Dualism and superposition in the analysis of English synthetic compounds ending in -er

1 Introduction

There is a longstanding debate concerning two mutually exclusive hypotheses, i.e., whether synthetic compounds (henceforth, SCs), such as E. heartbreaker, have to be analyzed either as agentive or instrumental derivations from a phrase [[heart break][-er]] or as compounds of a modifying noun and an agent/instrument noun [[heart] [break+-er]] (see Section 2). Inspired by Libben’s (2017) article “The quantum metaphor and the organization of words in the mind”, we use wave-particle dualism (or duality) and superposition of both conflicting analyses in quantum physics as a model to analogically account for English SCs ending in -er, yet going in a different direction from Libben’s (2017) article.

Einstein (1905) had problems accounting for light as a continuous field of waves or as a stream of particles (later called ‘photons’ as quanta of light). This dualism is also called ‘complementarity’ of the concepts of wave versus particle, the two concepts being mutually exclusive (Zeilinger et al. 2005). Dualism as such occurs in two forms: (a) in general, just one of the concepts can be right, the other must be wrong, (b) photons and other quanta can appear as waves or particles, depending on the experiment chosen, which indicates complementarity. Whereas dualism Type (a) was assumed earlier on, Einstein and Infeld (1938, cf. also Zeilinger 2005) assumed complementarity (dualism Type (b)) when they stated:

There seems no likelihood for forming a consistent description of the phenomena of light by a choice of only one of the two languages. It seems as though we must use sometimes the one theory and sometimes the other, while at times we may use either. (Einstein and Infeld 1938: 278)

This change was due to Heisenberg’s (1949 [1930]) arguments for explanatory complementarity of the two models and for the possibility of a superposition of both (G. Überlagerung), which can make it difficult to decide between the two, even under the same conditions (Heisenberg’s uncertainty principle, cf. also Nairz et al. 2003), which Einstein did not want to accept (cf. Zeilinger 2005). Such uncertainty presents a problem to all those who, following Noam Chomsky, make epistemological assumptions like classical physics, which only accepted dualism Type (a) (cf. Matthews 1993).

As for the existing models of SCs, the usual position is exclusive dualism (Type (a)), but see below for complementarity (Type (b)) in Marchand (1969: 16). In this study, we assume complementarity in certain (sub)families of SCs, but superposition (with different weights of compounding and derivation) in most of them.

A partially comparable syntactic analysis exists for superposition of an argument (complement) structure with an adjunct structure in Dowty (2000, in contrast to Grimshaw’s 1990 assumption of ambiguity), but without any epistemological discussion and with an incorrect inclusion of word-formation. Indeed, the interaction between morphological derivation and lexical storage is not a case of superposition in the sense of quantum physics and as intended here. Analogically, Pesetsky (1996) assumes a dual system of simultaneous representations of two conflicting syntactic derivations (most explicitly on p. 258), without mentioning the epistemological problem of indeterminism. Another possible syntactic case is discussed by van Riemsdijk (2000).

Simultaneous synchronic superposition of two different constructions (in our case of derivation and compounding) is not the same as the coexistence of two different mutually exclusive homophonous constructions. Superposition holds for one and the same SC in contrast to coexistence of derivation for some specific SCs and of compounding for other SCs. However, in diachrony, the sources of superposition always seem to be the confluence of distinct constructions. Thus, language change consists in changing interactions between multiple source constructions into their superposition.

Therefore, our approach is in clear contrast to the strict Saussurean separation of synchrony and diachrony as debated in Joseph and Janda’s (2003) handbook, similar to many works on analogy (cf. Anttila 2003; Hüning 2009; Mattiello 2017; Mattiello and Dressler 2018). In fact, even the mere explanation of a new word as analogous presupposes the inclusion of micro-diachrony into synchronic argumentation. This reference to analogy is particularly relevant, because the formation of new SCs within the same semantic subfamily appears to involve analogy and not simply the application of a productive rule.

This study deals with the internal structure of SCs (also called ‘verbal’, ‘deverbal’ or ‘argumental compounds with deverbal heads’), i.e., nominal compounds such as heart breaker. It focuses on compound families, i.e., compounds whose second element contains breaker, holder, killer, maker, manager, producer, or provider.^[1] In the study, we propose a fine-grained semantic categorization of these families and their subfamilies. From a theoretical point of view, we claim that different subfamilies are amenable to different types of structural analysis. The subfamilies will be defined on the basis of the interpretation of the verb underlying the second element in the specific compound. The study contains a diachronically motivated analysis of SCs, allowing multiple sources for each particular compound. Some compound families will be analyzed as instances of derivation (e.g., [[path break]-er]) and others as instances of compounding (e.g., [tie [break-er]]). In some cases, however, we will argue for superposition, where both analyses are acceptable: e.g., painkiller being analyzable as both [[pain kill]-er] and [pain [kill-er]].

When we discuss whether SCs are derived via suffixation from a verb phrase or via the compounding of two nouns, this may sound like the traditional analysis of devising a morphological construction from a phrasal construction (do-gooder, no hoper, hardliner, all-nighter) or like generative derivations from (near)identical deep/base structures. Our discussion can instead be translated into onomasiology-based or construction morphology, which assumes that “phrasal constructions may be similar in function to morphological constructions” (Booij 2012: 344; cf. Masini 2019).

The aim of our study is to show that English SCs have no single origin, but, depending on their history and semantics, they may have originated either from a phrase or by compounding after derivation has occurred, or even by the superposition of these two sources. Indeed, some -er SCs, such as bondholder [1823] and loan-holder [1823], suggest an explanation as compounds which combine a deverbal agent noun -holder (indicating ‘possession’) with a noun modifier (bond, loan), probably after the model of an earlier attested SC shareholder [1800], by analogy. Others, instead, suggest a derivation from phrases: e.g., heartbreaker [1674] and icebreaker [1825] may have derived from the respective verb phrases break one’s heart [c1385] and break the ice [1553]. Moreover, in spite of their shared second component -breaker, the different semantics that they involve (cf. ‘[metaphorically] causing rupture’ vs. ‘interrupting, causing the end of [silence]’) suggest that they belong to different subfamilies of the same formal family X-breaker, although the sense of both is metaphorical. Still other -er SCs are historically preceded by both the -er derivation of the second component and an underlying phrase. For instance, law breaker [c1440], whose meaning is instead connected with ‘violation, transgression’, chronologically follows both the derivation breaker (of the law) [c1384] and the verb phrase break the law [1325]. Therefore, we may envisage a twofold origin for this latter. In this study, our purpose is to show that there is a third possible explanation for this type of SCs, i.e., the superposition of (a) derivation followed by composition and (b) derivation from a phrase.

For this purpose, we will investigate and describe in detail formal word-formation families and semantic subfamilies of English SC nouns of the type heartbreaker, shareholder, painkiller, and watchmaker in the Oxford English Dictionary ([OED] 1989–2021) and in electronic corpora of English.

In particular, this study combines dictionary-based and corpus-based analyses in order to identify:

1. Formal synthetic-compound families and their semantic subfamilies – i.e., subfamilies of compound families which are semantically related – and

2. Agent/instrument compound nouns that serve as schema – i.e., concrete model (vs. abstract rules) – for the development of other analogy-based SCs belonging to the same subfamily.

Analogy as the basis for new word creations involves both synchronic and diachronic development, in that analogical words originate from a precise or schema model that chronologically precedes the target.

The study deals with both formal and semantic properties of -er SCs, covering formal issues related to orthography, to the boundary between compound and syntactic phrase, as well as issues related to compositionality and direct/indirect object referencing first elements.

From the theoretical perspective, this study provides a novel view of SCs as both derivations from phrases and compounds at the same time, which would not be accepted in any rule-based or generative-type approach, where they would be considered as either derivation or composition.

2 Previous research and theoretical background

SCs occurring in Germanic languages, such as German Tennisspieler, Geschirrspüler and their English equivalents tennis player, dish washer, are considered a controversial topic of morphological analysis since their first mention in the seminal works of early grammarians (Behaghel 1917; Wilmanns 1896), considering them cases of “Zusammenbildungen” (Leser 1990, ‘co-formations, lit. together-formations’), a term which is neutral with respect to the two opposing analyses. The term ‘synthetic compound’ can be traced back to a monograph by Leopold von Schroeder, who proposed the designation “synthetische Composita” for formations such as Machthaber ‘power holder’ (von Schroeder 1874: 206). This term suggests that they are subtypes of compounds.

Then, Leonard Bloomfield introduced the term ‘synthetic compound’ into the English literature on word-formation. Bloomfield (1933: 231–232) labelled denominal forms like blue-eyed and snub-nosed ‘synthetic compounds’ to be analyzed as [[blue eye]+-ed] and [[snub nose]+-ed], rather than as blue+eyed and snub+nosed, because *eyed and *nosed do not exist independently. By contrast, deverbal forms like nominal meat eater and adjectival meateating were considered ‘semi-synthetic compounds’, because the words eater and eating exist alongside the compounds.

Unlike Bloomfield, Marchand (1969: 15) applied the term ‘synthetic (or verbal nexus) compounds’ to “derivatives from verbs which form a direct syntagma with the determinant”, such as nominal watchmaker and adjectival heartbreaking. According to Marchand (1969: 79), morphologically speaking, verbal nexus compounds are suffixal extensions of combinations such as chimney sweep, with a missing agentive suffix -er (or a virtually present zero suffix according to other theoretical approaches, Matthews 1991: 123). He distinguished SCs both from “primary” compounds consisting of purely nominal constituents (e.g., steamboat) and from “bahuvrihi adjectives” of the type pale-faced or five-fingered, which are not analyzable into the immediate constituents pale+*faced or five+*fingered (Marchand 1969: 17–19). Remarkably, among nominal SCs, Marchand (1969: 16) identified two different types illustrated by deer hunter and watchmaker. According to him, the first type is analyzable as two independent lexical units (i.e., deer+hunter) in that hunter exists and is used independently, whereas the second type is not, because *maker is not a lexical word, but a “functional derivative” which renders the syntactic relation ‘he makes’. In his view, the type deer hunter and similar (e.g., ballet dancer, cigar smoker, language teacher, etc.) consists of two independent lexical entities, while the type watchmaker consists of an independent first element watch plus a functional derivative maker. This may be regarded as symptomatic of the ambiguous nature of SCs. Indeed, while from a purely formal viewpoint the two types seem to be equivalent, from a finer-grained analysis of their underlying structure they may have different diachronic origins and different synchronic sources, a case of mutually exclusive dualism in homophonous constructions, which is clearly different from the assumption of superposition, but represents complementarity (dualism Type (b)). This is comparable to Booij’s (2012: 345) assumption of different subschemas which characterize different subsets of compounds (Masini 2019 distinguishes between different subregularities).

Agent nouns and SCs referring to personal nouns have been the object of analysis in various languages, such as Dutch (Booij 1986), German (Gaeta 2010), Spanish (Rainer 2005), French (Rosenberg 2007), and others. Handbooks of English morphology are not lacking in descriptions of both personal derived words and verbal nexus combinations (Adams 2001; Bauer and Huddleston 2002; Jespersen 1942; Lieber 2004; Marchand 1969).

More recently, Bauer et al. (2013: 467; cf. Neef 2015; Lieber 2016b, who has also refuted several claims on limitations of base verbs) have claimed that agent/instrument nouns ending in -er often constitute the second elements of argumental compounds, such as object referencing air heater, alcohol abuser or prepositional object referencing adoption worker, asthma sufferer.^[2] Non-argumental compounds can be SCs as well, as illustrated by aerial observer, bakery worker, afternoon trainer, etc. (Bauer et al. 2013: 482), which are formally comparable to the air heater type. By contrast, Bauer et al. (2013: 511) have listed babysitter, basketballer, footballer, etc. among SCs that are obtained after -er suffixation of a compound base (i.e., babysit, basketball, football, etc.). Unlike the former group of SCs combining an agent/instrument derivative (e.g., heater) with a simplex object (air in this case), the latter group derives from a compound base, with the exception of the first of this list, which is a case of back-formation (i.e., babysit ← babysitter), rather than derivation following composition.

As for the referents realized by the derivational suffix -er, Kolbusz-Buda (2014: 105–106) has claimed that they are confined to the roles of Agent (bricklayer, bodybuilder) and Instrument (screwdriver, timekeeper), and only occasionally of Manner (sharp-shooter, bestseller) and Location (grass-hopper, rope-dancer, street-walker, church-goer). In our study, we only focus on agent/instrument referents, disregarding the others, which do not form families of SCs, only local or surface analogies (e.g., well-/best-wisher, well-/evil-doer).

Nowadays, there is still little consensus as to whether SCs should be treated as deriving from a verb phrase by suffixation and thus be left-branching (e.g., Kolbusz-Buda 2014: 91–95, 107; cf. in terms of construction morphology Booij 2010: 48–50), or rather as combining two independent words by compounding, the second being an agent/instrument noun (e.g., Lieber 2004: 47–48, 54–60). An issue of the Journal of Word Formation (2017) was entirely devoted to this topic, from different angles: i.e., from a lexicalist perspective (Olsen 2017), with categorizing aims (Iordăchioaia et al. 2017), and from a historical perspective (Werner 2017; also in Werner et al. 2020). In this issue, Iordăchioaia et al. (2017: 48) claim that there are two possible analyses of SCs, depending on whether they are viewed as first involving compounding and then derivation, or first derivation and then compounding. According to what they call “the Synthetic Compound Approach” (cf. also Borer 2013), SCs such as heart-breaker have a different make-up from that of root compounds: i.e., [[[heart]_N+[break]_V]+-er]_N. By contrast, according to “the Root Compound Approach”, synthetic and root compounds are obtained through similar morphosyntactic mechanisms, possibly, with a difference in interpretation that can be traced back to the base verb for SCs: i.e., [[heart]_N+[[break]_V+-er]_N]_N. Therefore, in the Root Compound Approach, heart combines with breaker, rather than with break and then with -er.

According to Gaeta and Zeldes (2017: 2), instead, previous approaches to SCs can be summarized as follows:

1. Incorporation, i.e., morphological derivation via suffixation of a verb, as in break heart → to heart-break → heart-break-er (cf. back-formation to baby-sit ← baby-sitter);

2. Morphological derivation and subsequent compounding, as in break → breaker; heart+breaker → heart-breaker (cf. Root Compound Approach above);

3. Morphological derivation via suffixation of a word group, as in break+heart+er → heartbreaker (cf. Synthetic Compound Approach above), which presupposes a false intermediate step of a potential, but non-existing verb to heart+break.

Since the first and the third solution only differ in the intermediate verbal step being actual or only potential, we will merge them into the derivational analysis (Section 7.5) when discussing differences between dualism and superposition.

Derivations applying not to a lexeme, but to a word group, stand out as being unusual or dispreferred, e.g., within Natural Morphology (cf. Dressler 1988) (cf. the derivation from univerbations such as ne’er-do-well-ish, stick-in-the-mud-ish). In this study, by contrast, it will be shown that derivation from a word group is not a peripheral but a major productive process obtaining large families of SCs to which the notion of headedness applies (cf. lawbreakers vs. *laws-breaker, co-occurring with a verb phrase break laws/a law/the law). The notions of headedness and productivity stress that SCs remain in the domain of morphology, not of syntax.

The existence of -breaker and similar complex elements that act as constituents in series could also support the second approach. In the literature, this position is also held by Reis (1983) and Olsen (2017). Adopting an analogical approach, Olsen (2017: 30–31) claims that some SCs belong to series formed by repeating a constituent: e.g., -maker (e.g., peacemaker, troublemaker, shoemaker), -killer (weed killer, man-killer, lady-killer), or -breaker (law-breaker, promise-breaker, heartbreaker). The existence of formal families (Mulder et al. 2014) and of semantic subfamilies of SCs sharing a constituent suggests that they can act as schema model for novel formations (cf. Mattiello and Dressler 2018 for analogical compounds; Hüning 2009).

From the theoretical viewpoint, these cases can be accommodated within extended analogy or analogy via schema (Mattiello 2017). A schema is a concept drawn from Köpcke (1993), who developed it for inflectional morphology, but which has been lately extended to the word-formation module (Mattiello 2017; Mattiello and Dressler 2018). A schema can refer either to a series of words sharing the same formation, or to a word family sharing some of the bases, as with SCs sharing the constituents -maker, -killer, and -breaker. The families X-maker, X-killer, and X-breaker (along with X-holder) will be the non-Latinate case studies investigated here from a diachronic and corpus-based perspective.

A psycholinguistic approach to SCs is adopted by Gary Libben (2017, 2019. Libben (2019) considers words as building blocks of language and words as being in lexical superstates. In his view, a word such as heartbreaker may be structured in the mind of an individual in very different ways, depending on whether or not it is decomposed (cf. undecomposed heartbreaker vs. fully decomposed [heart] [break] [-er]), or whether the first or the last two elements are grouped together (cf. [heartbreak] [-er] vs. [heart] [breaker]). Accordingly, SCs’ structure depends on the grouping and number of constituents (two vs. three) we identify in the SC. This may be viewed as superposition of lexical and morphological processing.

Our morphological contribution only investigates and contrasts the partial units, i.e., [heartbreak] [-er] versus [heart] [breaker], in order to discuss them as a parallel to the wave-particle dualism and superposition and to eventually compare the results of our analyses with the findings on the acquisition of SCs in seven languages (Dressler et al. 2019). It is worth noticing that the constituent -breaker of heartbreaker is not equivalent to the autonomous word breaker (also in the compositional analysis of [[heart] [breaker]]), since the two are only related but separate units (Libben 2014), a claim which holds both for the compositional and for the derivational analysis.

In addition, Libben (2017) claims that, for ease of processing, all token frequencies may have an impact, i.e., the frequencies of the SC heart-breaker, of the freestanding words heart and break, of the suffix -er, of the underlying phrase break heart, and of the complex second member breaker. In this study, we will adopt Libben’s (2017: 49) notions of “wave/particle duality” and “superposition” to investigate the relevance of derived agent/instrument nouns or phrases, or both, to the creation of SCs and of formal SC families and subfamilies. Of the three approaches discussed by Gaeta and Zeldes (2017: 2, see Sections 1–3), we will subsume 1 (i.e., derivation via compound verb suffixation) and 3 (i.e., derivation via suffixation of a word group) under the heading ‘derivation from a phrase’, and consider them morphological variants.

Other scholars have shown the importance of combining diachronic study with corpus linguistics (Kaunisto 2007; Renouf 2019). A corpus-based study can, for instance, help identify hapax legomena (i.e., words occurring once only), whose relevance to neology, word-formation, and productivity has been observed and demonstrated by Renouf (2019: 42) and earlier scholars.

By using corpus-based analysis combined with a diachronic investigation, we intend to demonstrate that all the above-mentioned approaches are relevant and applicable, depending on the SC, the existence of an underlying verb phrase, and its semantics. Our hypothesis is that the identification of families and subfamilies of SCs with an agent/instrument as their second constituent can help categorize SCs, and interpret or predict new ones (also including hapaxes) on account of their productivity. The importance of semantic subfamilies is a logical consequence of the relevance of the semantic relation that links the constituents of compounds and similar morphological constructions, as ascertained in psycholinguistic experiments (Gagné and Spalding 2014). Besides being a matter of frequency, semantic coherence, availability and profitability, in our approach we define productivity as the ability to extend families and subfamilies of compounds and, more significantly, to create new families and subfamilies (more in the general discussion, Section 7).

3 Data selection and methodology

In this article, we focus on families of SCs whose second and third morpheme form a deverbal -er agent/instrument noun. The literature offers several cases of this type (e.g., dish washer, tennis player, peacemaker, ladykiller, law breaker, ballet dancer, cigar smoker, meat eater, air heater, alcohol abuser, watchmaker, etc., all attested in the OED with this spelling) (cf. Section 2). All were taken into account when considering possible families of SCs to analyze. Our analysis is meant to compare SCs with the respective verb or noun phrases (VPs/NPs). In particular, synchronically, we aim at investigating whether or not a VP and/or NP co-occurs with a SC, and diachronically, we aim at finding which of the two phrases precedes the SC. In our analysis, we do not adhere to mutually exclusive dualism (= Type (a)) as in most literature on SCs, but sometimes to complementarity (= dualism Type (b)), as Marchand (1969) did for all SCs, and most of the times to superposition, which has never been assumed for word-formation and only rarely for syntax (yet without reflecting the epistemological problem).

The analysis carried out in the article is both dictionary based and corpus based. First, for the selection of some SC constituents that could be possible candidates for the corpus-based analysis of compound families, a lexicographic investigation in the online edition of the OED was conducted. This investigation allowed us to exclude from the analysis constituents such as eater, lover, or smoker, which, despite being constituents of existing SCs (e.g., bread-eater, music-lover, cigarette-smoker), are unambiguous in meaning (‘one who eats’, ‘one who loves’, ‘one who smokes’) and do not seem to be very productive in the formation of series of SCs. Still others were excluded because the OED did not offer any use in composites, i.e., preceded by modifying word: for instance, teacher, besides being unambiguous (‘one who teaches’), is not described in the OED as second constituent in formatives, although some uses are attested under the first constituent’s description (e.g., language teacher, s.v. language).

The next step of lexicographic investigation was meant to identify ideal candidates for the development of families and subfamilies of SCs. Some potential candidates (i.e., breaker, holder, killer, and maker) were selected and checked in the OED, because they appeared to be (a) possible second compound components and (b) sufficiently ambiguous in meaning to develop semantic subfamilies within compound families. For each of them, the OED’s meanings and examples of SCs were taken into consideration.^[3][4] Table 1 reports the results of our lexicographic search. This table is the basis for the analysis which will be conducted in Section 4.

Table 1 shows that all the constituent agent/instrument nouns checked are promising candidates to be explored from the morphosemantic viewpoint. All of them are used to convey different meanings (also concrete vs. metaphorical for breaker and killer), depending on their context of occurrence. Needless to say, only the meanings that were connected with the formation of SCs were included in the table. For instance, the meaning of breaker ‘one who combs wool’, attested in 1514 in the OED but now obsolete, seems not to have formed any corresponding SC. This is a further confirmation that autonomous breaker is not equivalent to -breaker as final constituent of SCs.

The dates in square brackets indicate the earliest attestation of the agent and instrument nouns in the OED of the meanings (second column) or of the SCs (third column). This preliminary investigation demonstrates that, in contrast with Marchand’s (1969: 16) claim, maker in watchmaker is an independent lexical unit currently used as synonymous with ‘manufacturer’. The OED investigation also suggests that different objects preceding the elements taken into consideration can activate different meanings or shades of meanings and, therefore, a variety of subfamilies of the SCs to be investigated.

The earliest attestations of all the three noun elements breaker, holder, killer as autonomous words precede the earliest attestations of the SCs containing these elements: the interval is between 25 years and more than two centuries. Only maker is attested 40 years later than bookmaker. In addition, if we divide these four relations into the nine meaning relations of column 2, then the meaning of the simplex nouns is attested six times earlier than the corresponding compound constituents and only three times later. Only considerable time intervals between the SCs and their derivational or compositional sources are meaningful. From this observation, we may infer that the greater attestation intervals may represent real time differences in usage, indicating diachronic precedence of the autonomous nouns over the semantically corresponding noun constituents, whereas the small attestation intervals between earlier SC constituents and later semantically corresponding autonomous nouns may be due to chance in attestation.

However, the corresponding verb phrases are usually attested earlier than the SCs (e.g., to break prison [c1300] vs. prison breaker [1704]), often with objects that are not exactly identical but only similar. Clearly, it is not surprising that verbs are attested before both agent/instrument nouns and SCs derived from them. Thus, our attention should be focused on the meaning differences between the SCs, their corresponding verb phrases, and the simplex agent/instrument nouns.

These preliminary results encouraged us to proceed with the corpus-based quantitative and qualitative analysis of the four compound constituent elements. For the quantitative analysis, an automatic search in the Corpus of Contemporary American English ([COCA] 1990–2017, currently 560 million words) was conducted. We checked token and normalized (per million words, henceforth, pmw) frequencies of the SCs belonging to the families X-breaker, X-holder, X-killer, and X-maker.

Some instances were excluded from the analysis because they were irrelevant to the above-mentioned formal SC families. They mainly consisted of brand names (e.g., Rainbreaker ‘name of a jacket designed to keep people dry’, DoubleKiller) and personal or place names (e.g., Burkholder, Schumaker). Other irrelevant examples were beholder or upholder, derived from the verbs behold and uphold, nouns whose first element was a neoclassical combining form (e.g., psycho-killer), and coordinative compounds (e.g., hunter-killer).

A qualitative study of the data was also conducted. In particular, contextualized examples from COCA were necessary, first, to disambiguate meanings and the semantic relations between the compound constituents and, second, to categorize SCs into subfamilies. For each formal family, subfamilies were classified and arranged chronologically, according to the earliest attestation of the meaning of the corresponding verb.

A diachronic investigation in the OED and in the Middle English Dictionary (1891) was then conducted for each meaning connected to SC subfamilies, and the corresponding autonomous simplex nouns/phrases displaying the same meanings. In addition, a micro-diachronic investigation in the Corpus of Historical American English ([COHA] 1810–2009, 400 million words) and in Google Ngram Viewer was conducted to verify the earliest attestation of the independent simplex nouns,^[5] some of the SCs, and related phrases in historical corpora of English. Finally, a search of the objects collocating with the verbs break, hold, kill, and make in the British National Corpus ([BNC] 1980–1990, 100 million words), using the tool Word Sketch of the corpus manager Sketch Engine, integrated previous results on the verb phrases related to SCs.

Hitherto, research on compound families has only dealt with formal families (see Section 2), which implies that all SCs ending in -breaker form just one big family. By contrast, following the results on German diminutives in Ransmayer et al. (2016) (cf. Schwaiger et al. 2017; Dressler et al. 2020; in a homogeneous set cf. Mattiello and Dressler 2018), we claim that formal word-formation families are a heterogeneous set. Therefore, for their thorough analysis, we have to focus on the much more homogeneous sets of semantic subfamilies, i.e., the subsets of SCs which share the meaning of the basic verb elements, i.e., of -break, -hold, -kill, -make respectively, including their basic selective restrictions in respect to human versus non-human patient arguments and to concrete versus metaphorical meaning relations. Usually, all members of a semantic subfamily are either agents or instruments, but this is not the case with subfamily 1 of X-killer and X-maker, nor with subfamily 4 of X-maker (Sections 4.3–4.4). This similarity is analogical to similarity of simplex agent/instrument nouns in -er in English (and in many other languages), whereas in syntax agent and instrument arguments behave very differently. This constitutes an argument for the compositional analysis of the SCs of these subfamilies.

As for metaphorical subfamilies, we claim that they may have two origins: 1) metaphorical meanings of the base verb or of the simplex agent/instrument noun were used first for a single case and then for a series of SCs, or 2) there was an isolated lexical metaphor, which was then generalized to a series of SCs. Lagerqvist (2016) has described how the introduction of metaphors can lead to the genesis of new morphologically complex words and word forms. His description may be thus extended to new metaphorical series of words and thus lead to new metaphorical subfamilies.

Another tricky issue requiring clarification concerns the size of semantic subfamilies in terms of constitutive members. Following the pioneering use of the term ‘family’ in Ludwig Wittgenstein’s (1953) Familienähnlichkeiten ‘family resemblances’, we assume that a subfamily must have the size of a core family, i.e., an absolute minimum of at least four members. In doubtful cases, “occasionalisms” (Chanpira 1966) – i.e., novel words that are attested only once or twice and have not been accepted as neologisms – can contribute as members of a subfamily. Indeed, the coinage of occasionalisms within a subfamily is symptomatic of its productivity, more than the attestation of hapax legomena (Gaeta and Ricca 2006; Hay and Baayen 2002).

As for the number of subfamilies in each formal family, because of the complex framework emerging from the lexicographic results reported in Table 1, we can predict a higher number of subfamilies for X-maker and X-breaker than for X-holder and X-killer. Moreover, some of the meanings of breaker, holder, and maker also seem to develop from phrasal verbs (e.g., break in, break up) or from phrases (break into a house, hold shares, make books), which contribute to increase the complexity of the X-breaker/-holder/-maker families and the intricacy of senses involved. At the same time, this suggests different origins for SCs belonging to the same family, but to different subfamilies, as an evidence of superposition. Therefore, in the analysis, we will have to investigate whether a semantic subfamily X-V+er (or only specific SCs) appears to be derived from a verb phrase or from an agent/instrument noun, or whether there is a superposition of both, i.e., simultaneous superposition of derivation from a phrase and of compounding with a deverbal agent/instrument noun.

For this purpose, in the next section, the corpus-based selection of SCs will be combined with:

1. A manual diachronic search in the OED and MED of the noun/verb phrases connected with the agent/instrument nouns (breaker/holder/killer/maker + (Preposition) + N) and the verbs (break, hold, kill, make + Noun X or Particle/Preposition Noun X);

2. An automatic micro-diachronic search in the BNC (1980–1990) of the collocations of both the V-er agent/instrument nouns and the respective verbs. Although BNC collocations do not show the diachronic development of SCs, they attest the use of some verb/noun phrases which may be connected with the consolidation and expansion of SC families and subfamilies.

3. The results of an additional search in Google Ngram Viewer ([Google Ngram Viewer] 1700–1800) of the agent nouns and of some of the SCs will also be reported in Sections 4.1–4.4. A bigger interval did not provide visibly noticeable results in Ngram Viewer diagrams.

Finally, a questionnaire was submitted to native speakers of English to investigate their interpretation of SCs as either verb phrases (‘N who/that X-s’) or agent/instrument nouns (‘X-er of N’), or both. For the questionnaire, we selected SCs which represented all seven formal families and most of the subfamilies identified for each of them. Another criterion for selection was the type of interpretation that we expected for these compounds, formed either after a verb phrase or after derivation followed by compounding, or even after both, as superposition implies. Metaphorical meaning was also represented by some of the SCs discussed.

4 Results of the corpus-based analysis

In this section, we will report on the results obtained from the automatic search of the strings -breaker, -holder, -killer, and -maker in COCA, with a preceding diachronic search in COHA. The results of the quantitative and qualitative analyses will then be integrated with cross-checks in the OED or in more recent web dictionaries, such as the Urban Dictionary (UD) or other web sources. The role of lexicographic checks will be central in validating the hypothesis that formal families, or rather semantic subfamilies of SCs are constantly expanding with new words, which may be well-established neologisms attested in dictionaries, or mere occasionalisms coined for specific occasions and unlikely to become a permanent part of English vocabulary (see Chanpira 1966; Christofidou 1994; Dressler and Tumfart 2017).

5 A look at Latinate base verbs and their corresponding SCs

To the best of our knowledge, it has not been observed so far that SCs are hardly formed from Latinate verb bases. The relatively few cases (in comparison with non-Latinate verb bases) that superficially look like SCs, are clearly N–N compounds, such as alcohol abuser or drug abuser, or Adj–N compounds, such as wide receiver or administrative receiver, whereas commonplacer is derived from the converted noun commonplace and neither from the verb to place nor from the noun placer. Many more are coordinate compounds, obtained from singer / pianist / conductor / trumpeter / lyricist / producer / writer, etc. and -composer.

Latinate SCs of the type oil producer or office manager have neither been borrowed nor extracted from borrowed material but have been newly developed within English by analogy with already existing non-Latinate SCs. Hence, we assume that they were formed differently from non-Latinate SCs, in the composition of which both derivation and compounding played a key role. We rather believe that Latinate SCs have an ambiguous nature, which can be resolved by assuming dualism and superposition of suffixing and compounding.

The most promising candidates for families are X-producer, X-manager, and X-provider.

Table 10 only reports part of the results obtained from the corpus-based search of the X-producer formal family. Coordinative compounds of the type writer-producer (60 occ.), actor-producer (15 occ.), or consumer-producer (4 occ.) were excluded from the table because they were irrelevant to a study of subfamilies of SCs. Since most of the remaining results consist of occasionalisms, meanings were not available, neither in the OED nor in other online dictionaries. Only contexts from COCA have been reported in the table.

Corpora collocations of the noun and verb phrases corresponding to each SC show that, for the X-producer family, numbers 1, 2, 5, 6, 7, 9, and 11 support a derivational analysis, 8 and 10 are mainly in favor of a derivational analysis, while 3 supports superposition.^[8] None of the results is in favor of a compositional analysis and, for number 4 (segment-producer), neither NPs nor VPs have been found.

Table 11 includes several names of computer programs, software, or other technological devices, such as AppManager or PageManager. This is confirmed by spelling (with camel case)^[9] and by the fact that the language used in the COCA contexts is specialized.

Corpora collocations of the noun and verb phrases corresponding to each SC show that, for the X-manager family, numbers 6, 9, 10, 12, 14, 15, and 16 (mainly) support a derivational analysis, 17 supports a compositional analysis and 11 is mainly in favor of a compositional analysis, while 2, 6, 7, 8, and 11 support superposition. For 1, 3, 4, 5, and 13 no underlying phrases have been found.

Unlike the previous Latinate families, for the X-provider family corpora (Table 12) collocations of the noun and verb phrases show that, numbers 3, 5, 6, 8, and 9 are clearly pro-derivation. All the others mainly support a derivational analysis, but do not exclude a compositional analysis, thus suggesting that superposition is again the case.

In contrast to SCs derived from non-Latinate verb bases, these formal SC families not only are very limited in number (and apparently hard to expand), but also cannot be split into semantic subfamilies and their meanings appear restricted to the specialized field of economics/business. Moreover, the chronological distance between the first attestation of the base verbs and the corresponding SCs is much more manifest than in the case of non-Latinate verbs. This is especially true for X-provider SCs, which seem to only appear in the twentieth century (vs. the fifteenth century of the base verb provide).

Therefore, SCs should be added to the list of morphological devices which uncommonly occur with Latinate vocabulary, similar to, e.g., the suffix -ness, whose attachment to Latinate adjectives is avoided in favor of the Latinate suffix -ity, as in *electric+ness versus electric+ity and rare opaque+ness versus more frequent opac+ity. This infrequency of Latinate bases is neither observable in N–N compounds (so-called root compounds) nor in simplex agent/instrument nouns in -er. Thus, this is evidence of the autonomy of SCs from other word-formation means.

In addition, the extension of the formation of SCs from non-Latinate to Latinate verb bases is a sign of the productivity of SC formation (more in Section 7.2), and of the increasing productivity of SCs in time.

6 Questionnaire

A questionnaire was additionally elaborated in order to test the rating of some SCs by native speakers of English. Lieber (2004: 60) had already observed that native speakers might differ or even waver in interpreting the same English SC as either a compound or a derivation. Since our pilot questionnaire had pointed out difficulty in distinguishing between different definitions of SCs, the 52 participants in the questionnaire were divided into two groups, both consisting of 26 informants. All informants came from English-speaking countries (Canada, USA, UK, Ireland, Jamaica, Virgin Islands).^[10] Each group was submitted a series of SCs along with a definition which presupposed either a derivational or a compositional analysis. The stimuli were the same for the two groups, but the definitions differed. The ratings ranged from 1 (= true) to 4 (= false), with intermediate 2 (= more true than false) and 3 (= more false than true). Some distractors which are not SCs were inserted in the questionnaire. This type of experimental approach, though not having statistical significance, was previously adopted by Lehrer (1996), e.g., for the quantitative analysis of blends via questionnaires on a smaller number of stimuli.

Figure 1 reports the 24 stimuli presented to the informants in our questionnaire, distractors excluded.

Figure 1:

Means of the ratings for ‘N who/that X-s’ versus ‘X-er of N’ interpretations.

The bar chart in the figure shows the means of the ratings for the two interpretations – i.e., ‘N who/that X-s’ versus ‘X-er of N’ – respectively supporting a derivational versus a compositional analysis.

The results could have been influenced by agent versus instrument interpretations (cf. Rappaport-Hovav and Levin 1992; Roeper 1987), in that instruments do not generally favor nor allow the ‘X-er of N’ interpretation. For this reason, it was at times necessary to add some details or a more complete explanation to the SCs’ definitions. For instance, a cupholder can be interpreted both as ‘a holder of cups’ (instrument) and as ‘someone who holds the cup’ (agent), thus, the two groups were presented more complete definitions: i.e., ‘A CUPHOLDER is a holder of cups, a device for keeping a drink upright’ / ‘A CUPHOLDER is someone who holds the cup, a team that won cup (prize) previous year’, to help the informants in their ratings and us to elicit their responses.

We considered the results to be pro-superposition if the two groups respectively agreed with either the derivational or the compounding analysis which were shown to them in balanced way. We considered the results to be pro-derivation if the interpretation ‘N who/that X-s’ prevailed, and, vice versa, we considered them to be pro-compounding if ‘X-er of N’ was the predominant option. Even if there was much variation among the participants’ ratings, we only found three cases of clear contradiction with the results of our corpus-based/diachronic analysis of Sections 4–5: i.e., for ladykiller, heartbreaker, and officeholder our analysis supported derivation, while the questionnaires supported compounding.

Of the other stimuli results, 6 clearly agreed with our analysis: i.e., moviemaker, peace-breaker, office-manager, painkiller (pro-superposition), peacemaker, icebreaker¹ (pro-derivation). Moreover, 2 stimuli ratings which were pro-superposition for the participants (shoemaker, penholder) corresponded to no clear results in our analysis, because we had found no arguments for either derivation or compounding, which suggested a kind of superposition.

In all the other cases, there was neither confirmation nor contradiction, in that either our analysis or the questionnaire’s results suggested superposition. Our analysis was pro-superposition (vs. pro-compositional ratings) for landholder, time-killer, lawbreaker, and (vs. pro-derivational ratings) for lawmaker, icebreaker², cupholder¹. By contrast, the questionnaire’s results were pro-superposition (vs. our pro-derivational analysis) for matchmaker, service-provider, kingmaker, record-breaker, wife-killer, housebreaker, oil-producer, and (vs. our pro-compositional analysis) for giant-killer.

This lack of correspondence between our diachronic analysis and the stimuli ratings by the participants in the questionnaire could be motivated by their synchronic reanalysis of the SCs under examination.

7 General discussion

8 Conclusions

This study has demonstrated the importance of formal SC families and their semantic subfamilies in the formation and interpretation of English SCs ending in -er. In addition, it has shown how some of these SCs can serve as a schema model for the development of other analogy-based SCs belonging to the same (sub)family. A corpus-based analysis was combined with qualitative synchronic and diachronic investigations in order to provide a fine-grained semantic categorization of the subfamilies of four formal families of English non-Latinate SCs. The results of the investigation of X-breaker, X-holder, X-killer, and X-maker in COCA and BNC, including occasionalisms, have demonstrated that these are rapidly expanding formal families. This finding has led us to conclude that a new property should be added to the definition of productivity of word-formation rules, namely, the emergence and development of new formal families and semantic subfamilies.

The results from the diachronic analysis of SCs and their semantic subfamilies in the OED and MED were compared both with the simplex elements breaker, holder, killer, and maker (of X), and with the corresponding verb phrases break/hold/kill/make X. The comparison has provided evidence of the emergence of SCs much later than their corresponding noun/verb phrases, with the attestation of verb phrases generally preceding both the simplex elements and the SCs.

It is evident from our analysis that a homogeneous approach to SCs is not sufficient to cover the varied patterns and meanings identified, neither in synchrony nor in diachrony, and that different types of SCs reflect different constructional types, supporting either a derivational analysis or a compositional analysis. This basic ambiguity can be resolved by assuming dualism and superposition of suffixing and composition, comparable to Einstein’s assumption of dualism and superposition of waves and particles in light.

The results from a questionnaire (cf. Section 6) submitted to native speakers of several varieties of English have confirmed our hypothesis that dualism and superposition can account for the SCs analyzed. These results are largely compatible with the results of our analysis, but show a greater propensity for superposition or slight superposition, which may reflect a general diachronic development of English SCs. In this way, we have also been innovative in our attempt to weight compounding versus derivation (cf. either one or the other analysis vs. a more probable one than the other analysis).

9 Outlook

In order to verify whether our type of analysis is also fruitful elsewhere, it should be expanded to SCs ending in other suffixes (as discussed in Lieber 2016b; Werner 2020; Werner et al. 2020) and to other languages, both in synchrony and diachrony. Our generalizations are based on four formal families of English non-Latinate SCs and three Latinate ones, but the analysis of other families of -er SCs could further corroborate our hypotheses. In particular, the differences between Latinate and non-Latinate SCs will be further elaborated in our future research.

Another area for future research involves psycholinguistic studies, beyond mere questionnaire rating. So far, we have combined different types of evidence, such as historical data and synchronic corpus data with data elicited from questionnaires. But the questionnaire study still needs to be analyzed using appropriate quantitative methods and some more statistical evidence is needed to support our conclusions. Therefore, psycholinguistic experiments will be our next step in the analysis of SCs. In this area, psycholinguistic offline and online experiments for testing our hypotheses and classifications will be elaborated with the expertise of Gary Libben. The need for such experiments is confirmed, in some cases, by the lack of correspondence between our diachronic analysis and the synchronic reinterpretation by native speakers. Our additional new criterion of morphological productivity also needs to be applied systematically to other areas of word-formation.

A theoretical issue to explore, which for reasons of space we have not dealt with in this article, would be to decide whether the basic criteria for subdividing formal families into semantic subfamilies are truly semantic in a linguistic sense or rather conceptual, as pioneered by Coseriu (1978) and experimentally studied by psycholinguists such as Gagné and Spalding (2014) (cf. Dressler et al. 2020).

Moreover, in the area of first language acquisition, Dressler et al. (2019) concluded that, in the first phases of the acquisition of SCs in seven languages (other than English), young children treat them as compounds rather than as derivations. In the light of our present analysis of the synchrony and diachrony of SCs in adult language, a reinterpretation of the above acquisitionist conclusions is available. In phases where children have not developed productive schemas into rules, they cannot yet have self-organized a rule of either deriving or compounding SCs. Therefore, later phases of development should be studied, when children have developed rules from schemas, and when dualism (complementarity) and superposition can begin. In our future study we should also include English SCs, whose behavior may depart from that of other languages (see Werner et al. 2020 for the acquisition of SCs that are action or result nouns).

Superposition can be compared with age-old dialectics, especially in Georg Friedrich Wilhelm Hegel’s “superposition” of thesis and antithesis in their synthesis (cf. Ritter 1972, especially pp. 164–226). However, Hegelian synthesis also includes complementarity (dualism Type (b)). It still needs to be studied whether this conception can be developed so as to overcome the “postmodern” quantum physics one, especially its further development of qubits, i.e., the simultaneity of polar opposed states (Williams 2011).