Unit completion for a computer-aided translation typing system

Abstract

This work is in the context of TRANSTYPE, a system that observes its user as he or she types a translation and repeatedly suggests completions for the text already entered. The user may either accept, modify, or ignore these suggestions. We describe the design, implementation, and performance of a prototype which suggests completions of units of texts that are longer than one word.

User Interface Aspects of a Translation Typing System

“…In this project we had to address the following problems: how to interact with the user and how to find appropriate multi-word units for suggestions that can be computed in real time. The former has been described by Langlais [6] but this article focuses on the latter. Fig.…”

“…Briskels were presented to the user as soon as a user selected a sentence. The briskels were determined by hand for the sentences of our experiment but Langlais [7] has shown that is possible to automatically compute longer units than one word. …”

Using idiolects and sociolects to improve word prediction

“…After this first data-driven improvement, numerous authors have shown that taking the contextof previously entered words into account improves prediction accuracy further. A simple approach to implementing context-sensitivity is applying the frequency list technique to word n-grams (Hunnicutt, 1987); in a string of work other statistical language modeling approaches have been proposed (Lesher et al, 1999;Langlais et al, 2000;Garay-Vitoria and Abascal, 2006;Tanaka-Ishii, 2007; Van den Bosch and Bogers, 2008).…”

Response-based Learning for Grounded Machine Translation

“…Computer assisted translation (CAT) subsumes several modes of interaction, ranging from binary feedback on the quality of the system prediction (Saluja et al, 2012), to human post-editing operations on a system prediction resulting in a reference translation (Cesa-Bianchi et al, 2008), to human acceptance or overriding of sentence completion predictions (Langlais et al, 2000;Barrachina et al, 2008;Koehn and Haddow, 2009). In all interaction scenarios, it is important that the system learns dynamically from its errors in order to offer the user the experience of a system that adapts to the provided feedback.…”