The TELA structure, a set of layered and linked lattices, and the notion of Similarity between TELA structures, based on the Edit Distance, are introduced in order to formalize Translation Memories (TM). We show how this approach leads to a real gain in recall and precision, and allows extending TM towards rudimentary, yet useful Example-Based Machine Translation that we call Shallow Translation.

We have developed a Japanese-to-English machine translation system for market flash reports called ALTFLASH. ALTFLASH is a hybrid translation system based on a combination of rule-based translation and template-based translation systems. The experimental results were that the system could achieve good translation for 90% of source sentences (70% of articles) in reports on the foreign section of the Tokyo Stock Exchange. In addition, we focused on account settlement flashes, which formed fixed patterns, and developed a new system to translate them. This system has been installed, by Nihon Keizai Shimbun (Nikkei) in March 1998 in their English translation service for news flashes on settlements of accounts. It is a fully automatic translation system that enables news flashes to be broadcast to the world without requiring human intervention.

This paper describes a Transfer-Driven Machine Translation (TDMT) system as a prototype for efficient multi-lingual spoken-dialog translation. Currently, the TDMT system deals with dialogues in the travel domain, such as travel scheduling, hotel reservation, and trouble-shooting, and covers almost all expressions presented in commercially-available travel conversation guides. In addition, to put a speech dialog translation system into practical use, it is necessary to develop a mechanism that can handle the speech recognition errors. In TDMT, robust translation can be achieved by using an example-based correct parts extraction (CPE) technique to translate the plausible parts from speech recognition results even if the results have several recognition errors. We have applied TDMT to three language pairs, i.e., Japanese-English, Japanese-Korean, Japanese-German. Simulations of dialog communication between different language speakers can be provided via a TCP/IP network. In our performance evaluation for the translation of TDMT utilizing 69-87 unseen dialogs, we achieved about 70% acceptability in the JE, KJ translations, almost 60% acceptability in the EJ and JG translations, and about 90% acceptability in the JK translations. In the case of handling erroneous sentences caused by speech recognition errors, although almost all translation results end up as unacceptable translation in conventional methods, 69% of the speech translation results are improved by the CPE technique.