Simultaneous Amalgamation - Computational Linguistics and Talking Robots - page 167

In summary, the DBS class definitions for all kinds of constructions, includ-
ing the infinitives, characterize complex contents by means of (i) schemata
consisting of pattern proplets, (ii) explicit lists specifying the concepts which
may serve as the matrix in the class schema (defined as restriction sets for the
variable ), and (iii) an explicit listing of the selectional constellations between

the concepts in a class schema, obtained from a corpus.
The DBS class schemata (e.g., 8.5.2, 8.6.4, 8.6.6, 8.6.8) constitute authentic
linguistic generalizations, constrained empirically by the restriction set on the
variable

α

and the n-tuples of selectional constellations. By storing the ele-
ments of such class definitions in an RMD Corpus Word Bank, they may be
used to support lexical selection, for example, in machine translation. They
may also be used to reduce the search space in artificial nonlanguage and lan-
guage recognition.
The described method of determining the selectional constellations in a cor-
pus depends crucially on the binary treatment of semantic relations in DBS.
The method can not be used by systems computing possible substitutions such
as Phrase Structure Grammar and Categorial Grammar. 26 This is because a
substitution-based derivation of phrases and clauses is inherently compelled
to use nonterminal nodes, resulting in non-binary (indirect) relations between
the terminal nodes.
In a big corpus, the selectional constellations represent the content word dis-
tributions of many domains and many authors all mixed up. This is instructive
for what might be regarded as the language as whole at a certain time. It is
equally instructive, however, to analyze the content word distributions in a
certain domain or in the work of a certain author.
Determining the selectional constellations for a big corpus, for a single do-
main, or for an author is based on the same routine in DBS. It consists of auto-
matic word form recognition, syntactic-semantic parsing, frequency analysis,
and storage in a Corpus Word Bank.

α

26 As proven by C. Gaifman in June 1959, bidirectional Categorial Grammars and context-free Phrase
Structure Grammars are weakly equivalent (Bar Hillel 1964, p. 103). See also Buszkowski (1988)
and FoCL'99, Sect. 9.2.