In this paper, I will address the question of what is compounding. In pursuing this question, I will adopt a terminological approach. In terminology, it is common to start from a concept with rather indeterminate boundaries and work towards a more precisely determined concept that maintains the basic nature of the original idea. For compounding, I will start with some dictionary and textbook definitions and discuss some often-used formal criteria to make the boundary more specific. Then I will present my own approach, which takes as its point of departure that compounds are used first of all as names for concepts. I will show how this approach can work in practice and end with an evaluation of the definition and of the general approach it illustrates.

Compounding is the formation of compounds. For a general understanding of these concepts, we can consult dictionaries or look up the treatment in morphology textbooks. Some typical definitions are given in (1).

(1)

a. Compound: a word made up of two or more existing words

b. Compound: a word formed from two existing words or combining forms

c. Compounding is a word-formation process that combines two or more roots inside the same word.

The first two definitions are from general dictionaries of English. The word compound has several meanings and only the one for the relevant linguistic sense is quoted here. The definition in (1a) is from COED (2011), the one in (1b) from Collins (2014). Although the two definitions broadly agree, (1b) insists on binary branching and focuses on the formation process, whereas (1a) takes the resulting word as the starting point and takes two words only as a lower limit. The statement in (1c) is from Fábregas & Scalise (2012: 111). It is the first sentence of the chapter on compounding. Although it focuses on compounding rather than on compound, it is strikingly similar to the dictionary definitions, in particular (1a). The difference is the reference to root instead of existing word. A common property of such definitions is that they are quite vague. For the following discussion, I will take symphony orchestra as a typical example of a compound.

There have been various attempts in the linguistic literature to make the concept of compounding more precise. Compounds occur in many languages. It is natural to align the concept across languages so that what is called compound in English corresponds to the concept in German, French, or Slovak. This does not mean that the translation of symphony orchestra into French is necessarily a French compound, but that when we discuss English compounding in an article written in French, the French translation of compound, mot composé, should identify the same set of items. This universal application is not entirely straightforward, because French discussions of mots composés take French examples as their starting point, whereas English discussions of compounding start from English examples. In evaluating proposed criteria, it is worth keeping in mind their application to different languages.

A frequently used criterion for compounding is stress. In (2), primary stress is indicated by an underline.

(2)

a. symphony orchestra

b. expensive orchestra

Whereas the compound in (2a) has a single primary stress on the first component, the A+N phrase in (2b) has two primary stresses, one for each word. Although there is more to say about the application of this criterion to English, its most immediate problem is that it cannot be generalized to languages such as Slovak, as illustrated in (3).

(3)

a. idem na obed (‘I_go for lunch’)

b. dovaril obed (‘he_finished_preparing lunch’)

Slovak has stress on the first syllable of a word. In (3b), each word has its own stress. In (3a), however, the preposition na takes the stress from the noun obed (‘lunch’). Thus, the English-based stress criterion would make na obed (‘for lunch’) a compound, which is highly counterintuitive.

Whereas stress is a criterion that is universally applicable at least in principle, other criteria that have been proposed are not. In German, Feld (‘field’) is pronounced with a final /t/, because it is at the end of the word. This is called Auslautverhärtung (‘final devoicing’). In the plural Felder (‘fields’), the grapheme ádñ is pronounced /d/, because it is followed by a vowel. In compounds, the first component is treated phonologically as a separate word. Therefore, Feldarbeit (‘field work’) is pronounced with a /t/ at the end of Feld, because the following vowel is separated by a word boundary. In English, this criterion cannot be used, because English does not have devoicing of word-final consonants. Similarly, in German Arbeitsfeld (‘working field’), an -s- appears between the two components. This is not an inflectional ending of Arbeit (‘work’), because, as a feminine noun, Arbeit never has such an inflectional ending. It is called a Fugenelement (‘linking element’). Although such linking elements are common (but not general) for compounds in German, they do not occur in English.

Another type of criterion concerns the morphological behaviour of the components. The effect of one such criterion can be seen in (4).

(4)

a. direttore musicale (‘music director’)

b. direttori musicali (‘music directors’)

In (4), we have the singular and plural of an Italian combination of a noun and a relational adjective. As the English translations suggest, the meaning is conveyed by a clear compound in English. The plural marker is attached to the head in English, but in Italian both the noun and the adjective are marked as plural. For some linguists, e.g. Matthews (1974: 34-35), this is a reason to exclude (4) in Italian from compounding. As English does not have plural marking on adjectives, the criterion does not really apply to cases such as musical instrument.

The phonological and morphological criteria exemplified so far have the advantage that they generally give clear results as far as they apply, but they have a number of disadvantages. One disadvantage is that they are highly language-specific. The properties they address are often not relevant in all languages (e.g. final devoicing) or they have entirely different effects (e.g. stress). Moreover, even within a single language, they often apply only to some cases. German devoicing does not play a role in Weltordnung (‘world order’), because Welt has a /t/ in all cases (cf. Welte, ‘worlds’). Similarly, compounds such as Feldarbeit and Weltordnung do not have a linking element. From a terminological perspective, this makes these properties less useful as candidates for conditions, because they are only usable as sufficient, not as necessary conditions. Proper terminological definitions are composed of necessary conditions (cf. ten Hacken 2018).

In view of the problems with criteria of the type discussed so far, we can understand the assessments of the possibility of defining compound in (5).

(5)

a. [W]e cannot always make a clean distinction between compound words on the one hand and derived words or phrases on the other.

b. [U]niversal definitions are not only theory-dependent […] but also cross-linguistically never watertight—in many languages there are exceptions or fuzzy transitions to non-compounding[.]

In their introduction to the Oxford Handbook of Compounding, discussing the problem of defining compound, Lieber & Štekauer (2009: 4) make the remark in (5a). Given the context of a handbook, it seems an appropriate summary of a certain consensus in the field. In (5b), Dressler (2006: 24) goes further and implies some assumptions that are quite problematic from a terminological perspective. The start of (5b) suggests that theory-dependence should be evaluated negatively in terminological definitions. In fact, terms are necessarily theory-dependent and their definitions must be embedded in a theoretical framework. Therefore, the theory-dependence is inevitable and should not be subject to any positive or negative evaluation. In the second part, Dressler mentions the existence of exceptions, which implies a higher authority than the definition determining that an expression is actually a compound. However, in terminology, the boundaries of a concept are determined by the definition. Therefore, if we want to use compound as a theoretical concept and make claims about it, we should reject (5b) and interpret (5a) as an incitation to develop a proper definition.

As a starting point for a proper terminological definition, we can analyse a few typical examples of compounds and consider which properties of them we want to retain in a definition. A preliminary observation in this respect is that we should aim for a cross-linguistically valid definition. Without this aim, there are too many shared properties, many of which are due to accidental features of an individual language. On the basis of this decision, we can exclude criteria such as stress, which is subject to different language-specific constraints, for the primary identification of compounds. After all, compounding is not primarily a phonological phenomenon, but a type of word formation. Given that word formation is driven by the need to provide new names for concepts, a good starting point is the relationship between the form and the meaning.

As a first example, let us return to symphony orchestra and consider how the meaning of the word is related to its form. In accordance with the dictionary definitions in (1), symphony orchestra consists of two words. The contribution of the two components to the meaning of the compound is asymmetric, in the sense that a symphony orchestra is an orchestra. Orchestra is the head and symphony the non-head. The role of the non-head can be studied by contrasting it to paradigmatic alternatives. Table 1 gives the most frequent compounds with orchestra in COCA (2008-2020).

We have now seen two approaches to defining compound, each with their strengths and weaknesses. Criteria of the type discussed in section 1 are easy to apply, but cannot be generalized to all languages. Criteria of the type discussed in section 2 give a better perspective of cross-linguistic application, but they are quite abstract. In order to come up with a definition procedure that is at the same time valid for all languages and practical in its application to individual cases, I propose a two-step procedure. In a first step, the properties taken as criteria in the definition of compound are used to identify language-specific constructions that qualify as compounding. In a second step, properties of such a language-specific construction are used to determine which expressions are compounds.

The basic idea of such a two-step procedure for the definition of compounding can be traced back to ten Hacken (1994). The advantages of the criteria of the type discussed in section 1, which I will call low-level criteria here, are not lost, because they are used as tests for constructions. The calibration of their use and interpretation is motivated by invoking the high-level criteria selected as part of the cross-linguistically valid definition. What exactly counts as a construction is a question that can be approached pragmatically. It is determined by how the high-level criteria can be applied and the low-level criteria selected. The notion of construction used in ten Hacken (1994) is not determined by Booij’s (2010) homonymous concept, although the two are probably largely compatible.

In a terminological approach to defining, the starting point is a number of typical examples. The next step is the selection of shared properties of these examples that are deemed crucial. On the basis of the discussion in sections 1 and 2, we can identify two sources of properties that can be used as criteria. On one hand, a compound behaves as a word with two components. On the other, compounds have a characteristic way of being assigned a meaning. The definition I propose is (8).

(8) Definition of compound

A compound is a word with a structure [X Y]_Z or [Y X]_Z, such that:

• The denotation of Z is a subset of the denotation of Y;

• The range of possible relations between X and Y depends on the semantics of X and Y;

• X does not have independent access to the discourse.

In line with standard terminological practice, e.g. Wüster (1979), (8) specifies a hyperonym and special conditions that distinguish the concept from its co-hyponyms. The hyperonym determines a binary structure. For symphony orchestra and piano trio, this condition is obviously fulfilled. If we find three basic elements, as in radio symphony orchestra, we can only fit it into the definition in (8) if we assume that it has one of the structures in (9).

(9)

a. [radio [symphony orchestra]]

b. [[radio symphony] orchestra]

Semantically, (9a) is more plausible. In this case, Y is instantiated as symphony orchestra, whose internal structure does not play a direct role in the analysis of (9a). The reason for giving both orders of X and Y in (8) is that the semantic and syntactic status of the two components is not equal, but the role is not universally correlated with their order. In discussing compounds in French, Arnaud (2016) gives the example of opéra rock (‘rock opera’), which illustrates that in French the order is reversed compared to English.

The first bullet point in (8) highlights the role of Y on the basis of its semantic contribution. Although we may use the term head for Y, we should keep in mind that it is not the same notion as Williams’s (1981) head. Williams defines head on the basis of morphosyntactic properties and assigns this status also to affixes. The reference to denotation in (8) implies that Y is a content word with a lexical rather than purely grammatical meaning. In English, Y is the right-hand component, i.e. orchestra in symphony orchestra and trio in piano trio.

The second bullet point in (8) addresses the semantic relation between the two components. The list of different compounds with orchestra as the second component in Table 1 illustrates how the meaning of the first component influences the range of possible interpretations. However, in view of cases such as piano trio, it cannot be stated that the relation is determined by the two components. As we saw in section 2, onomasiological coercion determines the meaning of the compound. The reference to a range of possible relations again distinguishes compounds from derivations. Viewer can refer to a person or an instrument, and the choice between them is determined by onomasiological coercion, but this is not a choice between different relations of the components. If there is a relation involved in viewer, it is one between view and viewer, not between view and -er.

The third bullet point in (8) is about the syntactic position of the non-head. Its impact can be seen in (10).

(10)

a. I went to hear a symphony_*i orchestra. It_i was very loud.

b. I went to hear a symphony orchestra_i. It_i was very loud.

c. I went to hear a piano_i trio. *It_i was out of tune.

By introducing symphony orchestra, one does not introduce symphony into the discourse. Therefore, it in (10a-b) cannot refer to symphony, but only to orchestra. Similarly, in (10c) it cannot refer to piano. As illustrated in (11), some proper names do not have this restriction, because they inherently refer to the object they name.

(11)

Brahms_i symphonies are generally considered too difficult for a youth orchestra. He_i requires too much technical skill and musical maturity.

Because Brahms in (11) refers to an individual, independently of its context of use, it can serve as the antecedent of he, even if embedded in a compound. Therefore, (11) cannot be used to exclude Brahms symphony as a compound.

In (8), we have a definition of compound that fulfills basic terminological criteria for a definition. The properties it builds on are widely accepted as properties of compounds. Thus, Allen (1978) argues for the IS A and Variable R conditions, which correspond to the first two bullet points and Lieber (2004: 46) lists properties corresponding to the first and third bullet points. The application of this definition was illustrated for some prototypical English N+N compounds in order to explain the individual criteria. At the same time, the less superficial nature of these criteria makes their cross-linguistic applicability plausible.

The N+N construction illustrated by symphony orchestra and piano trio can be taken as prototypical for compounding. Another construction that is often considered compounding is the Saxon genitive. Marchand (1969: 65-69) discusses cases such as driver’s seat as compounds. As it is generally accepted that spelling is a derivative property, the question may in principle be asked whether driver’s seat should not be analysed as a variant spelling of drivers seat, with a plural noun as non-head. Often, German linking elements are homonymous to plural endings. Thus in Blumenladen (‘flower shop’), the first noun is Blume. It has a plural form Blumen, but -en is also a common linking element. In the two-step procedure, we can choose the examples of a construction to be used for determining whether or not it is a compounding construction. Therefore, I will take women’s as the genitive noun, for which the morphophonological ambiguity of the ’s does not arise. In (12), we have an example of women’s magazine, taken from BNC (2007).

(12) One national women’s magazine recently carried an advertorial with no fewer than ten participating, but of course non-competing, manufacturers.

OED (2000-2020 [2011]: woman) describes women’s magazine as “a magazine designed primarily for women”, which indicates that it has a head magazine and the relation with the non-head was selected in a way so as to name a particular concept. The final clause of (8) can be applied by continuing with a pronoun after (12), e.g. They were enthusiastic about it. In this case, they can only refer to manufacturers, not to women. This can be explained because, as the non-head of a compound, women in (12) does not actually introduce any women in the discourse.

The conclusion that women’s magazine is a compound does not imply that all N’s+N are compounds. Another corpus example, taken from NOW (2016-2020) is (13).

(13) It’s 2015 and yet we still can’t separate women’s accomplishments from their looks.

In (13), women’s accomplishments cannot be a compound, because their refers to women. An easy way to separate the cases such as (12) and (13) is the position of adjectives. In (12), national comes in front of women. The alternative, *one women’s national magazine is ungrammatical. In (13), if we insert academic, it must come between the two nouns, women’s academic accomplishments. Therefore, the test of adjective placement can be used to distinguish compounds such as women’s magazine in (12) from syntactic constructs such as women’s accomplishments in (13).

This brief discussion of the Saxon genitive constitutes a good example of the two-step procedure for recognizing compounds. By means of the definition in (8), we can determine that there are N’s+N compounds in English. The compounding construction has to be distinguished from a similar syntactic construction, which can be achieved by a language-specific test based on the position of an adjective.

The affinity between genitive constructions and compounding is not only found in English. Many German linking elements can be analysed as (historical) genitive endings. Thus, in Handelshaus (‘trade house’), the linking element -s- is homophonous to the genitive ending of Handel. In Romance languages, the genitive is expressed by a periphrastic construction with a preposition, illustrated for French in (14).

(14)

a. usine d’automobiles (‘factory of cars’, i.e. car factory)

b. goutte d’eau (‘drop of water’, i.e. water drop)

c. bouteille de vin (‘bottle of wine’ or wine bottle)

The French preposition de, translated as ‘of’ in (14), has a meaning and function that corresponds to the morphological genitive in other languages. Ten Hacken (2013) compares English, with its Saxon genitive, Polish, with a morphological genitive as in fabryka samochodów (‘factory cars_gen’), and French. In all cases, we get an ambiguity of the type we saw in the contrast between (12) and (13). In (14c), this ambiguity is realized in a single expression as illustrated in (15).

(15)

a. une bouteille de vin ouverte (‘a_f bottle_f of wine_m open_f’)

b. une bouteille de vin blanc (‘a_f bottle_f of wine_m white_m’)

In (15), the glosses indicate gender of all words that are marked for it. In (15a), the adjective has feminine gender and modifies bouteille. This corresponds to an interpretation as ‘wine bottle’, which aligns it with compounds. In (15b), the adjective has masculine gender and modifies vin. This corresponds rather to the interpretation ‘bottle of wine’, i.e. as a syntactic construction in which bouteille indicates a quantity. What we can observe here is, then, a correlation between syntactic and semantic properties. How this correlation can be used to come up with a set of precise criteria for delimiting the compounding construction in French remains to be determined.

Another type of construction with a recognized affinity to compounding is the combination of relational adjectives with nouns (RA+N). As we saw in the discussion of (4) in section 1, Matthews (1974) excludes them from compounding because the adjective agrees with the noun. This criterion does not apply in English, because adjectives do not show agreement. Levi (1978) includes RA+N into her account, but she avoids the term compounding, using complex nominal instead, and only discusses English data. As argued by ten Hacken (2019b), there are good reasons for considering RA+N a compounding construction, because the way they receive their meaning is similar to that of prototypical N+N compounds.

The central properties that mark a relational adjective are that it corresponds to a noun and that its meaning is only determined by the relation to the designation of the noun. This is illustrated by pairs such as (16) and (17).

(16)

a. urban police

b. city police

(17)

a. urban development

b. city development

In (16a) and (17a), we have RA+N expressions. The noun corresponding to urban is city. Although it is possible that the two expressions in (16) and (17) are assigned different meanings, any such difference is due to onomasiological coercion, not to differences in the expression. A more likely scenario is that the pairs in (16) and (17) are competing expressions. The frequency in BNC (2007) suggests that, at least in British English, (16b) and (17a) have won this competition, although the other expressions are also attested.

A question that has to be addressed when RA+N expressions are treated as compounds is how to distinguish relational adjectives from other adjectives. The Italian contrast in (18) illustrates this issue.

(18)

a. banca popolare (lit. ‘bank popular’)

b. calciatore popolare (lit. ‘football_player popular’)

The Italian adjective popolare can be a relational adjective corresponding to popolo (‘people’), as in (18a), or a qualitative adjective expressing popularity, as in (18b). For (18a) there is no lexicalized corresponding expression in English, but in German there is the corresponding compound Volksbank, which has Volk (‘people’) as its non-head. There are various well-known tests that exploit the difference between these two uses. Two of them are illustrated in (19).

(19)

a. #banca molto popolare

b. calciatore molto popolare

c. #la banca è popolare

d. il calciatore è popolare

In (19a-b), the adjective popolare is modified by molto (‘very’). In the case of (19b), the modifier amplifies the property it conveys in (18b), but in (19a), it requires a reinterpretation of the adjective compared to the normal reading of (18a). The same contrast occurs in (19c-d), where the adjective is used predicatively with the copula è (‘is’). The # in (19) indicates the change in interpretation of the adjective with respect to (18). Clearly, the precise set of tests will have to be calibrated for each language, but the general idea behind it is that a relational adjective expresses a relation to an underlying noun, which, as urban in (16) and (17) illustrates, need not be morphologically related to the adjective. The meaning of an RA+N expression is determined in much the same way as for (other) compounds.

The dictionary definitions in (1) apply not only to typical compounds such as symphony orchestra and piano trio, but also to a range of other expressions. Bloomfield (1933) gives a typology of compounds that is based on earlier work, in particular in Sanskrit grammar. This typology has had a significant influence on subsequent accounts, mostly in a somewhat simplified format with three types, exocentric, determinative and copulative. Determinative compounds have a head and a non-head, so it is no surprise that all compounds discussed so far are of this type. Indeed, the definition in (8) takes the asymmetry between the head and the non-head as a defining criterion. It is therefore a legitimate question what happens with the expressions that have been labelled as exocentric or copulative compounds in the literature.

In (20), some examples of expressions that have been labelled as exocentric compounds are given. They represent different types of expression that have been assigned to this category.

(20)

a. blackcap

b. breakwater

c. spaceship

In (20a), we have an A+N noun. As such it contrasts with blackbird. Whereas blackbird refers to a kind of bird, blackcap does not refer to a kind of cap, but to someone or something marked by a black cap. The type in (20a) is often called bahuvrīhi. In both blackbird and blackcap, onomasiological coercion determines the actual meaning, but the meaning assignment is not parallel. In the case of blackbird, the meaning is a species of birds, Turdus merula, whose male specimens are marked by a black colour. In the case of blackcap, the meaning is another species of birds, Sylvia atricapilla, whose male specimens are marked by a black area on the top of their head. The two species are illustrated in Fig. 1 and Fig. 2.

In this paper, the question of the delimitation of compounding was approached from a terminological perspective. In the relevant sense, compound is a term of linguistics. The fact that it is used in other fields, e.g. chemistry, is irrelevant for the linguistic term. At the same time, compound, taken in the linguistic sense, is a word in the mental lexicon of linguists and other speakers who know it. In the mental lexicon, the meaning of words is prototype-based and speaker-specific. In terminology, the purpose is to delimit concepts more precisely and in such a way that different speakers arrive at the same conceptual boundaries. Hence the starting point is a concept with vagueness originating from two sources. On the one hand, for individual speakers (linguists), there are prototypical instances and a gradual transition to non-instances without a clear, natural borderline. On the other hand, different linguists each have their own prototypical concepts in their mental lexicon. The purpose of a terminological definition is to introduce a standard with precise boundaries. Such a standard is necessary for testing theoretical claims. The definition changes the nature of the concept. Instead of a structured concept with more or less typical instances, the terminological concept is determined exclusively by its boundary.

The result of the discussion in this paper can be viewed in two different ways. First, I proposed a particular definition of compound, formulated in (8), and explored its effects on a number of controversial and less controversial instances. At a more basic level, however, I demonstrated how a two-step procedure can be used for the definition of concepts such as compound. This two-step procedure takes into account that interesting theories in linguistics generally make cross-linguistic claims, but data are necessarily language-specific. Definition then involves a universal and a language-specific step, mediated by the level of what I call construction. The basic idea of this two-step procedure can also be applied to linguistic concepts such as adjective, subject, and diminutive.

In formulating the definition in (8), I made the assumption that what determines whether an expression is a compound or not is the way its meaning is assigned. Therefore, phonological and inflection-based properties were not retained in the cross-linguistic definition used to identify language-specific constructions, although they can play an important role in the second, language-specific step, mediating between constructions and individual expressions. The central intuition used for the language-independent definition is that compounds are words with a binary asymmetric structure that combine the two components without imposing a specific relation between them. As with all naming units, the meaning of a particular compound is ultimately determined by onomasiological coercion.

Compared to the definitions in (1), (8) is much more specific. Whereas the definitions in (1) might make without a compound, (8) clearly excludes this. In cases such as RA+N constructions, whether the definitions in (1) include them depends on one’s interpretation of word. In (8) they are included. The main difference is not that one definition uses other terms and the other one does not. The main difference is that (8) gives a number of specific properties that identify constructions cross-linguistically that share the way the expressions receive their meaning. Definitions such as (1) can be seen as deficient in the sense that they do not determine a boundary for the concept. They only give a property that compounds share, but do not indicate why this should be. The notion of word used in (8) is specified by the further conditions it imposes. As a result, (8) diverges in some cases from traditional classifications. However, by taking the way meaning is assigned as a starting point, the constructions identified by (8) share important properties. As such, (8) is a better definition from a terminological perspective.