(42) Automatic Acquisition of Bilingual Translations Using
Recursive Chain-link-type Learning from a Parallel Corpus
¡¡¡¡¡¡Ë̳¤³Ø±àÂç³Ø¹©³ØÉô¸¦µæÊó¹ð, Âè30¹æ,pp.165-190, 2004-2
This paper describes an automatic acquisition method of bilingual
translations (bilingual pairs of words or phrases) based on the learning
algorithm
that we call Recursive Chain-link-type Learning. The system with Recursive
Chain-link-type Learning automatically acquires many bilingual translations
from
various bilingual sentences without requiring any analytical knowledge or
a
large number of similar bilingual sentences. In this learning method,
bilingual translations and bilingual templates are reciprocally acquired
like a
linked chain. A bilingual template is a bilingual pair of generalized sentences
that the system acquires by replacing bilingual translations with variables.
By
using collocations based on variables and their adjoining words in
acquired bilingual templates, the system automatically acquires new bilingual
translations from various bilingual sentences. In the experimental results,
the
correct acquisition rate for all nouns and noun phrases in a parallel corpus
was 74.9%. Moreover, the acquisition rate of the correct equivalent for
unknown words in the commercial English-to-Japanese machine translation system
was 78.4%. These results show that the learning ability of the system with
Recursive Chain-link-type Learning is high and the automatic acquisition
method for bilingual translations using Recursive Chain-link-type Learning
is
effective.