Verb classes and alternations
We saw in the previous section how many psych-verb meanings (the meanings of verbs for mental states) can be realized by two variants, each of which selects different subject and object arguments: fear vs. frighten, like vs. please/appeal to, think vs. strike, and so on. Doublets like these, wide spread cross-linguistically, pose a problem for linking rules, since they allow quite different mappings from meaning to argument structure, only one of which is consistent with the thematic hierarchy. Because of these verbs, Culicover and Jackendoff (2005: 184) suggest that the the matichierarchy ‘does not apply to all combinations of thematic roles’.
In this section we will consider another problem-case for the standard account of argument structure and the thematic hierarchy: argument structure alternations. Many verbs show not just one argument structure, but several. Consider the verbs in (24). They can all appear in an array:
But, along with many other verbs, they can also appear in a so-called middle alternation, with as the only subcategorized argument:
Not all verbs allow the middle alternation. The verbs in (26), for instance, are impossible in the middle:
Are there any general principles governing which verbs allow the middle alternation and which don’t? Hale and Keyser (1987) use an archaic English verb, gally, to suggest an answer. If you’re like me, gally isn’t always on the tip of your tongue; in fact, this may well be the first time you’ve come across the word. Imagine two English speakers hearing the sentence The sailors gallied the whales for the first time (gally is a word particularly used in whaling jargon). One speaker might assume that gally means ‘see’, while the other might think it means ‘frighten’. Hale and Keyser point out that each assumption has certain consequences for what the two speakers will consider as possible alternations for the verb. If gally is assumed to mean frighten, then the middle alternation – the whales gallied easily – will be acceptable, whereas if it is thought to mean see, the middle alternation will not. This suggests that whether the middle alternation is possible depends on the verb’s meaning. (Gally means ‘frighten’, by the way, not ‘see’.)
What is the factor in the meaning that makes the difference? Hale and Keyser suggest that verbs which do allow the middle alternation all express the bringing about of a change of state in the verb’s object. This meaning is absent from the ones that don’t allow the middle, like see, consider, believe and notice.
We have, then, two hunches about the relation between semantics and argument structure:
• a verb’s meaning determines what syntactic alternations it participates in;
• verbs fall into semantically defined classes, which all show similar syntactic behaviour with respect to their alternations.
These ideas have been especially pursued by Levin and Hovav (Levin 1993; Levin and Hovav 1995, 2005).
Let’s continue our exploration of this by considering a couple of other alternations. The conative alternation is exemplified by each of the second sentences in (27):
As the term ‘conative’ (Latin ‘attempt’) suggests, this alternation conveys a reduced degree of effectiveness: (27a) suggests that the zombies directed the slashes at my face, but that their slashing didn’t succeed in putting me out of action completely – I could still slash back at them, perhaps. The reduced effectiveness is also true for the verbs in (27b–f).
As with the middle alternation, not all verbs allow the conative alter nation:
Now consider the body-part possessor ascension alternation. Many verbs can appear in this alternation:
QUESTION Can you find any regular meaning difference in the sentence pairs in (29a–f)?
QUESTION What other types of verb alternation can you think of?
If we examine (24) – (30) we can see a pattern. The verbs in these examples seem to cluster together in a way that allows us to predict what alternations they will appear in. Shatter and rip always behave the same as break: if break allows an alternation, so will the other two; if it doesn’t, rip and shatter won’t either.
The verbs in each class, Levin claimed, pattern in exactly the same way with respect to the middle, conative and body-part possessor ascension alternations (subject, of course, to dialectal differences). The patterns can be summed up in Table 10.2 (Levin 1993: 7).
In the spirit of Hale and Keyser’s discussion of gally, Levin suggested that the differences between these classes are basically semantic: what alternations a verb participates in is explained by its underlying semantic structure. These differences can be revealed by decomposing the verb’s meaning into a set of basic sub-events, involving primitives such as CAUSE, ACT, BECOME, like the Jackendoffian decompositions already discussed in 10.1.4. The particular way in which these subevents are present in the meaning of any given verb is known as that verb’s event structure.
We have already seen the meaning difference hypothesized to explain the middle alternation: cut and break verbs, which do manifest the alternation, include the idea of a change of state being brought about. We could represent their event structure as follows:
Hit and touch, on the other hand, do not decompose into an underlying change of state structure.
In the middle alternation, the cut and break verbs lose the idea of causing anything to happen, and jettison the ‘Agent’ argument as a result. This just leaves the ‘change of state’ idea intact:
Can we apply similar reasoning to the conative alternation, manifested by hit and cut verbs? Following a suggestion by Guerssel et al. (1985), Levin proposes that verbs of these classes involve two ideas: contact, and motion preceding the contact. Break verbs, on the other hand, lack both components: one can break something without coming into contact with it, and without the presence of any preceding motion. For instance, I can break an appliance by not turning it off, and a glass can be shattered by a high-pitched sound. Touch verbs do include a contact component, but they lack a motion one. Levin suggests that both motion and contact are necessary for the conative alternation to be possible. Confirmation for this idea comes from the fact that verbs involving motion alone don’t allow the alternation either:
If both motion and contact are required to qualify a verb for the conative alternation, the incompatibility of pure motion verbs with this alternation is exactly what we would expect.
What about a possible semantic basis for the body-part possessor ascension alternation?
This alternation is available to all the verb classes except the break verbs. Again, contact is the relevant component: hit, cut and touch classes all involve a notion of contact, but break verbs, as we just saw, don’t.
Confirmation of this analysis of the semantics of the four verb classes comes from another alternation, the causative/inchoative alternation. As we have seen, the meanings of both cut and break involve a change of state. Cut verbs, however, also involve notions of contact and motion. This semantic difference is correlated with a syntactic difference; only break verbs participate in the causative/inchoative alternation:
The causative/inchoative alternation appears to be confined to pure change of state verbs. Since hit and touch aren’t change of state verbs, they don’t display the alternation:
The key to the understanding of this is that the with variant entails the locative variant, but not vice versa. That is, (39a) entails (39b), but (39b) doesn’t entail (39a):
The reason that the entailments hold in one direction only is that the two variants differ in the extent to which the object of the verb is affected. If Seth loaded the cart with hay, the default interpretation is that the cart is fully loaded, whether or not all the hay has been transferred. But if Seth loaded hay onto the cart, no such implication holds. Similarly, if Ruth sprayed the wall with paint, we understand that the wall was entirely covered, whether or not all of the paint was used up. But if Ruth sprayed paint onto the wall, we have no information either about whether the paint was used up, or whether the entire wall was covered.
The core of the lexical semantic representations proposed by Rappaport and Levin (1988) for the locative alternation are shown in (40a) and (b), corresponding to (39a) and (b). The ‘x’ variable refers to the subject, the ‘y’ to the hay, and ‘z’ to the cart:
The differing degrees of affectedness are correlated with the linking of arguments to the direct object position: whichever argument is direct object is understood as wholly affected by the verb. Assuming a general principle like that of Jackendoff, according to which the first argument of a BE function is coded as direct object, the differing object assignments fall out from the representations in (40): in (40a) the cart is the first argument of the main (COME TO) BE function; in (40b) the hay is.
Levin and Hovav (2005: 206) observe that this analysis is supported by some other facts about English. There are some verbs – putting verbs – which only appear in locative-type structures, and some – filling and covering verbs – which only allow structures like the with variant:
We have, then, a situation where verbs’ syntactic behaviour seems to divide them into a small number of semantically differentiated classes. Levin and Hovav (2005: 18) observe that verb classes are similar in status to natural classes of sounds in phonology and the elements of meaning which serve to distinguish among the classes of verbs are similar in status to phonology’s distinctive features. Furthermore, since these grammatically relevant facets of meaning are viewed as constituting the interface between a full-fledged representation of meaning and the syntax, most researchers have assumed that, like the set of distinctive features, the set of such meaning elements is both universal and relatively small in size.
This approach to alternations shares with Jackendoff the idea that a verb’s syntactic possibilities are derived from the nature of its underlying semantic representation. This allows a far more economical description of the grammatical facts: instead of coding each verb separately with a list of the possible syntactic alternations in which it can appear, the analysis simply derives these possibilities directly from the details of the verbs’ meanings.
