Modern Machine Translation Systems do not have a mechanism for interpretation of the input messages on models of concepts systems.
Data Машинный перевод: причина неудач (Machine translation: reaso...
Probably is the lack of correct classification: the same word means different things for different fields, so machines fail. Mayby they should offer an option for declaring the field from the user.
Interpretation of the input message should begin with extraction of a meaning of the message. This can be done only when this message contains link to the model of the concepts system, on which the message is to be interpreted.
I believe that we must distinguish between the tasks of creating machine translation systems for well-defined subject areas (science, technology, etc.) and undefined areas.
Models and tools of message interpretation can be created for well-defined subject areas in a relatively short period of time.
Dear Vladimir
Current approaches to statistical machine translation assume that sentences in a text are independent, ignoring the property of connectedness present in virtually all discourse
Machine translation researchers widely agree that translation is a complex task that cannot be solved by looking at words and their immediate local neighborhood only; however, much of the existing work on machine translation depends on strong assumptions of locality for practical reasons
I suggest reading the article:
Callison-Burch, C. et al. 2012. Findings of the 2012 Workshop on Statistical Machine Translation. In Proceedings of the 7th Workshop on Statistical Machine Translation. Stroudsburg: Association for Computational Linguistics: 10-51. Available online: http://www.aclweb.org/anthology/W/W12/W12-3102.pdf
Dear Nelson,
Statistical machine translation is not designed for extraction of a meaning of the input message and construction of the output message, which has the same meaning.
Statistical machine translation is suitable for translation of messages for which subject areas are not defined.
Dear Lionel,
My question is about whether we can satisfactorily solve the problem of machine translation, without having models of the concepts systems and mechanism of message interpretation on that models.
Cheers!
Dear @Vladimir, interesting thread, thanks for sharing. In this book, as I could not read your paper (sorry), it is mentioned that three factors play a major role in machine Translation. They are: the characteristics of the text to be translated, the desired level of automation of translation process and the quality of translation acceptable to the user.
I accept that "we must distinguish between the tasks of creating machine translation systems for well-defined subject areas (science, technology, etc.) and undefined areas. " But I do not think that short period of time is needed for "well-defined subject areas" (relatively).
The following project might be interesting for this issue: Modelling Discourse Entities and Relations for Coherent Machine Translation!
http://books.google.rs/books?id=YUNLlurHNAEC&printsec=frontcover#v=onepage&q&f=false
http://www.idiap.ch/project/modern/
http://p3.snf.ch/project-147653
Dear @Ljubomir,
Models and tools of message interpretation can be created for well-defined subject areas in a relatively short period of time.
Dear @Ljubomir,
Well-defined subject area: a certain set of concepts and family of relations defined on this set.
I think that formalized models of well-defined subject areas can be constructed in a relatively short period of time.
Probably is the lack of correct classification: the same word means different things for different fields, so machines fail. Mayby they should offer an option for declaring the field from the user.
Dear @Vladimir, Machine Translation Systems are machines and are simulating a human work, so, they are improved but still there is always a space for better work.
Dear Colleagues,
Scheme of meaningful translation from language A to language B:
1. receiving and recognition of message in A;
2. extraction of meaning, using the systems of concepts in A;
3. search for suitable systems of concepts in B;
4. creating and sending message in B, which has the same meaning.
The steps 2. and 3. are absent in current technologies of machine translation.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Artificial Intelligence