Context-Sensitive Electronic Dictionaries

Gábor Prószéky, Balázs KisCOLING 2002

TL;DR

This paper introduces MoBiMouse, a context-sensitive electronic dictionary that provides instant translations for any text on a computer screen without user interaction. It achieves this through a novel three-phase process of text acquisition, morpho-syntactic analysis, and a restructured dictionary lookup, enabling it to understand and display only relevant translation information based on context.

What Problem Does This Paper Address?

In today's digital world, users frequently encounter foreign language text on their computers. While many tools exist to help with translation, they often require users to actively switch applications or copy-paste text, disrupting their workflow. The authors address the need for a seamless, background process that provides instant, contextually relevant translations without interrupting the user's current task. This involves overcoming challenges in acquiring text from any application, understanding the linguistic context of a word, and presenting only the most pertinent translation information from potentially large dictionaries.

Methods

MoBiMouse employs a three-phase approach:

  1. Text Acquisition: This phase captures text directly from the computer screen. It uses a combination of an OCR-like procedure for general compatibility and application-specific protocols for higher accuracy with known programs. Crucially, it operates without requiring explicit user input like mouse clicks or clipboard access.
  2. Linguistic Analysis and Dictionary Lookup: Once text is acquired, the system performs morpho-syntactic analysis on the context of the selected word. This involves stemming, spelling correction, and shallow parsing to identify potential multi-word expressions. The core innovation here is how dictionaries are structured: entries are split into smaller, indexed pieces, allowing the system to filter and retrieve only information relevant to the specific context. This is a departure from traditional dictionary structures.
  3. Rendering and Graphic Output: The relevant translation is displayed in a non-intrusive, bubble-shaped pop-up window that disappears when the mouse moves away.

Datasets

Not discussed in the paper. The paper focuses on the system's architecture and functionality, mentioning that dictionaries are adapted from printed versions or created using automatic/semi-automatic tools.

Key Insights and Ideas

  • Context is King for Dictionaries: The paper strongly advocates for context-sensitive translation, arguing that displaying entire dictionary entries is often overwhelming and irrelevant. By restructuring dictionaries and analyzing context, MoBiMouse delivers more targeted information.
  • Seamless Integration is Crucial: The "no mouse click" activation and pop-up display are key to minimizing user disruption, making the tool a background assistant rather than another application to manage.
  • Open Architecture for Extensibility: The modular design with documented APIs allows for easy replacement of core components, facilitating future development and integration of new linguistic technologies.
  • User Feedback as a Development Driver: The system's ability to log and report words it cannot translate provides a valuable mechanism for identifying dictionary gaps and improving linguistic resources.

Key Results

The paper highlights the system's unique features:

  • Linguistic Flexibility: Stemming, morphological analysis, and shallow parsing contribute to accurate context understanding.
  • Open Architecture: The modular design allows for easy replacement of architectural blocks.
  • Flexible User Interface: Replaceable dictionaries and direct user feedback mechanisms are supported.
  • Discontinuous Multi-word Expression Recognition: The system can identify and translate multi-word expressions even if they are not contiguous on the screen.
  • Performance Metrics (Examples): Text acquisition speed is cited at 1000 characters/s, stemming at 0.002 s/word-form, and average dictionary lookup at 0.02 s.

Limitations and Future Work

Limitations:

  • The text acquisition accuracy can decline with non-standard text sizes or spacing, though alternative methods are used for some applications.
  • While the system filters output for relevance, it doesn't currently filter different meanings of a single word or multiword lexeme based on context.

Future Work:

  • Improving the user interface is a primary focus.
  • Integrating a parser/translator engine to analyze and translate entire sentences or larger passages, moving beyond word-level translation.
  • Developing a more sophisticated language identification module (though they currently use a simpler approach of checking all installed dictionaries).
  • Expanding the range of supported languages and dictionary pairs.

Summary updated

BibTeX
@inproceedings{proszeky2002context,
  author = {Gábor Prószéky and Balázs Kis},
  title = {{Context-Sensitive Electronic Dictionaries}},
  booktitle = {International Conference on Computational Linguistics},
  year = {2002},
}
Figure