Machine Learning

In order to overcome the knowledge acquisition bottleneck, we are currently attempting to make ALT-J/E's translation rules less costly and more reliable through the use of inductive machine learning techniques. We use inductive machine learning algorithms to learn conditions of translation rules from examples of Japanese sentences and their English translations (Akiba et al.1995, Akiba et al.2000). The task is now turned into a search for some reasonable rules that explain the given training examples, where the search is handled automatically by a learning algorithm. We give an example in Figure 2.

Figure 2: An example of a learning task
An example of a learning task

Yasuhiro Akiba