Machine Translation (The MIT Press Essential Knowledge series) by Thierry Poibeau

Author:Thierry Poibeau [Poibeau, Thierry]
Language: eng
Format: azw3
Tags: ALPAC report, robots and artificial intelligence, segment-based approach, cognitive science, comprehension, statistic, technology, reasoning, linguistics, deep learning artificial intelligence, introduction artificial intelligence, Weizenbaum, computer science, ambiguity, artificial intelligence, language diversity, rule-based approach, turing test, Eliza system, applied artificial intelligence, Babelfish, semantics, computers, manual, expert systems and artificial intelligence, automatic dialoguing system, deep learning, computer artificial intelligence, paradigm, automatic, parallel corpora
Publisher: The MIT Press
Published: 2017-09-22T04:00:00+00:00

Back to the Roots of the Domain?

The IBM models are in some ways a return to the roots of the domain, since the techniques proposed very directly echo several of Weaver’s 1949 propositions. The fact that the module aiming at producing the translation is sometimes called a “decoder” is no coincidence. This is a reminder of the general model of communication, and the goal is to “decode” the source text by translating it into the target language (see chapter 5).

The models designed by IBM achieved phenomenal success. They were revised, modified, and improved. They remain the basis of most machine translation systems used today, although all the major players in the field are now moving towards a new kind of techniques called deep learning (see chapter 12).

However, these models also have their own limitations, the main one being that they require huge quantities of data to achieve reasonable performance. In the following chapter, we will take a look at recent developments around these models, but also at the situation of rare languages for which not enough data are available to achieve accurate statistical translation systems.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.