Encoding of basic syntactic information

1. The basic part of speech
2. What the word can be combined with
3. Certain syntactically relevant inherent properties, for example gender in the case of nouns, irregular verb forms

Feature based syntax:

Examples:

Lexeme Maedchen:
    <cat> = N
    <gender> = neut.

Lexeme love:
    <cat> = V
    <arg0 cat> = NP
    <arg0 case> = nom
    <arg1 cat> = NP
    <arg1 case> = acc.

Note [ DG ]

The notation is so-called `PATR II' notation, and is based on attribute-value pairs: `cat' is an attribute, and one of its possible values is `N'; `gender is an attribute, and one of its possible values is `neut'.

In Prolog, the attributes are not named explicitly, but are represented by positions in a Prolog term. So one way of encoding this information about `M"adchen' in Prolog would be as follows:

lexeme('Maedchen','N','neut').

argument 0 codes up information about the subject of a verb. The lexeme "love" combines with a nominative subject NP, like all other english verbs. argument 1 codes up information about the direct object of a verb. The lexeme "love" combines subcategorizes for an accusative object NP, therefore love is a transitive verb.

Lexeme give:
    <cat> = V
    <arg0 cat> = NP
    <arg0 case> = nom
    <arg1 cat> = NP
    <arg1 case> = acc
    <arg2 cat> = PP
    <arg2 pform> = to.

Four argument features are sufficient for a grammar of English.

• subject
• direct object
• prepositional phrase
• indirect object

These verb features will be used by rules such as the following (there appears to be an inconsistency in the text ...):

Rule
VP>V X1
    <V arg0> = X1
    <V arg1> = 0
    <V arg2> = 0.

Example: die, as in he died; cf. he died his dog.

Rule
VP>V X1 X2:
    <V arg0> = X1
    <V arg1> = X2
    <V arg2> = 0.

Example: love, as in he loved her; cf. he loved her to his uncle, ?he loved.

Rule
VP>V X1 X2 X3:
    <V arg0> = X1
    <V arg1> = X2
    <V arg2> = X3.

Example: give, as in he gave her the book; cf. he gave her; cf. also he gave the book, but with the meaning `donate'.

To shorten lexical entries abbreviations called macros are employed. (cf. recipe for vinaigrette: mixture1, mixture2)

Macro syn_iV:   (intransitive verbs like "die")
    <cat> = V
    <arg0 cat> = NP
    <arg0 case> = nom.

Macro syn_tV:   (transitive verbs like "eat")
    syn_iV
    <arg1 cat> = NP
    <arg1 case> = acc.

Macro syn_dtV:   (ditransitive verbs like "give")
    syn_tV
    <arg2 cat> = PP
    <arg2 pform> = to. 

Macro syn_datV:   (dative verbs like "hand")
    syn_dtV
    <arg3 cat> = NP
    <arg3 case> = acc.

This gives a lexicon like this:

Lexeme die:
    syn_iV.

Lexeme elapse:
    syn_iV.

Lexeme eat:
    syn_iV.

Lexeme eat:
    syn_tV.

Lexeme give: 
    syn_tV.

Lexeme give:
    syn_dtV.

Lexeme give:  
    syn_datV.

Lexeme hand:
    syn_dtV.

Lexeme hand: 
    syn_datV.

Lexeme love:
    syn_tV.

For example:

Lexical entry:

Lexeme give:
    syn_dtV.

Macro definitions:

Macro syn_dtV:   (ditransitive verbs like "give")
    syn_tV
    <arg2 cat> = PP
    <arg2 pform> = to.

Expanded lexical entry (macro expansion):

Lexeme give:
    <cat> = V
    <arg0 cat> = NP
    <arg0 case> = nom.
    <arg1 cat> = NP
    <arg1 case> = acc.
    <arg2 cat> = PP
    <arg2 pform> = to.

Many words have several lexical entries because they belong to various syntactic classes.

For example: laugh, as in We had a good laugh, as opposed to we laughed, or eat as intransitive vs. transitive.