Transforming English Interfaces to Other Natural Languages: An Experiment With Portuguese

Gabriel Pereira LopesACL 1984

TL;DR

This paper details the adaptation of an English natural language interface (CHAT-80) to Portuguese, creating a system called LUSO. The core contribution lies in demonstrating how logic programming and extraposition grammars can effectively handle the complexities of Portuguese, enabling a knowledge-based system (SSIPA) to interact intelligently in Portuguese.

What Problem Does This Paper Address?

Building natural language understanding systems is a significant undertaking. When such systems are developed for one language, like English, adapting them to other languages presents a substantial challenge. This isn't simply a matter of translation; it involves grappling with the unique grammatical structures, semantic nuances, and morphological complexities of each target language. The authors aim to address this by showing a practical methodology for transforming an existing English-based system into a functional Portuguese interface, thereby making knowledge-based systems more accessible to a wider range of users.

Methods

The paper describes the transformation of the CHAT-80 system into a Portuguese interface named LUSO. The key methodological choices include:

  • Logic Programming (Prolog): Used as both a conceptual tool for understanding and a practical implementation language.
  • Extraposition Grammars: Employed as the formalism for implementing both the Portuguese language analyzer and synthesizer. This formalism is highlighted as being particularly well-suited for handling complex syntactical environments and "move-alpha" rules common in Portuguese.
  • Syntactically Driven Morphological Analysis: A core hypothesis is that Portuguese morphological analysis is driven by syntax. The system attempts to match input words to grammatical categories, and if a direct match fails, it systematically tries word endings.
  • Lexically Driven Semantic Analysis: Linguistic semantic analysis is based on lexical (functional) information, drawing inspiration from Bresnam's work.
  • Cognitive Semantic Analysis: This level of analysis depends on the previously achieved syntactic and linguistic semantic analyses.
  • Knowledge Base Extension: The interface was coupled to a geography knowledge base, an extension of the original CHAT-80's.

Datasets

Not discussed in the paper. The paper focuses on the methodology and system development rather than specific datasets for training or evaluation.

Key Insights and Ideas

  • Logic as a Unifying Tool: Logic programming serves as a powerful mental aid and practical tool, enabling the unification of seemingly contradictory aspects of natural language processing: the descriptive theory of language and the active use of that language.
  • Adaptability of CHAT-80: The experiment demonstrates that existing English natural language interfaces can be successfully adapted to other languages, provided the right formalisms and methodologies are employed.
  • Portuguese Complexity: Portuguese presents significant syntactic, semantic, and pragmatic challenges comparable to other complex languages like Latin, Slavic, and Germanic languages, making it a valuable testbed for natural language processing research.
  • Incremental Development: The authors advocate for building complex models by starting with simpler versions and incrementally augmenting knowledge and capabilities, a principle demonstrated by the evolution of SSIPA.
  • Application-Independent Interfaces: The goal is to develop natural language analyzers that are application-independent, allowing them to be improved over time as more linguistic knowledge is incorporated, freeing up developers to focus on more subtle aspects of language and concept definition.

Key Results

The paper reports on the successful development of LUSO, a Portuguese natural language interface that allows the SSIPA system to:

  • Understand and respond to questions about geography in Portuguese.
  • Agree or disagree with user opinions on geographical knowledge.
  • Perform actions traditionally associated with computer monitors, such as creating, deleting, and manipulating files.
  • Engage in intelligent clarification dialogues.

While specific quantitative metrics are not provided, the qualitative results indicate a functional and intelligent Portuguese interface capable of complex interactions.

Limitations and Future Work

  • Complexity of Formalization: The authors acknowledge the inherent difficulty in formalizing the complex, structured nature of natural languages.
  • Incremental Improvement: The paper suggests that the natural language analyzer, planned to be application-independent, is continuously under improvement as more linguistic knowledge is incorporated.
  • Focus on Core Functionality: The experiment primarily focuses on the morpho-syntactical and basic semantic aspects of Portuguese. Future work would likely involve deeper cognitive semantic analysis and the integration of more sophisticated linguistic knowledge.
  • Practical Conclusion: The authors emphasize the practical conclusion that it is more efficient to adapt existing systems and build upon them rather than constructing new application-directed interfaces from scratch.

Summary updated

BibTeX
@inproceedings{lopes1984transforming,
  author = {Gabriel Pereira Lopes},
  title = {{Transforming English Interfaces to Other Natural Languages: An Experiment With Portuguese}},
  booktitle = {Annual Meeting of the Association for Computational Linguistics},
  year = {1984},
}
Figure