Structural Semantic Correspondence For Example-Based Machine Translation
The main key challenge in Machine Translation (MT) is to preserve the meaning from the original source sentence to the target translation. This remains the core problem in Machine Translation, which leads to a number of issues, such as: how to analyze the meaning of text, what information should be...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2016
|
| Subjects: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The main key challenge in Machine Translation (MT) is to preserve the meaning from the original source sentence to the target translation. This remains the core problem in Machine Translation, which leads to a number of issues, such as: how to analyze the meaning of text, what information should be captured as the representation of meaning, how the meaning of a sentence should be represented (into what form and structure), how do we derive and generate target sentence based on this meaning information, etc. As the natural language is ambiguous in nature, the task of meaning treatment in MT systems is hence difficult. The idea of this study is to introduce semantic knowledge to improve the selection and matching of best translation examples in the Example-based Machine Translation (EBMT) system, also to facilitate a deeper semantic similarity measurement and evaluation of the matching examples. The approach is by specifying a structural semantics or “meaning” explicitly to the translation examples representation structures. The creation of the structural semantics begins with the English examples in the existing Bilingual Knowledge Bank (BKB). The structural semantics of the English examples is transferred to other languages based on parallel alignment (aligned words and tree structures) in this BKB. With structural semantic annotation of both the source language (SL) and target language (TL), a Structural Semantic Correspondence between these aligned translation examples is created. This structurally synchronized “meaning” or semantics of the SL and TL allowed the EBMT system to perform semantic-based analysis, selection of translation examples based on semantics similarity, and finally target translation hypothesis derivation. |
|---|