Generative Lexicon

Generative Lexicon is a computational theory of linguistic semantics that focuses on the nature of compositionality. The theory is firstly proposed by James Pustejovsky in 1991 in his paper titled as “The Generative Lexicon”.

The Generative Lexicon has 4 levels:

1. Lexical Typing Structure

2. Argument Structure

3. Event Structure

4. Qualia Structure

Basic Issues

Pustejovsky discusses about how generative lexicon can contribute towards a classification of different nominal types. He distinguishes  the lexical semantic theory into four major areas of concern:

1. Distinction of complement taking behaviour of nouns and verbs.

2. How nominalizations and event-denoting nominals are distinguished from their corresponding verbal representations. Basically, what is the difference between events represented as a sentence and event represented as a NP?

3. The Representation of logical polysemy in nominals, such as kitap, kapı, pencere and how they are differ from relational nominals.

4. How does the semantics of nominals facilitate the richer compositional interpretation and what allows for co-compositional interpretation in natural language?

The first 2 issuses are related. In addition how nominals and verbs refer to events is  sufficient to show how grammaticalization of events differs in these syntactic domains. The third issue is about the nature of the dot objects and how argument and event parameters interact in the qualia expressions. The fourth issue is realted to the coercion and co-composition rules  and the variation of the expression of the complements between nouns and verbs.

With regards to nominals, the interpretation of the semantic type system can vary according to three dimensions:

1.Argument Structure: This concerns about how many argument nominals take and what is their type?(simple, unified or complex)

2. Event Structure: Events that the nominals refer to both explicitlly and implicitly.

3.Qualia Structure: What is the predicative power of the nominal is and what relational information is associated with the nominal.


Nominals with Unified Types

In order to create a lexical representation, lexical items are often placed in a lattice and they make a reference to a type in the lattice like semantic networks. This representation allows a lexical item to inherit from more than one parent.  Most of the old fashioned ways of conceptual hiearchies make heavy use of conceptual hiearchies. The conventional view on inheritance relations is shown as below:

Models like these suffers from the lexical ambiguity. There are 2 major problem with this approach:

1. They can not explain how to assign structure to lexical items.

2. They can not explain lexical relations between items in terms of links between respective lexical structures with only certain aspects .

On this view , a lexical item inherits information according to the qualia structure that it carries.

There is a need for typed inheritance for lexical inheritance theory. Pustejovsky et al (1993) proposed a way for representing relations of lexical items with qualia roles.

The type hierarchy should be constrained that only distinct qualia types which are also referred as orthogonal types unifiable. So that different qualia  may unify to form a unified-type.

Nominals with Complex Types

There is another important aspect in Pustejovsky’s approach which is at the heart of his work with nominals are refferred as dotted types or complex types. With the help of those dotted types and qualia structures we can analyze the inherently relational aspect of certain nominals without having to treat them as overtly relational.

One of the important aspect of dot objects that they incorporate the meaning of its simple types into complex  object. For example:

Book, door and novel.

Predicative property of a nominal like book(or kitap) is given by its type as a dotted object.