2.1 Some basic issues

One fairly reliable constraint on English noun-noun compounds is that N₂ is normally the semantic head of the compound: the compound denotes a particular subtype of the type denoted by N₂. Beef stew is a kind of stew, but stew beef is a kind of beef. I’ll call this the Head Principle; it will be stated more formally in section 2.4.

Three sorts of exceptions to the Head Principle are well known. One sort (dvandva compounds) equates the two components (4a). Another sort, exocentric (or bahuvrihi) compounds, have a tacit semantic head (4b), and sometimes N₂ is metaphoric (4c). A third sort, relatively rare left-headed compounds of various patterns (4d), are not noun-noun compounds, so I will set them aside here.

(4)

1. a. tractor-trailer = ‘something that consists of both a tractor and a trailer’

2. b. blockhead = ‘someone with a head like a block’ ≠ ‘a head like a block’

3. c. seahorse = ‘something that resembles a horse and lives in the sea’ ≠ ‘a kind of horse’

4. d. attorney general, mother-in-law, blowup, pickpocket

Section 2.7 will show how exocentric compounds fall under a suitably generalized form of the Head Principle.

Beyond these basic points, the interpretation of compounds is hugely varied. Consider the variety of meaning relations between N₁ and N₂ in (5).

(5)

1. a. chocolate cake = ‘a cake made with chocolate in it’

2. b. birthday cake = ‘a cake to be eaten as part of celebrating a birthday’

3. c. coffee cake = ‘a cake to be eaten along with coffee and the like’

4. d. marble cake = ‘a cake that resembles marble’

5. e. layer cake = ‘a cake formed in multiple layers’

6. f. cupcake = ‘a little cake made in a cup’

7. g. urinal cake = ‘a (nonedible) cake to be placed in a urinal’

This range of semantic relations is not confined to conventionalized compounds: Wisniewski and Gentner (Reference Wisniewski, Gentner and Simpson1991) show that a similarly wide range is found in people’s judgments of novel compounds.

To be sure, speakers store thousands of lexicalized compounds with semi-idiosyncratic meanings. But compounds cannot all be stored in the lexicon. Many of the examples in (2)–(3) are understood on the fly. Downing (Reference Downing1977) stresses the frequent coining of compounds in particular discourse situations, citing for instance bike girl being used for a girl who left her bike in the vestibule, and apple juice seat for a seat at which apple juice was set on the table. Some years ago, my daughter left something for our cats, accompanied by the note “Maggie and Peanut’s heat wave present”. Gleitman and Gleitman (Reference Gleitman and Gleitman1970), Brekle (Reference Brekle1986), and Sadock (Reference Sadock, Lapointe, Brentari and Farrell1998) make similar observations. Such creativity is not confined to adults: Clark, Gelman, and Lane (Reference Clark, Gelman and Lane1985) observe that children begin understanding novel compounds – and coining their own – between about 2½ and 3½ years of age (see Berman Reference Berman, Lieber and Štekauer2009 for cross-linguistic documentation). For instance, a more recent coinage of my daughter’s, understood immediately by her two-year-old twins, was rainbow car for a car painted in multicolour stripes.

There is no principled line between listed and generated compounds. For example, soccer ball is listed in my lexicon: I can connect it to particular physical objects. However, although I may not have ever heard bocce ball before, I can guess on the fly that it is a ball for playing bocce – without knowing what one looks like. But it certainly is listed in the lexicons of bocce players, so speakers may differ in whether they list or “generate” particular compounds.

However, the creativity of recursive compounding along the lines of (2) is rather fragile by the usual standards of productive grammar. Students in Introductory Linguistics enjoy building up a compound like (6a) piece by piece, in such a way that everyone understands it at the end. But if it were presented as a whole to a naive class, few would get it (cf. Gleitman and Gleitman Reference Gleitman and Gleitman1970). By contrast, the syntactic paraphrase (6b), while unwieldy, is nevertheless easier to comprehend, presumably because the addition of function words makes the semantic relations more explicit.

(6)

1. a. an inflectional morphology instruction manual software programming course

2. b. a course in programming the software that accompanies manuals that teach inflectional morphology

Similarly, (7a), from a restaurant menu, is hard to understand, because the preference for balanced prosody encourages one to parse it pairwise, like (7b). The prosody for the proper parsing (7c) is highly marked.

(7)

1. a. braised beef tongue toast

2. b. [[braised beef] [tongue toast]]Footnote ³

3. c. [[braised [beef tongue]] toast],

   i.e. ‘toast with [braised [tongue of beef]]’

The upshot is that the fragility of recursive compounding also has to be part of the account: semantic interpretation is highly dependent on the pragmatics of the words being combined and on the contextual specifics of use. The language user must home in on the intended meaning of a novel compound by making use of (a) the semantic details of the constituent words and (b) the discourse and extralinguistic context. Syntax does not give much support.

Such a conclusion should not be seen as too radical. After all, the semantic relations that link individual sentences into discourse are not marked syntactically either. Rather, hearers use the meanings of the sentences plus understanding of the context – including social context – to create the semantic linkages from sentence to sentence, whether spoken by the same or different speakers. Compounding is just the same sort of phenomenon writ small. (We return to this parallel in section 2.8.)

2.2 Compositionality and its limits

The analysis here is carried out in the framework of Conceptual Semantics (CS; Jackendoff Reference Jackendoff1983, Reference Jackendoff1990, Reference Jackendoff2002, Reference Jackendoff2010), which is concerned both with details of word meaning and with how these interact with the composition of phrase meanings. CS incorporates a great deal of what is usually called pragmatics: aspects of meaning that are not encoded in word meanings or in relations conveyed directly by syntactic structure. Given the rudimentary syntax of compounds, pragmatics plays a much larger role here than in canonical syntactic constructions.

The basic intuition, as in other approaches, is that the meaning of a compound is a function of the meanings of its constituents. Thus the problem is: given two nouns N₁ and N₂ that mean X₁ and Y₂ respectively, what is the function F(X₁, Y₂) that yields the meaning of the compound [N₁ N₂]?

Of course, it is important to recognize the limits of compositionality in compounds. For novel compounds (say, backgammon ball), compositionality should be all that is available. But lexicalized compounds usually also incorporate idiosyncratic information. No linguistic principle predicts that stereotypical soup bowls and fish bowls differ in shape, or that a boxcar is a kind of railroad car but a kiddy car is a kind of toy. And general principles cannot account for cranberry morphemes (underlined parts of (8a)) or what I like to call “strawberry” morphemes – real words within compounds that play no role in the compound’s meaning; for example, the underlined parts of (8b).

(8)

1. a. cranberry, basset horn, bogeyman, pratfall, fascia board, iceberg, fig newton, nightmare

2. b. strawberry, cottage cheese, polka dot, bobby pin, dogwood, horseradish, monkey wrench, honey moon

So sometimes lexicalized meaning has to ignore the semantics of one noun or the other, just as it ignores bucket in the idiom kick the bucket. Still, on the whole there is significant compositionality, and an analysis of compounds should recognize it, to the extent it exists.

One technique for analysing compounds is to establish a phrasal paraphrase and to attribute the meaning of the paraphrase to the compound. However, as Lees (Reference Lees1960) observes, it is sometimes impossible to establish a single best paraphrase for a compound:

Gleitman and Gleitman (Reference Gleitman and Gleitman1970: 95) make similar remarks: “We suspect that the person who says lion-house would consider it rather odd if someone asked him: ‘Did you mean a house for a lion, a house suitable for lions, or a house lions can live in?’ Obviously the speaker meant any of these indifferently.” Levi (Reference Levi1978) also discusses this problem at length.

The solution involves three different cases. First, the paraphrases may be pure semantic variants, as in Gleitman and Gleitman’s ‘house for a lion’ and ‘house suitable for a lion’, or felafel ball = ‘ball made of/created from felafel’. Here, a proper semantic analysis should abstract F(X₁, Y₂) away from the particular way it happens to be expressed in the paraphrase.

Slightly more complex multiple paraphrases arise in cases like (10).

1. (10) ticket window = ‘window at which tickets are bought/at which tickets are sold’

   toll booth = ‘booth at which tolls are paid/at which tolls are collected’

   movie theater = ‘theater where movies are seen/at which movies are shown’

Here the multiplicity arises from different ways of orienting the same connection between N₁ and N₂. Following Langacker (Reference Langacker1987) and Fillmore and Atkins (Reference Fillmore, Atkins, Lehrer and Feder Kittay1992), the two paraphrases express the same conceptualized “event schema”, changing only the perspective. Buy places the initiative with the recipient of goods, and sell with the original owner; while in a real transaction both must play an active role. Similarly with the payment and collection of tolls, and with the seeing and showing of movies. The solution, then, is that F(X₁, Y₂) likely consists of the simple event schema, bereft of perspective or focus. The paraphrases, however, create alternative perspectives, because the overt use of a verb forces us to choose a particular frame of thematic roles.

The following examples, like pontoon bridge, present a third situation.

1. (11) boxcar = ‘car that carries boxes/that resembles a box/that serves as a box’

   elevator shaft = ‘shaft that an elevator travels in/that is part of an elevator’

   file folder = ‘folder in which one places a file/that forms part of a file’

Unlike (10), these cases are not lexical or perspectival variants of the same relation. Resembling a box and serving as a box are quite distinct relations.

Lees, Levi, and the Gleitmans suggest that there is no fact of the matter about which paraphrase is correct. Someone learning these words is typically given no evidence (e.g. “This is called a box car because it looks like a box”): the relation is normally taken to be self-evident and without need of explanation (“That thing is called a boxcar”). Boxcar is not ambiguous: it picks out the same objects no matter which reading we take to be the “correct” one. It is not like football the game vs. football the ball. Nor is boxcar vague: it does not leave open a continuous range of possibilities on a scale, the way, say, cold does.

I propose that there is another way that boxcar can have multiple meanings besides being ambiguous or vague: it can have all the meanings in (11) simultaneously, in cooperation rather than in competition. We might call such a word promiscuous (by contrast with ambiguous). The claim, then, is that pontoon bridge and boxcar are promiscuous. A learner attempts all plausible strategies for combining N₁ and N₂ (presumably in parallel), and since there are multiple satisfactory strategies that do not conflict, any or all such semantic combinations may be encoded in memory as part of the meaning of the compound. (Of course, if asked to define the compound, speakers will likely give only one of the combinations and be satisfied with that.)

If such a solution seems like giving up, one must remember that a word meaning is an entity in a brain, not in a logical system, and the brain likes redundancy. A linguist seeking to analyse these compounds faces the same problem as the learner. The insistence on a single best solution is only a prejudice, which, I acknowledge, is well grounded in scientific and common-sense practice. But in dealing with brain processes, I believe it is sometimes counterproductive, and should be judiciously abandoned when inappropriate.

2.3 Aspects of compound meaning that come from the semantics of nominals

The semantics of compounding involves a number of distinct components. This section sketches three components implicated in the semantics of nouns in general; the next four sections add components that are specifically involved in compounding.

2.4 Semantic structure of (relatively) simple compounds

2.5 Thirteen basic functions

Here is a list of the (most prominent) basic functions that can fill out F in English noun-noun compounds.

2.6 Using material from the meanings of N₁ and N₂

If this were all there were to filling out the interpretation of compounds, the number of possible relations in compounds would be thirteen, or, allowing for all the variants and reversibility, somewhere in the twenties or thirties – clearly not enough. However, two other strategies come into play that create a larger range of relations: incorporating material from the internal semantic structure of the two nouns, and composing functions to enable more distant relations between N₁ and N₂. We take these up in turn.

Consider first the locative relations illustrated in (33)–(35). They appear to include only locations paraphrasable by at, in, or on – perhaps the least marked spatial relations. Yet the meanings of compounds can include other spatial relations.

Usually, the spatial relation in question comes from the proper function of one or the other of the nouns, usually N₂. For instance, the proper function of a fountain is for liquid to flow out of it (45a). We can “reprofile” this, as we did with enjoy the book in section 2.3.3, yielding (45b). We can then use the reprofiled function as F in the Modifier Schema, and plug WATER into the LIQUID argument, yielding the structure of water fountain, as shown in (45c). (Again the cocomposed function is notated in italics.)Footnote ⁹

(45)

1. a. fountain = [FOUNTAIN^α; [PF (FLOW (LIQUID, OUT-OF α))]]

2. b. reprofiled: [FLOW (LIQUID, OUT-OF (FOUNTAIN))]

3. c. water₁ fountain₂ = [FOUNTAIN₂^α; [PF (FLOW (WATER₁, OUT-OF α))]]

   ‘a fountain that water flows out of’

Similar cases are coal mine (‘dug out of’), gas pipe (‘flows through’), Charles River bridge (‘crosses over’), and Downing’s (Reference Downing1977) toe-web (‘extends between’). In all these cases, N₂ has a proper function that includes a spatial relation, and N₁ is an argument of the reprofiled function.

This approach accounts for large families of compounds such (46).

(46)

1. a. N₂ is a container:

   coffee₁ cup₂ = [CUP₂^α; [PF (HOLD (COFFEE₁, IN α))]]

   also photo album, carbarn, soapdish, fishtank

2. b. N₂ is a vehicle:

   cattle₁ car₂ = [CAR₂^α; [PF (CARRY (CATTLE₁, IN α))]]

   also baby carriage, garbage/icecream/oil truck

3. c. N₂ is an article of clothing:

   pinky₁ ring₂ = [RING₂^α; [PF (WEAR (INDEF^β, α, ON [PINKY(β)]₁))]]

   also face mask, necktie, ankle bracelet, skullcap, earring, backpack, wristwatch

4. d. N₂ is itself a location:

   liquor₁ store₂ = [STORE₂^α; [PF (BUY/SELL (INDEF, LIQUOR₁; IN α))]]

   also fruit market, movie theater (SHOW/SEE), law school (LEARN/TEACH)

5. e. N₂ is an incipient stage of something else:

   rose₁bud₂ = [BUD₂^α; [PF (BECOME (α, ROSE₁))]]

   also chick embryo, grass seed, dinosaur egg

This strategy also accounts for cases in which N₂ is agent or instrument of an action but is not derived from a verb. (The agentive and instrumental cases in (25b, d) thus also fall redundantly under this schema.)

(47)

1. a. N₂ is an agent or causer:

   silk₁ merchant₂ = [MERCHANT₂^α; [OCC (SELL (α, SILK₁))]]

   also eye doctor, pork butcher, sanitation engineer, locksmith, brick mason, car thief

2. b. N₂ is an artifact:

   glue₁ gun₂ = [Y₂^α; [PF (SHOOT (INDEF, GLUE₁, FROM α))]

   also fishnet, ant bait, bread knife, handsoap, snow shovel

When N₂ denotes an information-bearing item such as a song, N₁ can describe the topic of the information (what the information is about).

1. (48) love₁ song₂ = [SONG₂^α; [BE (INFORMATION (α), ABOUT LOVE₁)]]

   also Passion play, research paper, success story, fairy tale, horror film, newsletter

It is also possible to cocompose F with material from N₁. Consider cannonball, ‘ball whose proper function is to be shot from a cannon’ (49a). Reprofiled as (49b), it can be cocomposed with F to form (49c).

(49)

1. a. cannon = [CANNON^β; PF (SHOOT (INDEF, BALL, FROM β))]

2. b. reprofiled: SHOOT (INDEF, BALL, FROM CANNON)

3. c. cannon₁ball₂ = [BALL₂^α; [PF (SHOOT (INDEF, α, FROM CANNON₁))]]

   ‘a ball whose proper function is for people to shoot it from a cannon’

2.7 Generative schemata for F

So far we have seen cases in which N₁ is an argument of N₂ (union member, hair dryer, etc.), cases in which N₁ and N₂ are co-arguments of a basic function (gemstone, sunspot, etc.), cases in which N₁ is an argument of a modifier within the lexical meaning of N₂ (coffee cup, silk merchant, etc.), and cases in which the proper function of N₁ serves as a modifier of N₂ (cannonball). Suppose none of these possible relations makes sense. Then further combinatorial options have to come into play to create a more complex relation.

First, two of the basic functions can compose. A swordfish is a fish with a part that is like a sword, so it involves both the PART and the SIMILAR functions.

1. (50) sword₁fish₂ = [FISH₂^α; [PART ([Z^β; SIMILAR (β, SWORD₁)], α)]]

   also alphabet soup

For a second case, a street singer is not someone who sings streets (like a ballad singer), but someone whose singing is in the street. Here F comes from the occupation or characteristic activity of singing, and the basic locative relation is used to construct a modifier to F:

1. (51) street₁ sing₃-er₂ =

   [PERSON₂^α; [OCC/CHAR (SING₃^β(α); [BE (β, IN STREET₁)])]]

   ‘a person who occupationally or characteristically sings, and the singing is in the street’

   also skywriter, water skier (ON instead of IN), grasshopper (INSECT instead of PERSON)

Another such configuration is found in (52), steamboat. Here F is filled in from the lexical semantics of boat, ‘something whose proper function is to move in water’, and ‘move’ is modified by the basic function CAUSE, encoding the means by which the boat moves.

1. (52) steam₁boat₂ = [BOAT₂^α; [PF(MOVE^β (α); [CAUSE (STEAM₁, β)])]]

   ‘a boat that moves by steam causing its movement’

Next consider barbershop. Like store and market in (46d), a shop is a place whose proper function is for goods and services to be bought and sold in it (53a). However, a barber shop does not sell barbers (like a cheese shop), it sells what barbers do, namely cutting hair. In other words, the thing being sold is found in the proper function of barber, (53b). In order to compose (53b) with (53a), it has to be reprofiled as an activity, (53c). It can then form an argument of the proper function of shop, yielding the desired meaning, (53d).

(53)

1. a. shop = [SHOP^α; [PF (BUY/SELL (INDEF, INDEF; IN α))]]

2. b. barber = [PERSON^α; [OCC (CUT (α, HAIR))]]

3. c. reprofiled: [CUT (PERSON, HAIR)]

4. d. barber₁ shop₂ =

   [SHOP₂^α; [PF (BUY/SELL (INDEF, [CUT (PERSON₁, HAIR)]; IN α))]]

   ‘a shop in which one buys the action of someone cutting hair’

   also toll booth, ‘a booth at which one pays tolls’

Hence the function linking N₁ and N₂ uses material from the meanings of both nouns.

A still more complex case is piano bench, ‘bench on which one sits while playing the piano’. Sitting comes from the proper function of bench, and playing comes from the proper function of piano. These two functions are connected by the basic function of temporal location, ‘while’. Thus there are three independent components involved in linking N₁ and N₂, two of which come from the proper functions of pianos and benches, and one from a basic function.

1. (54) piano₁ bench₂ =

   [BENCH₂^α; [PF (SIT^β (PERSON^γ, ON α); [BE_temp (β, AT

   [PLAY (γ, PIANO₁)])])]]

   ‘a bench on which one sits, such sitting being while one plays a piano’

   also bass stool, bike helmet, lobster bib, dessert wine, coffee cake

A final case is exocentric compounds such as blockhead, which violate the Head Principle. These result from using a schema, not restricted to compounds, that allows one to refer to an object by using the name of something that resembles it. This schema is invoked in swordfish (50) to characterize the part of the fish that is like a sword.

1. (55) Metaphor coercion

   N₁ = [Z^α; SIMILAR (α, X₁)], ‘something that is similar to X’

But now consider what happens if this schema is the outermost material in linking X and Y together. In this case, the semantic head of the compound corresponds to Z, that is, it is unexpressed, in violation of the Head Principle. Aside from that, the meanings of N₁ and N₂ are incorporated just as in ordinary compounds, interweaving basic functions and cocomposition. Here are five examples, each slightly different in how the constituent nouns are incorporated. Since these are exocentric compounds, the head has to be lexically stipulated.

(56)

1. a. pig₁tail₂ = [HAIR^α; [SIMILAR (α, [TAIL₂ (PIG₁)])]]

   ‘hair that is similar to the tail of a pig’

2. b. canvas₁back₂ = [DUCK^α; [SIMILAR ([BACK₂ (α)], CANVAS₁)]]

   ‘duck whose back resembles canvas’

3. c. bird₁brain₂ = [PERSON^α; [SIMILAR ([BRAIN₂ (α)], [BRAIN (BIRD₁)])]]

   ‘person whose brain is similar to that of a bird’

4. d. sea₁ horse₂ = [ANIMATE^α; [SIMILAR (α, HORSE₂)]; [CHAR (BE (α, IN SEA₁))]]

   ‘animate entity similar to a horse that is characteristically in the sea’

5. e. coat₁tail₂ = [Z^α; [SIMILAR (α, TAIL₂)]; [PART (α, COAT₁)]]

   ‘something that is similar to a tail and that is part of a coat’

2.8 Closing remarks

By the time we get to the semantic structures in the last section, a large proportion of the meaning is connective tissue: unexpressed basic functions, bound variables, and cocomposed functions. Nevertheless, the overall result should be clear. The semantic relation between N₁ and N₂ arises by coercing extra functions into the structure, either in argument or in modifier positions, and by filling these functions out either with basic functions or with internal semantic structure from N₁ and N₂. The generative system for compound meanings can be summed up as follows:

1. (22) Head Principle

   [N₁ N₂] = [Y₂ (…); (…)]

1. (23) Argument Schema

   [N₁ N₂] = [Y₂ (…, X₁, …)] ‘an N₂ by/of/ … N₁’

1. (26b) Modifier Schema

   [N₁ N₂] = [Y₂^α; [F (…, X₁, …, α, …)]]

   ‘an N₂ such that F is true of N₁ and N₂’

1. (55) Metaphor coercion

   N₁ = [Z^α; SIMILAR (α, X₁)], ‘something that is similar to X’

1. (57) Auxiliary principles

   Cocomposition of F or an argument or modifier of F in (26b) with

   1. i. a basic function

   2. ii. a modifier of Y₂ (e.g. Y₂’s proper function)

   3. iii. a reprofiled modifier of X₁ (e.g. X₁’s proper function)

   4. iv. recursion on the above

This is fairly straightforward when there is only one function F undergoing cocomposition, as in sections 2.4 and 2.5, but complexity increases quickly with multiple loci of cocomposition, as in sections 2.6 and 2.7. The resulting number of options for the semantic relation between N₁ and N₂ also increases quickly, which seems consistent with the literature’s limited success at enumerating them. In the present approach, the repertoire of possible relations is created by a generative system which, aside from the rudimentary linking of N₁ and N₂ into the structure, is entirely within the semantics. So in a sense Jespersen is correct in saying “The analysis of the possible sense-relations can never be exhaustive” – and yet it is systematic.

As mentioned earlier, another important system in language has properties similar to compounding: discourse. The semantic connections among sentences in discourse are by definition not expressed in syntax, which is confined to the structure of individual sentences. As is well known, the semantic connections among sentences are highly dependent on world knowledge:

(58)

1. a. Max fell. Bill pushed him. (S₂ causes/explains S₁)

2. b. Max fell. Bill helped him up. (S₂ follows S₁)

(59)

1. a. The car collided with the building. The headlights broke.

   (S₂ is result of S₁)

2. b. The car collided with the building. The clutch broke.

   (S₂ causes/explains S₁)

Asher and Lascarides (Reference Asher and Lascarides2003) (from which (59) is taken) work out a formal logic of discourse connection that depends on about ten basic relations, which can link up sentences or groups of sentences in the discourse – and not always sequentially, but rather in a plethora of possible ways. (See also Clark Reference Clark1996 for the hierarchical complexity of connections among sentences in conversation.) Some of the relations posited by Asher and Lascarides parallel basic relations in compounding. For instance, cause/result appears in both systems, and their Elaboration function, where S₂ specifies details of S₁, can be construed as parallel to PART in the compound system: the event denoted by S₂ is a part of the event denoted by S₁. The system proposed here for compounding has parallel properties, but it relates nouns inside a compound instead of sentences in a discourse.

Consideration of discourse, then, makes it clear that only part of language understanding can be controlled by the orderly composition of lexical meanings, guided by syntactic structure. Discourse requires much freer “seat-of-the-pants” composition, guided only by semantics and pragmatics. This provides a point of comparison for the analysis of compounds, especially cases like those in sections 2.6 and 2.7, where a great deal of unexpressed semantic material is needed in order to connect the nouns.

The questions raised by this account are the same ones that have persisted in the literature, but they can perhaps now be couched more precisely. Here are five; other researchers will surely have more.

• What is the full set of basic functions, and how uniform are they cross-linguistically?

• To what extent are these functions generally available for pragmatics (including discourse) and non-linguistic conceptualization?

• To what extent are these functions special to language – or to English?

• How extensive can coerced and cocomposed functions be in compounds, and to what extent does the answer differ between lexicalized and novel compounds?

• What other basic morphosyntactic patterns must be added to the simple N₁N₂ structure in (22), (23), and (26b) in order to account for the quasi-syntactic elaborations in (3) (e.g. health and welfare fund), and what is their status vis-à-vis morphology and syntax?

The first and third of these questions may be answered in part by the typological descriptions such as those in Lieber and Štekauer (Reference Štekauer, Lieber and Štekauer2009); the fifth may be addressable in the framework of Construction Morphology (Booij Reference Booij2010a). Other traditional questions have been answered by the present account; for example, why the possible relations between N₁ and N₂ are so varied yet not altogether wild, and precisely how the meanings of the two nouns contribute to the meaning of the compound. Above all, because the framework of Conceptual Semantics liberates the generative capacity of semantics from that of syntax, it has been possible to give a semantically based account of compounds that is sufficiently formal to see what is going on, while keeping the syntax as absolutely simple as it looks.

3.1 The framework

The framework of Lieber (Reference Lieber2004) is designed to represent the lexical semantics of both simple and complex words. It decomposes the meanings of words into two basic parts: the semantic skeleton which contains those aspects of lexical and affixal meaning that are syntactically relevant, and a semantic body which contains all aspects of meaning that are encyclopedic in nature. The skeleton is a hierarchically organized structure of functions and their arguments (1), where functions consist of combinations of features, the most important of which for our purposes are listed in (2).

1. (1)

   a.	[F1 ([argument])]
   b.	[F2 ([argument], [F1 ([argument])])]

1. (2) Semantic featuresFootnote ¹

   • [+/−material]: the presence of this feature defines the conceptual category of substances/things/essences, the notional correspondent of the syntactic category Noun. The positive value denotes the presence of materiality, characterizing concrete nouns. Correspondingly, the negative value denotes the absence of materiality; it defines abstract nouns.

   • [+/−dynamic]: the presence of this feature signals an eventive or situational meaning, and by itself signals the conceptual category of situations. The positive value corresponds to an event or Process, the negative value to a state.

   • [+/−IEPS]: this feature stands for ‘Inferable Eventual Position or State’. Informally, we might say that the addition of [IEPS] to a skeleton signals the addition of a path. The positive value implies a directed path, and the negative value a random or undirected path.Footnote ²

   • [+/−Loc]: lexical items that bear the feature [Loc] for ‘Location’ are those for which position or place in time or space is relevant. For those items which lack the feature [Loc], the notion of position or place is irrelevant. Further, those which bear the feature [+Loc] will pertain to position or place. [–Loc] items will be those for which the explicit lack of position or place is asserted.

   • [+/−B]: this feature stands for ‘Bounded’. It signals the relevance of intrinsic spatial or temporal boundaries in a situation or substance/thing/essence. If the feature [B] is absent, the item may be ontologically bounded or not, but its boundaries are conceptually and/or linguistically irrelevant. If the item bears the feature [+B], it is limited spatially or temporally. If it is [–B], it is without intrinsic limits in time or space.

   • [+/−CI]: this feature stands for ‘Composed of Individuals’. The feature [CI] signals the relevance of spatial or temporal units implied in the meaning of a lexical item. If an item is [+CI], it is conceived of as being composed of separable similar internal units. If an item is [–CI], then it denotes something which is spatially or temporally homogeneous or internally undifferentiated.

   • [+/−Scalar]: this feature signals the relevance of a range of values to a conceptual category. With respect to [–dynamic] situations it signals the relevance of gradability. Those situations for which a scale is conceptually possible will have the feature [+scalar]. Those situations for which a scale is impossible will be [–scalar]. With respect to substance/thing/essences, the feature [scalar] will signal the relevance of size or evaluation. This will be the feature which characterizes augmentative/diminutive morphology in those languages which display such morphology.

Example (1a) illustrates a skeleton that has a single feature as its function, say [+material] and a single argument; for example, the simple noun dog ([+material ([ ])]. Example (1b) is a more complex skeleton in which one set of features and arguments is embedded under another. We will see examples of such skeletons in some detail below.

It is assumed in this framework that the features in (2) are ones that are useful for characterizing English, but that they might not be either necessary or sufficient for other languages. In Lieber (Reference Lieber, Lieber and Štekauer2009), I propose that there might be a universal inventory of features from which individual languages choose to compose their skeletons, with the remaining features being relegated to the semantic body. The semantic body, then, would be composed partly of the universal but syntactically inactive features in any given language and partly of random bits of encyclopedic information that are likely to differ from one language to the next and, indeed, from one speaker to the next. For example, the simplex word cat is illustrated in (3), where the features in angle brackets are the syntactically inactive features and the items without brackets are meant to represent the flotsam and jetsam that constitute the remainder of the body.

1. (3) cat

   [+material ([ ])]

   <+animate>

   <−human>

   has fur

   four legs

   tail

   meows

   domesticated pet

   uses litter box

   kills mice

   coughs up hair balls

Note that the framework of Lieber (Reference Lieber2004) follows Higginbotham (Reference Higginbotham1985) in assuming that nouns have arguments just as verbs do, the typical argument being what is referred to as the “R” or referential argument.

In the lexical semantic framework, bound morphemes may have skeletons just as free morphemes do, although they may have little or no content to their bodies. I illustrate affixal skeletons with the skeleton for the suffix -er in English, which forms agent nouns like writer, instruments like printer, and a variety of other sorts of non-event nominals.

1. (4) Skeleton for -er

   [+material, dynamic ([ ], <base>)]

Affixal skeletons must be referentially integrated with their bases, a process which is effected by the Principle of Coindexation (Lieber Reference Lieber, Lieber and Štekauer2009).

1. (5) Principle of Coindexation:

   In a configuration in which semantic skeletons are composed, coindex the highest non-head argument with the highest (preferably unindexed) head argument. Indexing must be consistent with semantic conditions on the head argument, if any.

The composed skeleton for the complex word writer is given in (6).

1. (6) [+material, dynamic ([_i], [+dynamic ([_i], [ ])])]

             -er                             write

The suffix -er has no semantic restrictions on its R argument, so the Principle of Coindexation coindexes this argument with the highest argument of the verb write. Indexation is what accounts for the tendency of words derived in -er to express the thematic role of the subject of their base verb (but see Lieber Reference Lieber2004 for a detailed analysis of the polysemy of -er in this framework).

There is obviously a great deal more to be said about the details of this framework, but this brief overview will be enough to allow us to look at compounding. The interested reader is referred to Lieber (Reference Lieber2004, Reference Lieber2006, Reference Lieber, Lieber and Štekauer2009, Reference Lieber, Scalise and Vogel2010, forthcoming) for further details.

3.2 Kinds of compounds

Scalise and Bisetto (Reference Scalise, Bisetto, Lieber and Štekauer2009) give us a useful classification with which to organize our discussion of English compounds in section 3.3. They divide compounds into three main categories – subordinate, coordinate, and attributive – each of which can in turn be divided between endocentric and exocentric types. Subordinate compounds are those in which one element of the compound bears an argumental relation to the other, as we find in English synthetic compounds like truck driver, cost containment, or city employee, or in Romance compounds like Italian lavapiatti (lit. ‘wash dishes’, i.e. dish washer). The English examples are endocentric subordinates, the Romance variety exocentric, since the compound as a whole is not a hyponym of either of its elements. Coordinate compounds are those in which the two compounded elements bear a relationship like ‘and’, ‘or’, or ‘between’ to each other, as we find, for example, in scholar-athlete, pass-fail, or parent-child (in parent-child relationship). Of these, the last is arguably exocentric, as parent-child denotes neither a kind of parent nor a kind of child. The final category of compounds is attributive, which Lieber (Reference Lieber, Lieber and Štekauer2009) concludes is a default category. In attributives, one member of the compound bears a relationship of loose modification to the other. So-called root or primary compounds in English are attributives in this classification, with compounds like dog bed or file cabinet being endocentric exemplars and compounds like air head or hard hat (when it denotes a construction worker) being exocentric.

3.3 The analysis of compounding in the lexical semantic framework

Whereas affixation in the lexical semantic framework involves the subordination of a base skeleton to an affixal skeleton, compounding involves the composition of bases (which can themselves be either simple or complex) without subordination of skeletons. Depending on the characteristics of the two skeletons, subordinate, coordinate, or attributive interpretations can emerge. What we will see is that subordinate compounds are those that involve indexation between one element of the compound and an argument of the base verb from which the other element is derived. Coordinate interpretations arise when the lexical semantic representations of head and non-head are very similar in terms of both skeleton and body. Attributive compounds then are a sort of default, where the modification relationship arises as a way of resolving the semantic incommensurability of the two bases.

3.4 Exocentricity

So far I have looked exclusively at types of compounds in English which receive endocentric interpretations. English, of course, also displays exocentric compounds, although as Bauer (Reference Bauer, Scalise and Vogel2010) argues, there are a number of different ways in which compounds can be exocentric; exocentricity is not a unitary phenomenon. Here I will look at three cases of exocentricity in English and discuss how the lexical semantic framework might handle them. My argument will be that nothing needs to be added to the framework as the core semantic interpretations of exocentrics can be represented within the theory just as they are in endocentrics, and the reasons for their exocentricity can be attributed to grammatical devices that are not specific to compounding.

Among the candidates for exocentric compounds in English are those in (23).

1. (23)

   a.	pickpocket, cutpurse
   b.	doctor-patient (as in a doctor-patient disagreement)
   c.	airhead, hardhat

Careful readers will note that the compounds in (23) include a subordinate type (23a), a coordinate type (23b) and an attributive type (23c). Bauer (Reference Bauer, Scalise and Vogel2010: 167) also lists items like black out or show off (as nouns) as possible candidates for exocentricity, but I will exclude them here as they seem likelier to me to be the result of conversion from verb plus particle combinations than types of compounding.

I will start with the compounds in (23c). It has been argued that compounds like airhead are metaphorical or metonymic, and therefore should be treated in exactly the same way that we treat any other semantic extension of this sort. In other words, the argument is that some analysis of metaphor and metonymy is necessary for simplex words as well as complex words and compounds, and that whatever the analysis is for simplex items can extend to compound examples as well. So, just as pig can be used metonymically to refer to a messy person or gun to the person who wields the gun (e.g. a hired gun), the compound airhead can be used to refer to a person whose head is empty. Indeed, hardhat can be either literal (‘a helmet used by a construction worker’) or metonymic (‘a construction worker’), so we would not want exocentrics of this sort to have a different sort of analysis than endocentrics.

As for the coordinate example in (23b), let us note first that we are dealing with the possibility of both syntactic and semantic exocentricity. I will deal with the syntactic question first. Whether we must analyse doctor-patient in (23b) as an adjective because we find it modifying a noun is a matter of syntactic analysis rather than semantics: if syntacticians allow for the possibility that nouns can modify nouns, such examples would not display syntactic exocentricity at all. But this is an issue that is largely orthogonal to the question of semantic analysis. As for the semantic analysis needed for doctor-patient, we can say, as with the attributives, that the lexical semantic framework does not treat these any differently from endocentric coordinate compounds; that is, that the lexical semantic framework simply makes available the representations of the two items – doctor and patient – as it does with the members of other coordinate compounds and that the relationship between the two members is inferred on the basis of both encyclopedic information and context. If the compound is predicated of another noun like disagreement, an endocentric interpretation is ruled out. But in another context, doctor-patient could potentially be interpreted exactly as I analysed scholar-athlete above, in other words, as denoting a single person who is both doctor and patient at once. Again, since the same item has the potential for either an endocentric or an exocentric interpretation, we would not want radically different lexical semantic analyses for the endocentric and exocentric readings.

Our last case is that of subordinate compounds like pickpocket and cutpurse. Since English items of this sort are rare and all quite lexicalized, it is not possible to find an example analogous to hardhat or doctor-patient, that is, where either an exocentric or an endocentric interpretation is possible for the same item. However, I would argue that these do not require any extension to the lexical semantic framework either. Note that compounds like pickpocket are structurally similar to NDVCs like scrub woman or row boat. The latter are clearly endocentric (a scrub woman is a kind of woman, a row boat is a kind of boat), but structurally they consist of a verb (or a deverbal noun) plus a noun in an argumental relationship. The same can be said for pickpocket and cutpurse. The latter two, however, might arguably be treated on analogy to airhead and hardhat in the sense that they can be seen as metaphorical or metonymic extensions. Again, whatever accounts for metaphoric or metonymic semantic extension in simplex items can be said to account for these as well.

3.5 Conclusion

In this chapter I have largely stayed away from comparisons between the lexical semantic treatment of compounds and other frameworks. In closing, I will just briefly touch on this subject. The most important point is that no good analysis of compounds can afford to ignore compound semantics in all its richness and complexity. This would seem to be an obvious point, but it is unfortunately a point that still needs to be made. As an example, frameworks like Distributed Morphology (Harley Reference Harley, Lieber and Štekauer2009) or Borer’s exocentric model (Borer Reference Borer2013) still try to account for compounding within purely syntactic frameworks. Significantly, they stick to a much narrower range of semantic generalizations than what even a single language like English tends to exhibit. Those semantic generalizations include the following that I have tried to account for here:

• compounds include subordinate, coordinate, and attributive types;

• any of these might have either endocentric or exocentric exemplars;

• subordinate compounds allow for any argumental relation to be expressed between the head and non-head members of the compound and are not restricted to object or adjunct interpretations;

• subordinate compounds (even in English) include types other than the canonical synthetic compounds that are found in English;

• coordinate compounds also display a range of interpretations consistent with any conceivable relationship between compound members (simultaneity, mixture, opposite, etc.);

• attributives are unrestricted in the relationships that can be expressed between compound members.

In this chapter I have tried to show that the lexical semantic framework of Lieber (Reference Lieber2004, Reference Lieber2006, Reference Lieber, Lieber and Štekauer2009, Reference Lieber, Scalise and Vogel2010, forthcoming) can capture all of these facts. The challenge for other frameworks is to do so as well.

4.1 Theoretical framework

Semantics has not played an important role in Western linguistics since the Bloomfieldean period for the major part of the twentieth century. Different from this, Saussurean tradition and the tradition of the Prague School of Linguistics (e.g. Mathesius Reference Mathesius1961; Vachek Reference Vachek1976) teach us that form and meaning are inseparable and any linguistic analysis should take into account their interrelation. This view was given very strong support from Cognitive Linguistics. One of its founders, Langacker (Reference Langacker and Rudzka-Ostyn1988: 49) maintains that “a description of grammatical structure that makes no reference to meaning is no more revealing than a dictionary providing only a list of undefined forms”. Actually, Langacker repeats what was pointed out by Marchand as early as 1965 who, when criticizing Lees’ (Reference Lees1960) formalism, points out that “grammatical criteria alone are clearly insufficient to describe the semantic side of compounds” (Reference Marchand1965: 58) and concludes that a semantic description is indispensable.

As far as word formation is concerned, its first comprehensive and systematic description from an onomasiological perspective was provided by Dokulil in his ingenious (Reference Dokulil1962) work and in a series of follow-up articles. Dokulil, at least a decade before the emergence of generative word formation, laid theoretical foundations for a number of topics that came to play a central role in the mainstream, semasiologically oriented word formation, including productivity, derivational paradigms, the place of word formation in the system of linguistics, the differences between morphological and word formation analyses, word formation motivation, the internal form of a word, headedness, lexicalization, and the notion of word formation type. Dokulil influenced several generations of morphologists in the former Czechoslovakia as well as a number of significant morphologists in Central and Eastern Europe. The approach outlined here draws on some basic principles of Dokulilean onomasiological theory of word formation as elaborated into a theory of word formation by Štekauer (mainly Reference Štekauer1998 and Reference Štekauer, Štekauer and Lieber2005b).

The central claim of Štekauer’s approach concerns the triad of relations between the extralinguistic reality (object to be named),Footnote ¹ the speech community (represented by a ‘coiner’ of a new complex word), and the word formation component (word formation rules), thus emphasizing the fact, ignored by the vast majority of mainstream word formation theories, that each new complex word results from a specific act of naming which responds to a very real and specific naming demand on the part of a member (members) of the speech community. This means that this approach lays emphasis on the active role of language users in the process of giving names to objects instead of presenting word formation as an impersonal system of rules detached from the objects named and from language users. Thus, the naming act is not a purely linguistic act. Complex words do not come into existence in isolation from factors, such as human knowledge, human cognitive abilities, experiences, habits, traditions, and a whole range of sociolinguistic factors, including age (Štekauer et al. Reference Štekauer, Chapman, Tomaščíková and Franko2005; Körtvélyessy Reference Körtvélyessy2010), gender (Säily Reference Säily2011), education (Gleitman and Gleitman Reference Gleitman and Gleitman1970; Ryder Reference Ryder1994), and language background (Keune Reference Keune2012; Klembárová Reference Klembárová2012; Körtvélyessy and Štekauer Reference Körtvélyessy, Štekauer, Lieber and Štekauer2014). These circumstances of word formation cannot but have significant influence upon some basic questions of word formation such as productivity as well as the examination of the preferences of language coiners in regard to forming semantically transparent vs. formally economical complex words.

The onomasiological model of word formation which underlies my considerations in this chapter has already been described in Štekauer (Reference Štekauer1998, Reference Štekauer, Štekauer and Lieber2005b). Therefore, I will concentrate on the core of this model, in particular the interrelation between the onomasiological structure and the onomatological structure, because it is this part of the model which is crucial for an onomasiological approach to ‘compounds’. The onomasiological structure results from conceptual analysis of the object to be named; the onomatological structure reflects the competition between two general tendencies in any language – the tendency toward semantic transparency and the tendency toward formal economy.

4.2 Basic principles of an onomasiological approach to word formation

In order to understand the basic difference between the traditional (structuralist as well as generativist) models, on the one hand, and the onomasiological model of word formation, on the other, it is useful to highlight the fundamental principles of the latter. An onomasiological model is a dynamic model which aims to account for the way new words come into existence. It takes a speaker’s cognitive abilities as a point of departure. Second, it does not take binary analysis into immediate constituents as a dogma. Instead, it introduces a ternary onomasiological structure as its prototypical structure for the cognitive category of substance. An onomasiological structure, arrived at by conceptualization of a class of objects to be labelled by a linguistic sign, is constituted by the relations between semantic categories like Agent, Patient, Instrument, Location, Time, and Manner. One constituent, the onomasiological base, stands for a class to which the object to be named belongs. The onomasiological mark (modifier in the mainstream approach), which specifies the base, consists of the determining constituent and the determined constituent. The determined constituent always stands for the cognitive category of action in one of its three modifications (Action proper, Process and State) and is, as it were, the crucial constituent in terms of interpretation/predictability of the meaning of novel complex words,Footnote ² because it has the capacity to relate the other two constituents of the onomasiological structure. All in all, the onomasiological structure is a cognitive foundation for the act of naming.

For illustration, let us suppose that there is no name for a ‘person who writes novels (professionally)’. A conceptual analysis results in an onomasiological structure in which the semantic category of Action (writing) as a determined constituent of the mark relates the Agent (person who performs the Action of writing) and the Result of the Action (novel).

1. (1) Result – Action – Agent

At the onomatological (morphological) level, the onomasiological structure is morphologically represented in accordance with the Morpheme-to-Seme-Assignment Principle (MSAP): this principle ensures that the individual constituents (semantic categories) of the onomasiological structure are assigned morphemes whose semantic facet corresponds with the particular, cognitively founded, semantic category of the onomasiological structure. In other words, this operation is based on matching the meaning facet of a potentially assignable morpheme (stored in the Lexicon) with the respective semantic category of the onomasiological structure.

What is crucial is that the MSAP operates both syntagmatically and paradigmatically. Paradigmatically, it scans the Lexicon with regard to the lexical and affixal morphemes that can be retrieved from it in order to represent the semantic categories of the onomasiological structure. Syntagmatically, it reflects any and all restrictions (both semantic and formal) imposed on the combinability of the individual lexical and affixal morphemes. While there are a number of options for the morphological representation of a particular onomasiological structure (suffice to think of the number of affixes that can stand for the category of Agent in our example), general productivity of word formation rules (i.e. onomasiological types), fashionable trends, and individual preferences determined by age, gender, education, and language background narrow down the actual number of options from which a speaker can choose. In any case, the possibility to choose from several options indicates an active role of a coiner. From this it follows that no new complex word comes into existence from a blind, predetermined process. The opposite is true. This phenomenon is labelled as Creativity Within Productivity Constraints.

To get back to our example, the way of representing the onomasiological structure determines the nature of novel complex words. This brings us to the notion of onomasiological type – an onomasiological counterpart of the formally anchored Word Formation Rule. The onomasiological type is defined by the interrelation between two levels of word formation: the onomasiological level and the onomatological level. One option of how to represent the onomasiological structure of a ‘person who writes novels (professionally)’ is (2).

1. (2)

   Result –	Action –	Agent
   novel –	write –	er

In this onomasiological type, all three constituents of a ternary onomasiological structure are represented by a corresponding morpheme. Let us label this type as Onomasiological Type 1 (OT1).

In Onomasiological Type 2 (OT2), the determining constituent is not represented at the onomatological level, as in (3).

1. (3)

   Result –	Action –	Agent
   0	write –	er

In Onomasiological Type 3 (OT3), it is the determined constituent of the mark that is left unexpressed, as in (4).

1. (4)

   Result –	Action –	Agent
   novel	0	ist

It should be noted that these three onomasiological types represent different naming strategies reflecting two universal contradictory tendencies, in particular, the tendency toward economy of expression and the tendency toward semantic transparency, which also manifest themselves at other levels of language (cf. assimilation, elision, reduction of weak words at the phonic level; elliptical constructions, one-element sentences, two-element verbless sentences at the level of syntax); in fact, the whole system of word formation is a paramount manifestation of the economy principle because it may be viewed as a shortcut representation of descriptive phrases.

OT1, illustrated in (2), features high semantic transparency thanks to which the predictability of the meaning of a new complex word is high owing to the presence – at the onomatological level – of an Actional semantic category which unambiguously relates the morphemes standing for the base and the determining constituent of the mark, that is, between -er and novel. The high degree of semantic transparency is obtained to the detriment of formal economy: the complex word is formally not economical because all three constituents are morphematically represented. By implication, this kind of complex word is listener/reader-friendly but speaker/writer-unfriendly.

OT3 is speaker/writer-friendly, because it is shorter, but it may cause serious interpretation problems, and the predictability of the meaning of a new complex word of this type is much lower: the relation between novel and -ist may in principle be interpreted in a number of ways; for example, ‘a person who writes novels’, ‘a person who sells novels’, ‘a person who proofreads novels’, ‘a person who publishes novels’, ‘a person who loves to read novels’, and dozens of other potential interpretations. Hence, this onomasiological type is not listener/reader-friendly.

OT2 is somewhere between the two extremes. It is economical, hence speaker/writer-friendly, and, at the same time, it is partly predictable thanks to the presence of the morpheme representing the Actional constituent of the mark, which makes this type seemingly acceptable to the listener/reader. However, the problem is that this type usually produces complex words that are too general, and therefore it does not identify the named object of extralinguistic reality sufficiently. Writer can write novels, poems, letters, articles, blogs, etc. The meaning predictability is thus much lower than in the case of OT1.

These three basic (ternary) onomasiological types clearly indicate a close interrelation between word formation and word interpretation. In this chapter, the original system of onomasiological types (Štekauer Reference Štekauer1998) is elaborated on in order to provide a more fine-grained system. It is represented in (5) where R indicates that a particular semantic category of the onomasiological structure is represented by a corresponding morpheme.

1. (5)

   OT 1	DingM –	DedM –	Base
   	R	R	R
   Example:	Instrument –	Action –	Agent
   	guitar	play	er
   OT 2	DingM –	DedM –	Base
   	0	R	R
   Example:	Instrument –	Action –	Agent
   	0	play	er
   OT3	DingM –	DedM –	Base
   	R	0	R
   Example:	Instrument –	Action –	Agent
   	guitar	0	ist

The new, more fine-grained proposal of onomasiological types presented here deviates from the original version, where Onomasiological Type 4 (OT4) represented all complex words with a binary onomasiological structure. These are words in which it is not possible to distinguish between the determining and the determined constituents, and the mark functions as a single whole. In the new system, it is labelled as OT7 because it is placed among binary structure types.

Similarly, the original OT5 represented all types of conceptual recategorization (conversion in the traditional terminology) without recognizing considerable structural differences between them. Here, it is split into two types: one of them is based on a ternary onomasiological structure (labelled here as OT4), the other one is based on a binary onomasiological structure (labelled here as OT8).

OT4 is of the type of action-to-substance recategorization. This type ranks among the most economical of all onomasiological types because a ternary structure is represented by a single morphematic unit. For example, cheat is ‘a person who cheats (someone)’. The base and the determined constituent of the mark are represented by a single morpheme, as in (6).

1. (6)

   

OT5 employs the same feature as OT4, that is, joint representation of the base and the action category. However, unlike OT4, the determining constituent of the onomasiological structure is represented by a morpheme. For illustration, this type is exemplified with a complex word for ‘a person who cheats tourists’ in (7).

1. (7)

   

OT6 is another extremely economical onomasiological type. Neither the base nor the determined mark are expressed, as is the case with exocentric compounds like redskin, that is, ‘a person who has red skin’. If viewed from the perspective of the Surface Structure, this type might raise suspicion as the internal structure of complex words has been (in various modifications) aptly accounted for by what Kastovsky (Reference Kastovsky1982) labels as an identification-specification scheme where the head (onomasiological base) is first identified as a class of objects to be named, and subsequently the modifier (onomasiological mark) restricts the scope of the class. This also applies to OT6. The base is an inherent part of the onomasiological structure (‘person’ in our example), and the mark restricts its scope to a person ‘who has red skin’). The absence of the base and the determined constituent of the mark at the onomatological level do not contradict this scheme. Their absence can only be explained by an effort of a coiner at maximum possible economy of expression. By implication, the meaning predictability of this onomasiological type is poor. Anything can have a red skin, as evidenced by another meaning of this word, that is, ‘potato which has a red skin’.

1. (8)

   OT6	DingM –	DedM –	Base
   	R	0	0
   Example:	Quality –	State –	Patient
   	red skin	0	0

In OT7, the mark cannot be structured into the determining and the determined parts, which yields a binary onomasiological structure, including a mark and a base. Both base and mark are morphematically represented.Footnote ⁵

1. (9)

   OT7	Mark –	Base
   	R	R
   Example:	Negation –	Quality
   	un	happy

In OT8, the mark of a binary onomasiological structure is not expressed. This can be illustrated with an example of substance-to-action conversion such as ‘to bridge (something)’.

1. (10)

   OT8	Mark –	Base
   	0	R
   Examples:	Object –	Action
   	0	bridge
   	Manner–	Action
   	0	laze

Some of these onomasiological types are employed in an onomasiological description of the ‘compounding’ system in section 4.

4.3 An onomasiological analysis of compounds

4.4 Conclusions

This chapter aims to demonstrate the following.

1. 1. An onomasiological method provides us with a completely different perspective of the process of coining new complex words. It does not replace the semasiological method. Both of these methods should be perceived as complementary in the description of the system of complex words of a particular language.

2. 2. The onomasiological model enables us to view the formation of complex words as a very specific act of naming by a concrete language user who takes an active role in choosing one of several naming strategies represented by various onomasiological types.

3. 3. Word formation is a competition between two contradictory tendencies, the tendency toward economy of expression and the tendency toward semantic transparency, the former of them being speaker/writer-friendly and the latter listener/reader-friendly. The above overview indicates that these tendencies are differently implemented by individual onomasiological types which specify different degrees of both semantic transparency and economy of expression. However, meaning predictability, which is significantly conditioned by the onomasiological type underlying a specific complex word, represents a continuum within each onomasiological type. It is in this sense that we can speak about the scalar nature of the transparency-economy continuum.Footnote ¹⁰

4. 4. An inclination of a language user toward one of these two tendencies is influenced by the productivity of onomasiological types but also by sociolinguistic factors (age, education, profession, language background), the psycholinguistic factor of verbal vs. non-verbal type of personality, as well as fashionable trends in the field of word formation.

These observations can be represented by means of the onomasiological model of complex words in Figure 4.1 reflecting the interconnection between word formation and word interpretation with regard to naming strategies (represented by the selection of a particular onomasiological type); two basic tendencies in the formation of complex words; and the fundamental principles of forming new words from an onomasiological perspective. Finally, the model aims to show that a comprehensive theory of complex words must reflect the role of sociolinguistic and psycholinguistic factors without which it cannot be complete.

Figure 4.1 An onomasiological model of complex words