COMPUTATIONAL COMPARATIVE STUDIES ON ROMANCE LAGUAGES: A linguistic comparison of lexicon-grammars

Annibale Elia, Yvette MathieuCOLING 1986

TL;DR

This paper introduces TRANSLEGS, a computational system for comparing lexicon-grammars of Romance languages. It addresses the challenge of formalizing linguistic comparisons by developing a method to map syntactic and lexical properties between languages, enabling deeper cross-linguistic analysis.

What Problem Does This Paper Address?

Comparing languages formally, especially at the lexicon-grammar level, is a complex task. Traditional linguistic analysis often relies on intuition, making it difficult to achieve rigorous, reproducible comparisons. The authors aim to bridge this gap by creating a computational framework that can systematically analyze and compare the syntactic and lexical structures of languages. Specifically, they focus on lexicon-grammars, which represent a comprehensive linguistic description of a language's words and their grammatical behavior. The challenge lies in developing a method to identify and represent formal correspondences between these lexicon-grammars, allowing for a more objective and detailed cross-linguistic study.

Methods

The core of the paper is the development of the TRANSLEGS system, a query-answer system designed for comparing lexicon-grammars. The system operates on a "query-answer" paradigm, allowing users to explore linguistic data banks. Key methodological aspects include:

  • Lexicon-Grammar Representation: The system utilizes a matricial form to represent lexicon-grammars, capturing nearly 300 syntactic properties for each word.
  • Formal Correspondences: TRANSLEGS identifies formal correspondences between structures and properties of different languages. This involves mapping "grammatical words" (like prepositions and determiners) and analyzing syntactic structures.
  • Distributional Syntactic Properties: The system focuses on comparing the distributional syntactic properties of words and their paraphrases across languages.
  • Proximity Degree: For cases where direct correspondences are not found, TRANSLEGS proposes "substitution" classes based on formal proximity degrees, allowing for nuanced comparisons.
  • Deduction Rules: The system employs deduction rules to handle situations where properties or classes exist in one language but not another, facilitating the creation of equivalent structures.
  • PROLOG Implementation: The system is implemented in PROLOG, a logic programming language, which is well-suited for symbolic manipulation and rule-based reasoning.

Datasets

The paper utilizes linguistic data banks for Italian and French, assembled by the Istituto di Linguistica of Salerno University and the Laboratoire Automatique Documentaire et Linguistique (C.N.R.S.-France). These data banks contain formalized grammars of the respective languages, forming the basis for the lexicon-grammars. No new benchmarks are introduced; rather, existing formalized linguistic data is leveraged.

Key Insights and Ideas

  • Formalizing Linguistic Comparison: The paper demonstrates that it's possible to move beyond intuitive linguistic comparisons to a more formal, computational approach by leveraging lexicon-grammars.
  • The Power of Lexicon-Grammars: Lexicon-grammars provide a rich source of information for detailed cross-linguistic analysis, capturing selectional restrictions, sub-categorization, and derivational morphology.
  • Handling Asymmetries: The system's ability to propose "substitution" classes based on formal proximity is crucial for dealing with linguistic asymmetries between languages, where direct one-to-one mappings don't exist.
  • User-Centric Linguistic Exploration: TRANSLEGS offers a flexible interface that allows both ordinary users to consult linguistic data and researchers to conduct in-depth comparative studies by modifying deduction rules or exploring specific linguistic phenomena.

Key Results

The paper doesn't present quantitative results in terms of accuracy metrics. Instead, it showcases the system's capabilities through examples of dialogues and comparisons.

  • Comparative Analysis of "investire/investir": The paper provides a detailed comparison of the Italian verb "investire" and its French correspondent "investir," illustrating how TRANSLEGS can map different uses, syntactic structures, and semantic nuances. For instance, it highlights how certain Italian structures have no direct French correspondent, and vice-versa, but TRANSLEGS can identify formal proximities.
  • Identification of Linguistic Differences: The system successfully identifies subtle differences in how verbs are used and structured across languages, such as the example of "douter" (French) and "dubitare" (Italian), where certain constructions are acceptable in one language but not the other.
  • Flexible Querying: The system demonstrates its flexibility by allowing users to query for specific classes of words, individual words, or even to explore the behavior of equivalent words across languages.

Limitations and Future Work

  • Limitations:
    • The paper acknowledges that the "very complex data organization is usable by specialists but not by ordinary users." This implies a need for more user-friendly interfaces and abstraction layers.
  • The formal proximity degree notion is presented as dependent on the researcher's theoretical choice, suggesting a degree of subjectivity in the comparison process.
  • Future Work:
    • The authors envision TRANSLEGS as a tool for both ordinary users to consult linguistic data banks comfortably and for researchers to conduct scientific descriptions and comparisons of languages.
  • They suggest the possibility of modifying deduction rules or delving deeper into lexical and grammatical information, implying ongoing development and expansion of the system's capabilities.
  • The system is currently focused on Romance languages, but the underlying principles could potentially be extended to other language families.

Summary updated

BibTeX
@inproceedings{elia1986computational,
  author = {Annibale Elia and Yvette Mathieu},
  title = {{COMPUTATIONAL COMPARATIVE STUDIES ON ROMANCE LAGUAGES: A linguistic comparison of lexicon-grammars}},
  booktitle = {International Conference on Computational Linguistics},
  year = {1986},
}

Loading related content...

Figure