Indexing Of Bilingual Knowledge Bank Based On The Synchronous Sstc Structure
Loading...
Date
2006-07
Authors
Ye, Hong Hoe
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The basic idea of Example-Based Machine Translation (EBMT) is to translate a
sentence by using similar translation examples. To increase memory of EBMT, one
simply needs to add new translation examples into a database. However, when the
example database becomes larger, it becomes more difficult to retrieve proper
examples as references for a translation.
In a previous work, translation examples were annotated with a flexible
structure called synchronous Structured String-Tree Correspondence (SSTC), and
were stored in a database called Bilingual Knowledge Bank. However, its word-based
indexing was not an efficient way of retrieving translation examples.
Extending on the previous work, we exploited correspondences (that include
mappings between source and target parts of translation examples) in the synchronous
SSTC structure to improve retrieval of translation examples. In addition, we
generalized translation examples to increase coverage of input text without increasing
the example database.
Based on the correspondences and generalization, two criteria, viz. word and
structure, were used to index the translation examples. Indexing using words gave us
a good coverage of input text while indexing using structures may be used to produce
well-formed translations. We classified structural indexes according to different types
and structures of examples. Structural indexes include a phrasal index and
generalized indexes (which in turn include template indexes and a rule index).
Besides indexing, we added some linguistic information (i.e. bilingual lexicon
and root forms) into our English-Malay EBMT system in order to further increase
coverage of input text.
In our translation process, given an input sentence, we first carried out lexical
matching using a word index and a phrasal index. Then, we performed structural
matching using generalized indexes to find translation examples that are structurally
close to the input sentence.
The effectiveness of our approach was evaluated against the previous work.
Our system outperforms the previous work in terms of well-formed ness and accuracy
of translation outputs.
Description
Keywords
Computer Sciences , The Synchronous Sstc Structure