Machine Translation презентация

Introduction
sub-field of computational linguistics that investigates the use of software to translate text or speech from one natural language to another (http://en.wikipedia.org/)
Use: translation of large amount of date in the shortest possible time
Standard documents
Instructions and manuals
Web sites, multilingual search
Reference information(addresses, recipes, etc.)
Aim: to understand the main contents of the document in a foreign language unknown to the user
NOT to be used instead of human translation !!!

Approaches to machine translation
Rule-based approach
Statistical
Example-based approach
Hybrid machine translation

Rule-based translation
Stages
Morphological analyses of source language
Parsing source language (syntactic groups)
Getting syntactic information about each word
Dictionary based translation
example:
A girl eats an apple. (Eng.-Ger.)
stages of translation:
1st: getting basic part-of-speech information of each source word: a = ind.art.; girl = n.; eats = v.; an = ind.art.; apple = n.
2nd: getting syntactic information about the verb “to eat”: here: eat – Pr. Simple, 3rd Pers. Sing., Act. V.
3rd: parsing the source sentence:(an apple) = the object of eat
4th: translate English words into Germana (category = indef.article) => ein (category = indef.article)girl (category = noun) => Mädchen…
5th: finding appropriate inflected forms: A girl eats an apple. => Ein Mädchen isst einen Apfel.

Statistical translation
Translations are generated according to probability distribution on the basis of statistical models whose parameters are derived from the analysis of bilingual text corpora
Benefits
Better use of resources
More natural translations
No programmers or linguists* involved
Shortcomings
Corpus creation can be costly for users with limited resources.
The results are unexpected. Superficial fluency can be deceiving.
Statistical machine translation does not work well between languages that have significantly different word orders

Статистический перевод
Основа - параллельный корпус
Вероятности назначаются подсчетом наиболее вероятного варианта перевода
Оценки вероятности зависят от объема и качества обучающего корпуса
Лингвистическая информация: разбиение на предложения, графематический анализ, морфология
При наличии корпуса простейшая система перевода может быть сделана на 2 недели

Rule-based vs. statistical
news:
document:

Rule-based translation
Types
Dictionary-based (direct)
Transfer-based
Interlingual

Dictionary-based (direct)
word by word translation
with or without morphological analysis or lemmatisation
Application
translation of long lists of phrases on the subsentential (i.e., not a full sentence) level, e.g. lists, inventories or simple catalogs of products and services.

Direct translation
example

Transfer-based machine translation
1. Analyzing the input text for morphology and syntax (and sometimes semantics)
2. Creating an internal representation
3. Generating translation using both bilingual dictionaries and grammatical rules
Sentence in a source language
Source language structure
Sentence in a target language
Target language structure
analysis
transfer
synthesis

Interlingua machine translation
the source language is transformed into an interlingua, i.e., an abstract language-independent representation
the target language is generated from the interlingua.

Transfer vs. interlingua

Hybrid machine translation
method of machine translation characterized by the use of multiple approaches within a single machine translation system.
Types:
RBMT guided by statistics
Statistical method guided by RBMT