Software Tools for the Environment of a Computer Aided Translation System

Daniel Bachut, Nelson VerasteguiCOLING 1984

TL;DR

This paper introduces three software tools – ATLAS, THAM, and VISULEX – designed to enhance the ARIANE-78 Computer Aided Translation (CAT) system. These tools aim to streamline the process of building and managing linguistic dictionaries, improve translator efficiency, and facilitate the comprehension of complex translation data.

What Problem Does This Paper Address?

Machine translation (MT) systems, like ARIANE-78, require extensive linguistic knowledge to function effectively. This knowledge is typically stored in coded dictionaries, which can be complex and difficult for linguists to manage. The process of creating and updating these dictionaries, as well as the overall translation workflow, can be time-consuming and error-prone. Users of the ARIANE-78 system needed tools to simplify dictionary creation, improve translator productivity, and make the underlying linguistic data more accessible and understandable.

Methods

The paper presents three distinct software tools, each addressing a specific aspect of the CAT system environment:

  • ATLAS: This tool acts as an aid for linguists to introduce new words and their associated codes into a CAT application's dictionary. It provides a structured way to build and interpret "indexing charts," which are essentially decision trees guiding the selection of linguistic codes. ATLAS allows for interactive editing of these charts, visualization in a tree-like format for documentation, and interpretation to derive linguistic codes for dictionary entries. It's written in PASCAL with assembler for screen handling.
  • THAM: Designed to assist human translators in Machine Aided Human Translation (MAHT), THAM integrates with a text editor and an uncoded dictionary. It offers a set of functions to improve translator efficiency by allowing simultaneous display and manipulation of source texts, translated texts, revised texts, and dictionaries. THAM facilitates tasks like updating dictionaries, inserting text segments, and reordering content, aiming to support a multi-stage translation and revision process.
  • VISULEX: This is a visualization tool for reassembling and clarifying information within a linguistic application's data base, particularly coded dictionaries. It addresses the challenges of dispersed information and obscure coding in large lexical data bases. VISULEX offers two levels of detail: one showing only comments associated with codes, and another providing a parallel listing of codes and their symbolic definitions. It can be used for documentation, debugging, and converting lexical bases.

Datasets

Not discussed in the paper. The paper focuses on the software tools themselves and their integration with the ARIANE-78 system, rather than specific datasets used for training or evaluation.

Key Insights and Ideas

  • Structured Dictionary Management: ATLAS demonstrates that complex linguistic coding can be managed through intuitive, tree-based indexing charts, making dictionary creation more systematic.
  • Translator Augmentation: THAM highlights the potential of integrating text editing and dictionary access to significantly boost human translator productivity by streamlining common tasks.
  • Data Comprehensibility: VISULEX addresses a critical need in MT research: making the internal linguistic data understandable. By offering different levels of detail, it bridges the gap between raw code and human comprehension, aiding both development and learning.
  • Modular System Design: The development of these three tools showcases a modular approach to enhancing a larger system (ARIANE-78), allowing for targeted improvements without overhauling the entire platform.

Key Results

The paper describes the functionality and intended use of each tool rather than presenting quantitative performance metrics. However, it notes that VISULEX output for a Russian-French application generated listings of around 150,000 lines each, which greatly facilitated the detection of indexing errors. The overall development effort for these tools is estimated at around two man-years, with a total source code size of approximately 15,000 PASCAL lines and 4,500 EXEC2/XEDIT lines. The ARIANE-78 system, extended by these tools, is described as more comfortable and homogeneous for users.

Limitations and Future Work

The authors acknowledge that these are the first versions of the tools and that they have already gathered many ideas for improvement from users and their own experience. Future work would likely involve further refinement of the user interfaces, expansion of functionalities, and potentially more extensive testing and validation across different language pairs and applications. The paper implies a desire for these tools to be easily adaptable to different dialogue languages, suggesting a direction for future multilingual support.

Summary updated

BibTeX
@inproceedings{bachut1984software,
  author = {Daniel Bachut and Nelson Verastegui},
  title = {{Software Tools for the Environment of a Computer Aided Translation System}},
  booktitle = {International Conference on Computational Linguistics},
  year = {1984},
}
Figure