Upload
todd-bruce
View
218
Download
1
Tags:
Embed Size (px)
Citation preview
Research on Semantic-based Passive Transformation in Chinese-English
Machine Translation
Wenfei Chang, Zhiying Liu, Yaohong Jin Institute of Chinese Information Processing
Beijing Normal University
Introduction1
Semantic analysis of passive voice2
Transformation rules and algorithm3
Experiments and Result Analysis4
Conclusions5
Outline
1. Introduction
Type Sentence number Proportion
Sentences with passive mark 390 39%
Sentences without passive mark
610 61%
2. Semantic analysis of passive voice
We have investigated 1000 sentences which should be transformed into English when translating.
Table 1. Classification of Passive Sentence
Sentences with passive mark in Chinese • 因此提交订单的交易者将被通知成交。 (Thereby the trader
that sent in the order will be informed about the deal.)
• 它不需要处理在第一排列单元所接收的订单。 (It does not need to handle the order that was received at the first ranking unit.)
2. Semantic analysis of passive voice
Passive mark BEI
Passive mark SUO
ALL_PASS
passive voice will be used in English
Verb+ Prep
2. Semantic analysis of passive voice Sentences without passive mark in Chinese
For example,
1、 “ V+NP” • 经由所述双向隧道转发分组。 (Packets are forwarded via the bi-
directional tunnel.) 2、 “ NP+V”• 固定的和旋转的磁鼓面对面地安装。 (The fixed and rotatable
drums are installed face to face.)
• 包套可滑动地安装在可弯曲管内。 (A sheath is slideably mounted inside the flexible pipe .)
• 这种组合物可以做成很薄、很小的产品。( The composition can be made into a very thin and small product.)
Component ellipsis in sentence.
“Verb+Prep” structure in sentence.
Effect Sentence
2. Semantic analysis of passive voice
A series of rules are drawn up according to several situations . The specific steps are as fellows:
3. Transformation rules and algorithm
Type Total number
Should be transformed
Transformed
Right transformed
RB 1000 632 540 481
Google 1000 632 515 430
System Precision Recall
RB 89.1% 76.1%
Google 83.4% 68.1%
Table 2. Types of data
Table 3. Result of transformation
4. Experiments and Result Analysis
• Rules have not covered all kinds of linguistic phenomenon.
• Knowledge base gives wrong property (“ALL_PASS[Y]”) to the verb.
• The verb is wrongly recognized, thus leading to wrongly match the transformation rules.
4. Experiments and Result Analysis
By analyzing errors in the result, we find there are mainly have three reasons:
Results show that our system has achieved a good effect.
In the future, we will make further improvements based on the errors.
5. Conclusions
23/4/18
Thank you !Thank you !