Лингвистические автоматы в современных гуманитарных технологиях. Беляева Л.Н. - 101 стр.

UptoLike

Составители:

Рубрика:

ФУНКЦИЯ АННОТИРОВАНИЯ (РЕФЕРИРОВАНИЯ) ТЕКСТА

101

America) and the speakers of majority languages, and the

potential preservation of endangered languages – over half

of the 6,000 presently existing languages worldwide. The

primary scientific challenge is the creation of MT systems

for languages of little economic importance at very low cost

per language, including the acquisition of linguistic

information with minimal preexisting bilingual corpora and

little or no previous linguistic analysis of the minority

language.

In order to address these needs, we are investigating

omnivorous MT systems, including statistical and example

based MT when some parallel training corpora can be

acquired, and machine learning of transferbased MT rules

when access to a native nonlinguist informant permits

partial elicitation of linguistic information, such as

translations of model sentences and lexicallevel bilingual

alignments. This paper focuses on this last objective of our

project:supervised learning of transfer rules with the aid of

an elicitation interface to a bilingual native speaker without

any assumptions regarding his or her linguistic

sophistication. While our technology is eventually aimed at

lowdensity languages, it is intended to be target language

independent. Hence, we are developing the system using

examples from various languages, such as Chinese, German,

Mapudungun

, and Swahili. For illustration purposes, we

present examples in these languages throughout the paper.

Mapudungun, spoken in Chile, is one of the minority languages we focus on.