Transcript Lecture 11

CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
HPSGs
How its done,
Examples,
Examples,
Examples
Instructor: Nick Cercone - 3050 CSEB - [email protected]
1
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
HPSG Feature Structure Descriptions – recall for pizza’s
pizza thing
pizza
topping set
CRUST
TOPPINGS
vegetarian
OLIVES
ONIONS
MUSHROOMS
non-vegetarian
SAUSAGE
PEPPERONI
HAM
Instructor: Nick Cercone - 3050 CSEB - [email protected]
2
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Instructor: Nick Cercone - 3050 CSEB - [email protected]
3
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
A type hierarchy....
• ... states what kinds of objects we claim exist (the
types)
• ... organizes the objects hierarchically into classes
with shared properties (the type hierarchy)
• ... states what general properties each kind of object
has (the feature and feature value declarations).
Instructor: Nick Cercone - 3050 CSEB - [email protected]
4
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
• Pizza Descriptions and Pizza Models
• How many pizza models (by definition, fully
resolved) satisfy this description?
Instructor: Nick Cercone - 3050 CSEB - [email protected]
5
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Answer: 2
{<CRUST , thick> , <TOPPINGS , { <OLIVES ,
+ > , <ONIONS, +> , <MUSHROOMS, −>}>}
{<CRUST , thick> , <TOPPINGS , { <OLIVES ,
+ > , <ONIONS, +> , <MUSHROOMS, +>}>}
Instructor: Nick Cercone - 3050 CSEB - [email protected]
6
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Pizza Descriptions and Pizza Models
How many pizzas-in-the-world do the pizza
models correspond to?
Answer: A large, constantly-changing number.
Instructor: Nick Cercone - 3050 CSEB - [email protected]
7
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
‘type’/‘token’ distinction applies to sentences as well
Instructor: Nick Cercone - 3050 CSEB - [email protected]
8
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Combining Constraints
Instructor: Nick Cercone - 3050 CSEB - [email protected]
9
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Combining Constraints
Instructor: Nick Cercone - 3050 CSEB - [email protected]
10
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Combining Constraints
Instructor: Nick Cercone - 3050 CSEB - [email protected]
11
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Combining Constraints
Instructor: Nick Cercone - 3050 CSEB - [email protected]
12
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Combining Constraints
Instructor: Nick Cercone - 3050 CSEB - [email protected]
13
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
A new theory of pizzas
Instructor: Nick Cercone - 3050 CSEB - [email protected]
14
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Combining Constraints
Instructor: Nick Cercone - 3050 CSEB - [email protected]
15
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Identity Constraints (tags)
Instructor: Nick Cercone - 3050 CSEB - [email protected]
16
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Combining Constraints
Instructor: Nick Cercone - 3050 CSEB - [email protected]
17
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Note
Instructor: Nick Cercone - 3050 CSEB - [email protected]
18
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Combining Constraints
Instructor: Nick Cercone - 3050 CSEB - [email protected]
19
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Why combine constraints?
• The pizza example illustrates how unification can
be used to combine information from different
sources.
• In our grammar, information will come from
lexical entries, grammar rules, and general
principles.
Instructor: Nick Cercone - 3050 CSEB - [email protected]
20
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Linguistic Application of Feature Structures: Making the
Mnemonic Meaningful
What do these CFG categories have in common?
NP & VP:
are both phrases
N & V:
are both words
NP & N:
are both ‘nouny’
VP & V:
are both ‘verby’
Instructor: Nick Cercone - 3050 CSEB - [email protected]
21
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
The Beginnings of our type hierarchy
Instructor: Nick Cercone - 3050 CSEB - [email protected]
22
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
A Feature for Part of Speech
Instructor: Nick Cercone - 3050 CSEB - [email protected]
23
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Type Hierarchy for Parts of Speech I
Instructor: Nick Cercone - 3050 CSEB - [email protected]
24
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Type Hierarchy for Parts of Speech II
Instructor: Nick Cercone - 3050 CSEB - [email protected]
25
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
A Feature for Valence
Instructor: Nick Cercone - 3050 CSEB - [email protected]
26
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Underspecification
Instructor: Nick Cercone - 3050 CSEB - [email protected]
27
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Another Valence Feature
Instructor: Nick Cercone - 3050 CSEB - [email protected]
28
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
SPR and Verbs
Instructor: Nick Cercone - 3050 CSEB - [email protected]
29
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
S and NP
• We created a monster
• our creation of a monster
Instructor: Nick Cercone - 3050 CSEB - [email protected]
30
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Type Hierarchy So Far
Instructor: Nick Cercone - 3050 CSEB - [email protected]
31
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Reformulating Grammar Rules 1
Which rules do these correspond to?
Instructor: Nick Cercone - 3050 CSEB - [email protected]
32
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Reformulating Grammar Rules 2
Instructor: Nick Cercone - 3050 CSEB - [email protected]
33
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Reformulating the Grammar Rules 3
Instructor: Nick Cercone - 3050 CSEB - [email protected]
34
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Advantages of the New Formulation
• Subject-verb agreement is stipulated only once
(where?)
• Common properties of verbs with different
valences are expressed by common features
• Parallelisms across phrase types are captured
Instructor: Nick Cercone - 3050 CSEB - [email protected]
35
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Disadvantages of the New Formulation
• We still have three head complement rules
• We still have two head specifier rules
• We only deal with three verb valences (which
ones? what are some others?)
• The non-branching rule does not really account
for any empirical work
• Others?
Instructor: Nick Cercone - 3050 CSEB - [email protected]
36
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Heads
Intuitive idea: A phrase typically contains a word that
determines its most essential properties, including
– where it occurs in larger phrases, and
– what is its internal structure
This is called the head
The term “head” is used both for the head word in a
phrase and for all the intermediate phrases containing
that word
• NB: Not all phrases have heads
Instructor: Nick Cercone - 3050 CSEB - [email protected]
37
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Formalizing the Notion of Head
• Expressions have a feature HEAD
• HEAD’s values are of type pos
• For HEAD values of type agr-cat, HEAD’s value
also includes the feature AGR
• Well-formed trees are subject to the Head
Feature Principle
Instructor: Nick Cercone - 3050 CSEB - [email protected]
38
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
The Head Feature Principle
• Intuitive idea: Key properties of phrases are
shared with their heads
• The HFP: In any headed phrase, the HEAD
value of the mother and the head daughter must
be identical.
• Sometimes described in terms of properties
“percolating up” or “filtering down”, but this is
just metaphorical talk
Instructor: Nick Cercone - 3050 CSEB - [email protected]
39
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
A Tree is Well-Formed if …
• It and each subtree are licensed by a grammar
rule or lexical entry
• All general principles (like the HFP) are satisfied.
• NB: Trees are part of our model of the language,
so all their features have values (even though
we will often be lazy and leave out the values
irrelevant to our current point).
Instructor: Nick Cercone - 3050 CSEB - [email protected]
40
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Question:
Do phrases that are not headed have HEAD
features?
Instructor: Nick Cercone - 3050 CSEB - [email protected]
41
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Instructor: Nick Cercone - 3050 CSEB - [email protected]
42
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Instructor: Nick Cercone - 3050 CSEB - [email protected]
43
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Reminder: Where We Are
• Attempting to model English with CFG led to
problems with the granularity of categories, e.g.
– Need to distinguish various subtypes of verbs
– Need to identify properties common to all verbs
• So we broke categories down into feature structures
and began constructing a hierarchy of types of
feature structures.
• This allows us to schematize rules and state
crosscategorial generalizations, while still making fine
distinctions.
Instructor: Nick Cercone - 3050 CSEB - [email protected]
44
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase
Structure Grammars
(HPSGs)
But it’s still not quite
right…
• There’s still too much
redundancy in the rules.
• The rules and features
encode the same
information in different
ways.
Instructor: Nick Cercone - 3050 CSEB - [email protected]
45
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Solution: More Elaborate Valence Feature Values
• The rules just say that heads combine with
whatever their lexical entries say they can (or
must) combine with.
• The information about what a word can or must
combine with is encoded in list-valued valence
features.
– The elements of the lists are themselves
feature structures
– The elements are “cancelled” off the lists
once heads combine with their complements
and specifiers.
Instructor: Nick Cercone - 3050 CSEB - [email protected]
46
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Complements
• This allows for arbitrary numbers of complements, but only
applies when there is at least one.
– Heads in English probably never have more than 3 or 4
complements
– This doesn’t apply where Head-Complement Rule 1 would.
• This covers lots of cases not covered by the old HeadComplement Rules 1-3.
Instructor: Nick Cercone - 3050 CSEB - [email protected]
47
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Question: What would the grammar change if English had
postpositions instead of prepositions?
Instructor: Nick Cercone - 3050 CSEB - [email protected]
48
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Specifiers - Head Specifier Rule (version 1)
• Combines the rules expanding S and NP.
• In principle also generalizes to other categories.
• Question: Why is SPR list-valued?
Instructor: Nick Cercone - 3050 CSEB - [email protected]
49
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
QUESTION: Why are these rightbranching?
That is, what formal property of our grammar forces the
COMPS to be lower inthe tree than the SPR?
Instructor: Nick Cercone - 3050 CSEB - [email protected]
50
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Another Question…
What determines the VAL value of phrasal nodes?
ANSWER: The Valence Principle
Unless the rule says otherwise, the mother’s values for the VAL
features (SPR and COMPS) are identical to those of the head
daughter.
Instructor: Nick Cercone - 3050 CSEB - [email protected]
51
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
More on the Valence Principle
• Intuitively, the VAL features list the contextual
requirements that haven’t yet been found.
• This way of thinking about it (like talk of “cancellation”) is
bottom-up and procedural.
• But formally, the Valence Principle (like the rest of our
grammar) is just a well-formedness constraint on trees,
without inherent directionality
Instructor: Nick Cercone - 3050 CSEB - [email protected]
52
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
So far, we have:
• Replaced atomic-valued VAL features with listvalued ones.
• Generalized Head-Complement and HeadSpecifier rules, to say that heads combine with
whatever their lexical entries say they should
combine with.
• Introduced the Valence Principle to “cancel”
things off the COMPS and SPR lists.
Instructor: Nick Cercone - 3050 CSEB - [email protected]
53
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
The Parallelism between S and NP
• Motivation:
– pairs like Chris lectured about syntax and Chris’s
lecture about syntax.
– both S and NP exhibit agreement
The bird sings/*sing vs. The birds sing/*sings
this/*these bird vs. these/*this birds
• So we treat NP as the saturated category of type noun and
S as the saturated category of type verb.
Instructor: Nick Cercone - 3050 CSEB - [email protected]
54
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Head Driven Phrase Structure Grammars (HPSGs)
Question: Is there any other reason to treat V as the
head of S?
• In standard English, sentences must have verbs.
(How about non-standard English or other
languages?)
• Verbs taking S complements can influence the
form of the verb in the complement:
I insist/*recall (that) you be here on time.
• Making V the head of S helps us state such
restrictions formally
Instructor: Nick Cercone - 3050 CSEB - [email protected]
55
CSE6339 3.0 Introduction to Computational Linguistics
Tuesdays, Thursdays 14:30-16:00 – South Ross 101
Fall Semester, 2011
Concluding Remarks
NAÏVE
Naive you are
if you believe
life favours those
who aren't naive.
Instructor: Nick Cercone - 3050 CSEB - [email protected]
56