The meaning of constructions
The accounts of argument structure and alternations that we have looked at so far are sometimes described as projectionist. This means that they are ultimately based in the verb’s semantic representation, which ‘projects’ (determines) its syntactic behaviour (either directly, or via theta-roles). Projectionist accounts give the individual lexeme a central role in the explanation of the clause: the argument structure and the alternations associated with a verb are always the product of its semantic representation.
Projectionist accounts seem plausible as long as the verb only participates in a limited number of alternations. The locative alternation, discussed above for the verbs spray and load, is a good example of an alternation plausibly handled in a projectionist account. There is a basic variant, the locative variant, whose underlying representation forms part of the with variant, as in (43):
On this account, load has two separate lexical entries, one for each variant. Load is, in other words, polysemous (5.3).
Let’s now think about the verb siren. This can appear with a number of quite distinct argument structures:
In (44a) the verb is intransitive and denotes the emission of a sound. In (44b) it still denotes sound-emission, but has an object, which denotes not the sound emitted, but the time of day which the siren marks. In (44c) the verb’s object denotes the entity caused to stop by the sirening. In (44d) siren is again intransitive, but this time seems to be primarily a verb of motion, and only secondarily one of sound-emission.
Different syntactic complement structures like these are usually taken to reflect differences in the verb’s semantic representation: siren has to have a number of polysemous senses to account for the different structures in (44). But Goldberg (1995) points out that this approach leads to a blow-out in the number of senses which we attribute to verbs. Take sneeze. We usually think of this as an intransitive verb. But Goldberg notes that it’s ditransitive in the phrase sneeze the napkin off the table, suggesting an entirely separate, polysemous sense.
For Goldberg, this analysis is undesirable. Siren and sneeze should not be credited with a large number of senses, one for each different argument structure pattern they display. In order to avoid this, Goldberg develops a different conception of the nature of grammar. On this conception, words are not the only meaning-bearing units in grammar. Semantic representations are also associated with constructions. In the most general terms, constructions are form–meaning pairs, just like words. The difference is that the forms involved in constructions are on a higher level than individual lexemes: they are particular grammatical patterns, which individual lexemes instantiate. Some examples of constructions are given in Table 10.3 (Goldberg 1995: 3–4).
Each construction can be instantiated by a large number of lexical items. This is most obvious with the intransitive motion construction. As well as appearing with verbs of motion (go, come) and manner of motion (run, limp, hobble, slide, fall, drop), the construction can be instantiated by verbs of sound emission (buzz, siren), by verbs of bursting (burst, pop) and even by verbs such as sweat (the runners sweated up the hill). The construction is thus a broad general pattern in the grammar.
He pried it apart/open/loose/free. (Goldberg and Jackendoff 2004: 559)
It came apart/open/loose/free. (Goldberg and Jackendoff 2004: 559)
Pat sliced the carrots into the salad. (Goldberg 2006: 7)
Pat sliced the box open. (Goldberg 2006: 7)
The descriptions of the syntax of each construction are essentially specifications of argument structure. An important feature of a constructional account of the syntax–semantics interface is that arguments can be sub categorized by the construction itself. In (45), for example, the highlighted phrases are all brought into the structure of the clause by the construction: they are not subcategorized by the verb:
Note that all the sentences are grammatical without the highlighted phrases. This means that the highlighted phrases are adjuncts: the verbs do not obligatorily select them as part of their argument structure. It is important to see that adding the highlighted phrases doesn’t simply add an argument to the verb; it also changes the basic meaning of the sentence. On its own, sneeze is simply an intransitive verb denoting a bodily emission. But when it is plugged into the caused motion construction, the construction sup plies two extra argument slots, filled in (45a) by the napkin and off the table, and is paraphrased ‘she caused the napkin to move off the table’. Complementation patterns are therefore the joint product of verbs and the constructions in which they are placed.
In this respect, constructions are like idioms. Consider an idiom like take (someone) to task or the let alone idiom, as in I wouldn’t do X, let alone (do) Y. These idioms are listed in the lexicon with a syntactic structure, a meaning, and a partially filled phonology. For example, the lexical entry for take to task would specify the structure [take NP to task], and include the information that the NP must be human. The let alone idiom could be described as [V NP, let alone (V) NP], with the specification that the first V NP component must have a negative interpretation. (For discussion of let alone, see Fillmore et al. 1988; for comparison of idioms and constructions, Goldberg and Jackendoff 2004.) Constructions are like this too, except that they are even less specified lexically. The intransitive motion construction, for instance, just specifies the structure V PP, and imposes certain constraints on what types of verb and prepositional phrase may instantiate it (more on this below). The conative construction, again, just specifies the structure V at NP. Constructions are thus clausal/phrasal shells, waiting to be filled with lexical material.
As we have already seen, the important difference between a constructional and a traditional account of argument structure is that the constructional account reduces the proliferation of verb-senses. Sneeze has exactly the same semantic structure in (45a) as it does in its ordinary intransitive use (someone sneezed loudly, say). We do not have to list sneeze as polysemous between the basic sense sneeze1 ‘involuntarily emit burst of air as result of nasal irritation’ and a sneeze2 sense (‘cause to move by sneezing1’). Instead, the extra meaning, ‘x causes y to move z’, comes from the caused motion construction itself, which we only need to state once. Similarly, we do not have to postulate a different polysemous sense of slice in order to account for the different complement configurations in which it figures. Rather, exactly the same lexical entry of slice is operative in each of the contexts below; it is different constructions which contribute the different arguments, and the particular semantic interpretations:
Not all verbs can appear in all constructions. Take the caused-motion construction. We can sneeze a napkin off the table, but we cannot use or waste a napkin off the table: use and waste are not compatible with the caused-motion construction. In the same way, the intransitive-motion construction cannot be used with verbs of sense-perception (*Ann smelled/noticed/listened into the room) or some verbs of striking (*She hit/knocked into the room), among others.
How can we account for these constraints? As Goldberg (1995: 24) puts it, constructions don’t just impose their meaning on ‘unsuspecting’ verbs; a verb’s meaning determines whether it is compatible with a given construction. She gives two general conditions governing which verbs can appear in which construction. The conditions turn on the question of what Goldberg calls the ‘event type’ of the verb and construction – whether the verb/construction concerns motion, change of state, causation, and so on. Here are the conditions, with ec standing for the event type designated by the construction, and ev for the event type designated by the verb.
In (47a) and (47c) the verb denotes the means by which the construction’s event type arises. ((47a) is a caused-motion event type, (47c) resultative.) (47b) denotes the manner, or perhaps the means, in which the fly entered the room – buzzing – and (47d) denotes a subtype of the category ‘directed action’ – here, obviously, kicking. Constraints like these, Goldberg claims, determine the range of constructions in which a verb can appear.
The constructional approach to language has a wide application. Goldberg states that ‘[e]ven basic sentence patterns of a language can be understood to involve constructions. That is, the main verb can be understood to com bine with an argument structure construction (e.g. transitive, intransitive, ditransitive, etc.)’ (2006: 6). We will illustrate this with the ditransitive construction, which we have not yet discussed. The ditransitive construction is the following:
These sentences don’t, of course, satisfy the semantic description ‘x causes y to receive z’ since the object doesn’t receive anything. Nevertheless, they count as instances of the construction. The justification for this is that Goldberg sees constructions as basically polysemous: they do not have a single sense, but a family of closely related senses, arranged according to the same principles of prototypicality as other radial sense networks (on polysemy and radial networks, see 7.2). In the ditransitive construction, the core sense is ‘actual transfer’ (Goldberg 1995: 32). But many instances of the construction, such as all of those in (49), don’t entail that the recipient actually receives the theme. Further, the ‘transfer’ involved may not be literal, but metaphorical, as in Claude taught the class French.
Thinking of the ditransitive as a construction with its own polysemous set of meanings is a radical departure from projectionist accounts of grammar, in which different argument options are interpreted as reflecting differences in the lexical entries of verbs themselves. For Goldberg, everything in language can be seen as a construction – a form–meaning pairing in which the phonology may be more or less fully specified. From this point of view, individual lexemes are themselves a type of construction. For Goldberg, an uttered expression ‘typically involves the combination of at least half a dozen different constructions’ (2006: 10). The centrality of constructions to grammar is not, of course, limited to English. As an example of an obvious construction in another language, Goldberg (2006: 7–8; examples from Chidambaram 2004) discusses the Russian data in (50):
The constructions express the meaning that motion occurred, without the presence of any verb whatsoever. Cases like these provide clear evidence that certain form–meaning pairs need to be credited with meaning in their own right, and that it should not be the verb which is automatically assumed to be the kernel of the meaning of the rest of the clause.
