embedded clauses in tag
DESCRIPTION
Embedded Clauses in TAG. S. NP VP. V S-bar. S. COMP NP VP. We think that they have left. Embedded Clauses. Matrix Clause. Embedded Clause. The cat seems to be out of the bag. There seems to be a problem. That seems to be my husband. - PowerPoint PPT PresentationTRANSCRIPT
Embedded Clauses in TAG
Embedded Clauses
We think that they have left.
COMP NP VP
S
V S-bar
NP VP
S
Embedded Clause
Matrix Clause
How we know that the semantic role assignments are different with Seem and Try
• The cat seems to be out of the bag.
• There seems to be a problem.
• That seems to be my husband.
• The doctor seemed to examine Sam.
• Sam seemed to be examined by the doctor.
• The cat tried to be out of the bag.
• *There tried to be a problem.
• That tried to be my husband.
• The doctor tried to examine Sam.
• Sam tried to be examined by the doctor.
Raising to subject
It seems that they have left.
COMP NP VP
S
V S-bar
NP VP
S
They seem to have left.
COMP
VP
V VP-bar
NP VP
S
They seem to have left.
COMP
VP
V VP-bar
NP VP
S
They seem e to have left.
NP VP
V S
NP VP
S
Two ways to represent that “seem” and “leave” share a subject.
Subj theyVerb seemComplement subj verb leave
Comparison• Second method:
– Allow empty strings as terminal nodes in the tree. – An empty string needs to take the place of the missing subject of the
lower clause.– The empty string is linked to the subject of the main clause to show that
the main and embedded clauses share a subject.– The tree represents: word order, constituent structure, grammatical
relations, semantic roles.• First method:
– No empty strings in the tree. – The tree represents only word order and constituent structure.– Grammatical relations and semantic roles are represented in a separate
structure.– Structure sharing in the representation of grammatical relations shows
that the two verbs share a subject. • Is one method simpler than the other?
– No. Both methods have to represent word order, semantic relations, grammatical relations, and semantic roles.
• People who argue that one is simpler are usually wrong – they don’t know how to count steps in a derivation.
They try to leave.
COMP
VP
V VP-bar
NP VP
S
They(i) try PRO(i) to leave.
NP VP
V S
NP VP
S
Two ways to represent that “try” and “leave” share a subject.
Subj theyVerb seemComplement subj verb leave
PRO is an empty string, but not the same kind of empty string as e
Coindexing indicates that PRO refers to “they”.
“Seem” type verbs in TAG
S
John to be happy
V AP
NP VP V VP
VP
seem
Initial Tree
Auxiliary Tree
These trees represent the number of arguments for each verb:
“Seem” has one argument, represented as a VP.
“To be happy” has one argument, “John”.
Adjunction site
S
NP VP
to be happy
V AP
VP
John
V VP
VP
seem
Adjunction site
to be happy
V AP
VP
S
NP VP
John
V VP
VP
seem
Adjunction
V VP
VP
seems
to be happy
V AP
S
NP
John
This tree shows word order and constituent structure.
It also shows that “John” is the subject of “seem.”
It doesn’t show that “John” is the subject of “to be happy.”
“Try” type verbs in TAG
S
V S
NP VP
John tried
S
NP VP
TO VP
PRO leave
Adjunction site
These trees show the number of arguments for each verb:
“Try” has two arguments.
“Leave” has one argument.
Initial Tree
Auxiliary Tree
S
NP VP
TO VP
PRO leave
SAdjunction site
S
V S
NP VP
John tried
S
NP VP
TO VP
PRO leave
SAdjunction site
S
V S
NP VP
John tried
S
NP VP
TO VP
PRO leave
S
V
NP VP
John tried
Adjunction is only allowed at the top S node so as not to mess up compositional semantics:
After you put together “try to leave” you don’t want to have to take it apart again by inserting another verb like “expected” as in:
John tried to expect to leave.
Inserting “seem” into the middle of the tree doesn’t require you to disassemble any of the semantic pieces that were already assembled?