6.2.1.2. Non Derivable Meaning

First, she notices that many of the verbs in these constructions are not causative independently of the construction. As examples she presents kick and hit:

The verbs in these examples do not have a causative interpretation. Yet, when used in the caused-motion construction, a causality is implied:

It is also the case that many verbs do not code motion independently of this construction:

In addition, many transitive verbs in this construction do not have the same semantic relation to their direct object as when they appear in a simple transitive sentence. Compare the difference in meaning:

The sentences in (6.74) do not entail those in (6.75). Furthermore, there are verbs in this construction that do not license a direct object on their own. Contrast:

One possibility, instead of postulating the existence of the construction, would be to defend a novel causative motion sense for each of the verbs, but that would have the disadvantage of multiplying the number of lexical entries. Another possibility which would avoid this rampant lexical polysemy would be to derive the meaning of the whole sentence from the meaning of its constituent parts, in this case from the combination of the meaning of the verb and the directional preposition. Gawron (1985, 1986), Pustejovsky (1991) and Hoekstra (1992) have proposed slightly differing accounts of how the meaning of the caused-motion can be compositionally derived from the meaning of the parts.

According to Goldberg, a general problem with compositional accounts lies in the fact that their reasoning is based on a model of interpretation, without taking into consideration production. It might be possible that one could infer what examples (6.62) to (6.67) mean just from knowing what their constituent parts mean, but one could not predict that constructions of this type exist. In order to illustrate this general objection, she uses Makkai’s (1972) distinction between "idioms of encoding" and "idioms of decoding". Examples of decoding idioms are fly-by-night and by and large. The listener would not be able to infer the meaning without having learned the idiom separately. Examples of encoding idioms are serial killer and sofa bed. The meaning might be predictable from its constituent parts, but there is no way that a speaker could know that they are possible in the language without previous exposure to them.

According to Goldberg, Gawron’s (1985, 1986) compositional account argues that in caused-motion sentences there are two predicates, the verb and the preposition, and both retain their normal meanings. The relation between them, if not determined by the semantics of the verb, can be pragmatically inferable from the possible relations that hold between predicates in a single clause. The notion of "co-predication" is introduced as a new way for complements to be semantically conjoined with verbs. The verb and the preposition act as co-predicators, combining semantically in inferable ways and sharing one argument.

Pustejovsky (1991) offers a similar account suggesting that the verbs involved are lexically transitive process verbs that are combined with independent PPs which are associated with their own event structures. They are states in this case. The composition of process plus state gives a transition (accomplishment) interpretation. A problem for this, according to Goldberg, is the existence of combinations of process plus state that are not interpreted as transitions, as in the case of the well known depictive predicates or standard PP adjuncts:

These are not given the interpretation that as a result of the sleeping or playing the subjects ended under the bridge or in the house, as Pustejovsky’s account would imply.

A further problem for Pustejovsky’s and also Gawron’s accounts, is the existence of intransitive verbs, such as sneeze above, that appear in this construction with a direct object. They would need to postulate once more a three-argument sense for such verbs, going back to the kind of polysemy that they sought to avoid with their compositional accounts.

Hoekstra (1992) deals mainly with the type of resultative found in (6.75) and (6.76) above, and in:

In these examples the verb and the direct object do not bear their usual relation (the soap is not being washed, for instance). Hoekstra analyzes these sentences as involving an intransitive process verb combined with a small clause stative predicate. The problem is how to join the small clause with the verb without postulating a new lexical entry for the verb.

The final possibility Goldberg considers is to attribute the meaning of the caused-motion construction to the preposition. Thus, the directional meaning of the preposition would be responsible for the motion meaning of the whole. The major problem for this is that many prepositions which appear in this construction favor a locative interpretation, as in:

It could be argued that these prepositions are systematically ambiguous in English with both a locative and a directional interpretation. Nevertheless, they are not ambiguous in all contexts. When fronted, only the locative interpretation is possible:

An account relying on the ambiguity of the prepositions would have to specify when and how the ambiguity arises.

It might be argued that the motion component of the meaning is attributable to either the verb or the preposition. The necessary condition being that one must lexically specify motion. Goldberg rejects this possibility with the following counterexamples:

Neither the verbs nor the prepositions above independently code motion.

Goldberg concludes that since the caused motion interpretation cannot be attributed to the verb or the preposition or their combination, the need for a construction that combines the verb and the directional preposition to yield a particular, conventionalized interpretation is demonstrated.

Since the construction must contain a directional PP, examples with locative PPs, such as the ones above, pose a problem that Goldberg solves with a process of coercion (Croft 1991, amongst others), by which the construction coerces the locative into a directional reading. In Goldberg’s view coercion is not purely pragmatic. Instead, it is only licensed by particular constructions. In order for it to be feasible, moreover, a relationship between the inherent meaning and the coerced interpretation must exist. In this particular case, the locational and the directional meanings are related because the interpretation given to the locative is as endpoint of a path to that location.