In part 1 I outlined what I take to be the primary motivations and diagnostics for phrase structure analysis. Here I take up some of the other main issues touched on in the reading, namely phrase structure rules, transformations, and relative clause formation. There is also a brief digression on flat vs. hierarchical structures.
Phrase-Structure Rules (pp. 155-159)
Phrase structure rules are simply statements about what constitutes a possible phrase structure. They can be language-specific or they can be stated more generally. Here we will only be concerned with phrase structure rules specific to English.
Each phrase structure rule makes a statement about immediate dominance and linear order. Dominance is a relation in a tree: a node dominates all the nodes below it in the tree; if you can trace a downward path through the branches of the tree from one node, say NP, to another node, say N, then that NP dominates that N. Immediate dominance is just what it sounds like: you only look at the node immediately below the one you start from. If there is some node between NP and N, then NP doesn't immediately dominate N.
Linear order is just the question of whether node precede or follow each other.
So, the following rule:
(1) NP -> D N
gives trees in which NP immediately dominates two nodes, D and N, and in which D precedes N. It cannot produce any other tree. However, since there are many words that can fill in for D and N, there are many noun phrases that can be built, e.g. the dog, some rabbits, a buffalo, etc. The following rule,
(2) NP -> D AP N
produces trees in which NP immediately dominates three branches, in a fixed order. One of those branches is phrasal, AP; we discussed two AP expansions in class:
(3) AP -> A
(4) AP -> Adv A
This generates one-word adjective phrases like raw and bloody, and two word adjective phrases like very raw and extremely bloody. Using rules (2) and (3) yields noun phrases like some raw rabbits, and using (2) in conjunction with (4) leads to noun phrases like an extremely bloody buffalo.
However, the rules discussed so far do not allow the noun phrase milk in a sentence like Buffalo don't drink milk, nor the noun phrase sweet milk in Sweet milk can be fermented. Thus we need two additional rules:
(5) NP -> N
(6) NP -> AP N
Rules (1), (2), (5), and (6) can be collapsed into a single line with the help of parentheses:
(5) NP -> (D) (AP) N
Since prepositional phrases and relative clauses are also optionally present in noun phrases, we can expand the rule further:
(6) NP -> (D) (AP) N (PP) (S')
Another convention that has been adopted is the use of an asterisk to indicate that a phrase is iterable, that is, it can be repeated. Since you can have as many adjectives and prepositional phrases in a noun phrase as you can be bothered to pronounce, we can expand the power of the rule further as follows:
(7) NP -> (D) (AP)* N (PP)* (S')
Notice that this rule produces very flat structures. I have argued in class that the structure of the noun phrase is actually not so flat. Hierarchical structures require different rules. We will explore those in depth at a later time. In class so far, we have explored a set of rules producing quite flat structures, including something like the following:
S -> NP (Aux) VP
S' -> C S
NP -> (D) (AP)* N (PP)* (S')
AP -> (Adv) A
PP -> P NP
VP -> V (NP) (PP)* (S')
While these are obviously inadequate, for example they do not allow sentence-level or VP adverbs at all, they do produce an impressive range of grammatical sentences, and just as crucially, they exclude a vast number of impossible ones.
Note that the expansion for NP includes PP, and the expansion for PP includes NP. This means that you can generate infinitely long structures of the form a picture of a man with a book about a painting of a battle between...
Similarly, the rule for VP includes S', the rule for S' includes S, and the rule for S includes VP. This gives rise to potentially infinite recursion of the type seen in Kim said that Andrea thinks that Leslie knows that Lee claimed that...
Flat versus hierarchical structures
Just as a brief preview of what is to come (cf. Aarts chapter 7), compare the two trees below and consider the different phrase structure rules needed to produce them:
The tree in (8) is produced by rules (7) and (3) above. However, the tree in (9) is not. It requires an additional rule:
(10) NP -> AP NP
Rule (10) is needed for the top level of (9), where NP immediately dominates AP and NP. The lower NP is produced by rule (7). But notice that rule (10) gives a bad result. It is now possible to produce noun phrases like *dark the night, which are not good noun phrases (check to make sure you understand why rule (10) produces this bad result). This means rule (10) must be wrong and the tree in (9) must not be the right tree.
This is a very important point so I will belabor it a little. The idea is that with the right phrase structure rules, we should be able to produce all the good noun phrases in English, but NONE OF THE BAD ONES. If (7) is the right phrase structure rule for English noun phrases, and if rule (10) does not exist, then a very strong claim is being made about possible English noun phrases; the claim is that they can never have any form other than the ones generated by the application of (7). Or, seen another way, (7) can be seen as a way to test English noun phrase structures; if they don't conform to (7), then they are no good.
Of course, (7) can't possibly really be right, but that's the basic idea behind phrase structure rules. They produce all and only the correct phrase structures. Now consider a more plausible hierarchical structure for the noun phrase dead dangerous animals, in (11).
Like (9), (11) requires a new rule; in fact, it requires three. They are stated in (12-14).
(12) NP -> N'
(13) N' -> AP N'
(14) N' -> N
This is the absolute minimum necessary to make the structure in (9) work; of course, they really should be fleshed out so that they can replace (7); for example, (12) should allow an optional determiner to the left of N'. Notice also that although (13) and (14) both give expansions of N', they cannot be collapsed, the way (1) and (2) and (5) and (6) could be.
This hierarchical structure is a better solution than the one in (9-10); it does not produce NPs like *dark the night. Make sure you understand why.
However, I have had to introduce three rules in order to make (11) work. That should make you skeptical. One advantage of the three rules in (12-14) is that they make it possible to simplify one thing from (7); I no longer need the "star" to let APs iterate. This is because of rule (13), which has N' (pronounced "en bar", with stress on the first syllable) on both sides of the rule. This is a straightforward case of recursion, like the examples noted above of NPs inside PPs inside NPs inside PPs. (Note too that (12) can be rewritten as NP -> (D) N', as I suggested just above, without falsely predicting that determiners should be iterative. This is because unlike the case in (13), the symbols on the left and right of (12) do not lead to recursion; NP is different from N', and N' doesn't contain D.)
We will return to these matters in depth later. Now I want to turn to the issue of surface structures, deep structures, and transformations.
Surface Structures, Deep Structures, and Transformations
As Finegan notes, the missing VP in the following sequence has only one possible reading.
(15) Mary won a prize but Alex didn't.
This can be captured by imagining that there is a VP there, with the same content as the one in the first half of the conjunct, at some abstract level of representation. Call this level Deep Structure; the level you pronounce is Surface Structure, and Deep Structure has to be uncovered through investigation (the meaning is not "deep" in the sense of 'profound,' but in the sense of 'beneath the surface').
So consider the Deep Structure of (15) might be (16).
(16) Mary won a prize but Alex didn't win a prize.
Now we can imagine a rule getting to (15) from (16). Why would we want to do such a thing? Well, for one thing, recall what I said above about phrase structure rules generating all and only the possible phrase structures. Consider the simple phrase structure rule for VP that we have discussed so far, given in (16):
(17) VP -> V (NP) (PP)* (S')
Notice that despite the fact that it collapses many different kinds of verb phrase into one compact rule, (17) cannot produce the VPs in (15). Make sure you understand why.
Perhaps you think there is no VP after Alex in (15). But consider what kind of structure you think there is; there are plenty of reasons to think that Alex is the subject of some kind of clause. For example, there is an auxiliary. According to the phrase structure rules we have seen so far, this can only be introduced by the rule expanding S:
(18) S -> NP (Aux) VP
In general, the material following the conjunct seems sentence-like. It can be made into a question, for example:
(19) Mary won a prize, but did Alex?
So, grant that there is a clause there, i.e. an S; but this suggests that it should contain a VP, and that there is something defective about the VP. If we have a rule deleting VPs, then we don't have to complicate the phrase structure rule in (17), or make VP optional in (18). We simply let the phrase structure rules generate (16), then we delete the VP. That it keeps the phrase structure rules simple is not a knock-down argument for adding a deletion rule to the grammar, but ultimately it turns out to be useful to adopt such rules.
Rules changing the output of phrase structure rules are called Transformations. Now, when I said above that phrase structure rules generate all and only the well-formed phrases, this will only apply to Deep Structures. Phrase structure rules generate Deep Structures. Sometimes Surface Structures look a lot like Deep Structures. Other times, Transformations apply, changing them.
Relative Clause formation
Relative clause formation is a good example of a transformation. The first thing to note about relative clauses is that they contain gaps, in places where you would expect to find noun phrases. The gap can be in subject position, in object position, or in some other position, for example complement of a preposition. In the following examples, I indicate the gap with an underscore.
(20) (I read) documents that _ revealed the names of the spies
(21) (I read) documents that you left _ on the table
(22) (I read) documents that somebody spilled coffee on _
English also allows relative clauses with wh-words as relative pronouns:
(23) documents which _ revealed the names of the spies
(24) documents which you left _ on the table
(25) documents which somebody spilled coffee on _
In each case, the gap is understood as if it contained something corresponding roughly to the relative clause head, which is the noun (or noun phrase) outside the relative clause.
Consider the question of whether the phrase structure rules we have looked at generate these noun phrases. The basic expansion needed here is given by (7) above, repeated here as (26):
(26) NP -> (D) (AP)* N (PP)* (S')
Assuming that the relative pronoun is a type of C, it is sister to S according to the rule for S', given in (27):
(27) S' -> C S
This means that every relative clause contains a sentence, which is expanded according to (18), repeated here as (28):
(28) S -> NP (Aux) VP
Now, it is already clear that (20) and (23) are not generated by the phrase structure rules given. You should check to confirm that (22) and (25) are not either, though (21) and (24) are. The reason that (21) and (24) are generated is because the phrase structure rules allow VPs with no objects; however, we allowed that option because of intransitive verbs, like walk and expire. (21) and (24) must be understood with a transitive meaning of left.
So, if we were to complicate the phrase structure rules in order to allow the relative clauses, we would have to add several options, and they would allow ungrammatical sentences like *Revealed the names of the spies and *Somebody spilled coffee on. Furthermore, we would have no representation for the gap which we perceive in the semantics of the relative clause.
The solution, then, is a transformation which takes ordinary sentence structures (Deep Structures) like the ones in (29-31), and turns them into the relative clauses in (20-22) or (23-25) (Surface Structures).
(29) (documents) revealed the names of the spies
(30) you left (documents) on the table
(31) somebody spilled coffee on (documents)
There are two ways to think of this, in terms of deletion of the word in parentheses, like the VP deletion rule mentioned earlier, or in terms of movement, where something moves out of the position of the gap. Here is a reason that makes some people prefer to think of relative clause formation in terms of movement:
(32) documents on which you spilled coffee _
Notice that the gap now corresponds to a prepositional phrase. The preposition itself shows up in between the relative clause head and the relative pronoun. We might imagine a Deep Structure like the one in (33):
(33) documents you spilled coffee on which
Relative clause formation would then be a rule selecting either which and moving it to C, giving (25) above, or selecting the PP on which and moving it to C, giving (33). That can be treated the same way, as a pronominal element starting in an NP position and moving to C.
Alternatively, you might prefer to think of the head noun, the word documents, as originating inside the relative clause. The Deep Structure for (22) might then look like the following:
(34) that you spilled coffee on documents
Here the word that is simply a complementizer. If this is the right Deep Structure, then the noun moves out of the relative clause entirely, producing (22). If wh-relative clauses are parallel, they might start out with the following form:
(35) you spilled coffee on which documents
But if that is correct, then two movements are needed: One which gets the noun out of the relative clause, as was needed for (34), and another that places which or on which in C, as was needed for (33).
In the weeks to come we will discuss relative clause formation and other transformations and how best to represent them.