Feature structures and unification

216

description

Feature structures and unification. Attributes and values. Attributes and values. The following object describes a class of persons:. Attributes and values. The following object describes a class of persons: age 22 gender M nationality Norwegian. Attributes and values. - PowerPoint PPT Presentation

Transcript of Feature structures and unification

Page 1: Feature structures and unification
Page 2: Feature structures and unification

Feature structures and unification

Page 3: Feature structures and unification

Attributes and values

Page 4: Feature structures and unification

The following object describes a class of persons:

Attributes and values

Page 5: Feature structures and unification

The following object describes a class of persons:

age 22gender Mnationality Norwegian

Attributes and values

Page 6: Feature structures and unification

The following object describes a class of persons:

age 22gender Mnationality Norwegian

Attributes and values

Attributes

Page 7: Feature structures and unification

The following object describes a class of persons:

age 22gender Mnationality Norwegian

Attributes and values

AttributesValues

Page 8: Feature structures and unification

The following object describes a class of persons:

age 22gender Mnationality Norwegian

Attributes and values

Let this be the class of persons described:

Page 9: Feature structures and unification

The following object describes a class of persons:

age 22gender Mnationality Norwegian

Attributes and values

Let this be the class of persons described:

Page 10: Feature structures and unification

The following object describes a class of persons:

age 22gender Mnationality Norwegian

Attributes and values

Let this be the class of persons described:Then remove a feature...

Page 11: Feature structures and unification

The following object describes a class of persons:

age 22

nationality Norwegian

Attributes and values

Let this be the class of persons described:Then remove a feature...

Page 12: Feature structures and unification

The following object describes a class of persons:

age 22

nationality Norwegian

Attributes and values

Let this be the class of persons described:Then remove a feature...and the class grows.

Page 13: Feature structures and unification

The following object describes a class of persons:

age 22

nationality Norwegian

Attributes and values

Let this be the class of persons described:Then remove a feature...and the class grows.Add a feature instead...

Page 14: Feature structures and unification

The following object describes a class of persons:

age 22gender Mnationality Norwegianeyecolour brown

Attributes and values

Let this be the class of persons described:Then remove a feature...and the class grows.Add a feature instead...

Page 15: Feature structures and unification

The following object describes a class of persons:

age 22gender Mnationality Norwegianeyecolour brown

Attributes and values

Let this be the class of persons described:Then remove a feature...and the class grows.Add a feature instead...and the class shrinks.

Page 16: Feature structures and unification

The following object describes a class of persons:

age 22gender Mnationality Norwegian

A grammar example:

Attributes and values

Page 17: Feature structures and unification

The following object describes a class of persons:

age 22gender Mnationality Norwegian

A grammar example:

cat NPnumber sgperson 3

Attributes and values

Page 18: Feature structures and unification

The following object describes a class of persons:

age 22gender Mnationality Norwegian

A grammar example:

cat NPnumber sgperson 3

This object describes a class of phrases:

Attributes and values

Page 19: Feature structures and unification

The following object describes a class of persons:

age 22gender Mnationality Norwegian

A grammar example:

cat NPnumber sgperson 3

This object describes a class of phrases:

Attributes and values

a manthe horsesome red carthe King’s manwaternice beer...

Page 20: Feature structures and unification

The following object describes a class of persons:

age 22gender Mnationality Norwegian

A grammar example:

cat NPnumber sgperson 3

This object describes a class of phrases:Remove a feature...

Attributes and values

a manthe horsesome red carthe King’s manwaternice beer...

Page 21: Feature structures and unification

The following object describes a class of persons:

age 22gender Mnationality Norwegian

A grammar example:

cat NP

person 3

This object describes a class of phrases:Remove a feature...

Attributes and values

a manthe horsesome red carthe King’s manwaternice beer...

Page 22: Feature structures and unification

The following object describes a class of persons:

age 22gender Mnationality Norwegian

A grammar example:

cat NP

person 3

This object describes a class of phrases:Remove a feature...and the class grows.

Attributes and values

a manthe horsesome red carthe King’s manwaternice beer...

menthe horsessome red carsthe King’s menwatersnice beers...

Page 23: Feature structures and unification

cat NPnumber sgperson 3

f1:

Feature structures as functions

Page 24: Feature structures and unification

cat NPnumber sgperson 3

• A set of ordered pairs (of attributes and values)

f1:

Feature structures as functions

Page 25: Feature structures and unification

cat NPnumber sgperson 3

• A set of ordered pairs (of attributes and values)• Never more than one occurrence of a given attribute

f1:

Feature structures as functions

Page 26: Feature structures and unification

cat NPnumber sgperson 3

• A set of ordered pairs (of attributes and values)• Never more than one occurrence of a given attribute• Never more than one value of a given attribute (but different attributes can have the same value)

f1:

Feature structures as functions

Page 27: Feature structures and unification

cat NPnumber sgperson 3

• A set of ordered pairs (of attributes and values)• Never more than one occurrence of a given attribute• Never more than one value of a given attribute (but different attributes can have the same value)• Hence such a structure can be considered as a function from attributes to values

f1:

Feature structures as functions

Page 28: Feature structures and unification

cat NPnumber sgperson 3

• A set of ordered pairs (of attributes and values)• Never more than one occurrence of a given attribute• Never more than one value of a given attribute (but different attributes can have the same value)• Hence such a structure can be considered as a function from attributes to values

Example:

f1(cat)=NPf1(number)=sgf1(person)=3

f1:

Feature structures as functions

Page 29: Feature structures and unification

cat NPnumber sgperson 3

• A set of ordered pairs (of attributes and values)• Never more than one occurrence of a given attribute• Never more than one value of a given attribute (but different attributes can have the same value)• Hence such a structure can be considered as a function from attributes to values

Example:

f1(cat)=NPf1(number)=sgf1(person)=3

Values can be atomic or complex:

f1:

Feature structures as functions

Page 30: Feature structures and unification

cat NPnumber sgperson 3

• A set of ordered pairs (of attributes and values)• Never more than one occurrence of a given attribute• Never more than one value of a given attribute (but different attributes can have the same value)• Hence such a structure can be considered as a function from attributes to values

Example:

f1(cat)=NPf1(number)=sgf1(person)=3

Values can be atomic or complex:

f1:

agreement

cat NPnumber singularperson third

Feature structures as functions

Page 31: Feature structures and unification

Subsumption

Page 32: Feature structures and unification

Subsumption

Page 33: Feature structures and unification

cat NP

Subsumption

Page 34: Feature structures and unification

cat NP

agreement

cat NP

number singular

Subsumption

Page 35: Feature structures and unification

cat NP

agreement

cat NP

number singular

agreement

cat NPnumber singularperson third

Subsumption

Page 36: Feature structures and unification

cat NP

agreement

cat NP

number singular

agreement

cat NPnumber singularperson third

agreement

cat NPnumber singularperson third

subjectnumber singularperson third

Subsumption

Page 37: Feature structures and unification

cat NP

agreement

cat NP

number singular

agreement

cat NPnumber singularperson third

agreement

cat NPnumber singularperson third

subjectnumber singularperson third

agreement

cat NPnumber singularperson third

subject

1

1

Subsumption

Page 38: Feature structures and unification

Not subsumption

Page 39: Feature structures and unification

agreement

cat NP

number singular1

Not subsumption

Page 40: Feature structures and unification

agreement

cat NP

number singular

agreement

cat NP

person third

1

2

Not subsumption

Page 41: Feature structures and unification

agreement

cat NP

number singular

agreement

cat NP

person third

1

2

1 2, 2 1

Not subsumption

Page 42: Feature structures and unification

agreement

cat NP

number singular

agreement

cat NP

person third

agreement

cat NP

number plural

1

2

3

1 2, 2 1

Not subsumption

Page 43: Feature structures and unification

agreement

cat NP

number singular

agreement

cat NP

person third

agreement

cat NP

number plural

1

2

3

1 2, 2 1

1 3, 3 1

Not subsumption

Page 44: Feature structures and unification

agreement

cat NP

number singular

agreement

cat NP

person third

agreement

cat NP

number plural

agreement

cat NPnumber singularperson third

1

2

3

4

1 2, 2 1

1 3, 3 1

Not subsumption

Page 45: Feature structures and unification

agreement

cat NP

number singular

agreement

cat NP

person third

agreement

cat NP

number plural

agreement

cat NPnumber singularperson third

1

2

3

4

1 2, 2 1

1 3, 3 1

1 4

Not subsumption

Page 46: Feature structures and unification

agreement

cat NP

number singular

agreement

cat NP

person third

agreement

cat NP

number plural

agreement

cat NPnumber singularperson third

1

2

3

4

1 2, 2 1

1 3, 3 1

1 4

2 4

Not subsumption

Page 47: Feature structures and unification

agreement

cat NP

number singular

agreement

cat NP

person third

agreement

cat NP

number plural

agreement

cat NPnumber singularperson third

1

2

3

4

1 2, 2 1

1 3, 3 1

1 4

2 4

1 2 = 4

Not subsumption

Page 48: Feature structures and unification

agreement

cat NP

number singular

agreement

cat NP

person third

agreement

cat NP

number plural

agreement

cat NPnumber singularperson third

1

2

3

4

1 2, 2 1

1 3, 3 1

1 4

2 4

1 2 = 4

1 3 = fail

Not subsumption

Page 49: Feature structures and unification

agreement

cat NP

number singular

agreement

cat NP

person third

agreement

cat NP

number plural

agreement

cat NPnumber singularperson third

1

2

3

4

1 2, 2 1

1 3, 3 1

1 4

2 4

1 2 = 4

1 3 = fail

Unification:a b = cif and only ifa c andb c andthere is no d such thata d andb d andd c

Not subsumption

Page 50: Feature structures and unification

Unification

Page 51: Feature structures and unification

cat NP

Unification

Page 52: Feature structures and unification

cat NP

agreement number singular

Unification

Page 53: Feature structures and unification

cat NP

agreement number singular

=agreement

cat NP

number singular

Unification

Page 54: Feature structures and unification

cat NP

agreement number singular

=

cat NP

agreement

cat NP

number singular

Unification

Page 55: Feature structures and unification

agreement

cat NP

number singular

cat NP

agreement number singular

=

cat NP

agreement

cat NP

number singular

Unification

Page 56: Feature structures and unification

agreement

cat NP

number singular

cat NP

agreement number singular

=

cat NP

agreement

cat NP

number singular

agreement

cat NP

number singular

=

Unification

Page 57: Feature structures and unification

Unification

Page 58: Feature structures and unification

Unification

Page 59: Feature structures and unification

agreement

cat NP

number singular

Unification

Page 60: Feature structures and unification

agreement

cat NP

number singular

=agreement

cat NP

number singular

Unification

Page 61: Feature structures and unification

agreement

cat NP

number singular

subject

=

agreement number singular

agreement number singular

agreement

cat NP

number singular

Unification

Page 62: Feature structures and unification

agreement

cat NP

number singular

subject

=

agreement number singular

agreement number singular

subject agreement person third

agreement

cat NP

number singular

Unification

Page 63: Feature structures and unification

agreement

cat NP

number singular

subject

=

agreement number singular

agreement number singular

subject agreement person third

agreement

cat NP

number singular

subject

agreement number singular

agreementnumber singular

person third

=

Unification

Page 64: Feature structures and unification

Unification

Page 65: Feature structures and unification

subject

agreement number singular

agreement number singular

Unification

Page 66: Feature structures and unification

subject

agreement number singular

agreement

1

1

subject

agreement number singular

agreement number singular

Compare with:

Unification

Page 67: Feature structures and unification

subject

agreement number singular

agreement

subject agreement

1

person third

1

subject

agreement number singular

agreement number singular

Unification

Compare with:

Page 68: Feature structures and unification

subject

agreement number singular

agreement

subject agreement

agreementnumber singular

person third=

1

person third

1

subject agreement 1

1

subject

agreement number singular

agreement number singular

Unification

Compare with:

Page 69: Feature structures and unification

Unification

Page 70: Feature structures and unification

subject

agreement

agreement

1

2f1:

Unification

Page 71: Feature structures and unification

subject

agreement

agreement

1

2f1:

f1(agreement) = f1(subject)(agreement)

Unification

Page 72: Feature structures and unification

subject

agreement

agreement

1

2f1:

f1(agreement) = f1(subject)(agreement)

1 2 3=

Unification

Page 73: Feature structures and unification

subject

agreement

agreement

1

2f1:

f1(agreement) = f1(subject)(agreement)

1 2 3=

subject

agreement

agreementf1:

3

3

Unification

Page 74: Feature structures and unification

subject

agreement

agreement

1

2f1:

f1(agreement) = f1(subject)(agreement)

1 2 3=

subject

agreement

agreementf1:

3

3

agreement

subject agreementf1:

Unification

Page 75: Feature structures and unification

subject

agreement

agreement

1

2f1:

f1(agreement) = f1(subject)(agreement)

1 2 3=

subject

agreement

agreementf1:

3

3

agreement

subject agreementf1:

Unification

Page 76: Feature structures and unification

subject

agreement

agreement

1

2f1:

f1(agreement) = f1(subject)(agreement)

1 2 3=

subject

agreement

agreementf1:

3

3

agreement

subject agreementf1:

Unification

Page 77: Feature structures and unification

Unification through constraints:

Page 78: Feature structures and unification

subject

agreement

agreement

1

2f1:

f1(agreement) = f1(subject)(agreement)

Unification through constraints:

Page 79: Feature structures and unification

subject

agreement

agreement

1

2f1:

f1(agreement) = f1(subject)(agreement)

Unification through constraints:

Alternative notation with paths:

Page 80: Feature structures and unification

subject

agreement

agreement

1

2f1:

f1(agreement) = f1(subject)(agreement)

Unification through constraints:

Alternative notation with paths:

‹agreement› = ‹subject agreement›

This means that the two paths have the same (unspecified) value.

Page 81: Feature structures and unification

subject

agreement

agreement

1

2f1:

f1(agreement) = f1(subject)(agreement)

Unification through constraints:

Alternative notation with paths:

‹agreement› = ‹subject agreement›

This means that the two paths have the same (unspecified) value.

A constraint may also specify a value:

Page 82: Feature structures and unification

subject

agreement

agreement

1

2f1:

f1(agreement) = f1(subject)(agreement)

Unification through constraints:

Alternative notation with paths:

‹agreement› = ‹subject agreement›

This means that the two paths have the same (unspecified) value.

A constraint may also specify a value:

‹agreement number› = sg

Page 83: Feature structures and unification

subject

agreement

agreement

1

2f1:

f1(agreement) = f1(subject)(agreement)

Unification through constraints:

Alternative notation with paths:

‹agreement› = ‹subject agreement›

This means that the two paths have the same (unspecified) value.

A constraint may also specify a value:

‹agreement number› = sg

We thus have two types of constraints:

Page 84: Feature structures and unification

subject

agreement

agreement

1

2f1:

f1(agreement) = f1(subject)(agreement)

Unification through constraints:

Alternative notation with paths:

‹agreement› = ‹subject agreement›

This means that the two paths have the same (unspecified) value.

A constraint may also specify a value:

‹agreement number› = sg

We thus have two types of constraints:

‹attribute path› = Atomic value (The path has the specified value)‹attribute path› = ‹attribute path› (The two paths have the same value)

Page 85: Feature structures and unification

Incorporating unification in a phrase structure grammar

Page 86: Feature structures and unification

VP

S

NPsleepsJohn

Phrase structure tree:

Incorporating unification in a phrase structure grammar

Page 87: Feature structures and unification

VP

S

NPsleepsJohn

Grammar:

S → NP VP

Lexicon:

John NPsleeps VPsleep VP

A one-rule grammar with lexicon:Phrase structure tree:

Incorporating unification in a phrase structure grammar

Page 88: Feature structures and unification

VP

S

NPsleepsJohn

Grammar:

S → NP VP

A one-rule grammar with lexicon:Phrase structure tree:

We incorporate features and unification to handle agreement.

Incorporating unification in a phrase structure grammar

Lexicon:

John NPsleeps VPsleep VP

Page 89: Feature structures and unification

VP

S

NPsleepsJohn

Grammar:

S → NP VP

A one-rule grammar with lexicon:Phrase structure tree:

We incorporate features and unification to handle agreement.

Grammar:

S -> NP VP ‹f:S› = ‹f:VP› ‹f:S subject› = ‹f:NP›

Incorporating unification in a phrase structure grammar

Lexicon:

John NPsleeps VPsleep VP

Page 90: Feature structures and unification

VP

S

NPsleepsJohn

Grammar:

S → NP VP

A one-rule grammar with lexicon:Phrase structure tree:

We incorporate features and unification to handle agreement.

Lexicon:

John NP ‹f:NP agreement number› = singular ‹f:NP agreement person› = third

sleeps VP ‹f:VP subject agreement number› = singular ‹f:VP subject agreement person› = third

sleep VP ‹f:VP subject agreement number› = plural

Grammar:

S -> NP VP ‹f:S› = ‹f:VP› ‹f:S subject› = ‹f:NP›

Incorporating unification in a phrase structure grammar

Lexicon:

John NPsleeps VPsleep VP

Page 91: Feature structures and unification

Incorporating unification in a phrase structure grammar

The rule now describes this subtree:

Page 92: Feature structures and unification

Incorporating unification in a phrase structure grammar

The rule now describes this subtree:

1subject

2

2

NP VP1

S

Page 93: Feature structures and unification

Incorporating unification in a phrase structure grammar

The rule now describes this subtree:

1subject

2

2

NP VP1

The lexical entries:

S

Page 94: Feature structures and unification

Incorporating unification in a phrase structure grammar

The rule now describes this subtree:

1subject

2

2

NP VP1

The lexical entries:

agreement

John sleepsnumber singular

person thirdagreement

number singular

person thirdsubject

NP VP

S

Page 95: Feature structures and unification

Incorporating unification in a phrase structure grammar

1subject

2

2

NP VP1

agreement

John sleepsnumber singular

person thirdagreement

number singular

person thirdsubject

NP VP

What happens if we insert ‘John’ as the NP daughter?

S

Page 96: Feature structures and unification

Incorporating unification in a phrase structure grammar

1subject

2

2

NP VP1

agreement

John sleepsnumber singular

person thirdagreement

number singular

person thirdsubject

NP VP

S

Page 97: Feature structures and unification

Incorporating unification in a phrase structure grammar

1subject

2

2

NP VP1

agreement

John sleepsnumber singular

person thirdagreement

number singular

person thirdsubject

NP VP

S

Page 98: Feature structures and unification

Incorporating unification in a phrase structure grammar

1subject

2

S

2

NP VP1

John

sleeps

agreementnumber singular

person thirdsubject

VP

agreementnumber singular

person third

Page 99: Feature structures and unification

Incorporating unification in a phrase structure grammar

1subject

2

S

2

NP VP1

John

sleeps

agreementnumber singular

person thirdsubject

VP

agreementnumber singular

person third

'sleeps' can now only be inserted if itsagreement-features are compatible with 'John'.

Page 100: Feature structures and unification

Incorporating unification in a phrase structure grammar

1subject

2

S

2

NP VP1

John

sleeps

agreementnumber singular

person thirdsubject

VP

agreementnumber singular

person third

Page 101: Feature structures and unification

Incorporating unification in a phrase structure grammar

1subject

2

S

2

NP VP1

John

sleeps

agreementnumber singular

person thirdsubject

VP

agreementnumber singular

person third

Page 102: Feature structures and unification

Incorporating unification in a phrase structure grammar

1subject

2

S

2

NP VP1

John sleeps

agreementnumber singular

person third

Page 103: Feature structures and unification

Feature structures inLexical-Functional Grammar

Page 104: Feature structures and unification

1. [S I forced him [S PRO to be kind]]

Phrase structure analyses in traditional transformational grammar:

Page 105: Feature structures and unification

1. [S I forced him [S PRO to be kind]]

2. [S I believed [S him to be kind]]

Phrase structure analyses in traditional transformational grammar:

Page 106: Feature structures and unification

1. [S I forced him [S PRO to be kind]]

2. [S I believed [S him to be kind]]

3. [S NP seems [S John to shout]]

Phrase structure analyses in traditional transformational grammar:

Page 107: Feature structures and unification

1. [S I forced him [S PRO to be kind]]

2. [S I believed [S him to be kind]]

3. [S NP seems [S John to shout]]

4. [S NP tends [S John to shout]]

Phrase structure analyses in traditional transformational grammar:

Page 108: Feature structures and unification

1. [S I forced him [S PRO to be kind]]

2. [S I believed [S him to be kind]]

3. [S NP seems [S John to shout]]

4. [S NP tends [S John to shout]]

5. [S Bill [VP killed John]]

Phrase structure analyses in traditional transformational grammar:

Page 109: Feature structures and unification

1. [S I forced him [S PRO to be kind]]

2. [S I believed [S him to be kind]]

3. [S NP seems [S John to shout]]

4. [S NP tends [S John to shout]]

5. [S Bill [VP killed John]]

6. [S NP [VP was killed John]]

Phrase structure analyses in traditional transformational grammar:

Page 110: Feature structures and unification

1. [S I forced him [S PRO to be kind]]

2. [S I believed [S him to be kind]]

3. [S NP seems [S John to shout]]

4. [S NP tends [S John to shout]]

5. [S Bill [VP killed John]]

6. [S NP [VP was killed John]]

1. [S I forced him [VP' to be kind]]

Phrase structure analyses in Lexical Functional Grammar:

Phrase structure analyses in traditional transformational grammar:

Page 111: Feature structures and unification

1. [S I forced him [S PRO to be kind]]

2. [S I believed [S him to be kind]]

3. [S NP seems [S John to shout]]

4. [S NP tends [S John to shout]]

5. [S Bill [VP killed John]]

6. [S NP [VP was killed John]]

1. [S I forced him [VP' to be kind]]

2. [S I believed him [VP' to be kind]]

Phrase structure analyses in Lexical Functional Grammar:

Phrase structure analyses in traditional transformational grammar:

Page 112: Feature structures and unification

1. [S I forced him [S PRO to be kind]]

2. [S I believed [S him to be kind]]

3. [S NP seems [S John to shout]]

4. [S NP tends [S John to shout]]

5. [S Bill [VP killed John]]

6. [S NP [VP was killed John]]

1. [S I forced him [VP' to be kind]]

2. [S I believed him [VP' to be kind]]

4. [S John tends [VP' to shout]]

Phrase structure analyses in Lexical Functional Grammar:

Phrase structure analyses in traditional transformational grammar:

Page 113: Feature structures and unification

1. [S I forced him [S PRO to be kind]]

2. [S I believed [S him to be kind]]

3. [S NP seems [S John to shout]]

4. [S NP tends [S John to shout]]

5. [S Bill [VP killed John]]

6. [S NP [VP was killed John]]

1. [S I forced him [VP' to be kind]]

2. [S I believed him [VP' to be kind]]

4. [S John tends [VP' to shout]]

6. [S John [VP' was killed]]

Phrase structure analyses in Lexical Functional Grammar:

Phrase structure analyses in traditional transformational grammar:

Page 114: Feature structures and unification

1. [S I forced him [S PRO to be kind]]

2. [S I believed [S him to be kind]]

3. [S NP seems [S John to shout]]

4. [S NP tends [S John to shout]]

5. [S Bill [VP killed John]]

6. [S NP [VP was killed John]]

1. [S I forced him [VP' to be kind]]

2. [S I believed him [VP' to be kind]]

4. [S John tends [VP' to shout]]

6. [S John [VP' was killed]]

Phrase structure analyses in Lexical Functional Grammar:

Phrase structure analyses in traditional transformational grammar:

How does LFG capture

Page 115: Feature structures and unification

1. [S I forced him [S PRO to be kind]]

2. [S I believed [S him to be kind]]

3. [S NP seems [S John to shout]]

4. [S NP tends [S John to shout]]

5. [S Bill [VP killed John]]

6. [S NP [VP was killed John]]

1. [S I forced him [VP' to be kind]]

2. [S I believed him [VP' to be kind]]

4. [S John tends [VP' to shout]]

6. [S John [VP' was killed]]

Phrase structure analyses in Lexical Functional Grammar:

Phrase structure analyses in traditional transformational grammar:

How does LFG capture•the difference between 1 and 2,

Page 116: Feature structures and unification

1. [S I forced him [S PRO to be kind]]

2. [S I believed [S him to be kind]]

3. [S NP seems [S John to shout]]

4. [S NP tends [S John to shout]]

5. [S Bill [VP killed John]]

6. [S NP [VP was killed John]]

1. [S I forced him [VP' to be kind]]

2. [S I believed him [VP' to be kind]]

4. [S John tends [VP' to shout]]

6. [S John [VP' was killed]]

Phrase structure analyses in Lexical Functional Grammar:

Phrase structure analyses in traditional transformational grammar:

How does LFG capture•the difference between 1 and 2,•the non-argument status of the subject of 3 and 4,

Page 117: Feature structures and unification

1. [S I forced him [S PRO to be kind]]

2. [S I believed [S him to be kind]]

3. [S NP seems [S John to shout]]

4. [S NP tends [S John to shout]]

5. [S Bill [VP killed John]]

6. [S NP [VP was killed John]]

1. [S I forced him [VP' to be kind]]

2. [S I believed him [VP' to be kind]]

4. [S John tends [VP' to shout]]

6. [S John [VP' was killed]]

Phrase structure analyses in Lexical Functional Grammar:

Phrase structure analyses in traditional transformational grammar:

How does LFG capture•the difference between 1 and 2,•the non-argument status of the subject of 3 and4,•and the semantic role of the subject of 6?

Page 118: Feature structures and unification

1. [S I forced him [S PRO to be kind]]

2. [S I believed [S him to be kind]]

3. [S NP seems [S John to shout]]

4. [S NP tends [S John to shout]]

5. [S Bill [VP killed John]]

6. [S NP [VP was killed John]]

1. [S I forced him [VP' to be kind]]

2. [S I believed him [VP' to be kind]]

4. [S John tends [VP' to shout]]

6. [S John [VP' was killed]]

Phrase structure analyses in Lexical Functional Grammar:

Phrase structure analyses in traditional transformational grammar:

How does LFG capture•the difference between 1 and 2,•the non-argument status of the subject of 3 and 4,•and the semantic role of the subject of 6?

Answer: Don’t operate on the trees,but annotate them with relevant informationabout syntactic functions and semantic arguments.

Page 119: Feature structures and unification

VP

V NP

S

NP

I

forced

kindbe

him

VP'

TO VP

AP

to

V

Page 120: Feature structures and unification

VP

V NP

S

NP

I

forced

kindbe

him

VP'

TO VP

AP

to

VP

V NP

S

NP

I

believed

kindbe

him

VP'

TO VP

V AP

to V

Page 121: Feature structures and unification

VP

V NP

S

NP

I

forced

kindbe

him

VP'

TO VP

V AP

to

VP

V NP

S

NP

I

believed

kindbe

him

VP'

TO VP

V AP

to

VP

V

S

NP

John

tends

shout

VP'

TO VP

V

to

Page 122: Feature structures and unification

VP

V NP

S

NP

I

forced

kindbe

him

VP'

TO VP

V AP

to

VP

V NP

S

NP

I

believed

kindbe

him

VP'

TO VP

V AP

to

VP

V

S

NP

John

tends

shout

VP'

TO VP

V

to

VP

V

S

NP

John

was

Bill

VP

V PP

NP

killed

by P

Page 123: Feature structures and unification

VP

V NP

S

NP

I

forced

kindbe

him

VP'

TO VP

V AP

to

VP

V NP

S

NP

I

believed

kindbe

him

VP'

TO VP

V AP

to

VP

V

S

NP

John

tends

shout

VP'

TO VP

V

to

VP

V

S

NP

John

was

Bill

VP

V PP

NP

killed

by P

INF

’FORCE ‹SUBJ OBJ XCOMP›’PRET

XCOMPOBJ

SUBJ

Page 124: Feature structures and unification

VP

V NP

S

NP

I

forced

kindbe

him

VP'

TO VP

V AP

to

VP

V NP

S

NP

I

believed

kindbe

him

VP'

TO VP

V AP

to

VP

V

S

NP

John

tends

shout

VP'

TO VP

V

to

VP

V

S

NP

John

was

Bill

VP

V PP

NP

killed

by P

INFINF

XCOMPOBJ

SUBJ

’FORCE ‹SUBJ OBJ XCOMP›’PRET

XCOMPOBJ

SUBJ

PRETBELIEVE ‹SUBJ XCOMP› OBJ’

Page 125: Feature structures and unification

VP

V NP

S

NP

I

forced

kindbe

him

VP'

TO VP

V AP

to

VP

V NP

S

NP

I

believed

kindbe

him

VP'

TO VP

V AP

to

VP

V

S

NP

John

tends

shout

VP'

TO VP

V

to

VP

V

S

NP

John

was

Bill

VP

V PP

NP

killed

by P

PRES

SUBJINF

INF

XCOMP

INF

XCOMPOBJ

SUBJ

’FORCE ‹SUBJ OBJ XCOMP›’PRET

XCOMPOBJ

SUBJ

TEND ‹XCOMP› SUBJ’

PRETBELIEVE ‹SUBJ XCOMP› OBJ’

Page 126: Feature structures and unification

VP

V NP

S

NP

I

forced

kindbe

him

VP'

TO VP

V AP

to

VP

V NP

S

NP

I

believed

kindbe

him

VP'

TO VP

V AP

to

VP

V

S

NP

John

tends

shout

VP'

TO VP

V

to

VP

V

S

NP

John

was

Bill

VP

V PP

NP

killed

by P

PRES

SUBJINF

INF

OBLag

XCOMP

INF

SUBJ

XCOMPOBJ

SUBJ

’FORCE ‹SUBJ OBJ XCOMP›’PRET

XCOMPOBJ

SUBJ

TEND ‹XCOMP› SUBJ’

PRET

KILL ‹OBLag SUBJ›’

BELIEVE ‹SUBJ XCOMP› OBJ’

Page 127: Feature structures and unification

The functional information in the annotations

is represented in a separate functional structure

(f-structure), in the form of an attribute-value graph:

Page 128: Feature structures and unification

SUBJPRED ’I’CASE nom

TENSE pret

OBJ

PRED ’HE’CASE oblNUM sg

XCOMPSUBJPRED ’LEAVE‹ SUBJ › ’

PRED ’FORCE‹ SUBJ OBJ XCOMP ›’

F-structure for I forced him to leave

f1 f2

f5

f6

Page 129: Feature structures and unification

SUBJPRED ’I’CASE nom

TENSE pret

OBJ

PRED ’HE’CASE oblNUM sg

XCOMPSUBJPRED ’LEAVE‹ SUBJ › ’

F-structure for I forced him to leave

f1 f2

f5

f6

PRED ’FORCE‹ SUBJ OBJ XCOMP ›’

Page 130: Feature structures and unification

SUBJPRED ’I’CASE nom

TENSE pret

OBJ

PRED ’HE’CASE oblNUM sg

XCOMPSUBJPRED ’LEAVE‹ SUBJ › ’

PRED ’FORCE‹ SUBJ OBJ XCOMP ›’

F-structure for I forced him to leave

f1 f2

f5

f6

Page 131: Feature structures and unification

SUBJPRED ’I’CASE nom

TENSE pret

OBJ

PRED ’HE’CASE oblNUM sg

XCOMPSUBJPRED ’LEAVE‹ SUBJ › ’

PRED ’FORCE‹ SUBJ OBJ XCOMP ›’

F-structure for I forced him to leave

f1 f2

f5

f6

Page 132: Feature structures and unification

SUBJPRED ’I’CASE nom

TENSE pret

OBJ

PRED ’HE’CASE oblNUM sg

XCOMPSUBJPRED ’LEAVE‹ SUBJ › ’

PRED ’FORCE‹ SUBJ OBJ XCOMP ›’

F-structure for I forced him to leave

f1 f2

f5

f6

Page 133: Feature structures and unification

Linking

A verb form contains information about the way in whichsemantic arguments are linked to syntactic functions:

Page 134: Feature structures and unification

Linking

A verb form contains information about the way in whichsemantic arguments are linked to syntactic functions:

"reparerer": reparere<agent, theme>

SUBJ OBJ

Page 135: Feature structures and unification

Linking

A verb form contains information about the way in whichsemantic arguments are linked to syntactic functions:

"reparerer": reparere<agent, theme>

SUBJ OBJ

"repareres": reparere<agent, theme>

SUBJ

Page 136: Feature structures and unification

Linking

A verb form contains information about the way in whichsemantic arguments are linked to syntactic functions:

"reparerer": reparere<agent, theme>

SUBJ OBJ

"repareres": reparere<agent, theme>

SUBJ

”like": like<experiencer, theme>

SUBJ OBJ

Page 137: Feature structures and unification

Linking

A verb form contains information about the way in whichsemantic arguments are linked to syntactic functions:

"reparerer": reparere<agent, theme>

SUBJ OBJ

"repareres": reparere<agent, theme>

SUBJ

”like": like<experiencer, theme>

SUBJ OBJ

”behage": behage<experiencer, theme>

OBJ SUBJ

Page 138: Feature structures and unification

Linking

If we assume a universal hierarchy of semantic roles and let the order ofthe arguments reflect the hierarchy, we don’t need to name the semantic roles:

"reparerer": reparere<agent, theme>

SUBJ OBJ

"repareres": reparere<agent, theme>

SUBJ

”like": like<experiencer, theme>

SUBJ OBJ

”behage": behage<experiencer, theme>

OBJ SUBJ

Page 139: Feature structures and unification

Linking

If we assume a universal hierarchy of semantic roles and let the order ofthe arguments reflect the hierarchy, we don’t need to name the semantic roles:

"reparerer": reparere<SUBJ, OBJ>

"repareres": reparere< , SUBJ>

”like": like< SUBJ, OBJ >

”behage": behage< OBJ, SUBJ >

Page 140: Feature structures and unification

Wellformedness constraints on functional structures:

SUBJ

PRED

OBJ

ADJUNCT

"the boy"

”the bike"

{”in the garage"}

repair<SUBJ, OBJ>"

”The boy repairs the bike in the garage":

Page 141: Feature structures and unification

Wellformedness constraints on functional structures:

SUBJ

PRED

OBJ

ADJUNCT

"the boy"

”the bike"

{”in the garage"}

repair<SUBJ, OBJ>"

1. Completeness: An f-structure must contain all grammaticalrelations mentioned in PRED’s subcategorization frame.

SUBJ

PRED

”the boy"

”use<SUBJ, OBJ>"

*”The boy uses":”The boy repairs the bike in the garage":

Page 142: Feature structures and unification

Wellformedness constraints on functional structures:

SUBJ

PRED

OBJ

ADJUNCT

"the boy"

”the bike"

{”in the garage"}

repair<SUBJ, OBJ>"

1. Completeness: An f-structure must contain all grammaticalrelations mentioned in PRED’s subcategorization frame.

2. Coherence: An f-structure cannot contain any subcategorizablegrammatical relations not mentioned in PRED’s subcategorization frame.

SUBJ

PRED

"gutten"

"sove<SUBJ>"

*"Gutten sover sykkelen":

OBJ "sykkelen"

”The boy repairs the bike in the garage":

Page 143: Feature structures and unification

Wellformedness constraints on functional structures:

SUBJ

PRED

OBJ

ADJUNCT

"the boy"

”the bike"

{”in the garage"}

repair<SUBJ, OBJ>"

1. Completeness: An f-structure must contain all grammaticalrelations mentioned in PRED’s subcategorization frame.

2. Coherence: An f-structure cannot contain any subcategorizablegrammatical relations not mentioned in PRED’s subcategorization frame.

3. Uniqueness: No grammatical relation (or other attribute) may occurmore than once in a functional structure.

SUBJ

PRED

”the boy"

”use<SUBJ, OBJ>"

*”The boy uses the bike the car ”:

OBJ "the bike"

OBJ ”the car"

”The boy repairs the bike in the garage":

Page 144: Feature structures and unification

SUBJPRED ’I’CASE nom

TENSE pret

OBJ

PRED ’HE’CASE oblNUM sg

XCOMPSUBJPRED ’LEAVE‹ SUBJ › ’

PRED ’FORCE‹ SUBJ OBJ XCOMP ›’

F-structure for I forced him to leave

f1 f2

f5

f6

Describing parts of the structureby means of equations

Page 145: Feature structures and unification

SUBJPRED ’I’CASE nom

TENSE pret

OBJ

PRED ’HE’CASE oblNUM sg

XCOMPSUBJPRED ’LEAVE‹ SUBJ › ’

PRED ’FORCE‹ SUBJ OBJ XCOMP ›’

F-structure for I forced him to leave

f1 f2

f5

f6

Describing parts of the structureby means of equations

f1 (TENSE) = pret

Page 146: Feature structures and unification

SUBJPRED ’I’CASE nom

TENSE pret

OBJ

PRED ’HE’CASE oblNUM sg

XCOMPSUBJPRED ’LEAVE‹ SUBJ › ’

PRED ’FORCE‹ SUBJ OBJ XCOMP ›’

F-structure for I forced him to leave

f1 f2

f5

f6

Describing parts of the structureby means of equations

f1 (TENSE) = pretf1 (SUBJ) = f2

Page 147: Feature structures and unification

SUBJPRED ’I’CASE nom

TENSE pret

OBJ

PRED ’HE’CASE oblNUM sg

XCOMPSUBJPRED ’LEAVE‹ SUBJ › ’

PRED ’FORCE‹ SUBJ OBJ XCOMP ›’

F-structure for I forced him to leave

f1 f2

f5

f6

Describing parts of the structureby means of equations

f1 (TENSE) = pretf1 (SUBJ) = f2f2 (CASE) = nom

Page 148: Feature structures and unification

SUBJPRED ’I’CASE nom

TENSE pret

OBJ

PRED ’HE’CASE oblNUM sg

XCOMPSUBJPRED ’LEAVE‹ SUBJ › ’

PRED ’FORCE‹ SUBJ OBJ XCOMP ›’

F-structure for I forced him to leave

f1 f2

f5

f6

Describing parts of the structureby means of equations

f1 (TENSE) = pretf1 (SUBJ) = f2f2 (CASE) = nomf1 (SUBJ)(CASE) = nom

Page 149: Feature structures and unification

SUBJPRED ’I’CASE nom

TENSE pret

OBJ

PRED ’HE’CASE oblNUM sg

XCOMPSUBJPRED ’LEAVE‹ SUBJ › ’

PRED ’FORCE‹ SUBJ OBJ XCOMP ›’

F-structure for I forced him to leave

f1 f2

f5

f6

Describing parts of the structureby means of equations

f1 (TENSE) = pretf1 (SUBJ) = f2f2 (CASE) = nomf1 (SUBJ)(CASE) = nom

f2

Page 150: Feature structures and unification

SUBJPRED ’I’CASE nom

TENSE pret

OBJ

PRED ’HE’CASE oblNUM sg

XCOMPSUBJPRED ’LEAVE‹ SUBJ › ’

PRED ’FORCE‹ SUBJ OBJ XCOMP ›’

F-structure for I forced him to leave

f1 f2

f5

f6

Describing parts of the structureby means of equations

f1 (TENSE) = pretf1 (SUBJ) = f2f2 (CASE) = nomf1 (SUBJ)(CASE) = nom

f2

Alternative notation:

(f1 TENSE) = pret(f1 SUBJ) = f2(f2 CASE) = nom(f1 SUBJ CASE) = nom

Page 151: Feature structures and unification

SUBJPRED ’I’CASE nom

TENSE pret

OBJ

PRED ’HE’CASE oblNUM sg

XCOMPSUBJPRED ’LEAVE‹ SUBJ › ’

PRED ’FORCE‹ SUBJ OBJ XCOMP ›’

F-structure for I forced him to leave

f1 f2

f5

f6

Describing parts of the structureby means of equations

f1 (TENSE) = pretf1 (SUBJ) = f2f2 (CASE) = nomf1 (SUBJ)(CASE) = nom

f2

Alternative notation:

(f1 TENSE) = pret(f1 SUBJ) = f2(f2 CASE) = nom(f1 SUBJ CASE) = nom

(f1 OBJ) = (f1 XCOMP SUBJ)

Page 152: Feature structures and unification

SUBJPRED ’I’CASE nom

TENSE pret

OBJ

PRED ’HE’CASE oblNUM sg

XCOMPSUBJPRED ’LEAVE‹ SUBJ › ’

PRED ’FORCE‹ SUBJ OBJ XCOMP ›’

F-structure for I forced him to leave

f1 f2

f5

f6

Describing parts of the structureby means of equations

f1 (TENSE) = pretf1 (SUBJ) = f2f2 (CASE) = nomf1 (SUBJ)(CASE) = nom

f2

Alternative notation:

(f1 TENSE) = pret(f1 SUBJ) = f2(f2 CASE) = nom(f1 SUBJ CASE) = nom

(f1 OBJ) = (f1 XCOMP SUBJ)

Page 153: Feature structures and unification

How to incorporatef-structure information

into a grammar

Page 154: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

Page 155: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

( SUBJ)

( OBJ)

( XCOMP)

Page 156: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

Page 157: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

VP

V NP

S

NPI

forced himVP'

to leave

Page 158: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

VP

V NP

S

NPI

forced himVP'

to leave

( SUBJ)

( OBJ)

( XCOMP)

Page 159: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

VP

V NP

S

NPI

forced himVP'

to leave

( SUBJ)

( OBJ)

( XCOMP)

( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

Page 160: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

VP:3

V:4 NP:5

S:1

NP:2I

forced himVP':6

to leave

( SUBJ)

( OBJ)

( XCOMP)

( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

Index the c-structure nodes

Page 161: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

VP:3

V:4 NP:5

S:1

NP:2I

forced himVP':6

to leave

( OBJ)

( XCOMP)

( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

(f1 SUBJ)

Instantiate the metavariables:Replace them with f-structurevariables based on the node indices.

Page 162: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

VP:3

V:4 NP:5

S:1

NP:2I

forced himVP':6

to leave

( OBJ)

( XCOMP)

( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

(f1 SUBJ)f2

Instantiate the metavariables:Replace them with f-structurevariables based on the node indices.

Page 163: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

VP:3

V:4 NP:5

S:1

NP:2I

forced himVP':6

to leave

( OBJ)

( XCOMP)

( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

(f1 SUBJ)f2 f1

Instantiate the metavariables:Replace them with f-structurevariables based on the node indices.

Page 164: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

VP:3

V:4 NP:5

S:1

NP:2I

forced himVP':6

to leave

( OBJ)

( XCOMP)

( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

(f1 SUBJ)f2 f1f3

Instantiate the metavariables:Replace them with f-structurevariables based on the node indices.

Page 165: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

VP:3

V:4 NP:5

S:1

NP:2I

forced himVP':6

to leave

( OBJ)

( XCOMP)

( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

(f1 SUBJ)f2 f1f3

f3

Instantiate the metavariables:Replace them with f-structurevariables based on the node indices.

Page 166: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

VP:3

V:4 NP:5

S:1

NP:2I

forced himVP':6

to leave

( OBJ)

( XCOMP)

( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

(f1 SUBJ)f2 f1f3

f3f4

Instantiate the metavariables:Replace them with f-structurevariables based on the node indices.

Page 167: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

VP:3

V:4 NP:5

S:1

NP:2I

forced himVP':6

to leave

( XCOMP)

( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

(f1 SUBJ)f2 f1f3

f3f4 (f3 OBJ)

Instantiate the metavariables:Replace them with f-structurevariables based on the node indices.

Page 168: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

VP:3

V:4 NP:5

S:1

NP:2I

forced himVP':6

to leave

( XCOMP)

( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

(f1 SUBJ)f2 f1f3

f3f4 (f3 OBJ)f5

Instantiate the metavariables:Replace them with f-structurevariables based on the node indices.

Page 169: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

VP:3

V:4 NP:5

S:1

NP:2I

forced himVP':6

to leave( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

(f1 SUBJ)f2 f1f3

f3f4 (f3 OBJ)f5 (f3 XCOMP)

Instantiate the metavariables:Replace them with f-structurevariables based on the node indices.

Page 170: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

VP:3

V:4 NP:5

S:1

NP:2I

forced himVP':6

to leave( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

(f1 SUBJ)f2 f1f3

f3f4 (f3 OBJ)f5 (f3 XCOMP)f6

Instantiate the metavariables:Replace them with f-structurevariables based on the node indices.

Page 171: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

VP:3

V:4 NP:5

S:1

NP:2I

forced himVP':6

to leave(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2 f1f3

f3f4 (f3 OBJ)f5 (f3 XCOMP)f6

Instantiate the metavariables:Replace them with f-structurevariables based on the node indices.

Page 172: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2 f1f3

f3f4 (f3 OBJ)f5 (f3 XCOMP)f6

The tree has done its job:Forget it.

Page 173: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

Collect the instantiated equationsinto an f-description

Page 174: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

Solve the equations in any orderto constuct an f-structure

Page 175: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

F-structure for I forced him to leave

Solve the equations in any orderto constuct an f-structure

Page 176: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

F-structure for I forced him to leave

Page 177: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

F-structure for I forced him to leave

f1f2

Page 178: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

F-structure for I forced him to leave

f1f2

Page 179: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

F-structure for I forced him to leave

f1f2

f3

Page 180: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

F-structure for I forced him to leave

f1f2

f3

Page 181: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

F-structure for I forced him to leave

f1f2

f3

f4

Page 182: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

F-structure for I forced him to leave

f1f2

f3

f4

Page 183: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

OBJ

F-structure for I forced him to leave

f1f2

f5

f3

f4

Page 184: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

OBJ

F-structure for I forced him to leave

f1f2

f3

f4

f5

Page 185: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

OBJ

XCOMP

F-structure for I forced him to leave

f1f2

f6

f3

f4

f5

Page 186: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

OBJ

XCOMP

F-structure for I forced him to leave

f1f2

f6

f3

f4

f5

Page 187: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

OBJ

XCOMP

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

F-structure for I forced him to leave

f1f2

f6

f3

f4

f5

Page 188: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

OBJ

XCOMP

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

F-structure for I forced him to leave

f1f2

f6

f3

f4

f5

Page 189: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

OBJ

XCOMP

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

F-structure for I forced him to leave

f1f2

f6

f3

f4

f5

Page 190: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

OBJ

XCOMP

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

F-structure for I forced him to leave

f1f2

f6

f3

f4

f5

Page 191: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

OBJ

XCOMP

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

F-structure for I forced him to leave

f1f2

f6

f3

f4

f5

Page 192: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

TENSE pret

OBJ

XCOMP

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

F-structure for I forced him to leave

f1f2

f6

f3

f4

f5

Page 193: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

TENSE pret

OBJ

XCOMP

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

F-structure for I forced him to leave

f1f2

f6

f3

f4

f5

Page 194: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

TENSE pret

OBJ

XCOMP SUBJ

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

F-structure for I forced him to leave

f1f2

f6

f3

f4

f5

Page 195: Feature structures and unification

(f4 PRED) = 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'(f4 TENSE) = pret(f4 OBJ) = (f4 XCOMP SUBJ)

(f1 SUBJ)f2f1f3f3f4(f3 OBJ)f5(f3 XCOMP)f6

SUBJ

TENSE pret

OBJ

XCOMP SUBJ

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

F-structure for I forced him to leave

f1f2

f6

f3

f4

f5

Notice: The f-structure hasfewer levels than the c-structurebecause of the nodes annotatedwith =↓

Page 196: Feature structures and unification

SUBJ

TENSE pret

OBJ

XCOMP SUBJ

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

The nodes in the tree and the elements of the f-structurenow stand in a many-to-one relation:

f1f2

f6

f3

f4

f5

VP

V NP

S

NPI

forced himVP'

to leave

( SUBJ)

( OBJ)

( XCOMP)

Page 197: Feature structures and unification

SUBJ

TENSE pret

OBJ

XCOMP SUBJ

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

The nodes in the tree and the elements of the f-structurenow stand in a many-to-one relation:

f1f2

f6

f3

f4

f5

VP

V NP

S

NPI

forced himVP'

to leave

( SUBJ)

( OBJ)

( XCOMP)

Page 198: Feature structures and unification

SUBJ

TENSE pret

OBJ

XCOMP SUBJ

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

The nodes in the tree and the elements of the f-structurenow stand in a many-to-one relation:

f1f2

f6

f3

f4

f5

VP

V NP

S

NPI

forced himVP'

to leave

( SUBJ)

( OBJ)

( XCOMP)

Page 199: Feature structures and unification

SUBJ

TENSE pret

OBJ

XCOMP SUBJ

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

The nodes in the tree and the elements of the f-structurenow stand in a many-to-one relation:

f1f2

f6

f3

f4

f5

VP

V NP

S

NPI

forced himVP'

to leave

( SUBJ)

( OBJ)

( XCOMP)

Page 200: Feature structures and unification

SUBJ

TENSE pret

OBJ

XCOMP SUBJ

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

The nodes in the tree and the elements of the f-structurenow stand in a many-to-one relation:

f1f2

f6

f3

f4

f5

VP

V NP

S

NPI

forced himVP'

to leave

( SUBJ)

( OBJ)

( XCOMP)

Page 201: Feature structures and unification

SUBJ

TENSE pret

OBJ

XCOMP SUBJ

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

The nodes in the tree and the elements of the f-structurenow stand in a many-to-one relation:

f1f2

f6

f3

f4

f5

VP

V NP

S

NPI

forced himVP'

to leave

( SUBJ)

( OBJ)

( XCOMP)

Page 202: Feature structures and unification

SUBJ

TENSE pret

OBJ

XCOMP SUBJ

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

The nodes in the tree and the elements of the f-structurenow stand in a many-to-one relation:

f1f2

f6

f3

f4

f5

VP

V NP

S

NPI

forced himVP'

to leave

( SUBJ)

( OBJ)

( XCOMP)

Page 203: Feature structures and unification

SUBJ

TENSE pret

OBJ

XCOMP SUBJ

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

The nodes in the tree and the elements of the f-structurenow stand in a many-to-one relation:

f1f2

f6

f3

f4

f5

VP

V NP

S

NPI

forced himVP'

to leave

( SUBJ)

( OBJ)

( XCOMP)

Page 204: Feature structures and unification

SUBJ

TENSE pret

OBJ

XCOMP SUBJ

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

The nodes in the tree and the elements of the f-structurenow stand in a many-to-one relation:

f1f2

f6

f3

f4

f5

VP

V NP

S

NPI

forced himVP'

to leave

( SUBJ)

( OBJ)

( XCOMP)

Page 205: Feature structures and unification

SUBJ

TENSE pret

OBJ

XCOMP SUBJ

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

The nodes in the tree and the elements of the f-structurenow stand in a many-to-one relation:

f1f2

f6

f3

f4

f5

VP

V NP

S

NPI

forced himVP'

to leave

( SUBJ)

( OBJ)

( XCOMP)

The relation is called a projection relation.

Page 206: Feature structures and unification

SUBJ

TENSE pret

OBJ

XCOMP SUBJ

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

The nodes in the tree and the elements of the f-structurenow stand in a many-to-one relation:

f1f2

f6

f3

f4

f5

VP

V NP

S

NPI

forced himVP'

to leave

( SUBJ)

( OBJ)

( XCOMP)

The relation is called a projection relation.A set of nodes which project the same f-structureare said to constitute a functional domain.

A functional domain

Page 207: Feature structures and unification

Let us now move from

I forced him to leave

to

I believed him to leave

SUBJ

TENSE pret

OBJ

XCOMP SUBJ

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

f1f2

f6

f3

f4

f5

Page 208: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

forced: ( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

VP

V NP

S

NPI

forced himVP'

to leave

( SUBJ)

( OBJ)

( XCOMP)

( PRED) = 'FORCE‹( SUBJ)( OBJ)( XCOMP)›'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

All we need to change is the lexical entry:

Page 209: Feature structures and unification

S -> NP VP

VP -> V (NP) (VP')

believed: ( PRED) = ’BELIEVE‹( SUBJ) ( XCOMP)›( OBJ)'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

( SUBJ)

( OBJ)

( XCOMP)

VP

V NP

S

NPI

believed himVP'

to leave

( SUBJ)

( OBJ)

( XCOMP)

( PRED) = ’BELIEVE‹( SUBJ) ( XCOMP)›( OBJ)'( TENSE) = pret( OBJ) = ( XCOMP SUBJ)

All we need to change is the lexical entry:

Page 210: Feature structures and unification

SUBJ

TENSE pret

OBJ

XCOMP SUBJ

PRED 'FORCE‹(f4 SUBJ)(f4 OBJ)(f4 XCOMP)›'

f1f2

f6

f3

f4

f5

This leads to the following change in the f-structure:

Page 211: Feature structures and unification

SUBJ

TENSE pret

OBJ

XCOMP SUBJ

PRED ’BELIEVE‹(f4 SUBJ)(f4 XCOMP)›(f4 OBJ)'

f1f2

f6

f3

f4

f5

This leads to the following change in the f-structure:

Page 212: Feature structures and unification

SUBJ

TENSE pret

OBJ

XCOMP SUBJ

PRED ’BELIEVE‹(f4 SUBJ)(f4 XCOMP)›(f4 OBJ)'

f1f2

f6

f3

f4

f5

This leads to the following change in the f-structure:

The only change is in the mapping between syntactic functionsand argument positions, as expressed in the value of PRED.The syntax as such is unchanged.

Page 213: Feature structures and unification

Constraint Equations

Consider these lexical entries:

ha V (↑PRED)='ha<(↑ SUBJ)(↑ XCOMP)>' (↑ XCOMP PTC)=perf

måtte V (↑PRED)='måtte<(↑ SUBJ)(↑ XCOMP)>' (↑ XCOMP VFORM)=inf

løpe V (↑PRED)='løpe<(↑ SUBJ)>' (↑ VFORM)=inf

løpt V (↑PRED)='løpe<(↑ SUBJ)>' (↑ PTC)=perf

Page 214: Feature structures and unification

Constraint Equations

Consider these lexical entries:

ha V (↑PRED)='ha<(↑ SUBJ)(↑ XCOMP)>' (↑ XCOMP PTC)=perf

måtte V (↑PRED)='måtte<(↑ SUBJ)(↑ XCOMP)>' (↑ XCOMP VFORM)=inf

løpe V (↑PRED)='løpe<(↑ SUBJ)>' (↑ VFORM)=inf

løpt V (↑PRED)='løpe<(↑ SUBJ)>' (↑ PTC)=perf

This enables us to derive:

gutten har løptgutten måtte løpe

Page 215: Feature structures and unification

Constraint Equations

Consider these lexical entries:

ha V (↑PRED)='ha<(↑ SUBJ)(↑ XCOMP)>' (↑ XCOMP PTC)=perf

måtte V (↑PRED)='måtte<(↑ SUBJ)(↑ XCOMP)>' (↑ XCOMP VFORM)=inf

løpe V (↑PRED)='løpe<(↑ SUBJ)>' (↑ VFORM)=inf

løpt V (↑PRED)='løpe<(↑ SUBJ)>' (↑ PTC)=perf

This enables us to derive:

gutten har løptgutten måtte løpe

But does it exclude the following?

*gutten har løpe*gutten måtte løpt

Page 216: Feature structures and unification

Constraint Equations

We need to change some equations into constraint equations:

ha V (↑PRED)='ha<(↑ SUBJ)(↑ XCOMP)>' (↑ XCOMP PTC)=c perf

måtte V (↑PRED)='måtte<(↑ SUBJ)(↑ XCOMP)>' (↑ XCOMP VFORM)=c inf

løpe V (↑PRED)='løpe<(↑ SUBJ)>' (↑ VFORM)=inf

løpt V (↑PRED)='løpe<(↑ SUBJ)>' (↑ PTC)=perf