke & Beuls Reference Van Eecke and Beuls2017). These modules make FCG the most appropriate computational tool for the study of usage-based approaches to construction grammar.

10.5.4 Static Constraints in Sign-Based Construction Grammar

SBCG shows the greatest promise for studying construction grammars from a purely mathematical perspective: Whereas FCG is strongly concerned with how constructions are processed, SBCG tries to describe construction grammars as process-neutral as possible through the form of static constraints (Sag & Wasow Reference Sag, Wasow, Borsley and Börjars2011). The influence of Ivan Sag and construction-based HPSG have resulted in a formal toolset that is finetuned for generative construction grammars that aim to describe all and only the well-formed structures of a language (van Trijp Reference van Trijp2013, Reference van Trijp2015), but an important research avenue would be to investigate how the SBCG approach can be generalized to also handle usage-based models.

SBCG uses ‘typed’ feature structures, which means that every SBCG grammar has a centralized ‘signature’ in which a ‘type declaration’ must be provided for the kinds of feature structures that are recognized by the grammar. For instance, a constructional analysis is modeled as a feature structure of type ‘sign’, whose type declaration is shown in example (31). The declaration states that the type sign has to be mapped onto a set of features such as phon, syn, and sem, which are typed themselves (and hence each feature is in turn mapped onto the feature structures that are declared to be appropriate for those types). Types are organized in a multiple inheritance type hierarchy, so typed feature structures inherit the type declarations of their parents.

The grammar signature allows grammar descriptions to remain concise. Example (32) shows a listeme for the verb cook, which only specifies the information that is specific to this verb. This feature structure is typed as trans-verb-lxm (transitive verb lexeme), so it inherits all constraints imposed on its parents.

Constructions are defined in SBCG as constraints on typed feature structures. Example (33) shows a ‘lexical class construction’ that can be read as follows: A feature structure of type main-v-lxm (main verb lexeme) should not be an auxiliary and should not occur in subject-verb-inversion constructs.

Even though feature structures could be used for representing all kinds of linguistic information, as shown in the previous section, SBCG has so far only used them for describing phrase-structural relations, so all other kinds of information are tied to a phrase-structural backbone. Moreover, Sag (Reference Sag, Boas and Sag2012) restricts the expressive power of constructions to only constrain local relations between a parent node and its immediate children, which effectively reduces SBCG to an ‘extended phrase structure grammar’ and which goes against the foundational principle that a construction should be able to access linguistic information at any level of depth (Fillmore Reference Fillmore1988). Fortunately, there are conceivable solutions for both issues. First, it is relatively easy to circumvent the locality restriction (Müller Reference Müller, Müller, Abeillé, Borsley and Koenig2021) in order to restore the expressive power of constructions. Secondly, related work on ‘linearization’ exists in HPSG (Kathol Reference Kathol2000; Daniels & Meurers Reference Daniels and Meurers2004) that shows how word order can be separated from phrase-structural relations. The same approach could be explored to investigate how SBCG could represent multiple dimensions on an equal footing instead of granting privileged status to constituent structure (see also Manning Reference Manning1995).

An open research question is how well SBCG’s formal approach could be used for usage-based models of construction grammar. At first sight, SBCG’s grammar signature seems incompatible with emergent grammars or with the assumptions of RCG. However, as demonstrated by Van Eecke (Reference Van Eecke2018) in FCG, it is entirely possible to develop a dynamically adaptive type hierarchy in which type declarations can evolve over time and be construction-specific. Constructionists may also have reservations about SBCG’s distinction between lexemes and phrasal constructions, but Michaelis (Reference Michaelis, Busse and Moehlig-Falke2019) has demonstrated how SBCG can nevertheless provide a uniform account of the syntax–lexicon continuum.

10.6.1 Generative Construction Grammars

Construction Grammar originated as an innovative stream in generative linguistics, and a significant amount of constructional research is still carried out within the assumptions of generative grammar. To repeat, a generative grammar is a formal and process-neutral description of a language that aims to account for all (and only) the well-formed structures of that language (Chomsky Reference Chomsky1965; Sag & Wasow Reference Sag, Wasow, Borsley and Börjars2011).

Generative construction grammars always take an ‘aggregate perspective’ on language. That is, they describe the linguistic knowledge of an idealized language user (Chomsky Reference Chomsky1965) based on acceptability judgments and corpus and experimental studies. Besides an idealized language user, the aggregate perspective also idealizes the temporal dimension by treating the diachrony of a language as a series of discrete states.

The constructional flavors that are most suited for generative studies are SBCG and CxG because these flavors tend to treat language more as a self-contained system than the other flavors do (see Section 10.6.3). To illustrate what that means, let us take a look at examples (34)–(35), taken from Goldberg (Reference Goldberg2019).

1. (34) ?? Explain me this.

1. (35) Explain this to me.

Speakers of English overwhelmingly prefer to use the verb to explain with a prepositional phrase over using it in a ditransitive pattern. Generative construction grammarians would qualify example (34) as an error of ‘overgeneration’ if the grammar would license such a construct and would, therefore, make the grammar more restrictive. For instance, in the lexicalist approach of SBCG, this could be reflected in the ARG-ST feature of the verb to explain. Usage-based construction grammars, on the other hand, will often appeal to other cognitive systems or processing preferences for explaining the same phenomena, as we will discuss further below.

As stated before, both CxG and SBCG can be considered to be ‘constraint-based’ approaches because constructions represent a number of constraints on what the grammar accepts as well-formed structures. One advantage of such approaches is that they are convenient for measuring the degree of acceptability simply by verifying how many constraints of the grammar have been satisfied or violated by a construct. This makes it possible to predict for instance that a construct such as ??explain me this may sound awkward to the ear for many speakers of English, but that it is still more acceptable than *this me explain.

10.6.2 Construction Grammars in the Middle

So far, we have treated the term ‘usage-based’ as a binary label that we can use to decide whether a construction grammar is usage-based or not. In reality, however, usage-based approaches vary a lot in terms of how dynamic they are and how many concepts they still inherit from generative linguistics. The term ‘usage-based’ was coined by Ronald Langacker (Reference Langacker1987, Reference Langacker and Rudzska-Ostyn1988) to emphasize that a language user’s linguistic inventory is not optimized for parsimony of storage (as assumed in Chomskyan linguistics) but, rather, for efficient language use. This means that instead of computing linguistic structures from scratch, language users may store and reuse many ‘lower-level’ constructions as ready-made chunks or exemplars. Besides storage of ‘redundant’ constructions, a usage-based model also typically reflects the frequency of constructions and their degree of entrenchment (Bybee & Hopper Reference Bybee and Hopper2001), which are all assumed to play a role in language change and productivity.

If the term ‘usage-based’ is used in this sense, any constructional flavor will do: CCG and RCG are outspokenly usage-based, while Berkeley CxG is equally appropriate with the added bonus of more formal precision. SBCG could be amended to incorporate usage-based storage effects as well. It is therefore possible for a generative construction grammar to be usage-based if it allows redundant storage and if it loosens the all-or-nothing grammaticality judgments in favor of acceptability judgments that take different degrees of entrenchment into account.

10.6.3 Dynamic Usage-Based Construction Grammars

The label ‘usage-based’ can also be understood in a broader sense of taking the situated communicative interactions or usage events into account (Clark Reference Clark1996), in which language users must establish joint attention and try to achieve their communicative intentions. In this sense, ‘usage-based’ means that language is embedded in other cognitive processes such as perception or construal operations. Such usage-based analyses are more difficult (but not impossible) to model in CxG and SBCG, while flavors such as CCG and RCG explicitly appeal to construal operations as part of language processing, as discussed in Sections 10.4.3 and 10.4.4.

Most usage-based accounts still take the aggregate perspective on language in the sense that they, too, abstract away from differences between individual language users. However, the distinction between synchronic and diachronic linguistics is blurred: Languages are treated as dynamic systems that are forever emergent (Hopper Reference Hopper, Aske, Beery, Michaelis and Filip1987). Dynamic usage-based models therefore do not consider utterances such as ??explain me this as a problem of overgeneration, because innovations are bound to happen (van Trijp Reference van Trijp2015). Instead, the preference for the construct explain this to me of example (35) is explained as a result of language usage. For instance, Goldberg (Reference Goldberg2019) argues that semantically similar constructions are competing with each other for attracting other constructions. In what she calls ‘statistical preemption’ (also known as ‘lateral inhibition’, Steels Reference Steels, Steels and Hild2012b), every observation of the verb explain with a prepositional phrase will not only strengthen their co-occurrence relation, but at the same time weaken the co-occurrence relation between explain and the competing Ditransitive construction. As such, language users will learn what not to say.

With the advent of more powerful computational technologies and the availability of more corpora, dynamic usage-based construction grammars can now also take a population perspective on language, which focuses on how language users interact with each other and how these local interactions may have global effects on the community language. In this view, the community language is effectively a ‘complex adaptive system’ (Steels Reference Steels, Schoenauer, Deb, Rudolph, Yao, Lutton, Merelo and Schwefel2000; Beckner et al. Reference Beckner, Blythe, Bybee, Christiansen, Croft, Ellis, Holland, Ke, Larsen-Freeman and Schoenemann2009) and can therefore be studied using tools from complex systems science such as multi-agent simulations. Particularly Luc Steels, together with his colleagues, has developed a multi-agent framework that operationalizes situated communicative interactions as ‘language games’ (Steels Reference Steels1995, Reference Steels2001, Reference Steels, Steels and Hild2012b), and he has designed FCG for supporting such simulations (Steels Reference Steels, Daelemans and Walker2004). More recently, novel experimental paradigms (Galantucci & Garrod Reference Galantucci and Garrod2010) and close examination of individual language user’s corpora (Petré & Van de Velde Reference Petré and Van de Velde2018) have taken the population perspective as well.

11.2.1 Syntactic Amalgams

Lakoff (Reference Lakoff and La Galy1974) introduces the term ‘syntactic amalgam’ as a label for sentential structures containing what we will refer to, following Fillmore (Reference Fillmore and Niepokuj1985), as ‘syntactic intrusions’. Discussing constructed examples like (6)–(8) below, taken from Matsuyama (Reference Matsuyama2015: 78), Lakoff defines a syntactic amalgam as “a sentence which has within it chunks of lexical material that do not correspond to anything in the logical structure of the sentence” (Lakoff Reference Lakoff and La Galy1974: 321). Assuming a set of implicational rules that he dubbed ‘transderivational constraints’ (Potts Reference Potts2001), Lakoff observes that the chunks in question “must be copied in from other derivations under specifiable semantic and pragmatic conditions” (Lakoff Reference Lakoff and La Galy1974: 321).

1. (6) John made [I don’t know how many mistakes].

1. (7) We ate [I don’t know what].

1. (8) John invited [you’ll never guess how many people] to his party.

Lakoff’s proposed transformational analysis of these structures, labeled ‘wh-syntactic amalgams’ by Matsuyama (Reference Matsuyama2015), involves the insertion of one sentence marker into another. This analysis is difficult to translate into Construction Grammar, a framework that does not use procedures to represent syntactic patterns. But Lakoff’s analysis offers several insights about use conditions – in particular, that the ‘inserted’ sentence (e.g., I don’t know how many mistakes) must have exclamatory force. Sentences like (6)–(8) function not only as assertions about some state of affairs (e.g., John’s making some number of mistakes) but also as commentary about the extreme value of some variable achieved by the entity under discussion. Sentence (6), for example, might be used to assert that John made a remarkably high number of mistakes. The wh-syntactic amalgam therefore seems to have something in common with the conflation pattern mentioned in Section 11.1: It expresses in one finite clause what might otherwise be expressed by two, for example, ‘John made some number of mistakes’ and ‘That number is remarkable’. There is, however, some reason to doubt that examples (6)–(8) in fact represent ‘compressed’ conjoined structures.

Matsuyama (Reference Matsuyama2015) takes issue with the ‘sluicing’ (ellipsis) analysis of wh-syntactic amalgams offered in Guimarães (Reference Guimarães2004), according to which, for example, the bracketed sentence in (6) is appropriately reconstructed as I don’t know how many mistakes John made. Matsuyama’s arguments include the observation that omission of the elided portion in a standard sluiced clause like (9) is optional, but is obligatory in the case of wh-syntactic amalgams, as shown in (10). We add to this point the observation that, in a wh-syntactic amalgam, the nominal head of the quantified wh-expression (in this case, how many letters) cannot be elided – a difference we might attribute to the fact that letters are under discussion (topical) in the context of (9) but discourse-new in the context of (10).

1. (9) She wrote several letters, but I can’t guess how many (letters she wrote).

1. (10) She wrote [you will never guess how many letters (*she wrote)].

Following Huddleston and Pullum (Reference Huddleston and Pullum2002), Matsuyama postulates that wh-syntactic amalgams are in fact ordinary predications containing a nominal argument that is a certain type of headless relative clause. In this type, the wh-word expresses a free variable. Under this analysis, (10), She wrote you’ll never guess how many letters, means something like (11), and the bracketed portion of (11) is similar in structure to the boldfaced bracketed portions of the attested sentences in (12) and (13), respectively.

1. (11) She wrote [some number of letters that you will never guess].

In (11)–(13), the NP head of the relative clause (some number of letters, the room, the stickler for details) is also an argument of the main verb: In (11) it is the direct object of the verb wrote, in (12) it is the object of the preposition at, and in (13) it is the subject of the VP might be grossly undercounting. Matsuyama assumes a derivation in which the head of the relative clause (e.g., how many letters, which contains a wh-degree marker) moves to its in situ position, as in (10), becoming the direct object of guess. While this account does represent some aspects of the meaning of such sentences, the monostratal, construction-based framework does not use syntactic derivations to capture interpretive facts. In a construction-based account, the wh-syntactic amalgam would be a construction that contains a syntactic intrusion – a piece of structure corresponding to you will never guess sequence in the bracketed portion of (10). Both Huddleston and Pullum (Reference Huddleston and Pullum2002) and Matsuyama (Reference Matsuyama2015) treat this sequence as a parenthetical expression. This analysis captures the intuition that the intrusive portion in (10), as in (12)–(13), provides commentary on the fit of words to world. So, (10) could be paraphrased as ‘She wrote a number of letters – a number that is impressively high’, (12) could be paraphrased as ‘He stopped at the guest room, or at least what some called the guest room’, and (13) as ‘The stickler for details – which is a term I once applied to Starr – was undercounting’. There are, however, at least two considerations that undermine the parenthetical analysis of wh-amalgams: (a) sequences like you will never guess in (8) lack the prosodic properties of parenthetical expressions (they do not constitute distinct intonation units when spoken), and (b) the wh-expression how many letters is not a possible complement of the main verb (write), as the ungrammaticality of (14) shows:

1. (14) *She wrote how many letters, you’ll never guess.

Construction Grammar, however, allows us to posit sign combinations in which one or more of the daughter signs are themselves daughters of daughter signs (non-locality). We can also posit mother signs whose syntactic category does not reflect the syntactic category of a head daughter (Michaelis Reference Michaelis, Francis and Michaelis2003). The reason is that sign combination is licensed by the construction rather than by a given lexical head or class of lexical heads. This makes Construction Grammar an ideal tool for describing the assembly of disparate classes of signs that amalgams appear to involve. Whatever the formal details (and we won’t pursue them further here), the Wh-Syntactic Amalgam construction must license a type of nominal sign (e.g., you’ll never guess how many letters) that can then serve as subject or object in any clause that contains it. One option is to treat that nominal expression as an internally headed relative clause. In this case, the licensed nominal expression in (10) would be you’ll never guess how many letters. The complex nominal sign licensed by the construction would consist of an epistemic clause, for example, you’ll never guess, and its wh-complement, for example, how many letters. It would have a nominal denotation roughly paraphrased as ‘letters so numerous that you would not guess their quantity’. The lesson here is that what appears to be a syntactic intrusion (e.g., you’ll never guess or you’ll never guess how many) might instead be licensed as part of a hierarchically structured rule of sign combination. The ‘non-local’ nature of this type of construction is astutely described by Fillmore (Reference Fillmore and Niepokuj1985) in the following passage:

If the phrasal sign licensed by the Wh-Syntactic Amalgam construction is in fact an internally headed relative clause, it is likely not a syntactic amalgam at all. The construction nonetheless features ‘niece licensing’ of a kind that is needed to capture syntactic and functional amalgams. To see this, we can reconsider the direct-object sign from (10): You will never guess how many letters. If we proposed that the Wh-Syntactic Amalgam construction licenses the two sister signs you’ll never guess and how many letters, respectively, we still must constrain the daughters of the latter constituent – it must consist of a nominal head sign and a wh-specifier sign.

It is an irony that the linguistic pattern to which the label ‘syntactic amalgam’ was first applied appears on closer inspection to be a ‘mere’ nominal sign. The two clauses said by Lakoff (Reference Lakoff and La Galy1974) to be combined by the Wh-Syntactic Amalgam construction are not in fact combined by that construction. Instead, sentences like (10), She wrote you’ll never guess how many letters, are composed by standard constructions in combination: the Subject-Predicate construction, which licenses the entire finite clause, and the Head-Complement construction, which licenses the VP headed by wrote. The direct object of wrote is the idiomatic nominal constituent you’ll never guess how many letters. As discussed, this constituent is not an amalgam, although it is a complex nominal structure that resembles an internally headed relative clause, as described by Cole (Reference Cole1987).

So how do we know a syntactic amalgam when we see one? Can we find more generalizable and intuitive criteria than those Lakoff (Reference Lakoff and La Galy1974) offers? Lambrecht’s (Reference Lambrecht, Axmaker, Jaisser and Singmaster1988) study of a non-standard English presentational construction (“There was a farmer had a dog”) is perhaps the most widely cited case study of a syntactic amalgam. But it offers little more than a general characterization of syntactic amalgams (in a footnote): “[In syntactic amalgams], pieces of structure are fitted together in a construction which are not expected to go together, given a compositional derivational model of generative syntax” (Lambrecht Reference Lambrecht, Axmaker, Jaisser and Singmaster1988: 338). In the spirit of Lambrecht’s intuitive definition, I offer that in (15):

1. (15) Syntactic amalgams are combinations of phrasal signs that pair two contiguous or overlapping syntactic sequences that are found elsewhere in the grammar but cannot otherwise be combined.

The intuition captured here is that syntactic amalgams combine familiar pieces in unfamiliar ways. These combinations are licensed by idiomatic (partially productive or partially lexically fixed) constructions. As it happens, a clear illustration of such a construction is found in the very same Chicago Linguistics Society volume that features Lakoff’s “Syntactic Amalgams” paper. John Lawler’s “Ample Negatives” (Reference Lakoff and La Galy1974) provides, among many other idiomatic expressions of negation, a good example of a syntactic amalgam involving double negation, which I will refer to, following Lawler’s terminology, as the Neg Dislocation construction. Examples (16)–(21) illustrate the Neg Dislocation construction. Lawler notes special intonational properties of Neg Dislocation that align with what is sometimes referred to as contrastive focus. I use small caps in rendering these examples, to indicate my intuition about where a point of prosodic prominence would likely fall in a spoken version:

We know that Neg Dislocation is not the product of a ‘fronting’ rule, because there is no possible in situ position for the initial negated constituent. This is shown, for example, by the impossibility of *You don’t not in my house. Instead, we should assume that the Neg Dislocation configuration is licensed by a construction – one that is pragmatically specialized. Lawler says that Neg Dislocation “cannot introduce new information” and that it “typically requires that the lexical item isolated with the initial negative be present in the context of the discourse, either verbally or in some situational way” (Reference Lawler and La Galy1974: 365). Neither of these claims is quite true. Neg Dislocation sentences do in fact introduce new (focal) information – in the pre-clausal constituent following not. This focal expression is the contextually active value of a variable (over entities, times, places, etc.) within a contextually restricted range. The speaker uses constituent negation to contrast the contextually active or proffered value of the variable, which the speaker thereby rejects, with other possible values of the relevant variable. For example, McCain’s Neg Dislocation response in (18) contrasts the interval recently with earlier times at which meeting with Clinton took place, apparently preserving the contextually active open proposition ‘You (McCain) talked with Senator Clinton at some point’. Attested examples like those in (16)–(19) suggest that the Neg Dislocation pattern is used to rebut a contextually active proposition by asserting it is invalid in one of its potential applications. For example, if I am asked whether I have sampled Colorado wines, I could respond with a version of (21): Not good ones I haven’t. R-based implicature strengthens this statement to a general denial: If I say that I have not sampled any good Colorado wines, it could also be true that I have not sampled any at all. This pairing of two (compatible) scopes of denial is achieved through the combination of two independently motivated constructions: a form of constituent negation that I will call the Negative Fragment construction, for example, Not in my house, Not without my daughter, and (negated) VP Ellipsis, for example, I don’t/won’t/haven’t. While the Negative Fragment portion denies the particular, the VP Ellipsis portion denies the more general proposition. Negative dislocation, as a syntactic amalgam, is well suited to perform its specialized discourse function, but it is not a functional amalgam. In Section 11.2.2 we will ask what makes an amalgam a functional amalgam.

11.2.2 Functional Amalgams

Functional amalgams are syntactic amalgams that provide (imperfect) solutions to encoding problems. Functional amalgams are important to investigators of language production because they reveal what kinds of adaptive strategies speakers use when they encounter problems mapping functional representations to morphosyntactic representations. One such problem is that of adhering to the discourse-pragmatic constraint described by Lambrecht (Reference Lambrecht1994: 185) as the Principle of Separation of Reference and Role, which he states in the form of a maxim: “Do not introduce a referent and talk about it in the same clause.” Speakers’ attempts to obey this mapping principle, founded on the Q-principle discussed in Section 11.1, while avoiding undue prolixity (an R-principle violation, as described in Section 11.1), give rise to what Prince has called “a conspiracy of syntactic constructions resulting in the nonoccurrence of NPs low on the [familiarity] scale in subject position” (Prince Reference Prince and Cole1981: 247). Several members of this conspiracy are amalgams, including the non-standard presentational construction exemplified in the boldfaced portions of (22)–(23).

This construction qualifies as an amalgam because it contains two finite verbs that are related neither by subordination nor coordination. It could be regarded as an instance of the zero-subject relative construction in which the second finite VP belongs to a non-restrictive relative clause, and is therefore subordinate. Under this analysis (22), for example, would express two distinct propositions: ‘I have one of my friends’ and ‘She is a designer also’. However, the speaker of (22) cannot reasonably be viewed as asserting both of these propositions, because there is no context in which the first (‘I have one of my friends’) is an informative statement. Further, (22) does not contain two grammatically complete clauses: Either the second ‘clause’ is just a VP (is a designer) or one of my friends is the subject of the second clause, and the finite verb preceding it (have) lacks a direct object. It appears, therefore, that (22) has two mutually incompatible syntactic representations: The NP one of my friends is the complement of the verb have, and at the same time it is the subject of the VP is a designer also. Lambrecht argues that the presentational amalgam is in fact biclausal, consisting of two S constituents, the latter of which lacks an overt subject. He postulates that the two clauses share an argument in the manner of an object-control structure, with the shared argument bearing a focus relation to the first predication and a topic relation to the second (Lambrecht Reference Lambrecht, Axmaker, Jaisser and Singmaster1988: 335). But it is unclear whether what is proposed is a type of coordinate structure or some other form of paratactic combination. In such cases, adherence to the Principle of Separation of Reference and Role, through the partitioning of the referent-introduction and predication functions, comes at the expense of the syntax.

In other cases, the syntax itself is the source of the problem. For example, as Zwicky (Reference Zwicky and Dowty1995) observes, the modification construction exemplified in (24)–(26), in which an adjective modified by an anaphoric degree word (e.g., so, how, too) precedes an indefinite NP whose head it modifies, is incompatible with NPs that have plural and mass-noun heads, as shown by (27)–(29).

1. (27) *How thick (a) foliage is it?

1. (28) *They had so good responses that they added another date.

1. (29) ??I didn’t pay too close (an) attention.

This construction, sometimes referred to as the Big Mess construction (Kay & Sag Reference Kay, Sag, Boas and Sag2012), may have triggered the development of the non-standard alternative illustrated in examples (30)–(33).

The partitive preposition of in (30) seems extraneous, because removing it would create a standard Big Mess token like that in (24), how long a drive. This is because drive is a count noun, and as such compatible with the indefinite article a, which the standard version of the Big Mess construction requires. Partitive of seems to play a different, more vital role in (31)–(33): In each of these cases, the nominal expression that the adjective (fast, good, close) is modifying, respectively (responses, grades, attention), is a plural or mass noun. Such nouns are not candidates for indefinite determination, but the use of the partitive pattern, for example, too fast of responses, allows the writer or speaker to evade the Big Mess construction’s indefinite NP constraint. This adapted version of Big Mess is more flexible, inasmuch as it accommodates more noun classes. It combines the Big Mess construction with a Partitive construction whose head is the nominal quantifier much or many and whose complement is a PP, for example (too) much of a difference. The resulting amalgam inverts the syntactic relationship of the modifier to its nominal sister, making the adjective the head and the nominal (or rather its NP projection) the complement of that adjective. The requirement that the modified nominal be indefinite applies only when the NP containing that nominal is a sister to the adjective, as it is in the Big Mess construction, so that constraint is eliminated when the NP becomes a daughter of a partitive PP complement, as in (31)–(33). However, elimination of the indefiniteness constraint comes at some cost to the grammar at large, and in particular to lexical selection, since non-relational adjectives do not otherwise license PP complements. If we view the partitive version of Big Mess as a grammar ‘fix’ – an imperfect solution to a morphosyntactic encoding problem – then the existence of these two formally and functionally similar grammatical patterns makes sense. Functional amalgams are imperfect solutions to encoding problems created by grammar.

What causes language change may be the most important question in the social science of language. While much research targets pronunciation and vocabulary changes, we don’t yet understand why grammar changes happen – particularly, rapid changes. Current models don’t give language users much credit for reasoning. Grammar changes are often viewed as the result of misapprehension or overgeneralization. Let us assume, by contrast, that grammar change is a form of innovation. Instead of asking how language users violate grammar rules, we can ask what they gain by doing so. For example, why do speakers say is twice, as in the attested example (34)?

I suggest that the answer to this seemingly small mystery sheds light on language evolution. Example (34) shows one of several strategies that speakers use to announce a forthcoming proposition (e.g., power involves controlling the resources …) by means of a ‘setup’ clause containing a low referentiality definite NP expressing a variable (e.g., the issue, my question, the thing, the problem). Brenier and Michaelis (Reference Brenier and Michaelis2005) refer to this kind of specificational copular clause as ISIS. ISIS contains two adjacent tensed forms of the verb be, the first of which bears stress. Why use one more verb than one has to? The answer given by Brenier and Michaelis (Reference Brenier and Michaelis2005) is that ISIS is a workaround. It allows the speaker to execute the ‘setup’ gambit without incurring intonation problems associated with the standard alternative, that is, Simplex (e.g., The thing is</sc> [pause] I always carry it with my checkbook). Simplex violates the tendency for pauses to align with the edges of syntactic constituents (Croft Reference Croft1995): The Simplex ‘setup’ (the thing is</sc>) is an intonation unit but not a syntactic unit, and its verb phrase is intonationally broken (is [pause] I always carry my checkbook). In addition, Simplex contains an accented function word, be. These anomalies come from the dual function that be is forced to perform in Simplex: a syntactic function (taking a clausal complement) and a discourse function (marking the forthcoming content as new). The ISIS solution distributes these functions across the two be verbs: ISIS contains a prosodically prominent be that flags forthcoming focal content, followed by an unbroken verb phrase with an unaccented is. But the ISIS solution also creates a syntactic paradox: What is the syntactic category of the ‘setup’? The moral of the story is that syntactic fixes aren’t perfect. The Q-based drive toward transparency, which might lead language users to disaggregate functions that would otherwise coalesce in a single copular-verb token, is countered by the R-based drive to fall back on entrenched routines wherever possible. It is relatively easy to find the more established appositive construction on which ISIS is based. Brenier and Michaelis (Reference Brenier and Michaelis2005) refer to it as Hypotactic Apposition; it is illustrated in the boldfaced portion of (35), an example from the Switchboard corpus.

In Hypotactic Apposition (HA), the ‘setup’ clause contains a cataphoric use of the demonstrative pronoun that and the low referentiality specificational NP (e.g., the main thing) that would ordinarily be a subject NP in post-verbal position (e.g., is the main thing). In a paratactic appositive clause, there would be nothing more than an intonational break between the set up clause (e.g., That’s the main thing) and the ‘payoff’ clause (e.g., I can’t tell). But the HA strategy is to ‘infill’ that space between setup and payoff with a finite form of the copula and, typically, a following that complementizer, resulting in a setup clause like That’s the main thing is (that). To offer a somewhat strained analogy, the HA construction is a poorly coordinated two-person horse costume: The front end doesn’t seem to know what the back end is doing. In this case, the back end is a VP (e.g., is that you can’t tell), but here it is not being predicated of a specificational nominal subject like the main thing. Just as in the presentational amalgam described earlier in this section, a single NP (in this case the main thing) does double duty, as both the complement of the setup clause’s presentational be and the subject of the payoff clause’s specificational be. Like many functional amalgams, including Lambrecht’s (Reference Lambrecht, Axmaker, Jaisser and Singmaster1988) presentational amalgam, HA is biclausal, but there is no standard syntactic rule providing for the combination of those two clauses. This is why a construction-based view of syntax is indicated in such cases: A construction-based grammar has many syntactic rules – as many as there are idiomatic patterns in the grammar – and the commitment to describe all such patterns is combined with the functionalist principle that patterns of sign combination are devised by language users to meet expressive needs. This is no less true at the level of lexical derivation, as we will see in the next section.

12.3.1 Constructions Sensitive to Sequential Position

Receipt Questions

To illustrate the potential of notating sequential information and its necessity for the appreciation of the dialogical nature of constructions, we will at the outset scrutinize a very distinct type of responsive construction, receipt questions (cf. ‘pro-repeats’ in Schegloff Reference Schegloff2007: 155). These are short, structurally minimal clausal interrogatives that respond to an informing from the co-participant and treat the new information as something that runs counter to the recipient’s expectations (see Norén Reference Norén, Lindholm and Lindström2010 on Swedish; Jørgensen Reference Jørgensen, Goldshtein, Hansen and Hougaard2021 on Danish; Thompson et al. Reference Thompson, Fox and Couper-Kuhlen2015: 90 on English; also ‘newsmarks’ in Couper-Kuhlen & Selting Reference Couper-Kuhlen and Selting2018: 279). We consider the phenomenon in Swedish interaction by concentrating on Excerpt (1), where we can find a receipt question in line 8. Speakers A and B play in amateur football teams, and they are discussing the teams’ success on the telephone. Speaker A names Hönnsta as the favorite team in an upcoming match (line 1), while B in line 7 informs A about his team’s latest victory. This is received by A with the minimal interrogative gjorde ni? ‘did you?’

Excerpt (1)Receipt question; telephone conversation between friends

01

A:

Hönnsta ä ju favori:t laget,

‘Hönnsta is of course the favorite team,’

02

(0.3)

03

mm: pt.h[h-

04

B:

[spöa:

‘beat’

05

(0.2)

06

A:

va¿

‘what¿’

07

B:

spöa

Kimmelby

me

tie

noll

ida,

beat.PST

NAME

with

ten

nil

today

‘(we) beat Kimmelby ten nil today,’

08 →

A:

gjorde

ni?

do.PST

you.PL

‘did you?’

09

(0.5)

10

A:

nä:

e

re

säkert?

‘no:

is

that

right?’

11

B:

j(h)a

‘yeah’

The receipt question at line 8 mirrors the tense in B’s previous turn, using the past tense form of the verb göra, which is a ‘pro-verb’ in the same sense as the English do. By using the receipt question format, A treats the information B has given as new. No confirmation ensues, and A then continues by producing a follow-up question that indicates (ritual) disbelief nä:e re säkert? ‘no: is that right?’ B responds to this in the positive in line 11.

Depending on the polarity of the sequence, the negating operator inte ‘not’ may be present in a receipt question, as seen in Excerpt (2). Speaker A asks here whether B has got new eyeglasses, which B bluntly denies in line 3. The denial runs counter to A’s expectations, which she marks with the negatively polarized receipt question har du inte ‘have you not’ in line 4. The contrast between A’s expectation, which is publicly available in the initial question, and B’s denial is apparently so great that A’s response bears tokens of embarrassment: The receipt question is produced with a smile voice, ending in laughter, and it signals generally ritualized disbelief. B also continues by explaining that her glasses, in fact, are quite old (line 5), which increases the gap with A’s expectation of the state of affairs. The explanation then is received with the conventional newsmark jaså du ‘I see’, which B reconfirms in line 7 (for an extended analysis of the extract, see Norén Reference Norén, Lindholm and Lindström2010: 60).

Excerpt (2)Receipt question with negation

01		A:	har_u fått nya glasögon
			‘have you got new glasses?’
02			(.)
03		B:	näe.
			‘no.’
04 →		A:	*har	du	inte* ((laughs))
			have-PRS	you	NEG
			‘*haven’t you*’
05		B:		dom e:: så rom börja bli för gamla nu.
			‘they’re getting to be too old now.’
06		A:		jaså [du¿]
		‘I see’
07		B:			[.ja:]
			‘yes’

(adapted from Norén Reference Norén, Lindholm and Lindström2010: 60)

From a morphosyntactic point of view, receipt questions consist of a (minimal) polar interrogative clause which includes a finite, tense-inflected ‘pro-verb’ (an auxiliary, a copula or a ‘do’-verb) and a pronominal NP in the second-person singular functioning as the subject. The tense of the verb is sensitive to the tense used in the preceding target turn, and a negating adverb is present depending on the polarity of the target turn, that is, a negated target receives a negated response (because it is the negated content that becomes topical). This general format could be described as a clausal construction (V+) in a morphosyntactic analysis as follows; for the sake of clarity, material from the constructional instantiation in Excerpt (2) is used.Footnote ¹

	har	du	(inte)
[V+	[Vfin	NPsubj	(Adv)]]
	cat=pro	cat=pro2	cat=neg
	tense=copy T1

It should be noted that the normally mandatory object complement to the verb ha ‘have, possess’ is most often absent in receipt questions (cf. a general interrogative: Har du inte nya glasögon? ‘Have you not (got) new glasses?’). Sometimes, though, such a complement can be pronominalized like the other items in the construction: Har du inte det? ‘Have you not (got) it/that?’

Receipt questions are characterized by their occurrence in second-position turns, that is, in responsive positions typical of many kinds of newsmarks, like oh, really, is that so. They thus work in the service of intersubjectivity and the management of epistemic relations: The speaker who uses a receipt question is clearly less knowledgeable than the co-participant. Hence, the questioner’s epistemic (knowledge) status is K− while the knowing participant is K+. The semantics of the construction is basically that of a question, but receipt questions do not necessarily make an answer conditionally relevant (as a bona fide question seeking information does). In both instantiations seen in the excerpts above, an immediate, simple token of confirmation is absent. However, a third position confirmation can be considered optional, and is also actively sought after in cases like in Excerpt (1), or the third turn slot can be occupied with an explanation, like in Excerpt (2). This then indicates that a pragmatically more specific function – that of marking the receipt of news – is derived from the general interrogative functional pattern associated with the clausal construction [V+ [Vfinite NPsubject (NPobject)]].

Taking together the constructional information accounting for grammatical form, function, and sequential context, we can establish regularities for Swedish receipt questions as shown in Figure 12.1.

Figure 12.1 The attribute–value matrix for a Swedish receipt question

In sum, the sequential second position, the speakers’ epistemic relations, sparsity of clausal form, and its pronominal shape strongly mark the responsive quality of receipt questions and their status as a distinct construction in the language. A prosodic dimension could be added to the matrix, as receipt questions are often delivered with some form of vivid prosody involving stress or pitch rise to high in the speaker’s range, which generally indicates the recipient’s surprised state.

Second Assessments

To cite another example of a sequentially dependent type of construction, we can consider the formatting of second assessments in interaction. These are evaluating utterances that follow up a first evaluating utterance by another speaker, for example, responding to a first assessment like It’s going to be a tough winter with a second assessment like I think so, too. Such an interactional and sequential position is intricate for the second assessor in that the possibility of saying something independently from the speaker in the first position is limited (Enfield Reference Enfield, Stivers, Mondada and Steensig2011). At the same time, there is a general social preference for expressing agreement with a preceding evaluation (Pomerantz Reference Pomerantz, Atkinson and Heritage1984). That is, assessing in second position means in effect dealing with the limits of the speaker’s social agency – saying something original and different from the first speaker on the one hand and willingness to display affiliation with the assessment put forward by the first-position speaker on the other. For a second assessor, it may be individually important to say something that is not heard as simply concurring with the first speaker (Heritage Reference Heritage, Ford, Fox and Thompson2002); but to say something that is overtly independent from the first assessment risks being heard as a disaffiliating, socially loaded move (Thompson et al. Reference Thompson, Fox and Couper-Kuhlen2015).

We can examine these intricacies in German exchanges in which assessments are formulated with the stance-taking verb finden ‘think, find’ (Auer & Lindström Reference Auer, Lindström, Lindström, Laury, Peräkylä and Sorjonen2021), beginning with Excerpt (3) taken from a dialogue in the Big Brother TV reality show. Interactants A and B take care of chickens, and people have been giving names to the chickens on postcards, which A considers to be a cute and funny idea (lines 1 and 2).

Excerpt (3)Second assessment, low affiliation; TV reality show, face to face

01 1a →		A:	ich find_s voll süss dass die die hühner taufen;
			‘I find it so cute that they give names to the chickens;’
02 1b →			des find ich ne total lustige [idee.
			‘that I find a totally funny idea.’
03 2 →		B:			[ja
					‘yes’
04			find	ich	auch.
			find	I	too
			‘I think so too.’

The first assessment comes in two parts (1a and 1b at the arrowed lines):Footnote ² the first part is in a full clausal form with straight word order, ich find’s voll süss …, and the second part that comments on the previous evaluation is a full clause as well, starting with an anaphoric pronoun des find ich ’ne total lustige idee. In contrast, the second-position assessment by B in line 4 is delivered in a syntactically reduced form, starting with the affirmative particle ja (line 3) and continuing with a clausal unit that includes only the stance-taking verb, the first-person subject pronoun and the additive adverb auch, which underscores that the assessment is a consequent to the preceding one. The prototypical clausal construction in such lowly affiliating German second assessments, seen in the instantiation in Excerpt (3), could thus be described as follows in the morphosyntactic domain; the initial affirmative particle (PAR) is considered optional and is rendered in parentheses:

(ja)	find	ich	auch
[(Par)]	[V+ [Vfin	NPsubj	Adv]]
cat=aff	cat=stance	cat=pro1	cat=add

Though less usual in second assessments, the actual construction may be filled in all principal clausal positions, including the pronominal object pronoun in topic position (i.e., as the first clausal constituent), which is the case in the latter part of the first assessment in line 2 (the word order of this TCU reflects thus its consequent position in A’s multi-unit turn). On the other hand, the second assessment format can be radically ellipted, just leaving the subject pronoun and the additive adverb present (ich auch ‘me too’). Relations between the first assessment and potentially elided (and understood) items in the second assessment can be illustrated as follows.

Turn in position 1	Das finde ich	eine total lustige Idee
Turn in position 2	( ) ( ) ich auch	( )

No explicit evaluative term is recycled or introduced in any of these second assessment formats; instead, such terms are referred to with the anaphoric pronoun das, or just implied by the fact that what is elided is in agreement. To account for all this surface variation in the morphosyntax (of the constructs), the second assessment construction should be described as follows, where optional (or ellipsis-prone) items are represented in parentheses:

(ja)	(das)	(find)	ich	auch
[(PAR)]	[V+ [NPobj	Vfin	NPsubj	Adv]]
cat=affirm	cat=pro3	cat=stance	cat=pro1	cat=add
	anaphoric

As in the case of Swedish receipt questions, the second assessment’s more condensed form is a reflection of its consequent position, but it also implies less agency from the assessor’s part. The speaker does not introduce an evaluative term of his/her own or even reuse evaluative terms from the first assessment. Nonetheless, the speaker concurs and displays affiliation with the previous turn but, due to the lack of an explicit evaluative item, to a somewhat lower degree. The above features constitute a recurrent pattern for weakly affiliating second assessments with the verb finden in German interaction and are represented in the constructional account in Figure 12.2.

Figure 12.2 The attribute–value matrix for weakly affiliating German second assessments

Second assessments of this kind are constructed in a way that reflects their responsivity (and second position): A full clausal form can be used when placing an anaphoric pronoun (das) in the topic position, but this topic item is frequently dropped, as in Excerpt (3), in which case the assessing verb, finden, initiates the clause and ties the turn structurally even closer to the first assessment. Sequential dependency is strongest in the non-clausal, maximally elliptic structure ich auch. It is a typical sequential consequence of low-affiliating second assessments that they are closing implicative, that is, the sparse but agreeing response seems to signify that there is not much to say about the topic (or the assessable) anymore.

The constructional format differs somewhat in second assessments where the speaker displays a higher degree of affiliation with the first assessment. An example of this is seen in Excerpt (4), taken from an informal face-to-face conversation. Ann assesses Sandra’s mascara positively (which is the reference of the pronoun die ‘it’ in lines 1 and 2). Sandra in turn concurs by producing a second assessment in a full clausal form with the same straight word order as in the first assessment.

Excerpt (4)Second assessment, high affiliation; informal conversation, face to face

01 1 → ANN:	aber ich find die voll gut.
	‘but I think it’s really great.’
02 2 → SAN:	ja ich find die auch voll (.) gut;
	yes I find.PRS it also really good
	‘yes I also think it’s really (.) great;’
03	und die war auch gar net so teuer;
	‘and it wasn’t even very expensive;’

Although the second assessment does not introduce a new evaluative term but recycles what is said in the first position (the adjective gut ‘great’ intensified by voll ‘really’), this move is explicitly agreeing (and perhaps even more so in a context where Ann’s assessment could be heard as complimenting Sandra’s possessions). It is significant that the syntax of the second-position turn mirrors the one in the first position; that is, it is constructed as if it was an original contribution and not so much a response that goes along with the previous speaker (Auer & Lindström Reference Auer, Lindström, Lindström, Laury, Peräkylä and Sorjonen2021). The word-order pattern and the presence of all clausal constituents strike a difference from the inherently responsive assessment in Excerpt (3). The sequential position, however, is lexically marked with the additive adverb auch. These constructional choices signify a high degree of affiliation and a claim to higher speaker agency, which is also evident in Sandra’s self-authored follow-up commentary, that is, line 3 on the price of the mascara (possibly also an orientation to the complimentary nature of the first assessment). This assessment format is not implicative of sequence closing either; on the contrary, the ongoing topic is naturally pursued further when a stronger interpersonal affiliation and speaker agency is displayed. The constructional schema for a highly affiliating second assessment of the kind in Excerpt (4) is presented in Figure 12.3.

Figure 12.3 The attribute–value matrix for highly affiliating German second assessments

By comparing the two constructional formats for German second assessments (Figures 12.2 and 12.3), we can conclude that speakers seem to claim less agency in second assessments by using reduced and markedly responsive clausal formats, while full formats with straight word order are used to claim higher agency and also to display a higher degree of affiliation.

12.3.2 Constructions Embodying Redirecting Moves

Meaning Negotiation Constructions

As noted in conversation analytic research, there are many general practices for carrying out repair (Schegloff Reference Schegloff and Givón1979), that is, correcting the course of talk and interaction. For example, self-repair can be done by cutting off the production of a problematic item and adding the adequate one (she was des- depressed) or by offering a substitute, flagged with, for example, I mean in English (she was desperate, I mean, depressed). Because talk is produced on the fly, it is not surprising that even more specific needs for repairing and revising what has been said occur, and one pragmatic environment for such redirecting moves is meaning negotiation (e.g., Norén & Linell Reference Norén and Linell2007; Deppermann & Schmidt Reference Deppermann and Schmidt2021). As these kinds of practices recur in language use, it is to be expected that they in some respects evolve into routinized patterns with constructional, templatic properties. One such example is ‘concessive repair’, which has been described by Couper-Kuhlen and Thompson (Reference Couper-Kuhlen, Thompson, Hakulinen and Selting2005) as a device for retracting from overstatements.

Another pattern of a related kind is the meaning-negotiating x-and-x construction in Swedish (see Linell & Lindström Reference Linell and Lindström2016). The construction’s focus on word meaning is easy to identify in contexts of self-repair, as in Excerpt (5). It comes from a parent/teacher meeting at a school. Here, the teacher (T) tells the parent (P) that the pupils do not get so much homework because they get things done and get along well in a small class. To underscore this, the teacher says in line 7 that nobody ever needs to “fix/sort out a fight” (fixa bråk). The parent then produces a continuer (näeh ‘no’), sensitive to the negative polarity in the teacher’s preceding contribution, after which the teacher starts a retrospective self-negotiation of meaning by repeating the verb fixa from her prior talk, i.e., fixa å fixa ‘fix and fix’ (line 9).

Excerpt (5)Meaning negotiation and self-repair

01	T:	men vi hinner ju läsa på här,
		‘but we manage to read a bit here,’
02	P:	mm
03	T:	man kanske inte behöver ha så mycke hemläxor.
		‘one does not perhaps need to have so much homework.’
04	P:	nä-eh (.) precis
		‘no-eh (.) right’
05	T:	utan vi har lugn å ro, å liten klass,
		‘so we’ve got peace and quiet, and a small class,’
06	T:	å duktia elever, å eh trevlit, å ingen eh (.)
		‘and good pupils, and em nice, and nobody um (.)’
07 →		behöver aldri fixa nån bråk eller nåt sånt där
		‘needs ever to fix a fight or something like that’
08	P:	näeh
		‘noeh’
09 →	T:	eller fixa å fixa, ORDNA UPP. (.) de gör vi aldri.
		‘or       fix and fix,   SETTLE.        (.) we never do that.’
10	P:	nä
		‘no’
11		(0.3)
12	T:	så vi har de bra.
		‘so we are doing great.’

(adapted from Lindström & Linell Reference Lindström, Linell, Engdahl and Londen2007)

The meaning negotiation seems to address the colloquial value of the verb fixa, which the teacher then revises to the more serious ordna upp ‘settle’ in line 9, perhaps a lexical choice that is more appropriate in her professional role. The coordinated pattern fixa å fixa ‘fix and fix’ focuses the problem source by repeating and reduplicating it,Footnote ³ which is then followed by a revised version of the problematic item. The sequential progression can thus be described as follows.

Utterance in position 1	Term that is possibly out of place in the context
Utterance in position 2	Reaction indicating a problem with the term proposed
Utterance 2 expansion	Revised version of the term

The coordination of these two identical items communicates a distributive meaning (‘is fixa correct or not?’) that is pragmatically understood as questioning. We can note that the disjunctive conjunction eller ‘or’ precedes the coordination in line 9, a use which recurs in self-repairs that introduce an alternative to a prior formulation. From the point of view of turn construction, the latter two sequential steps noted above form a one-turn constructional unit and together make the full reactive construction in a dialogical sense, that is, both utterance parts are needed in order to make meaning. In a morphosyntactic analysis, we could therefore say that x å x is a discourse-pragmatic, templatic frame occupying the ‘pre-front field’ in initial extraposition, and it projects a subsequent phrasal or clausal expansion:

(eller)	fixa å fixa,	ordna upp
[(PAR)]	[V+ [x å x],	[revised version of x]
cat=disj	cat=coord	cat=P/V+
	sem=distr	prag=relativization
	prag=questioning

Besides self-repairs, the construction is used in next-speaker-initiated meaning negotiations, whereby they appear as other-repairs. An example can be seen in Excerpt (6), from a Swedish TV talk show. The interactants are a political journalist (J) and a veteran politician (P), the latter being interviewed about his professional and personal life. The participants have talked about the many activities that P is constantly involved in. At the beginning of the excerpt, the journalist presents yet another example of P’s many interests: “and then you can speak a host of languages,” presuming them to be as many as eight. P reacts to this by repeating the verb tala ‘speak’ in line 3.

Excerpt (6)Meaning negotiation and other-repair

01 →	J:	å så kan du tala massa språk,
		‘and then you speak a host of languages,’
02		va e de? åtta?
		‘how many is it? eight?’
03 →	P:	tala å tala (.) ja kan nj- an:vända en ett så där
		‘speak and speak (.) I can use about’
04		åtta tietal språk ja.
		‘eight or ten languages yeah.’
05	J:	vikket håller du på å lära dej nu?
		‘which one are you learning now?’
06	P:	ja håller faktist på me kinesika (…)
		‘actually I’m busy with Chinese (…)’

(adapted from Linell & Lindström Reference Linell and Lindström2016)

The repetition of the verb tala ‘speak’ in the turn-initial coordinated X å X pattern signals some contestation about its appropriateness in the context. As we can see in the expansion of P’s contribution in lines 3–4, he says that he can “use” (använda) a number of languages corresponding to the number suggested by the journalist. The indirect message of this meaning negotiation is that speaking may imply the ability to master the languages fluently, which might be an overstatement in P’s case. P is thus relativizing the semantic intension of the verb tala, suggesting a lower level of proficiency which can involve the ability to speak and understand to some degree. The interactive constructional pattern for these kinds of meaning negotiations is summarized in Figure 12.4.

Figure 12.4 The attribute–value matrix for the Swedish meaning-negotiating x å x

The central sequential property of the construction is that it occurs in second-position utterances. That is, the construction is responsive – it picks up and repeats something specific (the term x) from the immediately preceding talk. As such, meaning negotiations suggest some concerns about the acceptability of the item that is picked up, the interactional move is disaffiliating and at the same time reflects the speaker’s high agency in the exchange. It is also true of most instances that the speaker in second position is more knowledgeable (K+) about the subject matter, although the first speaker must also have some knowledge in order to be able to say something (possibly controversial) in the first place. For example, the journalist in Excerpt (6) is fairly familiar with P’s career and interests, but she cannot have first-hand access to how well P masters different languages. When the terms for further talk have been negotiated, the speakers can continue to develop the topic at hand.

Meaning-negotiating constructions are a vehicle for promoting intersubjectivity in interaction. By focusing on an item that has surfaced in talk and elaborating on its appropriateness in the actual context of use, the speaker lays the ground for an acceptable degree of shared understanding between the participants (Linell & Lindström Reference Linell and Lindström2016). From a constructional point of view, we could say that the Swedish meaning-negotiating practice is a functionally specific grammatical construction (Fried & Östman Reference Östman, Östman and Fried2005) and is idiomatic in that the meaning of the coordinated pattern X å X is not transparent from its constituent parts. The construction could therefore be recognized as a ‘formal’ or ‘templatic’ idiom in a manner comparable to the English let alone and the X-er, the Y-er constructions, for example, the sooner, the better (Fillmore et al. Reference Fillmore, Kay and O’Connor1988).

Pseudo-clefts in Stance Shifts

The expression of speaker stance is a central interpersonal dimension of dialogic interaction (DuBois Reference DuBois and Englebretson2007), and there are various constructional resources to explicitly communicate a personal, subjective standpoint, for example, through finite clauses such as the English I think (e.g., Kärkkäinen Reference Kärkkäinen2003) or the German ich finde ‘I think, I find’ (e.g., Auer & Lindström Reference Auer and Lindström2016). Interestingly, there are also other, more generic grammatical constructions that are noticeably often recruited for the sequential organization of stance-taking in discourse and this should be recorded as one important facet of their functional potential and properties. One contender for this category of constructions is pseudo-clefts (or wh-clefts). Symptomatically, they are often exemplified with stance-taking predicates in the literature, like the English sentence What I like is champagne (Lambrecht Reference Lambrecht2001).

Pseudo-clefts have been described as one type of complex sentence which expresses a single proposition (e.g., I like champagne) in two clauses and with an informational structure that lays the focus on the element in the latter clause (see Lambrecht Reference Lambrecht2001). Studies of spoken interaction have further identified that pseudo-clefts recur in shifts of speaker orientation (e.g., Kim Reference Kim, Downing and Noonan1995), and moreover, the nature of such shifts tends to have to do with speaker stance (Günthner Reference Günthner, Günthner and Imo2006; Maschler & Pekarek Doehler Reference Maschler and Pekarek Doehler2022). We can see this happening in Excerpt (7), taken from a Swedish interaction (also analyzed in Lindström & Henricson Reference Lindström and Henricson2022). In the actual sequence, two high-school students discuss a pop song by a female artist. Before the excerpt, the participants have agreed that the song is good, and speaker A continues to evaluate details of the production and instrumentation in positive terms (lines 1–4). Then there is a gap of 0.9 seconds after which he makes a stance shift with the pseudo-cleft vad ja inte gillar e hennes nasala röst ‘what I don’t like is her nasal voice’, thus evaluating negatively the singer’s voice quality (which B bluntly objects to in line 8).

Excerpt (7)Stance-shifting pseudo-cleft; lay discussion, face to face

01

A:

mm (.) väldigt eh snyggt också=snyggt ljud,

‘mm (.) very um nice too =nice sound,’

02

många eh (.) duktiga:,

‘many um (.) talented:,’

03

(1.5)

04

>va heter de< musiker.

‘>what’s it called< musicians.’

05

(0.9)

06 →

å

vad

ja

inte

gillar

e

hennes

nasala

röst

and

what

I

NEG

like-PRS

COP

her

nasal-DEF

voice

‘and what I don’t like is her nasal voice’

07

som e lite: >för jobbig<.

‘that’s a bit too disturbing.’

08

B:

de: håller ja inte me om.

‘I don’t agree about that.’

Importantly, the pseudo-cleft construction offers an initial stance-taking frame for the upcoming action (Hopper & Thompson Reference Hopper, Thompson and Laury2008: 107), both delaying the gist of a negative, possibly risky evaluation, and alerting the other participant to such a shift of orientation. As in English, the initial alert is placed in the cleft clause, which is then linked to the subsequent cleft constituent with the finite copula verb (Figure 12.5).

Figure 12.5 The constituent parts of a pseudo-cleft

It is because of this argumentative property that pseudo-clefts tend to occur in different kinds of consultative talk (Hopper & Thompson Reference Hopper, Thompson and Laury2008). A telling example of this is seen in Excerpt (8) from an essay-writing tutorial in which a teacher (T) and a university student (S) participate. The teacher at first states that the student has produced a generally good text (line 1). After a continuer from the student and a gap, the teacher picks up a problem (line 6), which has to do with the student’s inconsistent use of pronouns with generic reference (du ‘you’, man ‘one’, vi ‘we’) and which the student recognizes in line 9 and following lines (for a further interactional analysis see Lindström & Henricson Reference Lindström and Henricson2022).

Excerpt (8)Stance-shifting pseudo-cleft; essay tutorial, face to face

01

T:

å här har ja då skrivi att de e en bra text.

and I’ve written here that it’s a good text.

02

(0.3)

03

S:

mm

04

(0.3)

05 →

T:

.h de

som

e

problem

här

språkligt

that

which

cop

problem

here

linguistically

.h the

thing

that

is problematic

here

linguistically

06 →

så

de

e

pronomenbruke(.)

PRT

it

be.PRS

pronoun.use-DEF

it’s the pronoun use (.)

07

[du har du] å du har man du har vi.

you’ve got you and you’ve got one and you’ve got we.

08

S:

[just de]

right

09

S:

*jå

just de*

((laughing)) .h

*yes

exactly*

.h

10

T:

mm

11

S:

ja

ja

blandar.

yeah

I’m

mixing.

In an analogous manner to Excerpt (7), the teacher’s stance shift from positive to negative is introduced with the pseudo-cleft: de som e problem här språkligt så de e pronomenbruke ‘the thing that is problematic here linguistically is the pronoun use’. It allows the speaker to flag a problem in the initial cleft clause and to state its specification in the following cleft constituent.

The overarching interactional properties of the pseudo-cleft construction are represented in Figure 12.6. Pseudo-clefts can also be considered a positionally subsequent phenomenon in that they occur in an argumentative context in which they respond to and alter trajectories in the preceding discourse. The general pragmatic operation is that of making a shift in the speaker’s orientation, more specifically the speaker’s stance shift in cases like (7) and (8). The responsive nature of clefts is displayed in the presence of an initial discourse particle that can be additive (e.g., ‘and’) or adversative (e.g., ‘but’) in meaning. The syntactic core of the construction is made of the cleft clause that introduces an assessment, which remarkably often is disaffiliative with what has been discussed before. The cleft constituent specifies and focuses the target of this evaluative and disaffiliative move.

Figure 12.6 The attribute–value matrix for the Swedish stance-shifting pseudo-cleft

Note that this interactionally sensitive constructional analysis does not contradict the way in which pseudo-clefts have been analyzed earlier from an information-structural point of view. Rather, it shows what the argumentative and interpersonal niche of the construction is: Yes, it places special focus on the cleft constituent (and semi-focus on the cleft clause), but that is done for good interactional reason. In the context of stance negotiation, the action implemented by the pseudo-cleft may be socially delicate, contrasting with what the speaker and other participants have said in prior talk. The verbally elaborate nature of the construction is on a par with such navigations in interpersonal waters. We can also note that stance negotiation is not the only, albeit a very characteristic, functional potential of pseudo-clefts in social interaction. Especially in English, its transitional function has been generalized to the degree that it is most typical with predicates like do, happen, and say (Hopper & Thompson Reference Hopper, Thompson and Laury2008; also Koops & Hilpert Reference Koops, Hilpert, Givón and Shibatani2009), that is, verbs that have to do with planning and narrating activities and the shifts in stages and events in them.