Browse Prior Art Database

A method to optimize the expression of machine translation

IP.com Disclosure Number: IPCOM000247574D
Publication Date: 2016-Sep-18
Document File: 6 page(s) / 141K

Publishing Venue

The IP.com Prior Art Database

Abstract

This disclosure introduces a method to optimize the emotion expression of machine translation, make it closer to the original language. It extracts sentence emotional characteristics from sentence to translate, through sentiment analysis, constructing emotion vector and comparing emotion vectors of original sentence with ones of machine translation result, to optimize the results of machine translation.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 33% of the total text.

Page 01 of 6

A method to optimize the expression of machine translation

Recently, machine translation is used in service platforms withcapability of language translation , such as instant messaging,social and other overseas travel APPs.

There are three phases in machine translation process :
- Firstly, analysis source sentence, identify phrases and syntax.

- Secondly, search for the maximum most probability phrase as aresult of translation phrase from Corpus .

- Finally, generate the target translation sentence according to the language model . Generally, machine translation's missionis to find the maximum probability of the translation sentence , among all possible translation in Corpus.

This technology does solve some basic communication problems . However, it does not take emotions and attitude of the original sentence into consideration. in most of time, it's difficult for machinetranslation to understand human emotions . When two persons speaking different languages are chatting in real-time machine translation messaging system, especially cross computer/mobile system without face to face clarification. The inappropriate emotional expression might cause misunderstanding or unexpected result .

This invention introduces a method to optimize the emotion expression of machine translation , make it closer to the original language.

Example:
Source sentence:我我我我我我我我,价价价价价价。只只只只只只只只只只,只太太只!

Machine Translation: I like the room very much. The price is very reasonable. It's just too far from the city center. It's a pity! Optimized Translation: I really like the room. The price is so reasonable. It's just too far from the city center. what a pity!

This invention won't change machine translation process, it extracts sentence emotional characteristics from sentence to translate , through sentiment analysis, to optimize the results of machine translation .

   Sentiment analysis can be considered as a classification process. - Firstly, identify whether one phrase has emotions.
- Secondly, identify emotion phrase is positive or negative .
- Thirdly, classify the emotional level of phrase.

When sentence sentiment analysis is done, emotion phrases are marked up. Search in emotional dictionary, get measurable emotional score for all marked emotional phrase. Emotion dictionary is a collection of all kinds of emotion words , emotional preference and emotional level. Each word has a score to show its emotional level in dictionary .

1



Page 02 of 6

According to the logical structure of the sentence , construct emotion vector with these scores.

Compare emotion vectors of original sentence with ones of machine translation result , determine if machine translation is appropriate to express the emotion of original sentence. Each emotion word maymap to many possible translations , all of them are translationcandidates. we can use optimization algorithm to run iterative matching calculation , select...