[PPT] - Grammar Formalisms Head-Driven Phrase Structure Grammar (HPSG) PowerPoint Presentation

SLIDE 1

Grammar Formalisms Head-Driven Phrase Structure Grammar (HPSG)

Laura Kallmeyer, Timm Lichte, Wolfgang Maier Universit¨ at T¨ ubingen 27.06.2007

HPSG 1

SLIDE 2

Outline

Feature Structure Basics The building blocks of HPSG Analyzing raising and control Analyzing long-distance dependencies

HPSG 2

SLIDE 3

Subsumption and Unification

2 6 6 6 4 attr1

val1

attr2

val2

...

attrn

valn

3 7 7 7 5

{<attr1,val1 >, <attr2,val2 >, ..., <attrn,valn >} subsumption ⊑ : A ⊑ B, iff if t ∈ A, then t ∈ B. unification ⊔ : A ⊔ B = C, iff C is the smallest feature structure such that A ⊑ C and B ⊑ C.

HPSG 3

SLIDE 4

Feature structures as Values

Feature structures as values: non-recursive: 2

6 4index

2 4

per

1 num

sg

gend

fem 3 5

3 7 5

recursive:

" subcat

fi

» subcat h ... i –

fl

#

= ⇒ Other than FTAG, HPSG uses recursive feature structures!

HPSG 4

SLIDE 5

Structure sharing (1)

Re-entrancies (or “structure sharing”): boxed numbers ( 1 , 2 , ...) indicate structure sharing within feature structures:

" attr1

1

attr2

1

# " attr1

1 val1

attr2

1

# » attr1

1

h

attr2

1

i

–

Re-entrancies produce cyclic structures!

HPSG 5

SLIDE 6

Structure sharing (2)

Correspondence between feature structures and graphs:

2 6 6 6 6 6 6 4 type1 attr1

1

attr2

1

2 4

type2 attr3

val

attr4

1

3 5

3 7 7 7 7 7 7 5

type1 type2 val

attr1 attr2 attr3 attr4 HPSG 6

SLIDE 7

Types and type hierarchy

Types: fixed bundles of attributes

2 6 6 6 4 index per

1

num

2

gend

3

3 7 7 7 5 2 6 4 synsem loc

1

nonloc

2

3 7 5

specified in the signature Type hierarchy: order on types ≈ attribute inheritance sign

" phon

phon

synsem

#

word

" phon

phon

synsem

#

phrase

2 6 4 phon

phon

synsem

dtrs

con-struc

3 7 5

specified in the signature

HPSG 7

SLIDE 8

Signatures, constraints, and grammars (1)

The signature Σ is a quadruple (in P&S,94): < types , type hierarchy , attributes , ((types × attributes) × types) > The signature licenses possible feature structures. (language independent) A constraint τ ∈ θ is based on Σ, but also contains variables, conjunction, disjunction, implication, negation ,... Other name: Principle. The constraints further restrict the set of well-formed feature structures. (language independent and dependent)

HPSG 8

SLIDE 9

Signatures, constraints, and grammars (2)

A grammar G is a signature Σ and a set of constraints θ. Each G-licensed feature structure must be type-resolved: each type is maximally specific. totally well-typed: exactly the legal attributes of a type must be present.

HPSG 9

SLIDE 10

Phenomena, Model, Formal Theory

from M¨ uller (2007)

Phenomenon: Ling. Objects Model: Feature Structures Formal Theory: Feature Descriptions models predicts licensed by the theory determines

HPSG 10

SLIDE 11

The architecture of linguistic feature structures

2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4 sign phon

phon

synsem

2 6 6 6 6 6 6 6 4

synsem local

2 6 6 4

local category

content

context

context 3 7 7 5

nonlocal

nonlocal 3 7 7 7 7 7 7 7 5

3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5

phon: The surface of a sign. synsem: The core of a sign. Contains syntactic and semantic information. local: Contains information local to the linguistic sign, such as agreement information nonlocal: Contains non-local references, e.g. for long-distance dependencies

HPSG 11

SLIDE 12

An example

2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4 word phon

˙

she¸ synsem

2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4

synsem local

2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4

local category

2 6 6 4

cat head

»

noun case

nom –

subcat

3

7 7 5

content

2 6 6 6 6 4

ppro index

1

2 6 6 4

ref per

3rd

num

sing

gend

fem 3 7 7 5 3 7 7 7 7 5

context

2 6 6 4

context backgr

8 < :

2 4 psoa reln

female

inst

1

3 5

9 = ; 3 7 7 5 3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5

nonlocal

. . . 3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5

3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5

HPSG 12

SLIDE 13

Content and Context

content: Contains semantic information

2 6 6 6 6 6 6 4 ppro index

1

2 6 6 4

ref per

3rd

num

sing

gend

fem 3 7 7 5

3 7 7 7 7 7 7 5

context: Contains pragmatic information

2 6 6 6 6 6 4 context backgr

8 > < > :

2 6 4 psoa reln

female

inst

1

3 7 5

9 > = > ;

3 7 7 7 7 7 5

HPSG 14

SLIDE 15

Abbreviations of feature structures (1)

2 6 6 6 6 6 6 6 6 6 6 6 6 4 synsem loc 2 6 6 6 6 6 6 6 6 6 4 cat 2 6 4head

»

noun case

case –

subcat 3 7 5 cont | index i 2 6 4 per

per

num

gend

3 7 5 3 7 7 7 7 7 7 7 7 7 5 3 7 7 7 7 7 7 7 7 7 7 7 7 5

NP i [per,num,gend][case]

2 6 6 6 6 6 6 6 6 4 synsem loc 2 6 6 6 6 6 4 cat

2 6 6 4

head

»

verb vform

vform –

subcat

D

synsem

E 3 7 7 5

cont

i

3 7 7 7 7 7 5 3 7 7 7 7 7 7 7 7 5

VP[vform]: i

HPSG 15

SLIDE 16

Abbreviations of feature structures (2)

" head

»

noun case

nom –

#

h

head

Noun[nom]

i 2 6 6 4cont

2 6 4

index

i

restr

»

rel arg1

1

– 3 7 5

3 7 7 5 " content

»

index i rel( 1 )

–

# " head

»

verb vform

fin –

#

h

head

Verb[fin]

i 2 6 6 6 6 4 cont

2 6 6 4

rel arg1

1

...

argn

n

3 7 7 5

3 7 7 7 7 5 h content

rel( 1 , ..., n )

i

HPSG 16

SLIDE 17

Abbreviations in action

2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4 word phon D walks E synsem | loc | cat 2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4 head " verb vform fin # subcat * 2 6 6 6 6 6 6 6 6 6 4 loc 2 6 6 6 6 6 6 6 6 6 4 cat 2 6 6 4 head " noun case nom # subcat

3

7 7 5 cont 2 4index " per 3 num sg # 3 5 3 7 7 7 7 7 7 7 7 7 5 3 7 7 7 7 7 7 7 7 7 5 + 3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5 3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5

2

6 6 6 6 6 6 4 word phon D walks E synsem | loc | cat 2 4 head Verb[fin] subcat D NP 1 [3sg][nom] E 3 5 3 7 7 7 7 7 7 5

HPSG 17

SLIDE 18

Phrases - Syntactic composition

= ⇒ Structures of type phrase have the feature dtrs.

2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4 phrase phon D Kim, walks E synsem | loc | cat " head Verb[fin] subcat

#

dtrs 2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4 head-comp-struc head-dtr 2 6 6 6 6 6 6 4 phrase phon D walks E synsem | loc | cat 2 4 head Verb[fin] subcat D NP[nom] E 3 5 3 7 7 7 7 7 7 5 comp-dtrs * 2 4phon D Kim E synsem NP[nom] 3 5 + 3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5 3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5

Tree notation:

" head Verb[fin] subcat

#

NP[nom]

C Kim

2 4 head Verb[fin] subcat D NP[nom] E 3 5

H walks

HPSG 18

SLIDE 19

Principles

The Word Principle

word ⇒ LE1∨ ... ∨ LEn

The Head Feature Principle

" synsem | loc | cat | head

1

dtrs | head-dtr | synsem | loc | cat | head

1

#

The Subcategorization Principle (binary version)

2 6 6 6 6 4 synsem | loc | cat | subcat D

2

E dtrs 2 4head-dtr | synsem | loc | cat | subcat

D

2 | 1

E

comp-dtr | synsem

1

3 5 3 7 7 7 7 5

HPSG 19

SLIDE 20

The Immediate Dominance Schemata (1)

The ID Principle Every headed phrase must satisfy exactly one of the ID schemata. ≈ phrase structure rules For example (from P&S,94):

Head-Subject Schema Head-Complement Schema Head-Subject-Complement Schema Head-Marker Schema Head-Adjunct Schema Head-Filler Schema

HPSG 20

SLIDE 21

The Immediate Dominance Schemata (1)

= ⇒ including Head Feature Principle and Subcategorization Principle! Head-Subject Schema:

" head

1

subcat

#

2

C

2 6 6 4 phrase head

1

subcat

D

2

E

3 7 7 5

H Head-Complement Schema:

2 4 head

1

subcat

D

2

E

3 5 2 4 head

1

subcat

D

2 , 3 ,..., n

E

3 5

H

3

C

...

C

n

C

HPSG 21

SLIDE 22

Raising and control (1) - A semantic distinction

(1) a. John seems to leave. b. John tries to leave. “seems”:

2 6 6 6 6 6 6 4 word cat

2 4

head

Verb[fin]

subcat

D

1 [3sg][nom], VP[inf,subcat< 1 >]: 3

E 3 5

cont

seem( 3 )

3 7 7 7 7 7 7 5

“tries”:

2 6 6 6 6 6 6 4 word cat

2 4

head

Verb[fin]

subcat

D

NP 1 [3sg][nom], VP[inf,subcat<NP 1 >]: 3

E 3 5

cont

try( 1 [ref ], 3 )

3 7 7 7 7 7 7 5

= ⇒ no PRO and no ECM! = ⇒ by and large: control and raising only differ semantically!

HPSG 22

SLIDE 23

Raising and control (2) - The other lexical entries

“to leave”:

2 6 6 6 6 6 6 4 word cat

2 4

head

Verb[inf]

subcat

D

NP 1

E 3 5

cont

leave( 1 )

3 7 7 7 7 7 7 5

“John”: 2

6 6 6 6 6 6 6 6 6 6 6 6 4 word cat

»

head

Noun[nom]

subcat

–

cont

2 6 6 6 4

index

1

2 4

per

3 num

sg

gend

masc 3 5

john( 1 )

3 7 7 7 5

3 7 7 7 7 7 7 7 7 7 7 7 7 5

HPSG 23

SLIDE 24

Raising and control (3) - The derivation

(2) John seems to leave.

" head Verb[fin] subcat

#

NP 1 [3sg][nom] C John

2 4 head Verb[fin] subcat D

2 [3sg][nom]

E 3 5

H

2 4 head Verb[fin] subcat D

2 [3sg][nom], VP[inf,subcat < 2 >]

E 3 5

H seems VP[inf,subcat <NP 1 >] C to leave

HPSG 24

SLIDE 25

Extraction - Traces

(3) Kim we know Sandy claims Dana hates . Trace (simplified version): 2

6 6 6 6 4 phon

synsem

2 6 4

local

1

nonlocal

»

slash

n

1

–

3 7 5

3 7 7 7 7 5

slash collects local-values of traces. local-values of traces and their fillers are structure shared.

HPSG 25

SLIDE 26

Extraction - Principle and ID Schema

Nonlocal Feature Principle (for head-comp-struc)

2 6 6 6 6 6 6 6 6 6 6 6 4 sysnem | nonloc | slash S

2 ... n

dtrs 2 6 6 6 6 6 6 6 6 4 head-comp-struc head-dtr | synsem | nonloc | slash

2

comp-dtrs * h synsem | nonloc | slash

3

i , ..., h synsem | nonloc | slash

n

i + 3 7 7 7 7 7 7 7 7 5 3 7 7 7 7 7 7 7 7 7 7 7 5

Head-Filler Schema (≈ P&S,94):

» synsem | nonloc | slash

2 \

n

1

–

h synsem | loc

1

i

F

2 6 6 6 4 loc | cat | head

»

Verb[fin] subcat

–

nonloc | slash

2

n

1 ,...

3

7 7 7 5

H

HPSG 26

SLIDE 27

Extraction - An example

(4) Kim we know Sandy claims Dana hates .

S

h nonloc|slash {} i

NP

h loc

1

i

Kim S

» nonloc|slash n

1

–

NP we VP

» nonloc|slash n

1

–

V know S

» nonloc|slash n

1

–

NP Sandy VP

» nonloc|slash n

1

–

V claims S

» nonloc|slash n

1

–

NP Dana VP

» nonloc|slash n

1

–

V hates NP

2 4local

1

nonloc | slash n

1

3

5

HPSG 27

SLIDE 28

Extraction - Ways to model islands (1)

1.) Finite sentences with complementizer (that, whether) (5) *Whoi did the elephant whisper that the emu saw

i ?

Whoi did the elephant say that the emu saw

i ?

Non-Bridge Verbs (whisper): specify their sentential complement having a empty slash-set:

" loc | head | subcat fi ..., h nonloc | slash

{}

i ,... fl #

Brigde Verbs (say): don’t specify the sentential complement’s slash-set.

HPSG 28

SLIDE 29

Extraction - Ways to model islands (2)

2.) Subjects from finite sentences with complementizer

(6) *Whoi did Alice say that

i left.

Whoi did Alice say

i left.

Trace Principle: synsem values of traces must be a noninitial member of a (substantive head’s) subcat list. implementation within the Head-Subject Schema:

" head

1

subcat

#

2

h nonloc | slash

{}

i

C

2 6 6 4 phrase head

1

subcat

D

2

E

3 7 7 5

H

HPSG 29

SLIDE 30

Extraction - Ways to model islands (3)

2.) Subjects from finite sentences with complementizer

(7) *Whoi did Alice say that

i left.

Whoi did Alice say

i left.

Subject Extraction Lexical Rule (SELR):

» synsem | loc | cat | head | subcat

D

...,S[unmarked],...

E

–

⇓

2 6 6 6 6 6 6 4 synsem 2 6 6 6 6 6 6 4 loc 2 6 6 4cat | head | subcat * ...,VP 2 6 4 loc | head | subcat fi h loc

1

i fl nonloc | slash {} 3 7 5 ,... + 3 7 7 5 nonloc | slash n

1

3

7 7 7 7 7 7 5 3 7 7 7 7 7 7 5

HPSG 30

SLIDE 31

Extraction - Ways to model islands (4)

3.) Adjuncts

(8) *[Which movie]i did Gorgette fall asleep after watching

i.

Head-Adjunct Schema:

" head

1

subcat

2

# " loc | cat | head | mod

3

nonloc | slash

{}

#

A

3

2 4loc | cat " head

1

subcat

2

# 3 5

H

HPSG 31

SLIDE 32

The lexicon

Lexical items are encoded as AVMs, such as the one we have already seen:

2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4 word phon D she E synsem 2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4 synsem local 2 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4 local category 2 6 6 6 6 4 cat head " noun case nom # subcat

3

7 7 7 7 5 content 2 6 6 6 6 6 6 4 ppro index

1

2 6 6 6 6 4 ref per 3rd num sing gend fem 3 7 7 7 7 5 3 7 7 7 7 7 7 5 context 2 6 6 6 6 6 4 context backgr 8 > > < > > : 2 6 6 4 psoa reln female inst

1

3 7 7 5 9 > > = > > ; 3 7 7 7 7 7 5 3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5 nonlocal . . . 3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5 3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5 HPSG 32

SLIDE 33

Generalization in the lexicon (1)

“Vertical” generalizations: Make use of underspecification possible through clever design of the type hierarchy

HPSG 33

SLIDE 34

Generalization in the lexicon (2)

“Horizontal” generalizations: Lexical rules which “automatically” license lexical entries based on the present ones. Example: Passivization (Kiss, 1992):

2 6 6 6 4 stem cat

2 4

head

verb

subcat

D

NP[nom], NP[acc] 1

E ⊕ 2 3 5

3 7 7 7 5 → 2 6 6 6 6 4 word cat

2 6 4

head

h

vform

passiv-part i

subcat

D

NP[nom] 1

E ⊕

2

3 7 5

3 7 7 7 7 5

HPSG 34

SLIDE 35

Summary

For a linguistic theory following HPSG, we need at least the following: Signature

Type hierarchy Feature geometry

Principles,

general principles, restrictions on linguistic objects ID schemata, licensing the combination of lexical items

Lexicon:

Lexical entries (AVMs) Lexical rules (AVMs), licensing more lexical items

HPSG 35

SLIDE 36

References

Please see the reference list on the course web page at http://www.sfb441.uni-tuebingen.de/emmy/gf/references.pdf

HPSG 36