Debonding and Clipping of Prefixoids in Germanic: Constructionalization or Constructional Change?

Abstract

This paper is concerned with the debonding of three Germanic prefixoids: Dutch kei ‘boulder’, German Hammer ‘hammer’, and Swedish kanon ‘cannon’. Drawing on an extensive corpus-based and statistical analysis, we compare the formal properties (construction types), semantics (degree of bleaching), collocational properties and productivity of bound and free uses of each prefixoid. We show that debonding of prefixoids is a productive process of lexical innovation in Germanic languages, which may lead to the creation of new intensifying adverbs or evaluative adjectives. In addition, we explore whether debonding of prefixoids can be fruitfully analysed from a constructional perspective. More in particular, we address the question of whether the observed changes accompanying debonding are best accounted for by Traugott and Trousdale’s concept of ‘constructionalization’, or by Hilpert’s concept of ‘constructional change’. To this end, we explore a variety of quantitative methods, including productivity measures and distinctive collexeme analysis. We conclude that the quantitative differences between the bound and the free forms of the three prefixoids studied in this paper allow us to consider them as two separate constructions, but that the distinction is a gradient one.

1 Introduction

This paper is part of our joint project on the debonding of prefixoids in the Germanic languages. We gratefully acknowledge the feedback provided by two anonymous reviewers, as well as by a number of colleagues who read an earlier draft of this paper: Malte Battefeld, Chris Ebert, Isa Hendrikx, Nikos Koutsoukos, Therese Lindström Tiedemann, Roland Pooth, Henrik Rosenkvist, Philip Shushurin, Sarah Sippach, and Lars Erik Zeige.

In this article, we will examine how the Germanic constructicon is expanded by a process we label ‘debonding’ – a composite change whereby bound morphemes (clitics, affixes, affixoids) in a specific context develop into free morphemes (Norde 2009: 186). Focus will be on debonding of three Germanic prefixoids that derive from nouns denoting a hard or high impact object: Dutch kei ‘boulder’, German Hammer ‘hammer’, and Swedish kanon ‘cannon’. Drawing on an extensive corpus-based analysis of bound and free uses of these three prefixoids, we aim to demonstrate that debonding of prefixoids is a productive process of lexical innovation in Germanic languages, which may lead to the creation of new intensifying adverbs or evaluative adjectives. In addition, we aim to explore whether debonding of prefixoids can be fruitfully analysed from a constructional perspective. More in particular, we will address the question of whether the semantic and formal changes accompanying debonding are best accounted for by Traugott and Trousdale’s (2013) concept of ‘constructionalization’, or by Hilpert’s (2013) concept of ‘constructional change’, which includes changes in frequency. In so doing, we will complement the study by Trousdale and Norde (2013) who examine two other types of degrammaticalization (degrammation and deinflectionalization) from a constructionist perspective.

The body of the paper is organized as follows. In the next section, we will outline our constructionist approach to debonding and clipping of prefixoids, as well as discuss the notions of constructionalization and constructional change. In Sect. 3, we will present the sources and method of our empirical studies. Sections 4, 5 and 6 will be dedicated to a detailed analysis of Dutch kei, German Hammer and Swedish kanon respectively. For each prefixoid, we will compare the formal properties (construction types), semantics (degree of bleaching), collocational properties and productivity of its bound and free uses. The results of these three case studies will be compared and analysed statistically in Sect. 7. We conclude the paper with a discussion of the central question: do the results of our data analysis allow us to treat the debonding of a bound prefixoid as an instance of constructionalization?

2 Theoretical Preliminaries

2.1 Debonding and Clipping of Prefixoids

Affixoids form a specific class of bound morphemes that have been characterized as morphemes “which look like parts of compounds, and do occur as lexemes, but have a specific and more restricted meaning when used as part of a compound” (Booij 2009: 208, see also Booij 2010: 55ff.).¹ They form a very productive class – Hoeksema (2012), for instance, lists 696 Dutch compounds with prefixoids, and his list is far from exhaustive. When affixoids develop out of left-hand members of head-final compounds, they typically acquire evaluative meaning (in [N-N] compounds) or intensifying meaning (in [N-ADJ] compounds), as in Dutch wereldwijf ‘fantastic woman’ (< wereld ‘world’), or bloedmooi ‘drop dead gorgeous’ (< bloed ‘blood’). The lexical basis of such modifying prefixoids is often a noun, but adjectives (e.g. dolverliefd ‘madly in love (< dol ‘mad’)), or verbs (e.g. loeistrak ‘tight as a drum’ (< loei- ‘to blast, roar’)) can develop into prefixoids, too.² In many cases, there exists a cline from determinative compounds via simile compounds to evaluative/intensifying compounds, as the following examples show (Norde and Van Goethem 2014: 259)³:

(1)	[N-N] compounds: German Riesenhand ‘hand of a giant’ (determinative) – Riesenbühne ‘giant stage’ (simile) – Riesenstimmung ‘great atmosphere’ (evaluative)
(2)	[N-ADJ] compounds: Swedish jättelik ‘giant-like’ (determinative) – jättestor ‘as big as a giant’ (simile) – jättegullig ‘very cute’ (intensifying)

On the formal side, prefixoids often have specific characteristics as well. For instance, some Dutch prefixoids are followed by a linking vowel [ə], e.g. bere- (< beer ‘bear’), reuze- (< reus ‘giant’), or rete- (< reet ‘ass’). Moreover, Dutch prefixoids can be emphasized either by vowel lengthening (reeeeetegoed ‘ass good > excellent’), or by reduplication (spek- en spekglad ‘bacon and bacon slippery > very slippery’). German prefixoids, on the other hand, are prosodically different from determinative compounds, compare the different stress patterns in ˈHammerklavier ‘hammer piano > fortepiano’ and ˈHammerklaˈvier ‘outstanding piano’ (Schlücker 2013: 457).

In this article, we will focus on another typical formal property of such evaluative/intensifying prefixoids, which is that they can be severed from their head and written as a separate word before its R1. Once separated from its head, the prefixoid may acquire new morpho-syntactic functions, such as taking scope over an entire NP (as in (3)), or modifying a verb (as in (4)). That these free morphemes developed out of the prefixoid, and not out of the corresponding noun, is not only evidenced by their evaluative or intensifying meaning, but, in some cases, also by their form: when the prefixoid had a linking vowel (example (5)), or consonant (example (6)), these are preserved in the free form. Conversely, if a vowel is dropped in the prefixoid, as in German end- (< Ende ‘end’), it is also absent if the prefixoid develops into a free morpheme, as in (7).⁴

(3)	das spiel hat eine riesen deutsche community wo man genug hilfe findet.
	‘the game has a huge German community where one can find plenty of support’
	(DECOW 2012)
(4)	min fredag startade kanon
	‘my Friday started wonderfully’
	(SECOW14AX)
(5)	Onze kinderen hebben zich reuze (*reus) vermaakt (…)
	‘Our kids enjoyed themselves tremendously (…)’
	(NLCOW 2012)
(6)	Ik hou ziels (*ziel) van jou meer dan van wie dan ook.
	‘I love you with all my soul, more than (I love) anyone’.
	www.quizlet.nl/chapters/1179678/part-155-zayn/
(7)	würd mich über eine antwort end (*Ende) freuen!!!
	‘(I) would be much looking forward to a reply!!!’
	(DECOW14AX)

The development of free morphemes out of erstwhile bound ones are examples of debonding (Norde 2009: 186). Debonding may affect clitics, affixes (both inflectional and derivational), and affixoids. Examples include the Northern Saami postposition haga ‘without’, which developed out of an abessive suffix, or the independent uses of English ish (Norde 2009: 186–227). Examples of debonding of prefixoids were discussed earlier in, for instance, Norde and Van Goethem (2014, 2015), Van Goethem and Hiligsmann (2014), Van Goethem and De Smet (2014), Van Goethem and Hüning (2015) and Battefeld et al. (2018). The specific linguistic context for debonding of prefixoids is one in which the bound prefixoid can be reanalysed as an attributive adjective (in case of [Prefixoid-N] constructions), as in (8), or an adverb (in case of [Prefixoid-ADJ] constructions), as in (9) (note that the adverb takes scope over two coordinated adjectives):

(8)	ich habe einen riesen fehler gemacht
	‘I have made a huge mistake’
	(DECOW 2012)
(9)	Het klinkt allemaal reuze leuk en aardig: vergeven en vergeten, streep eronder en doorgaan!
	‘It all sounds very nice and neat: forgive and forget, let bygones be bygones and move on!’
	(NLCOW 2012)

Debonding of prefixoids into adjectives is a gradual process – adjectives that develop in this way are usually not inflected, at least not initially (Van Goethem and De Smet 2014: 253). Furthermore, debonding of prefixoids is often ‘sneaky’ in the sense of De Smet (2012: 7), who defines this sneakiness as “apparently thriving on structural ambiguities and (possibly superficial) resemblances to existing patterns”. This may be illustrated by German riesen ‘giant’: in 19 out of 32 adjectival contexts,⁵ the adjectival suffix –en is required (Norde and Van Goethem 2014: 270), so that einen riesen Fehler in (8) above looks like a perfectly grammatical construction. Similarly, Dutch reuze can be interpreted as an adjective in –e, which is the suffix used in most attributive contexts (Van Goethem and Hiligsmann 2014: 60). At a later stage, the debonded adjectives may acquire adjectival inflections, such as the indefinite neuter singular in the German example in (10), or the comparative form in the Dutch example in (11):

(10)	Ein rieses Dankeschön nochmal
	‘A huge thank you, once more’
	(DECOW2012)
(11)	Ik krijg ineens een leuk idee: een nog reuzere mergpijp op taartformaat!
	‘I suddenly have this great idea, an even bigger (lit. more giant) marrow-bone (kind of Dutch pastry) the size of a pie!’
	http://forum.deleukstetaarten.nl/viewtopic.php?id=30091

However, free uses of erstwhile prefixoids may also be the result of clipping in a specific (predicative) context (Van Goethem and De Smet 2014, Van Goethem and Hiligsmann 2014, Van Goethem and Hüning 2015, Norde and Van Goethem 2015, Battefeld et al. 2018). For instance, in example (12) stapel derives from stapelgek (lit. ‘pile mad’), whereas piep in (13) derives from piepjong (lit. ‘squeak young’). In these cases, the meaning of the full compound is transferred to the free prefixoid. Note that the corresponding attributive constructions in (14) and (15) are ungrammatical (Van Goethem and Hiligsmann 2014: 58).

(12)	Ik ben stapel op mooi gemaakte kinderfilms
	‘I am mad about beautifully made children’s movies’
	(NLCOW 2012)
(13)	En met 54 jaar ben je ook niet meer zo piep
	‘And at 54 one is not the youngest anymore’
	(NLCOW 2012)
(14)	*Een stapele jongen ‘a mad boy’
(15)	*Een piep meisje ‘a very young girl’

The use of clipped prefixoids in predicative position may be favoured by the existence of the structurally similar predicative bare noun construction, which is found in all three languages studied. In these constructions, illustrated in (16) and (17), the bare noun expresses a general quality, e.g. a profession or nationality, and has become less noun-like in the sense that it cannot be modified by an adjective, and more adjective-like because it can be modified by an adverb (cf. Berman 2009: 99–101)⁶:

(16)	Er ist (*netter) Lehrer
	‘He is a (nice) teacher’
(17)	Er ist ganz Lehrer
	‘He is the typical cliché of a teacher’

These constructions are structurally very similar to free prefixoids in predicative position, which can likewise be modified by an adverb:

(18)	Ja wir wissen das Spiel sieht total mist aus.
	‘Yes we know, the game looks totally crap’
	(DECOW2014AX)

In some works (e.g. Pittner and Berman 2006, Berman 2009) the predicative bare noun construction is even considered the only source of evaluative adjectives that are homophonous with prefixoids. On the other hand, Van Goethem and Hüning (2015) have shown that, e.g., Dutch top and German spitze (both meaning ‘top’) have developed out of two source constructions: debonding of the corresponding prefixoids and noun to adjective change in predicative position (on multiple source constructions see further Trousdale 2013, Van de Velde et al. 2013).

2.2 Constructionalization and Constructional Change

In this article we will discuss debonding and clipping of prefixoids from a constructionist perspective. Since prefixoids are neither free morphemes nor prefixes, they are best analysed in a framework which treats symbolic form-meaning pairings (i.e. constructions) as the basic unit of analysis, without putting constraints on the category of a construction or one of its parts. In constructional approaches to language and language change, words and larger units are defined by their formal and semantic properties, and these properties link them both to constructions that are similar in meaning and/or form, and to more schematic constructions that generalize over similar types. In the case of prefixoids, this means that e.g. Dutch bere- and reuze- are laterally linked because of their similarity in function, whereas both are vertically linked to (or ‘sanctioned by’ in construction grammar terms) the more general schema for intensifying prefixoids with an adjectival base in (19) (see further Norde and Van Goethem 2015; Norde and Morris 2018):

(19)

[<a > [b]Ai]Aj

⇔ [[very [SEM]i]j

Recently there has been a growing interest in changes in constructions and the emergence of new constructions (cf. among others Bergs and Diewald 2008; Hilpert 2013; Traugott and Trousdale 2013; Trousdale and Norde 2013; Barðdal et al. 2015; Van Goethem et al. 2018; a recent survey of the field is given in Noël 2016). From this usage-based view of language change, it is taken for granted that changes do not occur in isolation, but that linguistic context is highly relevant (e.g., Bergs and Diewald 2009). From such a diachronic point of view, Hüning and Booij (2014) have convincingly argued that the ‘umbrella’ notion of ‘constructionalization’ (Traugott and Trousdale 2013) is more appropriate than the notions of grammaticalization and lexicalization to account for hybrid and context-dependent changes, such as the rise of affixoids. Hüning and Booij (2014: 600) argue that “[T]he general concept of constructionalization (…) offer[s] a way out of the problems associated with the “element based view” and with the idea of a “cline”, (…)” because in Construction Grammar the idea of a cline can be replaced by a taxonomic network of related constructions (Trousdale 2008: 172).

Another basic tenet of the Construction Grammar approach is that the underlying mechanisms of change are analogical thinking and subsequent neoanalysis⁷ (cf. Traugott and Trousdale 2013). In 2.1, we have illustrated that language users recognize the formal and semantic connections between evaluative modifiers in compounds or intensifying prefixoids, on the one hand, and evaluative adjectives or intensifying adverbs, on the other, and that this may trigger attraction to these other constructions in the network and ultimately the shift from bound to free morphemes.

For all these reasons, we believe it is interesting to apply the constructional perspective to debonding. To this end, we first need to briefly discuss the two main constructional approaches to language change, as advanced by Traugott and Trousdale (2013) and Hilpert (2013) respectively.

Traugott and Trousdale (2013) define the notion of ‘constructionalization’ as follows:

Constructionalization is the creation of form_new-meaning_new (combinations of) signs. It forms new type nodes, which have new syntax or morphology and new coded meaning, in the linguistic network of a population of speakers. It is accompanied by changes in degree of schematicity, productivity, and compositionality.⁸ (Traugott and Trousdale 2013: 22).

Traugott and Trousdale (2013: 22) specify that

[M]inimally, constructionalization involves neoanalysis of morphosyntactic form and semantic/pragmatic meaning (…). Formal changes alone, and meaning changes alone cannot constitute constructionalization.

When the change affects only the semantic or the formal pole of the construction, but no new construction is created (which would imply both formal and semantic change), Traugott and Trousdale (2013: 26) call this a ‘constructional change’ instead of a constructionalization:

A constructional change is a change affecting one internal dimension of a construction. It does not involve the creation of a new node. (Traugott and Trousdale 2013: 26)

Constructional changes that typically ‘feed’ constructionalization are pragmatic inferences, semanticization of those inferences, form-meaning mismatches and some small distributional changes. Constructionalization may be followed by further constructional changes, such as expansion of collocations and, in some cases, morphological and/or phonological reduction.

Whereas the development of new constructions is mostly gradual, or “a succession of micro-steps preceding the creation of a new node” (Traugott and Trousdale 2013: 29), some (lexical) micro-constructions arise “with no prior constructional changes discernible” (ibid). This distinction is important when comparing debonding to clipping: both processes involve the free use of bound morphemes, but whereas debonding is gradual, clipping is instantaneous. As has been shown in Van Goethem and Koutsoukos (forthcoming), the Dutch compound member luxe ‘lit. luxury; luxurious’ (e.g. luxehotel ‘luxury hotel’) was first used as a free form in attributive constructions (e.g. een erg luxe hotel ‘a very luxurious hotel’), where it still collocates with a noun, before gradually expanding to other contexts such as the predicative one (e.g. het hotel is erg luxe ‘the hotel is very luxurious’), whereas a clipped prefixoid such as piep ‘very young’ (< piepjong lit. ‘squeak young’) did not develop in such a bridging context.

In the second major introduction to Diachronic Construction Grammar, Hilpert (2013) does not use the term of ‘constructionalization’ but refers to the emergence of new constructions as ‘constructional change’. In his view, constructional change not only manifests itself through form and meaning change, but through changes in frequency or distribution as well:

Constructional change selectively seizes a conventionalized form-meaning pair of a language, altering it in terms of its form, its function, any aspect of its frequency, its distribution in the linguistic community, or any combination of these. (Hilpert 2013: 16).

An important difference between Traugott & Trousdale’s definition of constructional change and Hilpert’s is thus that Hilpert includes frequency as a third level of change: “[E]ven if a change does not create new functions or new structures, a rearrangement of relative frequencies still brings about a constructional change” (Hilpert 2013: 17). Changes in frequency may refer to text frequency, but also to the relative frequency of the functional and structural variants of the construction.

The central aim of this paper, then, is to investigate whether debonding can best be accounted for by Traugott and Trousdale’s (2013) concept of ‘constructionalization’, or by Hilpert’s (2013) concept of ‘constructional change’. According to the definition, this implies changes in formal, semantic and/or distributional properties. We will investigate this question through three different cases of debonding, which will be subject to a detailed qualitative and quantitative analysis of their formal and semantic properties, as well as of their change in collocational properties and productivity. Before examining the case studies, we will present the corpora and discuss some methodological issues in Sect. 3.

3 Sources and Methods

Data for this study are drawn from COW14 (Corpora from the Web),⁹ a gigatoken database of tagged and lemmatized texts from the web, compiled at the FU Berlin in 2011 and 2014 and released in 2014–2015 (Schäfer 2015). This corpus is perfectly suited to the study of language change in progress from a comparative perspective: it provides similar data sets from different languages, among them Dutch, German and Swedish, and a substantial portion of these data come from informal sources,¹⁰ which is the typical locus of recent and innovative constructions. The subcorpora used for this study are given in Table 1. If no alternative source is given, all examples mentioned in the remainder of this paper are drawn from one of these three subcorpora.

Table 1 Subcorpora used in this study

To collect our data, we proceeded as follows: using the Colibri² query interface, we carried out a search for both the bound and free forms of Dutch kei, German H/hammer and Swedish kanon. Since written corpus data do not allow us to check phonological criteria such as stress, the distinction between bound and free forms has been exclusively based on spelling. The results were imported into Excel, and we used the RAND function in Excel to shuffle them. The first results in the concordance were checked manually and all irrelevant hits were discarded until we had a sample of 1000 relevant occurrences for each construction, which made 6000 occurrences in all.

In the case of bound kei, we removed irrelevant examples such as keizer ‘emperor’ and determinative compounds such as keisteentjes ‘cobble-stones’ or keileem ‘boulder clay’. With respect to free kei, all occurrences of the noun kei, in its literal or figurative meaning (e.g. een kei in wiskunde ‘a crack mathematician’), were excluded, so that the comparison between the bound and free forms is exclusively based on the use of kei as prefixoid or part of a simile compound and its debonded uses. Similarly, for bound Hammer-/hammer-, we removed determinative compounds (e.g. Hammerschlag ‘hammer blow’), classifying compounds (e.g, Hammerhai ‘hammerhead shark’) and other irrelevant results such as hammermäßig ‘hammer-wise’.¹¹ For the free hammer construction, we took a random sample of 1000 tokens of lower case hammer. We did not include Hammer (with upper case H) in this analysis, because a pilot study revealed that only very few upper case Hammer constructions were relevant tokens.¹² From the raw hammer data, we discarded irrelevant hits such as the noun hammer (erroneously written in lower case), or hammer as short for haben wir ‘have we’ (colloquial). As far as our third prefixoid, Swedish kanon, is concerned, we discarded irrelevant examples (e.g. kanonen ‘the cannon; cool’, kanonkula ‘cannonball’, kanonisk ‘canonical’, kanonjär ‘cannonneer’). For free kanon we had to remove all examples where kanon was the original noun ‘cannon’, or a homographic noun meaning ‘canon’.

The remaining examples were tagged for R1 (i.e. the second compounding element in bound constructions, or the first word to the right in free constructions), part of speech of the R1, semantic type (e.g. simile) and particular properties such as reduplication. For H/hammer we furthermore noted whether the prefixoid was spelled with upper case or lower case, and whether the free form of the prefixoid was preceded by an article (definite or indefinite).

This database forms the basis for the quantitative analyses in the sections below. For each bound and free form, we will analyse four properties: (i) construction type (the part of speech the bound or free prefixoid collocates with), (ii) (in case of [Prefixoid-ADJ] constructions) semantic bleaching (the proportion of simile and intensifying constructions), (iii) collocational properties (R1 types and tokens) and (iv) productivity. We use two measures for productivity: type/token ratio (TTR) and Potential Productivity (PP). The latter is discussed in Baayen (2009), and is calculated by dividing the number of hapax legomena of a particular word formation pattern in the corpus by the total number of tokens of that pattern (Baayen 2009: 902). This ratio will allow us to compare the potential growth rate of the bound and free morphemes in both languages.

In Sect. 7, we will offer statistical analyses of all three case studies in order to assess whether formal, semantic, and collocational differences as well as differences in productivity between bound and free kei, hammer and kanon are significant. If the probability that an attested difference between the bound and the free form is due to chance is smaller than 0.05, we will argue that we are dealing with a constructional change. Since all four properties that we examine in this paper are quantifiable, constructional change, or the absence thereof, can be calculated with the help of associative statistics. A more tricky issue however, is how we can identify constructionalization. As Hilpert (2015: 134) aptly puts it: “Just after how many constructional changes exactly do we have a construction that counts as a new node?” We will return to this question in Sect. 8.

4 Dutch kei

The Dutch noun kei means ‘boulder, cobble-stone’ and is attested as left-hand member of simile compounds at least since the nineteenth century. According to Van der Sijs (2010), the compound keihard ‘rock-hard, as hard as (a) stone’ goes back to 1872; its first attestation in the Dictionary of Dutch language (WNT, s.v. keihard) is from 1921 (20). The same dictionary lists an occurrence of the simile compound keidood ‘stone-dead’ from 1803 (s.v. kei ^I) (21).

(20)	De kluiten droogden glashard op, (…), waardoor een bovenlaag van keiharde knikkers … verkregen werd (1921)
	‘The clods got as hard as glass when they dried up, (…), as a result of which an upper layer of rock-hard marbles … was obtained’
(21)	Hij viel van de stelling en hij was keidood (1803)
	‘He fell from the stand and he was stone-dead’

In the same period, keihard is already attested with a metaphorical meaning, too (‘very hard, with a lot of power’). In these cases, kei can be analyzed as an intensifying prefixoid meaning ‘very’:

(22)	Een keihard schot (met den voetbal) (1872)
	‘A powerful shot (with the football)’
(23)	Wanneer een voorwaarts op een achterspeler toeloopt en deze laatste trapt den bal keihard in het gelaat van den toeloopenden voorhoedespeler (…) (1909)
	‘When a forward runs into a back player and the latter kicks the ball at full speed in the face of the vanguard player who is running in his direction (…)’.

The WNT dictionary does not mention the use of kei in combination with other adjectives than hard or dood nor the use of kei as a free intensifying morpheme. In the sections below, we will examine in more depth the formal (4.1) and semantic (4.2) properties of bound and free kei, and its collocational properties and productivity as both a bound and free form (4.3).

4.1 Construction Types

We analysed a corpus sample of 1000 relevant tokens of both bound and free kei. For bound kei, this sample includes occurrences in which kei forms part of a simile or an intensified compound, either written as one word (983 occurrences) or hyphenated (only 17 compounds). The sample of free kei includes 1000 instances in which kei is written as a separate word and still preserves its simile or intensifying meaning. As indicated in Sect. 3, occurrences of the noun kei have been excluded since these are not debonded uses of the compound member.

Table 2 presents the construction types of bound and free kei as observed in both corpus samples.

Table 2 Bound and free kei – construction types

Table 2 indicates that bound kei mostly combines with adjectives (24–25) or adverbs (26); some examples do not contain sufficient context to determine whether the compound head was an adjective or an adverb (27). In one example only, bound kei combines with a nominal head (28).

(24)	Vissen leven in een keiharde, stressvolle wereld onder water waar het constant “eten of gegeten worden” is.
	‘Fish live in a tough, stressful underwater world with a constant threat of “to eat or to be eaten”.’
(25)	Laminaat ligt, maar ik ben ook keikapot …
	‘Laminate is ready, but I’m exhausted also …’
(26)	Ik was keihard aan het meezingen met het liedje I will always love you van Dolly Parton.
	‘I was singing loudly along with Dolly Parton’s song “I will always love you”.’
(27)	En gelukkig overal keihard , volume op standje 10
	‘And luckily everywhere very loud(ly), at volume 10’
(28)	Hij is echt een keisukkel (…)!
	‘He is really a complete idiot (…)!’

Free kei is used in a broader range of construction types than bound kei. Besides having scope over adjectives (29) and adverbs (30), which are still the most frequent heads, free kei can also intensify quantifiers (31) and verbs (32). Scope over a noun, as in (33), is still marginal. In 6 examples, kei is used as a predicative adjective without modifying a head (34): this signals that debonded kei has undergone “flexibilization”, i.e. an increase in syntactic freedom (Norde 2009: 131). Moreover, as shown by the examples (35–36), kei may also have scope over (adjectival or noun) phrases, which signals “scope expansion” (Norde 2009: 131) compared to bound kei.

(29)	Heb het wel al miljoen keer gezegd maar ik ben echt kei en kei trots op je !!!
	‘Have said it a million times already but I’m really super proud of you!!!’
(30)	Om 13 u verwachten ze me in het Jan Palfijn om een biopsie te nemen van mijn pancreas via mijn maag (…) En ik moet daarom dus kei lang nuchter zijn!
	‘At 1 p.m. they expect me in the Jan Palfijn hospital for a biopsy of my pancreas through my stomach (…) And therefore my belly needs to be empty for such a long time!’
(31)	Ik heb dagen dat ik echt helemaal niets eet en dagen dat ik kei veel eet !
	‘There are days that I am really eating nothing and days that I eat very much!’
(32)	Nogmaals kei bedankt dat ik hier mag wonen ik voel me echt thuis.
	‘Thanks again very much that I can live here, I feel really at home.’
(33)	‘s middags voelde ik me goed, ‘s avonds kei keelpijn en ziek … dag erna dood ziek
	‘In the afternoon I felt good, in the evening terrible sore throat and sick … next day sick as a dog’
(34)	(…) uitdagingen, die hij tof, kei , en hip moest vinden.
	‘(…) challenges, which he had to find nice, cool, and hip.’
(35)	Ze komen dus kei te laat , maar dat moeten ze maar op de koop toenemen.
	‘So they are arriving way too late, but they have to put up with that.’
(36)	En zeg nou zelf, 50 cent is echt kei geen geld.
	‘And let’s face it, 50 cents is really no money at all.’

Free kei mostly acts as an adverb modifying an adjectival, adverbial (including quantifier) or verbal head. On the other hand, kei has an adjectival function when it has scope over a noun or noun phrase, or when it is used predicatively. Its adjectival status manifests itself when kei is coordinated with other adjectives, as illustrated in example (34). This shows that the debonding of kei also involves “recategorialization” (Norde 2009: 131). To sum up, it is shown that free kei has some “innovative” uses in relation to bound kei that result from debonding, although they are not (yet) very frequent.

4.2 Semantic Properties

Table 3 provides an overview of the semantics of bound and free kei. We only include examples where the simile reading is potentially available, i.e. where kei has scope over adjectives, adverbs and quantifiers; when kei modifies verbs, nouns or is used predicatively, it always has an intensifying or evaluative value.

Table 3 Bound and free kei (R1 = Adj(P)/Adv/Quant) – semantics

Bound kei acts as an intensifying prefixoid in the great majority of occurrences (91.29%). The remaining instances are all examples of keihard used as a simile compound (‘as hard as stone’). Examples (37) and (38) illustrate the two semantic types: keihard is used as a simile in (37) and as an intensifying compound in (38):

(37)	Chocola uit Ghana heeft – om smelten in tropische temperaturen te voorkomen – een lager vetgehalte en is dus keihard.
	‘Chocolate from Ghana – in order to prevent melting in tropical temperatures – has a lower fat content and that’s why it is so hard.’
(38)	Dat laatste lieg ik keihard , want ik ga daar als eerste van genieten !!!
	‘About that last thing I’m just really kidding, cause I will be the first to have fun!!!’

As Table 3 and example (39) show, the simile interpretation of kei hard is still available for the free form, albeit more exceptionally. We suggest to regard these 16 occurrences as deviant spellings of simile compounds. In all other occurrences, free kei has an intensifying meaning, as in example (40).

(39)	De klei wordt kei hard en neemt gewoon geen vervuiling op
	‘The clay gets rock hard and just does not take any pollution’
(40)	United brands … een muts, sjaal en wanten ( kei Ally McBeal-achtig dus) van hetzelfde
	‘United brands … a hat, scarf and mittens (very Ally McBeal-ish) of the same type’

Surprisingly, in a number of occurrences kei even intensifies intrinsically non-gradable adjectives, such as vrijwillig ‘voluntary’ (41) and vatbaar (voor) ‘susceptible, prone (to)’ (42).

(41)	Bij het binnenkomen hadden we ons ‘ kei vrijwillig’ opgegeven voor het kauwgom-bellen-blazen (…)!
	‘When we entered we signed up ‘very voluntarily’ for chewing gum-blowing!’
(42)	Ben kei vatbaar voor dit soort klote dingen.
	‘[I] am very prone to this kind of fucked up things.’

Both bound and free kei can form part of an emphatic reduplicative construction (see 2.1); three different types of reduplication occur in the corpus sample:

(43)	Daten is kei-en keihard, en je kunt meedogenloos tegen de keien worden gesmeten
	‘Dating is extremely tough, and you can end up thrown ruthlessly against the rocks’
(44)	Jayh – Doe de thing is echt keikeikeihard.
	‘Jayh – Do the thing is really amazing.’
(45)	Het is een nieuw soort horrorfilm, (…), met 2 dames als hoofdrolspeelster, die allebei keihard (maar dan ook KEIhard) moeten vechten voor iets dat ze graag willen (…).
	‘It’s a new kind of horror movie, (…), with two women as protagonists, who both have to fight really hard (and I mean REALLY hard) for something that they want badly (…)’

The emphatic reduplicative construction is not restricted to purely intensifying morphemes; it is also available for simile compounds,¹³ and remarkably, we even observed reduplication of free similative kei (46). We regard these cases as instances of simile compounds written as separate words (instead of the standard spelling kei- en keihard).

(46)	Niet lekker, het koekje was kei en kei hard , ze zijn zo de vuilnisbak ingegaan, jammer.
	‘Not tasty, the cookie was rock-hard, they ended up in the bin, too bad.

In a number of instances, the compound keihard seems to have undergone semantic extension. In examples such as (47) and (48), kei still functions as an intensifying prefixoid, but the adjective or adverb keihard is used in contexts where the use of hard alone would not fit. The meaning of this ‘lexicalized’ keihard can be described as ‘obvious(ly), loud and clear’.

(47)	Het staat er keihard (*hard) , zwart-op-wit.
	‘It is loud and clear, black-on-white.’
(48)	Dat ze het doen weet je nu ook keihard (*hard) !!
	‘That they are doing it, now at least you know it loud and clear’

This lexicalized use is also attested for separated kei hard (49). Example (50) is a particular use of lexicalized kei on its own. This case probably illustrates a clipped form of keihard which retains the meaning of the entire compound form (cf. Norde and Van Goethem 2015).

(49)	En misschien is daar ooit iemand in getrouwd, die nu kei hard (*hard) gescheiden is.
	‘And perhaps someone ever got married there, who is now irrevocably divorced’
(50)	Maar het tegendeel werd mij op dat moment kei bewezen.
	‘But the opposite was clearly proved to me at that time.’

We can conclude that, at the semantic level, bound and free kei are used with a simile or an intensifying meaning. Free kei has not undergone any extension to new meanings or resemanticization. Language users tend to associate the simile interpretation with the compound keihard (written as one word).

4.3 Collocational Properties and Productivity

Table 4 shows the different types of adjectival and adverbial heads of bound kei; the occurrence keisukkel ‘great idiot’ (nominal head) has been excluded in order to make the data set uniform for the calculation of the productivity measures. As already suggested by the preceding examples, bound kei combines in the vast majority of cases with the adjective/adverb hard ‘hard’: keihard even covers 89.39% of the total corpus sample. Keigoed ‘very good’, keigaaf ‘absolutely great’, keileuk ‘very nice’ and keigezellig ‘very cosy’ complete the top 5, but their total token ratio only amounts to 6.40%. All the other types occur less than 5 times in the corpus, including 21 hapax legomena. Because of the small number of different adjectival/adverbial types (only 35) and the fact that kei almost exclusively combines with a single type, bound kei has a very low type-token ratio (0.04). Its potential productivity (0.02) is extremely low as well: it indicates that compounding with kei is running a “high risk of saturation”, in Baayen’s (2009) words. Potential expansion to new heads is therefore implausible.

Table 4 Bound kei (R1 = Adj/Adv) – collocates and productivity

Tables 5, 6 and 7 contain information about the collocational properties and productivity of free kei used as a modifier with scope over adjectives/adjectival phrases, adverbs and quantifiers (Table 5), verbs (Table 6) and nouns/noun phrases (Table 7).

Table 5 Free kei (R1 = Adj(P)/Adv/Quant) – collocates and productivity

Table 6 Free kei (R1 = V) – collocates and productivity

Table 7 Free kei (R1 = N(P)) – collocates and productivity

It is worth noting that, among the 155 different adjectival/adverbial types, 10 types are loanwords from English (cool, cute, happy, awesome, etc.), covering a total of 23 occurrences, while only one compound with bound kei contains an English loanword (keichill). Language users possibly tend to write combinations with English loanwords as two separate words because this is the standard spelling for English compounds.

Given the low number of tokens of kei with scope over verbs and nouns, we will not calculate its productivity for these construction types. With respect to the adjectival/adverbial/quantifier types, the adjective/adverb hard is clearly still the most frequent type (27.93%), but its frequency is much lower than that observed in combination with bound kei (89.39%). Instead, the distribution of free kei is spread over a far greater number of types (TTR = 0.16). Thanks to a higher number of hapax legomena, unbound kei also shows a greater potential productivity than bound kei (PP = 0.09). Both the type/token ratio and the potential productivity measures indicate that free kei is much more productive than bound kei. This does however not imply that bound kei is not a productive form with respect to its absolute frequency for instance, but the ratios indicate that it is almost saturated by one single type. The combination kei with hard reflects the oldest stage of the simile compound and is so high in token frequency that there may be a tendency to lexicalization (see 4.2) and univerbation, whereas more recent types with lower token frequency are more often orthographically separated from kei.

5 German Hammer

German Hammer ‘hammer’ is used as a noun referring to the tool, as well as in exclamations and predicative constructions to express a series of emotions, ranging from frustration and indignation to surprise and appreciation. In Grimms’ German dictionary (s.v. Hammer), these emotive uses are associated with the mythological hammer of the god Thor, a source of both fear and admiration. The following historical examples¹⁴ illustrate metaphorical and exclamative uses of Hammer:

(51)	Mein Wort ist ein Hammer /der die Felsen zerschmettert. (DTA 1603)
	‘My word is a hammer that crushes the rocks’ (DTA 1760)
(52)	Daß dich der Hammer !
	‘OMG!’

An early example of the simile compound hammerhart ‘hard as a hammer’ is given in (53). The quotation marks may imply that the expression had not yet been conventionalized at the time.

(53)	Der Stein […] war so groß wie eine Tischplatte, feinkörnig “hammerhart” […]. (Die Zeit 1949)
	‘The stone […] was as large as a tabletop, finegrained, and iron-hard […]’

These figurative uses of Hammer, which have been around for centuries, have most likely played a part in the rise of Hammer-/hammer- as an evaluative and intensifying prefixoid. In the sections below, we will discuss formal, semantic and collocational properties of Hammer-/hammer-, in the same way as we have done for kei. To ease reading, we will henceforth write hammer to refer to both upper case and lower case spellings.

5.1 Construction Types

German compounds are head-final, which implies that the second compound member determines the part of speech. Accordingly, formations with hammer should be written in upper case when R1 is a noun, and with lower case when R1 belongs to a part of speech other than nouns, according to German spelling rules. When R1 is an adjective initial capitals may be used as well, but only in combination with a hyphen, as in example (54). Hyphens may furthermore be used in compounds to emphasize the first compound member,¹⁵ which may account for the hyphen in (55). However, hyphens are not consistently used, as the contrastive examples in (55) and (56) show. This suggest that users are uncertain about the morphological status of hammer, although spelling inconsistencies may also be due to the informal register represented by the COW corpus. With nouns, there is a lot of variation – most examples of lower case nouns are spelling errors (other nouns in the context of these examples lack upper case initials as well). The frequencies of all spelling variants are listed in Table 8.

Table 8 Spelling variants of bound Hammer-/hammer- according to part of speech of R1

(54)	Das Game is so Hammer-geil
	‘The game is so totally cool’
(55)	Vor uns liegt ein echtes Hammer-Wochenende
	‘We’ve got a really great weekend ahead of us’
(56)	Dann steht einem Hammerwochenende nichts mehr im wege.
	‘Then nothing can prevent us from a great weekend.’

As Table 9 shows, bound hammer is most frequently used with adjectives (57), followed by nouns (58), adverbs (59), and quantifiers (60). Interestingly, the prefixoid may also collocate with other prefixoids, as in (61).

Table 9 Bound and free hammer – construction types

(57)	Naja, jetzt bin ich ja zufrieden mit der Vichy Nutrilogie 2, die für hammertrockene Haut gedacht ist;).
	‘Well, at the moment I am happy with Vichy Nutrologie 2, meant for extremely dry skin;)’.

(58)	Außerdem gefällt mir der Kandidat, der bei TV Total gecastet wurde, der hat echt eine Hammer-Stimme.
	‘Moreover I like the candidate that had been casted by TV Total, he really does have an awesome voice.’
(59)	Goku wurde von dem cyborg hammerhart geknebeld.
	‘Goku was tied up by the cyborg very rigidly’.
(60)	Ich finde 300 g Futter + Snack hammerwenig.
	‘I find 300 grams of (dog) food plus a snack very little.’
(61)	für so einen preis muß bei mir das mu [makeup] hammersuper sein.
	‘for such a price I think makeup has to be absolutely amazing’

From Table 9 it is evident that there is more variation in free hammer constructions than there is in bound hammer- constructions. Like its bound equivalent, free hammer may collocate with adjectives (62), adverbs (63), quantifiers (64), and nouns (65).

(62)	Nächste Folge wird hammer spannend (…)
	‘The next episode is going to be absolutely thrilling’.
(63)	matt du kannst hammer gut küssen *:) *
	‘Matt, you kiss extremely well’
(64)	aber das is bestimmt hammer viel arbeit …
	‘but that is clearly going to be a whole lot of work’
(65)	Ganz ehrlich, ich glaube Dragon Age ist ein hammer Spiel mit hammer Story (…)
	‘Quite frankly, I find Dragon Age an awesome game with an awesome story (…)’

In examples (62–64) hammer can be substituted by canonical intensifying adverbs such as sehr ‘very’ or furchtbar ‘terribly’, suggesting it is functionally similar to adverbs (but adverbs are not formally distinguished from the corresponding adjectives, so it is not possible to establish whether full conversion has occurred). On the other hand, when hammer precedes a noun, as in (65), it is clear that it has not (yet) become a fully-fledged adjective, as hammer does not formally agree with the noun (in which case we would expect hammeres and hammere respectively). Nevertheless, there is evidence that some speakers interpret hammer as an adjective like any other, with the correct inflections such as the indefinite neuter nominative singular (66), which is furthermore modified by the adverb total ‘totally’, the definite masculine dative singular in (67), or the comparative and superlative in (68), followed by the interesting meta-comment that this is not a grammatical construction, which suggests that the speakers are well aware that they are using innovative forms.

(66)	oh man ein total hammeres ende
	‘Oh man, a totally awesome ending’
	www.fanfiktion.de/r/s/4a22e2fb0000e22606705dc0/date/0/1
(67)	Alles in allem ein super Gesamtpaket zu einem Hammeren Preis.
	‘All in all, a super package deal for a terrific price.’
	http://www.fat-burners.org/in-den-wissenschaftlichen-ueberpruefungen-zeigte-sich-eine/
(68)	das cover, hammer ! die story, hammerer ! der mann, am hammersten !!! (ich weiß das das grammatikalisch falsch ist, mir egal!)
	‘The cover, awesome! The story, even more awesome! The man, most awesome! (I know this is grammatically incorrect, don’t care!)’
	https://www.amazon.de/Mad-Love-Tower-Don-Both/product-reviews/3945164346?pageNumber=6

As Table 9 also shows, however, that free hammer occurs in more construction types than bound hammer. It may modify a verb (69), or occur in predicative position. The latter construction is very frequent, and out of these 685 tokens, 335 have a definite article (example (70)), 20 have an indefinite article (example (71)), whereas 330 have no article at all (example (72)). Battefeld et al. (2018) note that the presence or absence of the indefinite article does not make a semantic difference. Morphosyntactically, however, hammer preceded by an article behaves more like a noun than bare hammer, as shown in (73), where hammer is modified by an adjective.

(69)	“you are not alone” von michael, er hat es hammer gesungen
	‘Michael’s “You are not alone”, he sung it magnificently’
(70)	düsseldorf war der hammer !!!
	‘Düsseldorf was fantastic!!!’
(71)	(…) bin ich seit einem jahr in pension – mit nur 62% der bezüge. Ist schon ein finanzieller hammer !
	‘(…) since a year I am retired – with only 62% retirement benefits. A financial blow for sure!’
(72)	Wir leben inner leistungsgesellschaft, und der Druck ist hammer.
	‘We live in a meritocracy, and the pressure is enormous.’
(73)	aber der absolute hammer is der dirrty look mit den blond/schwarzen haaren.
	‘but the real smasher is the dirty look with blond/black hair.’

According to Pittner and Berman (2006: 241), predicative hammer is the result of noun to adjective “conversion” in predicative position (cf. examples (16–17) in Sect. 2.1), and they explicitly rule out the role of prefixoid constructions¹⁶ such as hammerhart ‘very hard’ in the emergence of adjectival hammer. Example (72) however, casts doubt on this claim, since hammer may be a clipped form (compare hammerharter Druck ‘very strong pressure’). A more likely scenario is one in which debonding, clipping and noun to adjective conversion in predicative contexts all contributed to the rise of free hammer constructions and possibly reinforced one another (see Van Goethem and Hüning 2015 for a similar analysis of Dutch top and German spitze).

5.2 Semantic Properties

When hammer collocates with an adjective, adverb or quantifier it can, like Dutch kei, have either simile or intensifying meaning. Table 10 shows a substantial difference between bound and free hammer in this regard: whereas 39 occurrences of bound hammer represent the simile construction hammerhart, free hammer is exclusively intensifying.

Table 10 Bound and free hammer (R1 = Adj(P)/ Adv/Quant) – semantics

Hammerhart ‘hard as a hammer’ makes up more than a third of bound hammer constructions, but we only count them as similes when they refer literally to the substance of objects or body-parts (74), sound (75), or impact (76). In most cases however, hard is used in a metaphorical sense, e.g. ‘cool’ (77), ‘serious’ (78), or ‘difficult’ (79). Since these examples do not express a comparison to the physical properties of a hammer, we count them as intensifying.¹⁷

(74)	Ich nahm mir einen großen Hammer, prüfte ihn auf Härte und stellte fest, daß er hammerhart war.
	‘I took a large hammer, checked its hardness and concludes it was hard as a hammer.’
(75)	Ich höre gerne laut Musik und lege Wert auf ordentlichen Klang – also glasklare Höhen und hammerharte Bässe …
	‘I love listening to loud music and I appreciate a good sound, meaning crystal-clear highs and very loud basses...’

(76)	Er bekommt einen hammerharten Faustschlag ins Gesicht.
	‘He gets an iron hard punch in the face.’
(77)	Die Coverart ist auf jeden Fall hammerhart !
	‘In any event, the cover art is totally cool!’
(78)	Natürlich kann es sich in Einzelnen Fällen um hammerharte Allergien gegen Fischeiweiß usw handeln.
	‘Naturally, in some cases, this may be due to serious fish protein allergies etc.’
(79)	Eigentlich müsste ich richtig viel lernen, da ich demnächst ne hammerharte Klausur schreibe (…)
	‘Actually I need to study a lot, because I will have a very tough exam soon (…)’

When hammer has an intensifying function, it may collocate with a variety of adjectives (cf. Tables 13 and 14). The meta-comment in example (80) shows moreover that hammer is a strong booster, as hammergeil is considered the final point on a rating scale:

(80)	Von spitze über geil, echt geil, einfach geil, voll geil bis zu hammergeil reichen die Bewertungen
	‘The scores range from top via cool, really cool, simply cool, fully cool to hammer cool’

To further increase its intensifying function, hammer may co-occur with one or more other prefixoids for emphatic effect, both when the R1 is an adjective (81) or a noun (82)¹⁸:

(81)	Er sieht einfach hammer-mega-geil aus.
	‘He just looks absolutely totally cool’.
(82)	Ich investiere meine Zeit in eine Sache, die mir vorab von RTL als der Hammer-Super-Knaller-Mega-Event verkauft wird, und erlebe so was …
	‘I invest my time in something RTL claimed was going to be a fantastic event, and now this …’

When R1 is a noun, hammer may enhance the positive meaning of the noun, as in (83), or conversely its negative meaning, as in (84). In most cases, however, it has ameliorative function, as in (85).

(83)	wenn ihr mal ein eigenes, richtig gutes Hammer -Schnäppchen […] habt
	‘in case you would happen to have a really good real bargain yourself’
(84)	Ein Hammer -Gewitter: Regen, wie aus Eimern geschüttet […]
	‘An intense thunderstorm: rain by the buckets […]’

(85)	ich hab fast ein jahr in kapstadt gewohnt und es ist eine hammerstadt
	‘I have lived in Cape Town for almost a year and it’s a great city’

5.3 Collocational Properties and Productivity

In this section, we list the most frequent R1s of both bound and free hammer as well as their type/token ratio and potential productivity. With nouns (Tables 11 and 12) hammer is quite productive – with both free and bound forms, more than half of the tokens are hapax legomena, and the most frequent tokens in absolute numbers form a relatively modest set in terms of relative frequency. Moreover, although ranked differently, many of the most frequent nouns with bound hammer are among the most frequent collocates of free hammer as well.

Table 11 Bound hammer (R1 = Noun) – collocates and productivity

Table 12 Free hammer (R1 = Noun) – collocates and productivity

As far as adjective constructions are concerned, we already noted in the previous section that hammerhart occurs with far higher token frequency in the bound hammer sample. Furthermore, Tables 13 and 14 show that bound and free hammer have the same #1 adjective R1, geil, and some similarities in lower ranking adjectives as well. Productivity is low with both constructions.

Table 13 Bound hammer (R1 = Adjective, Adverb, Quantifier) – collocates and productivity

Table 14 Free hammer (R1 = Adjective, Adverb, Quantifier) – collocates and productivity

Finally, Table 15 lists the verbs that may occur as R1 with free hammer. It is clear that these do not represent a very productive construction type, and therefore we did not calculate its type/token ratio or potential productivity.

Table 15 Free hammer (R1 = V) – collocates

6 Swedish kanon

The Swedish noun kanon ‘cannon, which ultimately derives from Italian cannone (< Latin canna ‘tube, cane’ + the augmentative suffix --one), was borrowed into Swedish in the seventeenth century. One of the earliest examples given in the Dictionary of the Swedish Academy (SAOB, s.v. kanon) is (86).

(86)	Stenbock … låter spela på Slottet medh Canoner. (1656)
	‘At the castle, Stenbock had cannons fired.’

The noun soon came to be used metaphorically as a curse, e.g. in Bomber och Canoner! ‘bombs and cannons!’ (1791). In the 1920s, it started to appear in sports journalism (Lundbladh 2002: 30) to refer to high speed or impact (kanonskott ‘canonball shot’) or exceptional strength (kanonform ‘top condition’). The first attested [kanon-ADJ] formation is kanonfull (first attested 1909, according to SAOB),¹⁹ literally meaning ‘cannon drunk’.²⁰ In this sense, kanon could also be used independently, as in dricka sig fullständigt kanon ‘to get wasted (lit. drink oneself cannon)’. Clearly, the typical properties of a cannon (strength, impact and loudness) invited metaphorical extension, which in turn led to the evaluative and intensifying properties discussed in Sect. 6.2. The first occurrence of an evaluative [kanon-N] construction, kanonväder ‘great weather’ is mentioned in the 1986 edition of the Swedish Word List (SAOL) and labelled ‘colloquial’; the first intensifying [kanon-ADJ] construction, kanonbra ‘very good’ appears in the 1998 edition, likewise with the addition ‘colloquial’.

6.1 Construction Types

As for the other two prefixoids, we selected 1000 bound and 1000 free forms of kanon. Of the bound forms, 11 are hyphenated (e.g. kanon-blogg ‘great blog’, kanon-kul ‘really cool’), 989 are written as one word. As Table 16 shows, the distributional differences between bound and free kanon are substantial.

Table 16 Bound and free kanon – construction types

Bound kanon is most frequently used with adjectives as R1 (87), followed by nouns (88), adverbs (89), quantifiers (90). Example (91) illustrates that kanon can also be followed by other morphemes, such as the English suffix --ish (91):

(87)	Verkar vara en kanonfin häst!
	‘This seems to be a very fine horse!’
(88)	hoppas du haft en kanonkväll på stan!
	‘(I) hope you had a great evening in town!’

(89)	Nu ikväll kom jag på att det var kanonlänge sen jag skrev en sån.
	‘Tonight I realized it’s been ages (lit. very long) since I wrote one of those’
(90)	Jag jobbade ju hos Lotta o vi hade kanonmycke folk (…)
	‘I was working at Lotta’s and we had a lot of guests (…)’
(91)	Kommer bli kanonish !
	‘It’s going to be fantastic (ish)!’

Free kanon occurs in the same constructions as bound kanon, i.e. in collocations with adjectives (92), adverbs (93), quantifiers (94) and nouns (95), but is it also found in other construction types.

(92)	Jag älskar jul och traditioner tycker det är kanon mysigt, gör inte ni?
	‘I love Christmas and traditions, I think it’s really cosy, don’t you?’
(93)	I lördags blev det babysim som vanligt och det gick kanon bra !
	‘On Saturday we went baby-swimming as usual and it went really well!’
(94)	Annars var de kanon mycket gott.
	‘Apart from that they were really very tasty.’
(95)	tummen upp även för fotografen, som gjort ett kanon jobb !
	‘Thumbs up for the photographer too, who has done a terrific job!’

Additionally, free kanon collocates with verbs (96),²¹ and it is part of the fixed expression att ha kanon ‘to have a great time’ (97):

(96)	Själv sov jag kanon , men jag blev fruktansvärt sur när jag vaknade av min väckarklocka.
	‘I myself slept very well, but I was very irritated when the alarm woke me up.’
(97)	Barnen har hur roligt som helst och vi vuxna har det ocksa kanon.
	‘The kids are having a blast and we adults are also having a great time.’

Most frequently of all, however, free kanon is found in predicative position, as in examples (98–99), where it translates as ‘great, fantastic’. In these constructions, kanon may be a clipped adjective (kanonbra ‘very good’), but it may also have developed out of debonded kanon, which spread from attributive to predicative position.

(98)	Och tillsammans med parmesan blir det kanon.
	‘And together with parmesan it is going to be delicious.’
(99)	Två av böckerna är riktigt kanon !
	‘Two of the books are really great!’

It is difficult to determine the part of speech of predicative kanon. Unlike in Dutch and German, predicative adjectives are inflected in Swedish so that, if kanon were a fully-fledged adjective, we would expect a neuter form kanon t in (98) and a plural form kanon a in (99). The absence of inflection is however in line with Van Goethem and De Smet’s (2014) observation that debonding of affixoids is gradual, and that inflectional properties may be acquired at a later stage.²² We do in fact find inflected forms elsewhere, but they appear to be very rare:

(100)	Det kunde vara kanont om ochså personen kunde spela på saxofon.
	‘I would be great if this person could also play the saxophone.’
	www.danslogen.se › Forum › Dansband
(101)	Poolerna var kanona att simma i.
	‘The pools were great to swim in.’
	eyesoffinland.blogg.se/2008/may/minnen-fran-kreta.html

Other constructions which suggest adjectival status of kanon are ones in which kanon is modified by an adverb (or two adverbs, as in (102)), or in which kanon is coordinated with another adjective, as in (103):

(102)	Nu har jag kommit hem och denna dagen har varit helt jävla kanon !
	‘Now I have come home and this day has been totally bloody great!’
(103)	Min jul har vart kanon och mysig
	‘My Christmas has been awesome and comfy’

On the other hand, there is evidence that kanon is still interpreted as a noun by some speakers, since it may also occur in the definite form (104), and even in the genitive plural (105).²³ In neither case is there a difference in meaning.

(104)	Denna helgen ska bli kanonen !
	‘This weekend is going to be fantastic!’
(105)	Sen strör man på lite kanel & socker så blir det kanoners ! 25
	‘Then sprinkle with a bit of cinnamon and sugar and it will be fantastic!’

Summing up, predicative kanon has both adjectival and nominal properties. While this may be problematic for a grammatical theory which requires discrete categories, it is not for the constructional approach, in which categories are considered gradient (Traugott and Trousdale 2013: 74).

6.2 Semantic Properties

As we saw at the beginning of this section, the first adjective which kanon collocates with is full ‘drunk’, in which kanon is already used metaphorically. This may explain why kanon, unlike kei and hammer, is not used in simile constructions (adjectives meaning ‘drunk’ do not occur in the sample either). Kanonhård ‘very hard’ is attested only once, and in a figurative sense (två kanonhårda kamper ‘two very tough matches’). In other words, in Adjective/Adverb/Quantifier contexts both bound and free kanon have a purely intensifying function. For instance, adjectives like duktig ‘good at, clever’, snygg ‘cute’ or trött ‘tired’ clearly have no link to physical properties typically associated with cannons. A further striking example is (106), in which kanon is attached to an adverb with a negative prefix (o-troligt ‘in-credibly’).

(106)	Benpasset gick kanonotroligt bra
	‘The leg exercises went really incredibly well’

When kanon collocates with nouns or verbs, it almost always has ameliorative meaning (cf. examples (88) and (95) above), the only two exceptions being formations in which R1s are inherently negative, viz. kanonhuvudvärk ‘splitting headache’, and (possibly) kanonfylla ‘state of being stone drunk.’ The meta-comment in (107) furthermore shows that kanon is considered gradable.

(107)	Hur kanon är det på en skala?
	‘How awesome is this on a scale?’

Summing up this section, it is clear that kanon as a prefixoid has generally lost the association with the original noun ‘cannon’, and that semantic differences between bound and free forms are very small.

6.3 Collocational Properties and Productivity

Looking at [kanon-N] constructions, we note that many of them are temporal nouns, such as ‘day’, ‘evening’, or ‘weekend’, or nouns referring to events or accomplishments, such as ‘race’, ‘match’, or ‘job’. The association of loud noise or an explosion with a terrific time is not uncommon, compare English blast, which can be used in all three senses, or the Dutch and Swedish intensifying prefixoid knal(l) ‘bang’. In other collocations, however, the association with typical properties is less obvious, e.g. ‘blog’, or ‘picture’. From Tables 17 and 18 it also emerges that, at first glance, the R1 preferences of bound and free kanon are very similar, but free kanon is slightly more productive.

Table 17 Bound kanon (R1 = N) – collocates and productivity

Table 18 Free kanon (R1 = N) – collocates and productivity

As we saw in the previous section, kanon- has predominantly intensifying function in constructions with adjectives, adverbs and quantifiers. As far as bound kanon is concerned, the top four correspond neatly to the token frequencies of the adjectives in the SECOW14AX corpus as a whole,²⁵ which corroborates our earlier observation that kanon in Adjective/Adverb/Quantifier constructions is substantially bleached. Again, the collocational properties of bound and free kanon are very similar, and free kanon is more productive than its bound counterpart (Tables 19 and 20).

Table 19 Bound kanon (R1 = Adjective/Adverb/Quantifier) – collocates and productivity

Table 20 Free kanon (R1 = Adjective/Adverb/Quantifier) – collocates and productivity

When R1 is a verb, finally, productivity is low. Moreover, most verbs that collocate with (free) kanon are semantically close to predicative constructions, e.g. ‘to go’, ‘to feel’, or ‘to suit’. Only in a few cases, e.g. ‘sing’, or ‘sleep’, does kanon function as a manner adverb (Table 21).

Table 21 Free kanon (R1 = V) – collocates and productivity

7 Contrastive Statistical Analysis

In this chapter, we summarize the quantitative results from the three preceding sections in order to assess whether there exist statistical differences between the bound and free constructions of kei, hammer and kanon for all four properties mentioned in Sect. 2.2: construction type (7.1.), semantic bleaching (the proportion of simile and intensifying functions in [Prefixoid-ADJ] constructions) (7.2.), collocational properties (7.3.), and productivity (7.4.).

7.1 Construction Types

As far as construction types are concerned, we distinguish four categories that may occur as R1: AAQ (which comprises adjectives, adverbs and quantifiers), N(ouns), V(erbs), and predicative constructions. We merge adjectives, adverbs and quantifiers into a single category because the function of the prefixoid is similar when it modifies one of these. This furthermore enables us to calculate the χ² scores in R, because otherwise there would have been too many cells with a frequency < 5, in which case the χ² approximation may be incorrect.

Figure 1 shows the distribution of R1 construction types across the bound and free variants of the three prefixoids. It is clear that there is very little variation where kei is concerned, whereas variation is more substantial with the other two prefixoids.

Fig. 1

Construction types – Part of speech of the R1

In Table 22, we see that all results are significant (p < 0.05), but whereas hammer and kanon get the smallest possible p given by the standard χ² function in R, the p for kei is much larger. Moreover, the effect size²⁶ is large for hammer and kanon, but lower than 0.10 for kei.

Table 22 Results of the χ² test for construction types

7.2 Bleaching

For all three prefixoids, we measured the degree of semantic bleaching by comparing the proportion of similes to the proportion of intensifying constructions when R1 is an adjective, adverb or quantifier. Fig. 2 only represents these proportions of simile and intensifier meanings for bound and free kei and hammer; as mentioned in 6.2., Swedish kanon is not used in simile constructions in our sample²⁷ and is therefore not included in the figure. For both kei and hammer, we observe loss of the literal meaning in favour of the intensifying meaning, which we interpret as the result of bleaching.

Fig. 2

Bleaching

In order to compare degrees of bleaching when the prefixoid collocates with an adjective, adverb or quantifier we also performed χ² tests, the results of which are given in Table 23. With both kei and hammer differences between the bound and the free forms are significant (p < 0.05), but the effect size is small.

Table 23 Results of the χ² test for bleaching

7.3 Collocations

In order to establish to what degree the free and bound variants are used in different contexts, we performed a distinctive collexeme test, as described in Levshina (2015: 242ff.). This test was originally designed to compare the collocational preferences of two near-synonymous constructions in a corpus (Gries and Stefanowitsch 2004), and compares the observed frequency of a specific slot filler (R1) to the expected frequency of that R1. The purpose of this test is to compute if specific R1s are attracted to one of the two constructions.²⁸ We used the pv.Fisher.collostr()function of the Rling package (Levshina 2014), which computes the Fisher exact p-values for all R1s; these are subsequently log-transformed (using the negative base 10 logarithm). If the observed frequency is smaller than the expected frequency, the log-transformed score will remain negative. Conversely, if the observed frequency is larger than the expected frequency, the log-transformed score will become positive. The cut-off value was set at 1.3, which corresponds to a p-value of 0.05. In our case studies we use the free construction as a base. Hence, if the log-transformed p-value was >1.3, the R1 is distinctive for the free prefixoid, if it is < −1,3, it is distinctive for the bound prefixoid.

Table 24 gives the total number of types for kei, hammer and kanon in combination with an adjective, adverb or quantifier, as well as the distinctive R1s for the free and the bound constructions respectively. The distinctive R1s are given in Tables 25, 26 and 27. The differences in collocational preferences are most pronounced in Dutch kei, with 19 distinctive R1s for free kei and one (keihard) for bound kei; note also that the logp for keihard is very high, meaning strong attraction (see Table 25). Differences in German are much smaller, with five distinctive R1s for the free form and 1 for the bound form. In German, too, the simile hammerhart is distinctive for the bound form, but the attraction is smaller than in Dutch (see Table 26). For Swedish kanon, finally, there is only one distinctive R1 for the bound form (kanonfin ‘very fine’) and none at all for the free form (Table 27). If we divide the total number of distinctive collexemes by the number of types, we arrive at very low ratios (see Table 24). We are not aware of a threshold above which differences in collexeme distributions between two constructions can be said to be statistically significant, but since the number of distinctive collexemes relative to the number of types corresponds to a ratio ranging between 0 (no distinctive collexemes) and 1 (all collexemes are distinctive), we may assume the same ranges as Cramér’s V for effect size (see footnote 27). We may thus conclude that there is a small effect for kei, and none for hammer or kanon.

Table 24 Distinctive collexemes for R1 = AAQ

Table 25 Distinctive AAQ collexemes for free kei and bound kei- (shaded)

Table 26 Distinctive AAQ collexemes for free hammer and bound hammer- (shaded)

Table 27 Distinctive AAQ collexemes for bound kanon- (shaded)

We also looked at distinctive collexemes for hammer and kanon with R1 as a noun (kei hardly collocates with nouns at all), but we found only two distinctive nouns for free hammer and none at all for kanon, so we will not further discuss them here.

7.4 Productivity

Figures 3 and 4, finally, show differences in productivity between bound and free kei, hammer and kanon. Both with nouns and adjectives/adverbs/quantifiers, we see an overall increase in productivity, both type/token ratio and potential productivity. The differences are largest with hammer, followed by kei and kanon.

Fig. 3

Type/token ratio and Potential Productivity (R1 = AAQ)

Fig. 4

Type/token ratio and Potential Productivity (R1 = Noun)

8 Summary and Conclusions

A summary of the preceding section is given in Table 28. From this table, it is evident that only Dutch kei has undergone all four constructional changes that we examined in this paper. With hammer the differences in construction types are larger, but the number of distinctive collexemes is too low to conclude that free hammer is sufficiently distinct from bound hammer as far as collocational preferences are concerned. In the case of kanon, only two constructional changes have occurred: there is a large difference in construction types and an increase in productivity, but no bleaching, and no difference in collocation preferences.

Table 28 Constructional changes – summary

Let us return now to Hilpert’s rhetorical question: how many of these constructional changes can be said to suffice in order for the free form to qualify as an instance of constructionalization? As we have seen, Traugott and Trousdale’s (2013: 22) definition of constructionalization requires a change in both form and meaning, but these authors are not very explicit on which changes they consider formal, and which changes they consider semantic. Moreover, they do not tell us how to quantify changes in form or meaning. Nevertheless, if we consider the first constructional change in Table 28 as a formal change (after all, collocations with different parts of speech suggest changes in morphosyntactic properties), we note that all three prefixoids have undergone formal change (even in the absence of phonological change). Bleaching in [Prefixoid-ADJ] constructions, which can be considered a semantic change, is only attested for kei and hammer, but the effect size (Cramér’s V) is small with both. Kanon is exclusively used as an intensifier in our sample. Distinctive collexeme analysis, which we used to test whether there has been a change in collocational preferences, shows only a small effect for kei, and none for hammer and kanon. Finally, all three debonded prefixoids show an increase in productivity as compared to the bound forms. According to Traugott & Trousdale’s definition of constructionalization, then, only kei and hammer have developed a new node in the network, whereas kanon has only changed formally. This is somewhat problematic, however, because it implies that there is a single node for bound and free kanon, in spite of clear differences in construction types and productivity. It should also be noted that free kanon is not just an orthographic variant of the bound form, as we have shown for Swedish jätte ‘giant, very’ (Norde and Van Goethem 2014).²⁹ For this prefixoid, distributional differences between bound and free jätte were statistically insignificant. The proportion of bleaching was similar, and unlike free kanon, free jätte does not modify verbs nor does it occur in predicative position. On the other hand, we would have to assume two separate nodes for bound and free hammer, even though there are no significant differences in distribution.

Another complicating factor is that the free prefixoids may have developed out of more than one source construction – apart from debonding in [Prefixoid-N] and [Prefixoid-ADJ] constructions, clipping of specific [Prefixoid-ADJ] constructions and predicative use of bare nouns may have played a role in the rise of free forms as well. If kanon is ‘only’ a formal change, then which source node(s) has changed?

Summing up, it seems that Traugott & Trousdale’s framework with its focus on distinguishing constructionalization from constructional changes is difficult to apply to this particular case study. Rather than concern ourselves with the question of whether or not the emergence of free uses of prefixoids is constructionalization, we feel it is more insightful to look at observable changes at different levels, considering as many factors as possible that can be operationalized quantitatively, including collocational properties and productivity. Debonding of kei, hammer and kanon can therefore be considered an instance of ‘constructional change’ according to Hilpert’s approach, because he (2013: 7) explicitly states that “frequency changes (…) are no less indicative of constructional change than are developments in meaning or the phonological and morphosyntactic substance of a construction.”

To conclude, for all three case studies, we feel the quantitative differences discussed in Sect. 7 justify the conclusion that the bound and the free forms are different constructions (and hence distinct nodes in the constructicon). Although there is a tendency to write compound words as two words in Dutch, German and Swedish (probably under the influence of English), the free form is not merely an orthographic variant of the bound form. For in that case, we would not expect any distributional differences at all.