FFF CONFERENCE CTF07

Christian Horn & Christof Rumpf - Conceptual noun types: grammar and automatic classification

Nouns can be classified into four different conceptual types with the two binary features functional and relational. Löbner differentiates conceptual types into functional nouns, relational nouns, individual nouns, and sortal nouns (cf. Löbner 1979, 1985, 1998). Functional nouns have unique reference to objects in the world, like proper names (Peter), measure expressions (size of X) or unique relative expressions (mother of X). Relational nouns have possessor arguments like the X-positions in the last two examples. Nouns that are neither functional nor relational describe sortal concepts. The following table gives an overview:


 

 

+relational     

–relational

+functional

functional concepts

individual concepts

–functional

relational concepts

sortal concepts

 

The identification of the concept type of a noun occurrence in a text can be carried out by inspection of morphosyntactic context features like determination and the presence of possessor arguments. Definite determination indicates functionality, the presence of possessor arguments relationality. We work with the hypothesis that every noun has an underlying lexicalized concept type that can be changed by an appropriate context. For example, brother of X is lexicalized as a relational concept but becomes a functional concept in the context youngest brother of X. We assume that the lexicalized concept type of a noun is the most frequently used type for each noun.   We develop a system for the automatic classification of concept types in texts based on a maximum entropy model (cf. Ratnaparkhi 1998). This is trained with a corpus in which the concept type of each noun is manually annotated and the morphosyntactic context features are automatically annotated with a parser (Connexor’s MSYN). The training results in a model where the context features are weighted by their evidence as indicators for each concept type. This model can then be used for the automatic classification of new texts that have been preprocessed with the parser.

 

References

Löbner, Sebastian (1979) Intensionale Verben und Funktionalbegriffe. Tübingen. Narr.

 

Löbner, Sebastian (1985) Definites. Journal of Semantics 4. 279-326.

 

Löbner, Sebastian (1998) DefiniteAssociative Anaphora. ms.
http://web.phil-fak.uni-duesseldorf.de/~loebner/publ/DAA-03.pdf

 

Ratnaparkhi, Adwait (1998) Maximum Entropy Models for Natural Language Ambiguity Resolution. Ph.D. thesis, University of Pennsylvania, Philadelphia, PA.