Constructing a Domain Model Using Statistical Analyses for Forming a Partial Formal Grammar

When faced with a corpus describing an unfamiliar domain, it may be difficult to structure the domain [1]. Statistical analyses, that have originally been developed to be used in deciphering ancient scripts [8, 9], are applied to analyse the structure of the corpus by starting to form a formal grammar. Context-free grammars used out of different types of formal grammars are

• the simplest grammars with sufficient power for some kind of description of natural language, and
• the most complex grammars with a relatively easily manageable syntax.

Correct inference of a complete and unique context-free grammar from a text with no additional information is impossible [3]. However, additional information is expected to be collected by analyzing combinatorial properties of the corpus. In addition, the grammar need not be complete to reveal structural features of the corpus.

In the following sections, describing the analyses, a term may mean either a single word or a nonterminal symbol presenting a sequence of terms or a group of alternative terms.
 

• The Syntagmatic Approach
• The Paradigmatic Approach
• Harris's Tree Structure Analysis
• Application of the analyses
• Using Interview Transcripts as a Corpus

Sat Oct 14 22:52:14 EEST 2000