Syntax – Week 3, Lecture 1
1
Phrase Structure Grammar
Over the next four weeks we look at phrase structure grammar in much
more detail. This week we start by examining and expanding our phrase
structure rules (PSR’s). By understanding their limitations, we extend our
phrase structure grammar to make it more comprehensive.
What do we want a PSG to be?
● Descriptive
a PSG should clearly represent the constituency information that is
hidden in sentences. That is, it should identify different word classes
and phrases.
● Generative
a PSG should be able to generate new grammatical sentences and
rule out ungrammatical ones.
1. An English example
1. Tali danced
2. The child danced
3. The tall child danced
4. Tali danced
5. Yesterday, the child danced
6. The boy danced
7. *The Tali danced
8. The child faced Tali
2. Substitution and movement
Recall the techniques of substitution and movement we’ve talked about
Syntax – Week 3, Lecture 1
2
previously:
● Substitution is used to identify word-level categories (N, V etc.)
● Substitution and movement is used to identify phrase level
categories (NP’s etc.)
Compare 1,2 from our corpus above (repeated below):
1. Tali danced
2. The child danced
Here, Tali can be swapped with The child. So they are both the same kinds
of constituents/phrases (spoiler: Noun Phrases).
Similarly, compare 6,2: boy and child are interchangeable
Compare 2,3: tall is optional
Compare 3,5: yesterday is optional
On the basis of this kind of evidence, we can sort our words into
categories, e.g. boy and child belong in the same category. And because
we know the meaning of these words, we can assign each word category
a meaningful label like ‘Noun’, ‘Verb’, etc.:
Noun = N = {boy, child}, Proper Noun = PN = {Tali}
Verb = V = {dance, face}, Determinative = Dtv = {the}
Adjective = Adj = {tall}, Adverb = Adv = {yesterday}
Syntax – Week 3, Lecture 1
3
3. PSR’s for sentences 1-8 above
By studying the position of the words and phrases in the sentences above,
we can identify the following preliminary Phrase Structure Rules (PSR’s):
S → (Adv) NP V (NP)
NP → {
Dtv (Adj) N
PN
}
Reminders:
• The curved brackets ( ) show an optional element. E.g. not all
sentences have an adverb at the start, but #5 does.
• The curly brackets { } show different possibilities, e.g. a NP can be
made up of either a Dtv (Adj) N , or just a PN.
• The PSR’s should generate all of the grammatical sentences in our
corpus, but none of the ungrammatical ones.
4. Evidence for a VP
Compare 2,8:
2. The child danced S → NP V
8. The child faced Tali S → NP V NP
danced is interchangeable with faced Tali – this is evidence for a phrase.
Additional Evidence: The ordinary coordination test:
The child faced Tali and danced
The child danced and gave Mary the roses
Also, we can use the variant of the pronominalization test:
The child faced Tali and Bob did too (i.e, he faced Tali)
Syntax – Week 3, Lecture 1
4
5. Updated PSR’s for sentences 1-8 above (incorporating the new VP):
S → (Adv) NP VP
NP → {
Dtv (Adj) N
PN
}
VP → V (NP)
6.1 Creating a Lexicon
1. Look at the data
2. Identify constituents by looking for minimal pairs
3. Examine morphology for patterns
6.2 Write the grammar
1. list word categories
2. rules for combining words into phrases and sentences
3. rules describing when to use inflections
4. predict a new grammatical sentence
7. PSG for sentences 1-8:
Lexicon
N = {boy, child}
PN = {Tali}
Adj = {tall}
Adv = {yesterday}
V = {dance-, face-}
Dtv = {the}
Morphology
Vmorph = {-s, -d}
Notes
If Adv is yesterday, then the
inflection of V must be -d.
Syntax – Week 3, Lecture 1
5
PSR’s
S → (Adv) NP VP
NP → {
Dtv (Adj) N
PN
}
VP → V (NP)
New Grammatical Sentence
Yesterday, Tali danced.
8. Labelled Bracketed Strings
The information above can be used to draw labelled bracketed strings
and/or phrase structure trees (PST’s). Both labelled bracketed strings and
PST’s represent the same information.
Labels are conventionally subscripts, placed to the bottom right of an
opening bracket:
[S [NP [Dtv the][Adj tall][N child] ] [VP [V saw][NP [PN Tali]] ]]
Alternatively, the labels can be placed after the closing bracket of the
constituent which they label:
[ [ [the]Dtv [tall]Adj [child]N ]NP [ [saw]V [ [Tali]PN ]NP ]VP ]S
A PST (phrase structure tree) showing the same sentence:
Syntax – Week 3, Lecture 1
6
Note: The bracketed strings combined with free online tree renderers is
the way I usually create these trees. The bracketed string above when
entered into the text box at http://ironcreek.net/phpsyntaxtree/ gives the
tree above:
9. Expanding the PSR’s for English
1. Copula Constructions
Tali was a clown
Tali was (very) tall
We see that some verbs (i.e., copula verbs such as be, seem, appear etc.)
can be followed by either a NP or an AdjP. As such, we need to modify our
VP rule:
VP → V {
(NP)
(AdjP)
}
2. Adjectives form the head of AdjP’s:
The alien
The spooky alien
The very spooky alien
The most important alien
http://ironcreek.net/phpsyntaxtree/
Syntax – Week 3, Lecture 1
7
The four phrases above are all NP’s headed by the noun alien. This noun is
modified by an adjective phrase headed by an adjective (spooky and
important). In turn, adjectives can be modified by adverbs like very, most,
more, etc.
AdjP → (Adv) Adj
This also means we have to update our NP rule:
NP → {
Dtv (AdjP) N
PN
}
3. Adjectives can be stacked:
The cat
The ugly cat
The ugly dead cat
These stacked adjectives can be represented notationally using a Kleene
star operator. The Kleene star means that the expression associated with
it can be repeated zero or more times. For example, the PSR’s to
represent the “the ugly dead cat” could be:
AdjP → (Adv) Adj*
NP → {
Dtv (AdjP) N
PN
}
But actually, this isn’t quite right. What’s really going on is that we’re
stacking AdjP’s, not adjectives. Consider:
Syntax – Week 3, Lecture 1
8
Analysis 1: Stacked Adjs Analysis 2: Stacked AdjP’s
“The very ugly dead cat” “The very ugly dead cat”
The second analysis is to be preferred here because the second adjective
can also be modified:
The [AdjP very expensive] [AdjP mostly complete] car.
We will return to these types of constructions later, but for now we can
rewrite our NP and AdjP rules as:
AdjP → (Adv) Adj
NP → {
Dtv (AdjP)* N
PN
}
4. Pronouns
Like proper nouns and
common nouns, pronouns
can also function as the
subject and object of a
clause:
Singular Plural
Subject Object Subject Object
1st I me we us
2nd you you you you
3rd masc he him they them
3rd fem she her they them
3rd neu it it they them
John nursed the small baby.
He nursed her.
Syntax – Week 3, Lecture 1
9
Updated NP rule:
NP → {
Dtv (AdjP)* N
PN
pro
}
5. Prepositional Phrases (PP’s)
PP’s have the following form:
PP → P NP
Like AdjP’s, PP’s can be stacked one after another:
She saw him in the library after lunch with her friends
The simplest analysis is to add the PP’s into the VP:
VP → V {
(NP)(PP*)
(AdjP)
}
However, consider the contrast in meaning between the following
sentences:
I studied the man [PP with concern].
I studied the man [PP with a bad haircut].
The information “with a bad haircut” says something about the man –
and so it should be part of the NP which contains “man”. The PP “with
concern” says something about the event more generally – and so, it
should appear with the VP:
Syntax – Week 3, Lecture 1
10
VP → V NP PP NP → Dtv N PP
Semantically this makes sense – but it also makes a prediction. In the first
sentence, “the man with concern” is not a constituent – it is two separate
phrases that happen to be next to one another (a NP and a PP). However,
in the second sentence, “the man with a bad haircut” is a constituent.
This means we would predict that “the man with a bad haircut” can be
moved to the front of the sentence, whereas “the man with concern”
can’t be:
The man with a bad haircut I studied.
*The man with concern I studied.
Also clefting:
It was the man with a bad haircut that I studied.
*It was the man with concern that I studied.
Given these two analyses, what would be the best analysis for:
I studied the man with a bad haircut in the morning.
Syntax – Week 3, Lecture 1
11
10. Recursion
Now our PSG can make a prediction. Note the following two rules (the NP
rule has been updated given our observations above):
PP → P NP
NP → {
Dtv (AdjP*) N (PP*)
PN
pro
}
A NP can have a PP inside it, and a PP can have a NP inside it. In other
words, NP’s and PP’s can generate each other. This is called recursion, and
our PSG predicts that English sentences can have this property (see tree
below).
In later weeks we will return to these PSR’s and further modify them to
reflect a deeper understanding of constituents in English sentences.
Syntax – Week 3, Lecture 1
12
11. Updated PSR’s for English
S → (Adv) NP VP
VP → V {
(NP)
(AdjP)
} (PP*)
NP → {
Dtv (AdjP*) N (PP*)
PN
pro
}
PP → P NP
AdjP → (Adv) Adj
Need to add lexicon:
N = {cat, courage, …}
Dtv = {the, a}
PN = {Tali, …}
PRO = {she, he, I, …}
V = {see, face, run, sleep, …}
P = {to, with, of, …}
Adj = {tall, dead, …}
Adv = {mostly, very, yesterday, …}
12. Limitations of PSR’s
1. Our PSR’s don’t include functional information (i.e., SUBJ, OBJ etc – who
did what to whom.)
We could fix this by including functional information in the PSR’s:
S → (Adv) NP VP
SUBJ
VP → V (NP) (PP)*
OBJ
Note: we don’t expect you to do this in exercises/assessment unless we
explicitly ask for functional information.
2. Syntactic Problems
How do we prevent sentences like:
*The child faced.
*The man slept a car in the garage.
*The man saw very tall.
Syntax – Week 3, Lecture 1
13
Solution:
We establish subcategories of verbs. Verbs are subcategorised according
to the frames they enter into. Roughly:
Vi verbs that appear in this VP frame: [______] (e.g., sleep)
Vt verbs that appear in this VP frame: [ ____ NP] (e.g., see)
Vdi verbs that appear in this VP frame: [ ____ NP NP] (e.g., give)
Vcop verbs that appear in this VP frame: [ ____ AdjP] (e.g., be)
3. Morphological Problems
How do we account for number agreement between verbs and subjects:
The man sees the elephant
*The man see the elephant
*The men sees the elephant
The men see the elephant
Solution: Include morphological information separately. (We saw this
earlier, where morphology was noted in a separate section of the Phrase
Structure Grammar.)
4. Semantic Problems
Finally, assuming that we get our subcategorisation and morphology
correct, how can we rule out syntactically correct, but semantically
meaningless sentences like those below:
#Colourless green ideas sleep furiously.
#Harry frightens greed.
#The rock kicked the horse.
Syntax – Week 3, Lecture 1
14
Solution: Selectional Restrictions
We add further information to the lexicon that details any semantic
restrictions that may co-occur with different phrases. For example, the
verb to frighten requires that the object of the verb be animate, because
inanimate entities can’t be frightened. This is represented by the
following lexical entry for frighten:
Word Class Subcat Frame Selectional Restrictions
frighten, V, [ ____ NP], NP ____ NP
[+animate]
Note that selectional restrictions are non-local, they can refer to any part
of the clause. For example, frighten needs an animate object, but not
necessarily an animate subject (e.g. The recession frightens Harry). Other
verbs may require an animate subject (e.g. Harry knew the answer, #The
tree knew the answer).