We’ve concerned ourselves so far with making abstract and broad rules for language. In this chapter, we’re concerned with more specific things. In particular, what explains why certain words occur in certain environments. As we’ll discover below, a word’s syntactic environment (what it appears with) is highly idiosyncratic—within bounds. It’s still the case that the category ‘verb’ has a particular syntactic distribution, as does noun, complementizer, etc.

First a terminological note. We’ll be discussing verbs heavily in this chapter, but sometimes I’ll use the term predicate. Predicates are, loosely, what subjects are “about.” This notion of predicate groups together verbs, adjectives, and even prepositional phrases. So in the terminology defined below, verbs can “select,” adjectives can “select,” prepositions can “select,” etc.

C(ategory)-selection

Right now, our theory will generate an infinite number of sentences in any human language. But we’ve done very little to constrain our theory. For instance, nothing we’ve said so far will explain why (1) and (2) are not possible sentences in English.

The examples in (1) seem to be wrong for syntactic reasons (marked ungrammatical with “*”). The verb run doesn’t take an object and kiss must take an object.  On the other hand,  (2) seem to be wrong for semantic reasons (marked infelicitous with “#”). The verbs run and kiss  need an animate subject.

There’s nothing in our theory so far that directly accounts for these patterns. That is, there is nothing built into our theory that makes sure that a verb like run appears without an object and with an animate subject.

Here’s another example: Some verbs allow DP and CP complements, while others don’t.

There are some patterns that we’ve left implicit up until now, but if we want a coherent theory, we need to make everything as explicit as possible. What the above data teach us is that verbs make different requirements on what the rest of the structure can be. A verb like run dictates that there is only one other DP argument, and that argument is animate. A verb like kiss dictates that there are two other DP arguments, one of which is animate.

The idea that we’ll flesh out this week is that on top of our X’-schema, there’s a certain amount of information that is associated with each head, in particular, what it needs to combine with.  We call this information the selectional properties of the head. Selection comes in two “flavors.”

For example, we say about the verb run that it c-selects for a DP subject and it s-selects for an animate subject. This means that it must have a DP as a subject, and that DP must refer to an animate individual. We say about the verbs kiss and kick that they c-select for a DP subject and a DP object. Both verbs require a DP subject and object, and in both cases, only the subject DP needs to be animate. Fear c-selects for a DP subject and either a DP or a CP object, etc.

As a rule, c-selection is concerned with the syntactic category of the arguments of a predicate, like whether the object must be a CP or a DP. C-selection is also concerned with the number of arguments that a predicate has. For instance, eat and consume have different c-selectional properties: it’s possible for eat to only occur with a subject, while consume must occur with both a subject and an object.^[1]

In short: each word (really morpheme) dictates the syntax around it, both in terms of the category of arguments, and also the number of arguments.

S(emantic)-selection

Every predicate (really, every head) dictates what it can, and cannot combine with. So a verb like run cannot combine with a DP object, but a verb like kiss must combine with both a DP subject and a DP object.

However, sometimes things go wrong even when the c-selectional criteria are met. For instance, the following sentences sound incorrect, even though all the c-selectional requirements are met.

These sentences sounds off for semantic reasons. Trees don’t run and walls don’t punch. This suggests that in addition to the categorical information, verbs also dictate the “semantic features” of what they combine with. For instance, run and punch both need animate DP subjects. We call this requirement Semantic selection — it’s the semantic counterpart to c-selection.

When we talk about s-selection with respect to verbs, we’re typically (though not always) talking about thematic roles.

Theta-roles are ways to classify how each argument is meaningfully related to the predicate. For instance, the subject of both run and punch is a “doer.” It’s performing an action, thus needs to be animate. We call this role Agent. So the subject of both run and punch is an Agent, or has the thematic role of Agent. 

On the other hand, the object of punch isn’t performing an action, it’s affected by the action. We call this role Patient. So the object of punch has the thematic role of Patient. Every argument of a verb has a thematic role.

The point here is that, just like c-selection, different verbs have different requirements on what they s-select for.  Some verbs need Agents, some need Patients, some need Experiencers, etc. Notice that, just like with c-selection, order matters. It’s not just that punch needs to combine with an Agent and a Patient. The Agent must be the subject, and the Patient must be the object. We call this ordering of thematic roles the thematic structure of a verb.

So we say about the thematic structure of run that, “run has an Agent subject.” And we say about the thematic structure of punch that, “punch has an Agent subject and a Patient object.”

It’s important to keep in mind that  s-selection is a bit broader than just distribution of thematic roles. For instance, the following sentences sound off, but it’s not because the thematic structure is wrong (and it’s not because the c-selectional requirements haven’t been met).

The reason that (8) sound wrong is that gather and surround need a plural subject, while multiply needs a plural object. This is, again, an s-selectional requirement of these verbs, not a c-selectional requirement, because the ungrammaticality stems from a semantic “mistake,” not a problem with the category of the arguments.

Uniformity of Theta assignment hypothesis

If we spent a long time looking at thematic structure, we’d come up with a pretty striking correspondence between thematic structure and syntactic structure: Particular thematics roles are correlated with particular syntactic positions in a language. For instance, Agents are almost always subjects. Patients are almost always objects. Instruments are almost always in prepositional phrases or otherwise “oblique.” This old observation led to a hypothesis: the Uniformity of Theta Assignment Hypothesis or UTAH.

We’re going to assume for this class that UTAH is basically right — although we’ll revisit the question in a few chapters. In short, we’re going to assume that Agents always appear in the subject position (for now assumed to be specifier of TP). Patients always appear in the object position (complement to V). Instruments always appear in a prepositional phrase, etc.

Finally, we say that auxiliary verbs do not select for subjects. Their only requirement is that they select for a VP complement. Why? Well, the properties of a subject are never determined by the auxiliary, but rather by lower, main verb. In the examples below, the verbs start and cry determine what is an appropriate subject.

The observation that subjects are selected by the main verb, not an auxiliary, is actually extremely important, and it’s something that we’ll return to in chapter 10.

Where we’re going

One purpose of determining selectional properties is to discover classes of words. For instance, we can talk about the class of verbs that only select for a single DP: the intransitive verbs. This are distinguished from the class of verbs that only select for two DPs: the transitive verbs.

We can likewise talk about the class of verbs that take CP complements. It surely cannot be a coincidence that in no language does the verb kick combine with a CP, but basically every language allows say to combine with a CP. In the end, we’d like to come up with an explanation for why certain classes of verbs exist. But sometimes, this is impossible. For instance, eat and consume both mean, roughly, the same thing, but only one obligatorily c-selects for a DP object.

The second reason why it’s important recognize that heads dictate the syntax around them, is that we can use that information to diagnose movement. For instance, recall the topicalization constituency test.

Intuitively, we think that Susan has moved to the front of the sentence. Why? Because she’s who Rachel is hugging. Susan is the object of the verb. Implicit in this intuition is selection. We know (again, perhaps intuitively) that Susan is a Patient in this sentence, because hug is the kind of verb which dictates that its object is a Patient. Or stated differed, hug is the kind of verb that needs an object (due to c-selection), and so we understand that the DP Susan fills that syntactic role.

This is our first taste of phrasal movement, meaning, naturally, the movement of phrases. (As opposed to head-movement, which is the movement of heads, discussed in the previous chapter.) Because we know that heads dictate what they must and must not combine with, we conclude that the DP Susan must have combined with hug to satisfy the verb’s selectional requirement, and then moved to the front of the sentence.

Advanced